Commit | Line | Data |
---|---|---|
bcea3f96 | 1 | // SPDX-License-Identifier: GPL-2.0 |
7a8e76a3 SR |
2 | /* |
3 | * Generic ring buffer | |
4 | * | |
5 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
6 | */ | |
28575c61 | 7 | #include <linux/trace_recursion.h> |
af658dca | 8 | #include <linux/trace_events.h> |
7a8e76a3 | 9 | #include <linux/ring_buffer.h> |
14131f2f | 10 | #include <linux/trace_clock.h> |
e6017571 | 11 | #include <linux/sched/clock.h> |
0b07436d | 12 | #include <linux/trace_seq.h> |
7a8e76a3 | 13 | #include <linux/spinlock.h> |
15693458 | 14 | #include <linux/irq_work.h> |
a356646a | 15 | #include <linux/security.h> |
7a8e76a3 | 16 | #include <linux/uaccess.h> |
a81bd80a | 17 | #include <linux/hardirq.h> |
6c43e554 | 18 | #include <linux/kthread.h> /* for self test */ |
7a8e76a3 SR |
19 | #include <linux/module.h> |
20 | #include <linux/percpu.h> | |
21 | #include <linux/mutex.h> | |
6c43e554 | 22 | #include <linux/delay.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
7a8e76a3 SR |
24 | #include <linux/init.h> |
25 | #include <linux/hash.h> | |
26 | #include <linux/list.h> | |
554f786e | 27 | #include <linux/cpu.h> |
927e56db | 28 | #include <linux/oom.h> |
7a8e76a3 | 29 | |
c84897c0 | 30 | #include <asm/local64.h> |
79615760 | 31 | #include <asm/local.h> |
182e9f5f | 32 | |
6695da58 SRG |
33 | /* |
34 | * The "absolute" timestamp in the buffer is only 59 bits. | |
35 | * If a clock has the 5 MSBs set, it needs to be saved and | |
36 | * reinserted. | |
37 | */ | |
38 | #define TS_MSB (0xf8ULL << 56) | |
39 | #define ABS_TS_MASK (~TS_MSB) | |
40 | ||
83f40318 VN |
41 | static void update_pages_handler(struct work_struct *work); |
42 | ||
d1b182a8 SR |
43 | /* |
44 | * The ring buffer header is special. We must manually up keep it. | |
45 | */ | |
46 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
47 | { | |
c0cd93aa SRRH |
48 | trace_seq_puts(s, "# compressed entry header\n"); |
49 | trace_seq_puts(s, "\ttype_len : 5 bits\n"); | |
50 | trace_seq_puts(s, "\ttime_delta : 27 bits\n"); | |
51 | trace_seq_puts(s, "\tarray : 32 bits\n"); | |
52 | trace_seq_putc(s, '\n'); | |
53 | trace_seq_printf(s, "\tpadding : type == %d\n", | |
54 | RINGBUF_TYPE_PADDING); | |
55 | trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
56 | RINGBUF_TYPE_TIME_EXTEND); | |
dc4e2801 TZ |
57 | trace_seq_printf(s, "\ttime_stamp : type == %d\n", |
58 | RINGBUF_TYPE_TIME_STAMP); | |
c0cd93aa SRRH |
59 | trace_seq_printf(s, "\tdata max type_len == %d\n", |
60 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
61 | ||
62 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
63 | } |
64 | ||
5cc98548 SR |
65 | /* |
66 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
67 | * allocated for each CPU. A writer may only write to a buffer that is | |
68 | * associated with the CPU it is currently executing on. A reader may read | |
69 | * from any per cpu buffer. | |
70 | * | |
71 | * The reader is special. For each per cpu buffer, the reader has its own | |
72 | * reader page. When a reader has read the entire reader page, this reader | |
73 | * page is swapped with another page in the ring buffer. | |
74 | * | |
75 | * Now, as long as the writer is off the reader page, the reader can do what | |
76 | * ever it wants with that page. The writer will never write to that page | |
77 | * again (as long as it is out of the ring buffer). | |
78 | * | |
79 | * Here's some silly ASCII art. | |
80 | * | |
81 | * +------+ | |
82 | * |reader| RING BUFFER | |
83 | * |page | | |
84 | * +------+ +---+ +---+ +---+ | |
85 | * | |-->| |-->| | | |
86 | * +---+ +---+ +---+ | |
87 | * ^ | | |
88 | * | | | |
89 | * +---------------+ | |
90 | * | |
91 | * | |
92 | * +------+ | |
93 | * |reader| RING BUFFER | |
94 | * |page |------------------v | |
95 | * +------+ +---+ +---+ +---+ | |
96 | * | |-->| |-->| | | |
97 | * +---+ +---+ +---+ | |
98 | * ^ | | |
99 | * | | | |
100 | * +---------------+ | |
101 | * | |
102 | * | |
103 | * +------+ | |
104 | * |reader| RING BUFFER | |
105 | * |page |------------------v | |
106 | * +------+ +---+ +---+ +---+ | |
107 | * ^ | |-->| |-->| | | |
108 | * | +---+ +---+ +---+ | |
109 | * | | | |
110 | * | | | |
111 | * +------------------------------+ | |
112 | * | |
113 | * | |
114 | * +------+ | |
115 | * |buffer| RING BUFFER | |
116 | * |page |------------------v | |
117 | * +------+ +---+ +---+ +---+ | |
118 | * ^ | | | |-->| | | |
119 | * | New +---+ +---+ +---+ | |
120 | * | Reader------^ | | |
121 | * | page | | |
122 | * +------------------------------+ | |
123 | * | |
124 | * | |
125 | * After we make this swap, the reader can hand this page off to the splice | |
126 | * code and be done with it. It can even allocate a new page if it needs to | |
127 | * and swap that into the ring buffer. | |
128 | * | |
129 | * We will be using cmpxchg soon to make all this lockless. | |
130 | * | |
131 | */ | |
132 | ||
499e5470 SR |
133 | /* Used for individual buffers (after the counter) */ |
134 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 135 | |
499e5470 | 136 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 | 137 | |
e3d6bf0a | 138 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 139 | #define RB_ALIGNMENT 4U |
334d4169 | 140 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 141 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
adab66b7 SRV |
142 | |
143 | #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS | |
144 | # define RB_FORCE_8BYTE_ALIGNMENT 0 | |
145 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
146 | #else | |
147 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
148 | # define RB_ARCH_ALIGNMENT 8U | |
149 | #endif | |
150 | ||
151 | #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT) | |
649508f6 | 152 | |
334d4169 LJ |
153 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
154 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
155 | |
156 | enum { | |
157 | RB_LEN_TIME_EXTEND = 8, | |
dc4e2801 | 158 | RB_LEN_TIME_STAMP = 8, |
7a8e76a3 SR |
159 | }; |
160 | ||
69d1b839 SR |
161 | #define skip_time_extend(event) \ |
162 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
163 | ||
dc4e2801 TZ |
164 | #define extended_time(event) \ |
165 | (event->type_len >= RINGBUF_TYPE_TIME_EXTEND) | |
166 | ||
bc92b956 | 167 | static inline bool rb_null_event(struct ring_buffer_event *event) |
2d622719 | 168 | { |
a1863c21 | 169 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
170 | } |
171 | ||
172 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
173 | { | |
a1863c21 | 174 | /* padding has a NULL time_delta */ |
334d4169 | 175 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
176 | event->time_delta = 0; |
177 | } | |
178 | ||
34a148bf | 179 | static unsigned |
2d622719 | 180 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
181 | { |
182 | unsigned length; | |
183 | ||
334d4169 LJ |
184 | if (event->type_len) |
185 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
186 | else |
187 | length = event->array[0]; | |
188 | return length + RB_EVNT_HDR_SIZE; | |
189 | } | |
190 | ||
69d1b839 SR |
191 | /* |
192 | * Return the length of the given event. Will return | |
193 | * the length of the time extend if the event is a | |
194 | * time extend. | |
195 | */ | |
196 | static inline unsigned | |
2d622719 TZ |
197 | rb_event_length(struct ring_buffer_event *event) |
198 | { | |
334d4169 | 199 | switch (event->type_len) { |
7a8e76a3 | 200 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
201 | if (rb_null_event(event)) |
202 | /* undefined */ | |
203 | return -1; | |
334d4169 | 204 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
205 | |
206 | case RINGBUF_TYPE_TIME_EXTEND: | |
207 | return RB_LEN_TIME_EXTEND; | |
208 | ||
209 | case RINGBUF_TYPE_TIME_STAMP: | |
210 | return RB_LEN_TIME_STAMP; | |
211 | ||
212 | case RINGBUF_TYPE_DATA: | |
2d622719 | 213 | return rb_event_data_length(event); |
7a8e76a3 | 214 | default: |
da4d401a | 215 | WARN_ON_ONCE(1); |
7a8e76a3 SR |
216 | } |
217 | /* not hit */ | |
218 | return 0; | |
219 | } | |
220 | ||
69d1b839 SR |
221 | /* |
222 | * Return total length of time extend and data, | |
223 | * or just the event length for all other events. | |
224 | */ | |
225 | static inline unsigned | |
226 | rb_event_ts_length(struct ring_buffer_event *event) | |
227 | { | |
228 | unsigned len = 0; | |
229 | ||
dc4e2801 | 230 | if (extended_time(event)) { |
69d1b839 SR |
231 | /* time extends include the data event after it */ |
232 | len = RB_LEN_TIME_EXTEND; | |
233 | event = skip_time_extend(event); | |
234 | } | |
235 | return len + rb_event_length(event); | |
236 | } | |
237 | ||
7a8e76a3 SR |
238 | /** |
239 | * ring_buffer_event_length - return the length of the event | |
240 | * @event: the event to get the length of | |
69d1b839 SR |
241 | * |
242 | * Returns the size of the data load of a data event. | |
243 | * If the event is something other than a data event, it | |
244 | * returns the size of the event itself. With the exception | |
245 | * of a TIME EXTEND, where it still returns the size of the | |
246 | * data load of the data event after it. | |
7a8e76a3 SR |
247 | */ |
248 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
249 | { | |
69d1b839 SR |
250 | unsigned length; |
251 | ||
dc4e2801 | 252 | if (extended_time(event)) |
69d1b839 SR |
253 | event = skip_time_extend(event); |
254 | ||
255 | length = rb_event_length(event); | |
334d4169 | 256 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
257 | return length; |
258 | length -= RB_EVNT_HDR_SIZE; | |
259 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
260 | length -= sizeof(event->array[0]); | |
261 | return length; | |
7a8e76a3 | 262 | } |
c4f50183 | 263 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
264 | |
265 | /* inline for ring buffer fast paths */ | |
929ddbf3 | 266 | static __always_inline void * |
7a8e76a3 SR |
267 | rb_event_data(struct ring_buffer_event *event) |
268 | { | |
dc4e2801 | 269 | if (extended_time(event)) |
69d1b839 | 270 | event = skip_time_extend(event); |
da4d401a | 271 | WARN_ON_ONCE(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 272 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 273 | if (event->type_len) |
7a8e76a3 SR |
274 | return (void *)&event->array[0]; |
275 | /* Otherwise length is in array[0] and array[1] has the data */ | |
276 | return (void *)&event->array[1]; | |
277 | } | |
278 | ||
279 | /** | |
280 | * ring_buffer_event_data - return the data of the event | |
281 | * @event: the event to get the data from | |
282 | */ | |
283 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
284 | { | |
285 | return rb_event_data(event); | |
286 | } | |
c4f50183 | 287 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
288 | |
289 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 290 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 | 291 | |
b23d7a5f NP |
292 | #define for_each_online_buffer_cpu(buffer, cpu) \ |
293 | for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask) | |
294 | ||
7a8e76a3 SR |
295 | #define TS_SHIFT 27 |
296 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
297 | #define TS_DELTA_TEST (~TS_MASK) | |
298 | ||
e20044f7 SRV |
299 | static u64 rb_event_time_stamp(struct ring_buffer_event *event) |
300 | { | |
301 | u64 ts; | |
302 | ||
303 | ts = event->array[0]; | |
304 | ts <<= TS_SHIFT; | |
305 | ts += event->time_delta; | |
306 | ||
307 | return ts; | |
308 | } | |
309 | ||
66a8cb95 SR |
310 | /* Flag when events were overwritten */ |
311 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
312 | /* Missed count stored at end */ |
313 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 314 | |
abc9b56d | 315 | struct buffer_data_page { |
e4c2ce82 | 316 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 317 | local_t commit; /* write committed index */ |
649508f6 | 318 | unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */ |
abc9b56d SR |
319 | }; |
320 | ||
bce761d7 TSV |
321 | struct buffer_data_read_page { |
322 | unsigned order; /* order of the page */ | |
323 | struct buffer_data_page *data; /* actual data, stored in this page */ | |
324 | }; | |
325 | ||
77ae365e SR |
326 | /* |
327 | * Note, the buffer_page list must be first. The buffer pages | |
328 | * are allocated in cache lines, which means that each buffer | |
329 | * page will be at the beginning of a cache line, and thus | |
330 | * the least significant bits will be zero. We use this to | |
331 | * add flags in the list struct pointers, to make the ring buffer | |
332 | * lockless. | |
333 | */ | |
abc9b56d | 334 | struct buffer_page { |
778c55d4 | 335 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 336 | local_t write; /* index for next write */ |
6f807acd | 337 | unsigned read; /* index for next read */ |
778c55d4 | 338 | local_t entries; /* entries on this page */ |
ff0ff84a | 339 | unsigned long real_end; /* real end of data */ |
f9b94daa | 340 | unsigned order; /* order of the page */ |
abc9b56d | 341 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
342 | }; |
343 | ||
77ae365e SR |
344 | /* |
345 | * The buffer page counters, write and entries, must be reset | |
346 | * atomically when crossing page boundaries. To synchronize this | |
347 | * update, two counters are inserted into the number. One is | |
348 | * the actual counter for the write position or count on the page. | |
349 | * | |
350 | * The other is a counter of updaters. Before an update happens | |
351 | * the update partition of the counter is incremented. This will | |
352 | * allow the updater to update the counter atomically. | |
353 | * | |
354 | * The counter is 20 bits, and the state data is 12. | |
355 | */ | |
356 | #define RB_WRITE_MASK 0xfffff | |
357 | #define RB_WRITE_INTCNT (1 << 20) | |
358 | ||
044fa782 | 359 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 360 | { |
044fa782 | 361 | local_set(&bpage->commit, 0); |
abc9b56d SR |
362 | } |
363 | ||
45d99ea4 ZY |
364 | static __always_inline unsigned int rb_page_commit(struct buffer_page *bpage) |
365 | { | |
366 | return local_read(&bpage->page->commit); | |
367 | } | |
368 | ||
34a148bf | 369 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 370 | { |
f9b94daa | 371 | free_pages((unsigned long)bpage->page, bpage->order); |
e4c2ce82 | 372 | kfree(bpage); |
ed56829c SR |
373 | } |
374 | ||
7a8e76a3 SR |
375 | /* |
376 | * We need to fit the time_stamp delta into 27 bits. | |
377 | */ | |
bc92b956 | 378 | static inline bool test_time_stamp(u64 delta) |
7a8e76a3 | 379 | { |
bc92b956 | 380 | return !!(delta & TS_DELTA_TEST); |
7a8e76a3 SR |
381 | } |
382 | ||
15693458 SRRH |
383 | struct rb_irq_work { |
384 | struct irq_work work; | |
385 | wait_queue_head_t waiters; | |
1e0d6714 | 386 | wait_queue_head_t full_waiters; |
7e9fbbb1 | 387 | long wait_index; |
15693458 | 388 | bool waiters_pending; |
1e0d6714 SRRH |
389 | bool full_waiters_pending; |
390 | bool wakeup_full; | |
15693458 SRRH |
391 | }; |
392 | ||
fcc742ea SRRH |
393 | /* |
394 | * Structure to hold event state and handle nested events. | |
395 | */ | |
396 | struct rb_event_info { | |
397 | u64 ts; | |
398 | u64 delta; | |
58fbc3c6 SRV |
399 | u64 before; |
400 | u64 after; | |
fcc742ea SRRH |
401 | unsigned long length; |
402 | struct buffer_page *tail_page; | |
403 | int add_timestamp; | |
404 | }; | |
405 | ||
a389d86f SRV |
406 | /* |
407 | * Used for the add_timestamp | |
408 | * NONE | |
7c4b4a51 SRV |
409 | * EXTEND - wants a time extend |
410 | * ABSOLUTE - the buffer requests all events to have absolute time stamps | |
a389d86f SRV |
411 | * FORCE - force a full time stamp. |
412 | */ | |
413 | enum { | |
7c4b4a51 SRV |
414 | RB_ADD_STAMP_NONE = 0, |
415 | RB_ADD_STAMP_EXTEND = BIT(1), | |
416 | RB_ADD_STAMP_ABSOLUTE = BIT(2), | |
417 | RB_ADD_STAMP_FORCE = BIT(3) | |
a389d86f | 418 | }; |
a497adb4 SRRH |
419 | /* |
420 | * Used for which event context the event is in. | |
b02414c8 SRV |
421 | * TRANSITION = 0 |
422 | * NMI = 1 | |
423 | * IRQ = 2 | |
424 | * SOFTIRQ = 3 | |
425 | * NORMAL = 4 | |
a497adb4 SRRH |
426 | * |
427 | * See trace_recursive_lock() comment below for more details. | |
428 | */ | |
429 | enum { | |
b02414c8 | 430 | RB_CTX_TRANSITION, |
a497adb4 SRRH |
431 | RB_CTX_NMI, |
432 | RB_CTX_IRQ, | |
433 | RB_CTX_SOFTIRQ, | |
434 | RB_CTX_NORMAL, | |
435 | RB_CTX_MAX | |
436 | }; | |
437 | ||
10464b4a SRV |
438 | struct rb_time_struct { |
439 | local64_t time; | |
440 | }; | |
10464b4a SRV |
441 | typedef struct rb_time_struct rb_time_t; |
442 | ||
8672e494 SRV |
443 | #define MAX_NEST 5 |
444 | ||
7a8e76a3 SR |
445 | /* |
446 | * head_page == tail_page && head == tail then buffer is empty. | |
447 | */ | |
448 | struct ring_buffer_per_cpu { | |
449 | int cpu; | |
985023de | 450 | atomic_t record_disabled; |
07b8b10e | 451 | atomic_t resize_disabled; |
13292494 | 452 | struct trace_buffer *buffer; |
5389f6fa | 453 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 454 | arch_spinlock_t lock; |
7a8e76a3 | 455 | struct lock_class_key lock_key; |
73a757e6 | 456 | struct buffer_data_page *free_page; |
9b94a8fb | 457 | unsigned long nr_pages; |
58a09ec6 | 458 | unsigned int current_context; |
3adc54fa | 459 | struct list_head *pages; |
6f807acd SR |
460 | struct buffer_page *head_page; /* read from head */ |
461 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 462 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 463 | struct buffer_page *reader_page; |
66a8cb95 SR |
464 | unsigned long lost_events; |
465 | unsigned long last_overrun; | |
8e012066 | 466 | unsigned long nest; |
c64e148a | 467 | local_t entries_bytes; |
e4906eff | 468 | local_t entries; |
884bfe89 SP |
469 | local_t overrun; |
470 | local_t commit_overrun; | |
471 | local_t dropped_events; | |
fa743953 SR |
472 | local_t committing; |
473 | local_t commits; | |
2c2b0a78 | 474 | local_t pages_touched; |
31029a8b | 475 | local_t pages_lost; |
2c2b0a78 | 476 | local_t pages_read; |
03329f99 | 477 | long last_pages_touch; |
2c2b0a78 | 478 | size_t shortest_full; |
77ae365e | 479 | unsigned long read; |
c64e148a | 480 | unsigned long read_bytes; |
10464b4a SRV |
481 | rb_time_t write_stamp; |
482 | rb_time_t before_stamp; | |
8672e494 | 483 | u64 event_stamp[MAX_NEST]; |
7a8e76a3 | 484 | u64 read_stamp; |
2d093282 ZY |
485 | /* pages removed since last reset */ |
486 | unsigned long pages_removed; | |
438ced17 | 487 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
9b94a8fb | 488 | long nr_pages_to_update; |
438ced17 | 489 | struct list_head new_pages; /* new pages to add */ |
83f40318 | 490 | struct work_struct update_pages_work; |
05fdd70d | 491 | struct completion update_done; |
15693458 SRRH |
492 | |
493 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
494 | }; |
495 | ||
13292494 | 496 | struct trace_buffer { |
7a8e76a3 SR |
497 | unsigned flags; |
498 | int cpus; | |
7a8e76a3 | 499 | atomic_t record_disabled; |
8a96c028 | 500 | atomic_t resizing; |
00f62f61 | 501 | cpumask_var_t cpumask; |
7a8e76a3 | 502 | |
1f8a6a10 PZ |
503 | struct lock_class_key *reader_lock_key; |
504 | ||
7a8e76a3 SR |
505 | struct mutex mutex; |
506 | ||
507 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 508 | |
b32614c0 | 509 | struct hlist_node node; |
37886f6a | 510 | u64 (*clock)(void); |
15693458 SRRH |
511 | |
512 | struct rb_irq_work irq_work; | |
00b41452 | 513 | bool time_stamp_abs; |
139f8400 TSV |
514 | |
515 | unsigned int subbuf_size; | |
2808e31e | 516 | unsigned int subbuf_order; |
139f8400 | 517 | unsigned int max_data_size; |
7a8e76a3 SR |
518 | }; |
519 | ||
520 | struct ring_buffer_iter { | |
521 | struct ring_buffer_per_cpu *cpu_buffer; | |
522 | unsigned long head; | |
785888c5 | 523 | unsigned long next_event; |
7a8e76a3 | 524 | struct buffer_page *head_page; |
492a74f4 SR |
525 | struct buffer_page *cache_reader_page; |
526 | unsigned long cache_read; | |
2d093282 | 527 | unsigned long cache_pages_removed; |
7a8e76a3 | 528 | u64 read_stamp; |
28e3fc56 | 529 | u64 page_stamp; |
785888c5 | 530 | struct ring_buffer_event *event; |
139f8400 | 531 | size_t event_size; |
c9b7a4a7 | 532 | int missed_events; |
7a8e76a3 SR |
533 | }; |
534 | ||
139f8400 | 535 | int ring_buffer_print_page_header(struct trace_buffer *buffer, struct trace_seq *s) |
10464b4a | 536 | { |
d5cfbdfc | 537 | struct buffer_data_page field; |
10464b4a | 538 | |
d5cfbdfc TSV |
539 | trace_seq_printf(s, "\tfield: u64 timestamp;\t" |
540 | "offset:0;\tsize:%u;\tsigned:%u;\n", | |
541 | (unsigned int)sizeof(field.time_stamp), | |
542 | (unsigned int)is_signed_type(u64)); | |
10464b4a | 543 | |
d5cfbdfc TSV |
544 | trace_seq_printf(s, "\tfield: local_t commit;\t" |
545 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
546 | (unsigned int)offsetof(typeof(field), commit), | |
547 | (unsigned int)sizeof(field.commit), | |
548 | (unsigned int)is_signed_type(long)); | |
10464b4a | 549 | |
d5cfbdfc TSV |
550 | trace_seq_printf(s, "\tfield: int overwrite;\t" |
551 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
552 | (unsigned int)offsetof(typeof(field), commit), | |
553 | 1, | |
554 | (unsigned int)is_signed_type(long)); | |
10464b4a | 555 | |
d5cfbdfc TSV |
556 | trace_seq_printf(s, "\tfield: char data;\t" |
557 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
558 | (unsigned int)offsetof(typeof(field), data), | |
139f8400 | 559 | (unsigned int)buffer->subbuf_size, |
d5cfbdfc | 560 | (unsigned int)is_signed_type(char)); |
10464b4a | 561 | |
d5cfbdfc | 562 | return !trace_seq_has_overflowed(s); |
10464b4a SRV |
563 | } |
564 | ||
c84897c0 | 565 | static inline void rb_time_read(rb_time_t *t, u64 *ret) |
10464b4a SRV |
566 | { |
567 | *ret = local64_read(&t->time); | |
10464b4a SRV |
568 | } |
569 | static void rb_time_set(rb_time_t *t, u64 val) | |
570 | { | |
571 | local64_set(&t->time, val); | |
572 | } | |
10464b4a | 573 | |
a948c69d SRV |
574 | /* |
575 | * Enable this to make sure that the event passed to | |
576 | * ring_buffer_event_time_stamp() is not committed and also | |
577 | * is on the buffer that it passed in. | |
578 | */ | |
579 | //#define RB_VERIFY_EVENT | |
580 | #ifdef RB_VERIFY_EVENT | |
581 | static struct list_head *rb_list_head(struct list_head *list); | |
582 | static void verify_event(struct ring_buffer_per_cpu *cpu_buffer, | |
583 | void *event) | |
584 | { | |
585 | struct buffer_page *page = cpu_buffer->commit_page; | |
586 | struct buffer_page *tail_page = READ_ONCE(cpu_buffer->tail_page); | |
587 | struct list_head *next; | |
588 | long commit, write; | |
589 | unsigned long addr = (unsigned long)event; | |
590 | bool done = false; | |
591 | int stop = 0; | |
592 | ||
593 | /* Make sure the event exists and is not committed yet */ | |
594 | do { | |
595 | if (page == tail_page || WARN_ON_ONCE(stop++ > 100)) | |
596 | done = true; | |
597 | commit = local_read(&page->page->commit); | |
598 | write = local_read(&page->write); | |
599 | if (addr >= (unsigned long)&page->page->data[commit] && | |
600 | addr < (unsigned long)&page->page->data[write]) | |
601 | return; | |
602 | ||
603 | next = rb_list_head(page->list.next); | |
604 | page = list_entry(next, struct buffer_page, list); | |
605 | } while (!done); | |
606 | WARN_ON_ONCE(1); | |
607 | } | |
608 | #else | |
609 | static inline void verify_event(struct ring_buffer_per_cpu *cpu_buffer, | |
610 | void *event) | |
611 | { | |
612 | } | |
613 | #endif | |
614 | ||
6695da58 SRG |
615 | /* |
616 | * The absolute time stamp drops the 5 MSBs and some clocks may | |
617 | * require them. The rb_fix_abs_ts() will take a previous full | |
618 | * time stamp, and add the 5 MSB of that time stamp on to the | |
619 | * saved absolute time stamp. Then they are compared in case of | |
620 | * the unlikely event that the latest time stamp incremented | |
621 | * the 5 MSB. | |
622 | */ | |
623 | static inline u64 rb_fix_abs_ts(u64 abs, u64 save_ts) | |
624 | { | |
625 | if (save_ts & TS_MSB) { | |
626 | abs |= save_ts & TS_MSB; | |
627 | /* Check for overflow */ | |
628 | if (unlikely(abs < save_ts)) | |
629 | abs += 1ULL << 59; | |
630 | } | |
631 | return abs; | |
632 | } | |
a948c69d | 633 | |
efe6196a SRV |
634 | static inline u64 rb_time_stamp(struct trace_buffer *buffer); |
635 | ||
636 | /** | |
637 | * ring_buffer_event_time_stamp - return the event's current time stamp | |
638 | * @buffer: The buffer that the event is on | |
639 | * @event: the event to get the time stamp of | |
640 | * | |
641 | * Note, this must be called after @event is reserved, and before it is | |
642 | * committed to the ring buffer. And must be called from the same | |
643 | * context where the event was reserved (normal, softirq, irq, etc). | |
644 | * | |
645 | * Returns the time stamp associated with the current event. | |
646 | * If the event has an extended time stamp, then that is used as | |
647 | * the time stamp to return. | |
648 | * In the highly unlikely case that the event was nested more than | |
649 | * the max nesting, then the write_stamp of the buffer is returned, | |
650 | * otherwise current time is returned, but that really neither of | |
651 | * the last two cases should ever happen. | |
652 | */ | |
653 | u64 ring_buffer_event_time_stamp(struct trace_buffer *buffer, | |
654 | struct ring_buffer_event *event) | |
655 | { | |
656 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[smp_processor_id()]; | |
657 | unsigned int nest; | |
658 | u64 ts; | |
659 | ||
660 | /* If the event includes an absolute time, then just use that */ | |
6695da58 SRG |
661 | if (event->type_len == RINGBUF_TYPE_TIME_STAMP) { |
662 | ts = rb_event_time_stamp(event); | |
663 | return rb_fix_abs_ts(ts, cpu_buffer->tail_page->page->time_stamp); | |
664 | } | |
efe6196a | 665 | |
a948c69d SRV |
666 | nest = local_read(&cpu_buffer->committing); |
667 | verify_event(cpu_buffer, event); | |
668 | if (WARN_ON_ONCE(!nest)) | |
669 | goto fail; | |
670 | ||
efe6196a | 671 | /* Read the current saved nesting level time stamp */ |
a948c69d | 672 | if (likely(--nest < MAX_NEST)) |
efe6196a SRV |
673 | return cpu_buffer->event_stamp[nest]; |
674 | ||
a948c69d SRV |
675 | /* Shouldn't happen, warn if it does */ |
676 | WARN_ONCE(1, "nest (%d) greater than max", nest); | |
efe6196a | 677 | |
a948c69d | 678 | fail: |
c84897c0 | 679 | rb_time_read(&cpu_buffer->write_stamp, &ts); |
efe6196a SRV |
680 | |
681 | return ts; | |
682 | } | |
683 | ||
2c2b0a78 SRV |
684 | /** |
685 | * ring_buffer_nr_pages - get the number of buffer pages in the ring buffer | |
686 | * @buffer: The ring_buffer to get the number of pages from | |
687 | * @cpu: The cpu of the ring_buffer to get the number of pages from | |
688 | * | |
689 | * Returns the number of pages used by a per_cpu buffer of the ring buffer. | |
690 | */ | |
13292494 | 691 | size_t ring_buffer_nr_pages(struct trace_buffer *buffer, int cpu) |
2c2b0a78 SRV |
692 | { |
693 | return buffer->buffers[cpu]->nr_pages; | |
694 | } | |
695 | ||
696 | /** | |
b7085b6f | 697 | * ring_buffer_nr_dirty_pages - get the number of used pages in the ring buffer |
2c2b0a78 SRV |
698 | * @buffer: The ring_buffer to get the number of pages from |
699 | * @cpu: The cpu of the ring_buffer to get the number of pages from | |
700 | * | |
701 | * Returns the number of pages that have content in the ring buffer. | |
702 | */ | |
13292494 | 703 | size_t ring_buffer_nr_dirty_pages(struct trace_buffer *buffer, int cpu) |
2c2b0a78 SRV |
704 | { |
705 | size_t read; | |
31029a8b | 706 | size_t lost; |
2c2b0a78 SRV |
707 | size_t cnt; |
708 | ||
709 | read = local_read(&buffer->buffers[cpu]->pages_read); | |
31029a8b | 710 | lost = local_read(&buffer->buffers[cpu]->pages_lost); |
2c2b0a78 | 711 | cnt = local_read(&buffer->buffers[cpu]->pages_touched); |
31029a8b SRG |
712 | |
713 | if (WARN_ON_ONCE(cnt < lost)) | |
714 | return 0; | |
715 | ||
716 | cnt -= lost; | |
717 | ||
2c2b0a78 SRV |
718 | /* The reader can read an empty page, but not more than that */ |
719 | if (cnt < read) { | |
720 | WARN_ON_ONCE(read > cnt + 1); | |
721 | return 0; | |
722 | } | |
723 | ||
724 | return cnt - read; | |
725 | } | |
726 | ||
42fb0a1e SRG |
727 | static __always_inline bool full_hit(struct trace_buffer *buffer, int cpu, int full) |
728 | { | |
729 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
730 | size_t nr_pages; | |
731 | size_t dirty; | |
732 | ||
733 | nr_pages = cpu_buffer->nr_pages; | |
734 | if (!nr_pages || !full) | |
735 | return true; | |
736 | ||
623b1f89 SRG |
737 | /* |
738 | * Add one as dirty will never equal nr_pages, as the sub-buffer | |
739 | * that the writer is on is not counted as dirty. | |
740 | * This is needed if "buffer_percent" is set to 100. | |
741 | */ | |
742 | dirty = ring_buffer_nr_dirty_pages(buffer, cpu) + 1; | |
42fb0a1e | 743 | |
623b1f89 | 744 | return (dirty * 100) >= (full * nr_pages); |
42fb0a1e SRG |
745 | } |
746 | ||
15693458 SRRH |
747 | /* |
748 | * rb_wake_up_waiters - wake up tasks waiting for ring buffer input | |
749 | * | |
750 | * Schedules a delayed work to wake up any task that is blocked on the | |
751 | * ring buffer waiters queue. | |
752 | */ | |
753 | static void rb_wake_up_waiters(struct irq_work *work) | |
754 | { | |
755 | struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work); | |
756 | ||
757 | wake_up_all(&rbwork->waiters); | |
ec0bbc5e | 758 | if (rbwork->full_waiters_pending || rbwork->wakeup_full) { |
1e0d6714 | 759 | rbwork->wakeup_full = false; |
ec0bbc5e | 760 | rbwork->full_waiters_pending = false; |
1e0d6714 SRRH |
761 | wake_up_all(&rbwork->full_waiters); |
762 | } | |
15693458 SRRH |
763 | } |
764 | ||
7e9fbbb1 SRG |
765 | /** |
766 | * ring_buffer_wake_waiters - wake up any waiters on this ring buffer | |
767 | * @buffer: The ring buffer to wake waiters on | |
151e34d1 | 768 | * @cpu: The CPU buffer to wake waiters on |
7e9fbbb1 SRG |
769 | * |
770 | * In the case of a file that represents a ring buffer is closing, | |
771 | * it is prudent to wake up any waiters that are on this. | |
772 | */ | |
773 | void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu) | |
774 | { | |
775 | struct ring_buffer_per_cpu *cpu_buffer; | |
776 | struct rb_irq_work *rbwork; | |
777 | ||
7433632c SRG |
778 | if (!buffer) |
779 | return; | |
780 | ||
7e9fbbb1 SRG |
781 | if (cpu == RING_BUFFER_ALL_CPUS) { |
782 | ||
783 | /* Wake up individual ones too. One level recursion */ | |
784 | for_each_buffer_cpu(buffer, cpu) | |
785 | ring_buffer_wake_waiters(buffer, cpu); | |
786 | ||
787 | rbwork = &buffer->irq_work; | |
788 | } else { | |
7433632c SRG |
789 | if (WARN_ON_ONCE(!buffer->buffers)) |
790 | return; | |
791 | if (WARN_ON_ONCE(cpu >= nr_cpu_ids)) | |
792 | return; | |
793 | ||
7e9fbbb1 | 794 | cpu_buffer = buffer->buffers[cpu]; |
7433632c SRG |
795 | /* The CPU buffer may not have been initialized yet */ |
796 | if (!cpu_buffer) | |
797 | return; | |
7e9fbbb1 SRG |
798 | rbwork = &cpu_buffer->irq_work; |
799 | } | |
800 | ||
801 | rbwork->wait_index++; | |
802 | /* make sure the waiters see the new index */ | |
803 | smp_wmb(); | |
804 | ||
39a7dc23 SRG |
805 | /* This can be called in any context */ |
806 | irq_work_queue(&rbwork->work); | |
7e9fbbb1 SRG |
807 | } |
808 | ||
15693458 SRRH |
809 | /** |
810 | * ring_buffer_wait - wait for input to the ring buffer | |
811 | * @buffer: buffer to wait on | |
812 | * @cpu: the cpu buffer to wait on | |
e1981f75 | 813 | * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS |
15693458 SRRH |
814 | * |
815 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
816 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
817 | * it will wait for data to be added to a specific cpu buffer. | |
818 | */ | |
13292494 | 819 | int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full) |
15693458 | 820 | { |
3f649ab7 | 821 | struct ring_buffer_per_cpu *cpu_buffer; |
15693458 SRRH |
822 | DEFINE_WAIT(wait); |
823 | struct rb_irq_work *work; | |
7e9fbbb1 | 824 | long wait_index; |
e30f53aa | 825 | int ret = 0; |
15693458 SRRH |
826 | |
827 | /* | |
828 | * Depending on what the caller is waiting for, either any | |
829 | * data in any cpu buffer, or a specific buffer, put the | |
830 | * caller on the appropriate wait queue. | |
831 | */ | |
1e0d6714 | 832 | if (cpu == RING_BUFFER_ALL_CPUS) { |
15693458 | 833 | work = &buffer->irq_work; |
1e0d6714 | 834 | /* Full only makes sense on per cpu reads */ |
2c2b0a78 | 835 | full = 0; |
1e0d6714 | 836 | } else { |
8b8b3683 SRRH |
837 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
838 | return -ENODEV; | |
15693458 SRRH |
839 | cpu_buffer = buffer->buffers[cpu]; |
840 | work = &cpu_buffer->irq_work; | |
841 | } | |
842 | ||
7e9fbbb1 | 843 | wait_index = READ_ONCE(work->wait_index); |
15693458 | 844 | |
e30f53aa | 845 | while (true) { |
1e0d6714 SRRH |
846 | if (full) |
847 | prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE); | |
848 | else | |
849 | prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE); | |
e30f53aa RV |
850 | |
851 | /* | |
852 | * The events can happen in critical sections where | |
853 | * checking a work queue can cause deadlocks. | |
854 | * After adding a task to the queue, this flag is set | |
855 | * only to notify events to try to wake up the queue | |
856 | * using irq_work. | |
857 | * | |
858 | * We don't clear it even if the buffer is no longer | |
859 | * empty. The flag only causes the next event to run | |
860 | * irq_work to do the work queue wake up. The worse | |
861 | * that can happen if we race with !trace_empty() is that | |
862 | * an event will cause an irq_work to try to wake up | |
863 | * an empty queue. | |
864 | * | |
865 | * There's no reason to protect this flag either, as | |
866 | * the work queue and irq_work logic will do the necessary | |
867 | * synchronization for the wake ups. The only thing | |
868 | * that is necessary is that the wake up happens after | |
869 | * a task has been queued. It's OK for spurious wake ups. | |
870 | */ | |
1e0d6714 SRRH |
871 | if (full) |
872 | work->full_waiters_pending = true; | |
873 | else | |
874 | work->waiters_pending = true; | |
e30f53aa RV |
875 | |
876 | if (signal_pending(current)) { | |
877 | ret = -EINTR; | |
878 | break; | |
879 | } | |
880 | ||
881 | if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) | |
882 | break; | |
883 | ||
884 | if (cpu != RING_BUFFER_ALL_CPUS && | |
885 | !ring_buffer_empty_cpu(buffer, cpu)) { | |
886 | unsigned long flags; | |
887 | bool pagebusy; | |
42fb0a1e | 888 | bool done; |
e30f53aa RV |
889 | |
890 | if (!full) | |
891 | break; | |
892 | ||
893 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
894 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
42fb0a1e SRG |
895 | done = !pagebusy && full_hit(buffer, cpu, full); |
896 | ||
2c2b0a78 | 897 | if (!cpu_buffer->shortest_full || |
3b19d614 | 898 | cpu_buffer->shortest_full > full) |
2c2b0a78 | 899 | cpu_buffer->shortest_full = full; |
e30f53aa | 900 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
42fb0a1e | 901 | if (done) |
e30f53aa RV |
902 | break; |
903 | } | |
15693458 | 904 | |
15693458 | 905 | schedule(); |
7e9fbbb1 SRG |
906 | |
907 | /* Make sure to see the new wait index */ | |
908 | smp_rmb(); | |
909 | if (wait_index != work->wait_index) | |
910 | break; | |
e30f53aa | 911 | } |
15693458 | 912 | |
1e0d6714 SRRH |
913 | if (full) |
914 | finish_wait(&work->full_waiters, &wait); | |
915 | else | |
916 | finish_wait(&work->waiters, &wait); | |
e30f53aa RV |
917 | |
918 | return ret; | |
15693458 SRRH |
919 | } |
920 | ||
921 | /** | |
922 | * ring_buffer_poll_wait - poll on buffer input | |
923 | * @buffer: buffer to wait on | |
924 | * @cpu: the cpu buffer to wait on | |
925 | * @filp: the file descriptor | |
926 | * @poll_table: The poll descriptor | |
42fb0a1e | 927 | * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS |
15693458 SRRH |
928 | * |
929 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
930 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
931 | * it will wait for data to be added to a specific cpu buffer. | |
932 | * | |
a9a08845 | 933 | * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers, |
15693458 SRRH |
934 | * zero otherwise. |
935 | */ | |
13292494 | 936 | __poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu, |
42fb0a1e | 937 | struct file *filp, poll_table *poll_table, int full) |
15693458 SRRH |
938 | { |
939 | struct ring_buffer_per_cpu *cpu_buffer; | |
940 | struct rb_irq_work *work; | |
941 | ||
42fb0a1e | 942 | if (cpu == RING_BUFFER_ALL_CPUS) { |
15693458 | 943 | work = &buffer->irq_work; |
42fb0a1e SRG |
944 | full = 0; |
945 | } else { | |
6721cb60 SRRH |
946 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
947 | return -EINVAL; | |
948 | ||
15693458 SRRH |
949 | cpu_buffer = buffer->buffers[cpu]; |
950 | work = &cpu_buffer->irq_work; | |
951 | } | |
952 | ||
42fb0a1e SRG |
953 | if (full) { |
954 | poll_wait(filp, &work->full_waiters, poll_table); | |
955 | work->full_waiters_pending = true; | |
1e0cb399 SRG |
956 | if (!cpu_buffer->shortest_full || |
957 | cpu_buffer->shortest_full > full) | |
958 | cpu_buffer->shortest_full = full; | |
42fb0a1e SRG |
959 | } else { |
960 | poll_wait(filp, &work->waiters, poll_table); | |
961 | work->waiters_pending = true; | |
962 | } | |
963 | ||
4ce97dbf JB |
964 | /* |
965 | * There's a tight race between setting the waiters_pending and | |
966 | * checking if the ring buffer is empty. Once the waiters_pending bit | |
967 | * is set, the next event will wake the task up, but we can get stuck | |
968 | * if there's only a single event in. | |
969 | * | |
970 | * FIXME: Ideally, we need a memory barrier on the writer side as well, | |
971 | * but adding a memory barrier to all events will cause too much of a | |
972 | * performance hit in the fast path. We only need a memory barrier when | |
973 | * the buffer goes from empty to having content. But as this race is | |
974 | * extremely small, and it's not a problem if another event comes in, we | |
975 | * will fix it later. | |
976 | */ | |
977 | smp_mb(); | |
15693458 | 978 | |
42fb0a1e SRG |
979 | if (full) |
980 | return full_hit(buffer, cpu, full) ? EPOLLIN | EPOLLRDNORM : 0; | |
981 | ||
15693458 SRRH |
982 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || |
983 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
a9a08845 | 984 | return EPOLLIN | EPOLLRDNORM; |
15693458 SRRH |
985 | return 0; |
986 | } | |
987 | ||
f536aafc | 988 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
989 | #define RB_WARN_ON(b, cond) \ |
990 | ({ \ | |
991 | int _____ret = unlikely(cond); \ | |
992 | if (_____ret) { \ | |
993 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
994 | struct ring_buffer_per_cpu *__b = \ | |
995 | (void *)b; \ | |
996 | atomic_inc(&__b->buffer->record_disabled); \ | |
997 | } else \ | |
998 | atomic_inc(&b->record_disabled); \ | |
999 | WARN_ON(1); \ | |
1000 | } \ | |
1001 | _____ret; \ | |
3e89c7bb | 1002 | }) |
f536aafc | 1003 | |
37886f6a SR |
1004 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
1005 | #define DEBUG_SHIFT 0 | |
1006 | ||
13292494 | 1007 | static inline u64 rb_time_stamp(struct trace_buffer *buffer) |
88eb0125 | 1008 | { |
bbeba3e5 SRV |
1009 | u64 ts; |
1010 | ||
1011 | /* Skip retpolines :-( */ | |
1012 | if (IS_ENABLED(CONFIG_RETPOLINE) && likely(buffer->clock == trace_clock_local)) | |
1013 | ts = trace_clock_local(); | |
1014 | else | |
1015 | ts = buffer->clock(); | |
1016 | ||
88eb0125 | 1017 | /* shift to debug/test normalization and TIME_EXTENTS */ |
bbeba3e5 | 1018 | return ts << DEBUG_SHIFT; |
88eb0125 SR |
1019 | } |
1020 | ||
f3ef7202 | 1021 | u64 ring_buffer_time_stamp(struct trace_buffer *buffer) |
37886f6a SR |
1022 | { |
1023 | u64 time; | |
1024 | ||
1025 | preempt_disable_notrace(); | |
6d3f1e12 | 1026 | time = rb_time_stamp(buffer); |
d6097c9e | 1027 | preempt_enable_notrace(); |
37886f6a SR |
1028 | |
1029 | return time; | |
1030 | } | |
1031 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
1032 | ||
13292494 | 1033 | void ring_buffer_normalize_time_stamp(struct trace_buffer *buffer, |
37886f6a SR |
1034 | int cpu, u64 *ts) |
1035 | { | |
1036 | /* Just stupid testing the normalize function and deltas */ | |
1037 | *ts >>= DEBUG_SHIFT; | |
1038 | } | |
1039 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
1040 | ||
77ae365e SR |
1041 | /* |
1042 | * Making the ring buffer lockless makes things tricky. | |
1043 | * Although writes only happen on the CPU that they are on, | |
1044 | * and they only need to worry about interrupts. Reads can | |
1045 | * happen on any CPU. | |
1046 | * | |
1047 | * The reader page is always off the ring buffer, but when the | |
1048 | * reader finishes with a page, it needs to swap its page with | |
1049 | * a new one from the buffer. The reader needs to take from | |
1050 | * the head (writes go to the tail). But if a writer is in overwrite | |
1051 | * mode and wraps, it must push the head page forward. | |
1052 | * | |
1053 | * Here lies the problem. | |
1054 | * | |
1055 | * The reader must be careful to replace only the head page, and | |
1056 | * not another one. As described at the top of the file in the | |
1057 | * ASCII art, the reader sets its old page to point to the next | |
1058 | * page after head. It then sets the page after head to point to | |
1059 | * the old reader page. But if the writer moves the head page | |
1060 | * during this operation, the reader could end up with the tail. | |
1061 | * | |
1062 | * We use cmpxchg to help prevent this race. We also do something | |
1063 | * special with the page before head. We set the LSB to 1. | |
1064 | * | |
1065 | * When the writer must push the page forward, it will clear the | |
1066 | * bit that points to the head page, move the head, and then set | |
1067 | * the bit that points to the new head page. | |
1068 | * | |
1069 | * We also don't want an interrupt coming in and moving the head | |
1070 | * page on another writer. Thus we use the second LSB to catch | |
1071 | * that too. Thus: | |
1072 | * | |
1073 | * head->list->prev->next bit 1 bit 0 | |
1074 | * ------- ------- | |
1075 | * Normal page 0 0 | |
1076 | * Points to head page 0 1 | |
1077 | * New head page 1 0 | |
1078 | * | |
1079 | * Note we can not trust the prev pointer of the head page, because: | |
1080 | * | |
1081 | * +----+ +-----+ +-----+ | |
1082 | * | |------>| T |---X--->| N | | |
1083 | * | |<------| | | | | |
1084 | * +----+ +-----+ +-----+ | |
1085 | * ^ ^ | | |
1086 | * | +-----+ | | | |
1087 | * +----------| R |----------+ | | |
1088 | * | |<-----------+ | |
1089 | * +-----+ | |
1090 | * | |
1091 | * Key: ---X--> HEAD flag set in pointer | |
1092 | * T Tail page | |
1093 | * R Reader page | |
1094 | * N Next page | |
1095 | * | |
1096 | * (see __rb_reserve_next() to see where this happens) | |
1097 | * | |
1098 | * What the above shows is that the reader just swapped out | |
1099 | * the reader page with a page in the buffer, but before it | |
1100 | * could make the new header point back to the new page added | |
1101 | * it was preempted by a writer. The writer moved forward onto | |
1102 | * the new page added by the reader and is about to move forward | |
1103 | * again. | |
1104 | * | |
1105 | * You can see, it is legitimate for the previous pointer of | |
1106 | * the head (or any page) not to point back to itself. But only | |
6167c205 | 1107 | * temporarily. |
77ae365e SR |
1108 | */ |
1109 | ||
1110 | #define RB_PAGE_NORMAL 0UL | |
1111 | #define RB_PAGE_HEAD 1UL | |
1112 | #define RB_PAGE_UPDATE 2UL | |
1113 | ||
1114 | ||
1115 | #define RB_FLAG_MASK 3UL | |
1116 | ||
1117 | /* PAGE_MOVED is not part of the mask */ | |
1118 | #define RB_PAGE_MOVED 4UL | |
1119 | ||
1120 | /* | |
1121 | * rb_list_head - remove any bit | |
1122 | */ | |
1123 | static struct list_head *rb_list_head(struct list_head *list) | |
1124 | { | |
1125 | unsigned long val = (unsigned long)list; | |
1126 | ||
1127 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
1128 | } | |
1129 | ||
1130 | /* | |
6d3f1e12 | 1131 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
1132 | * |
1133 | * Because the reader may move the head_page pointer, we can | |
1134 | * not trust what the head page is (it may be pointing to | |
1135 | * the reader page). But if the next page is a header page, | |
1136 | * its flags will be non zero. | |
1137 | */ | |
42b16b3f | 1138 | static inline int |
6689bed3 | 1139 | rb_is_head_page(struct buffer_page *page, struct list_head *list) |
77ae365e SR |
1140 | { |
1141 | unsigned long val; | |
1142 | ||
1143 | val = (unsigned long)list->next; | |
1144 | ||
1145 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
1146 | return RB_PAGE_MOVED; | |
1147 | ||
1148 | return val & RB_FLAG_MASK; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * rb_is_reader_page | |
1153 | * | |
1154 | * The unique thing about the reader page, is that, if the | |
1155 | * writer is ever on it, the previous pointer never points | |
1156 | * back to the reader page. | |
1157 | */ | |
06ca3209 | 1158 | static bool rb_is_reader_page(struct buffer_page *page) |
77ae365e SR |
1159 | { |
1160 | struct list_head *list = page->list.prev; | |
1161 | ||
1162 | return rb_list_head(list->next) != &page->list; | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
1167 | */ | |
6689bed3 | 1168 | static void rb_set_list_to_head(struct list_head *list) |
77ae365e SR |
1169 | { |
1170 | unsigned long *ptr; | |
1171 | ||
1172 | ptr = (unsigned long *)&list->next; | |
1173 | *ptr |= RB_PAGE_HEAD; | |
1174 | *ptr &= ~RB_PAGE_UPDATE; | |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * rb_head_page_activate - sets up head page | |
1179 | */ | |
1180 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
1181 | { | |
1182 | struct buffer_page *head; | |
1183 | ||
1184 | head = cpu_buffer->head_page; | |
1185 | if (!head) | |
1186 | return; | |
1187 | ||
1188 | /* | |
1189 | * Set the previous list pointer to have the HEAD flag. | |
1190 | */ | |
6689bed3 | 1191 | rb_set_list_to_head(head->list.prev); |
77ae365e SR |
1192 | } |
1193 | ||
1194 | static void rb_list_head_clear(struct list_head *list) | |
1195 | { | |
1196 | unsigned long *ptr = (unsigned long *)&list->next; | |
1197 | ||
1198 | *ptr &= ~RB_FLAG_MASK; | |
1199 | } | |
1200 | ||
1201 | /* | |
6167c205 | 1202 | * rb_head_page_deactivate - clears head page ptr (for free list) |
77ae365e SR |
1203 | */ |
1204 | static void | |
1205 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
1206 | { | |
1207 | struct list_head *hd; | |
1208 | ||
1209 | /* Go through the whole list and clear any pointers found. */ | |
1210 | rb_list_head_clear(cpu_buffer->pages); | |
1211 | ||
1212 | list_for_each(hd, cpu_buffer->pages) | |
1213 | rb_list_head_clear(hd); | |
1214 | } | |
1215 | ||
1216 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
1217 | struct buffer_page *head, | |
1218 | struct buffer_page *prev, | |
1219 | int old_flag, int new_flag) | |
1220 | { | |
1221 | struct list_head *list; | |
1222 | unsigned long val = (unsigned long)&head->list; | |
1223 | unsigned long ret; | |
1224 | ||
1225 | list = &prev->list; | |
1226 | ||
1227 | val &= ~RB_FLAG_MASK; | |
1228 | ||
08a40816 SR |
1229 | ret = cmpxchg((unsigned long *)&list->next, |
1230 | val | old_flag, val | new_flag); | |
77ae365e SR |
1231 | |
1232 | /* check if the reader took the page */ | |
1233 | if ((ret & ~RB_FLAG_MASK) != val) | |
1234 | return RB_PAGE_MOVED; | |
1235 | ||
1236 | return ret & RB_FLAG_MASK; | |
1237 | } | |
1238 | ||
1239 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
1240 | struct buffer_page *head, | |
1241 | struct buffer_page *prev, | |
1242 | int old_flag) | |
1243 | { | |
1244 | return rb_head_page_set(cpu_buffer, head, prev, | |
1245 | old_flag, RB_PAGE_UPDATE); | |
1246 | } | |
1247 | ||
1248 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
1249 | struct buffer_page *head, | |
1250 | struct buffer_page *prev, | |
1251 | int old_flag) | |
1252 | { | |
1253 | return rb_head_page_set(cpu_buffer, head, prev, | |
1254 | old_flag, RB_PAGE_HEAD); | |
1255 | } | |
1256 | ||
1257 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
1258 | struct buffer_page *head, | |
1259 | struct buffer_page *prev, | |
1260 | int old_flag) | |
1261 | { | |
1262 | return rb_head_page_set(cpu_buffer, head, prev, | |
1263 | old_flag, RB_PAGE_NORMAL); | |
1264 | } | |
1265 | ||
6689bed3 | 1266 | static inline void rb_inc_page(struct buffer_page **bpage) |
77ae365e SR |
1267 | { |
1268 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
1269 | ||
1270 | *bpage = list_entry(p, struct buffer_page, list); | |
1271 | } | |
1272 | ||
1273 | static struct buffer_page * | |
1274 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
1275 | { | |
1276 | struct buffer_page *head; | |
1277 | struct buffer_page *page; | |
1278 | struct list_head *list; | |
1279 | int i; | |
1280 | ||
1281 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
1282 | return NULL; | |
1283 | ||
1284 | /* sanity check */ | |
1285 | list = cpu_buffer->pages; | |
1286 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
1287 | return NULL; | |
1288 | ||
1289 | page = head = cpu_buffer->head_page; | |
1290 | /* | |
1291 | * It is possible that the writer moves the header behind | |
1292 | * where we started, and we miss in one loop. | |
1293 | * A second loop should grab the header, but we'll do | |
1294 | * three loops just because I'm paranoid. | |
1295 | */ | |
1296 | for (i = 0; i < 3; i++) { | |
1297 | do { | |
6689bed3 | 1298 | if (rb_is_head_page(page, page->list.prev)) { |
77ae365e SR |
1299 | cpu_buffer->head_page = page; |
1300 | return page; | |
1301 | } | |
6689bed3 | 1302 | rb_inc_page(&page); |
77ae365e SR |
1303 | } while (page != head); |
1304 | } | |
1305 | ||
1306 | RB_WARN_ON(cpu_buffer, 1); | |
1307 | ||
1308 | return NULL; | |
1309 | } | |
1310 | ||
bc92b956 | 1311 | static bool rb_head_page_replace(struct buffer_page *old, |
77ae365e SR |
1312 | struct buffer_page *new) |
1313 | { | |
1314 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
1315 | unsigned long val; | |
77ae365e SR |
1316 | |
1317 | val = *ptr & ~RB_FLAG_MASK; | |
1318 | val |= RB_PAGE_HEAD; | |
1319 | ||
00a8478f | 1320 | return try_cmpxchg(ptr, &val, (unsigned long)&new->list); |
77ae365e SR |
1321 | } |
1322 | ||
1323 | /* | |
1324 | * rb_tail_page_update - move the tail page forward | |
77ae365e | 1325 | */ |
70004986 | 1326 | static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, |
77ae365e SR |
1327 | struct buffer_page *tail_page, |
1328 | struct buffer_page *next_page) | |
1329 | { | |
77ae365e SR |
1330 | unsigned long old_entries; |
1331 | unsigned long old_write; | |
77ae365e SR |
1332 | |
1333 | /* | |
1334 | * The tail page now needs to be moved forward. | |
1335 | * | |
1336 | * We need to reset the tail page, but without messing | |
1337 | * with possible erasing of data brought in by interrupts | |
1338 | * that have moved the tail page and are currently on it. | |
1339 | * | |
1340 | * We add a counter to the write field to denote this. | |
1341 | */ | |
1342 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
1343 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
1344 | ||
2c2b0a78 | 1345 | local_inc(&cpu_buffer->pages_touched); |
77ae365e SR |
1346 | /* |
1347 | * Just make sure we have seen our old_write and synchronize | |
1348 | * with any interrupts that come in. | |
1349 | */ | |
1350 | barrier(); | |
1351 | ||
1352 | /* | |
1353 | * If the tail page is still the same as what we think | |
1354 | * it is, then it is up to us to update the tail | |
1355 | * pointer. | |
1356 | */ | |
8573636e | 1357 | if (tail_page == READ_ONCE(cpu_buffer->tail_page)) { |
77ae365e SR |
1358 | /* Zero the write counter */ |
1359 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
1360 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
1361 | ||
1362 | /* | |
1363 | * This will only succeed if an interrupt did | |
1364 | * not come in and change it. In which case, we | |
1365 | * do not want to modify it. | |
da706d8b LJ |
1366 | * |
1367 | * We add (void) to let the compiler know that we do not care | |
1368 | * about the return value of these functions. We use the | |
1369 | * cmpxchg to only update if an interrupt did not already | |
1370 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 1371 | */ |
da706d8b LJ |
1372 | (void)local_cmpxchg(&next_page->write, old_write, val); |
1373 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
1374 | |
1375 | /* | |
1376 | * No need to worry about races with clearing out the commit. | |
1377 | * it only can increment when a commit takes place. But that | |
1378 | * only happens in the outer most nested commit. | |
1379 | */ | |
1380 | local_set(&next_page->page->commit, 0); | |
1381 | ||
70004986 SRRH |
1382 | /* Again, either we update tail_page or an interrupt does */ |
1383 | (void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page); | |
77ae365e | 1384 | } |
77ae365e SR |
1385 | } |
1386 | ||
b4b55dfd | 1387 | static void rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, |
77ae365e SR |
1388 | struct buffer_page *bpage) |
1389 | { | |
1390 | unsigned long val = (unsigned long)bpage; | |
1391 | ||
b4b55dfd | 1392 | RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK); |
77ae365e SR |
1393 | } |
1394 | ||
7a8e76a3 | 1395 | /** |
d611851b | 1396 | * rb_check_pages - integrity check of buffer pages |
7a8e76a3 SR |
1397 | * @cpu_buffer: CPU buffer with pages to test |
1398 | * | |
c3706f00 | 1399 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
1400 | * been corrupted. |
1401 | */ | |
b4b55dfd | 1402 | static void rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1403 | { |
8843e06f MO |
1404 | struct list_head *head = rb_list_head(cpu_buffer->pages); |
1405 | struct list_head *tmp; | |
308f7eeb | 1406 | |
8843e06f MO |
1407 | if (RB_WARN_ON(cpu_buffer, |
1408 | rb_list_head(rb_list_head(head->next)->prev) != head)) | |
b4b55dfd | 1409 | return; |
7a8e76a3 | 1410 | |
8843e06f MO |
1411 | if (RB_WARN_ON(cpu_buffer, |
1412 | rb_list_head(rb_list_head(head->prev)->next) != head)) | |
b4b55dfd | 1413 | return; |
77ae365e | 1414 | |
8843e06f | 1415 | for (tmp = rb_list_head(head->next); tmp != head; tmp = rb_list_head(tmp->next)) { |
3e89c7bb | 1416 | if (RB_WARN_ON(cpu_buffer, |
8843e06f | 1417 | rb_list_head(rb_list_head(tmp->next)->prev) != tmp)) |
b4b55dfd | 1418 | return; |
8843e06f | 1419 | |
3e89c7bb | 1420 | if (RB_WARN_ON(cpu_buffer, |
8843e06f | 1421 | rb_list_head(rb_list_head(tmp->prev)->next) != tmp)) |
b4b55dfd | 1422 | return; |
7a8e76a3 | 1423 | } |
7a8e76a3 SR |
1424 | } |
1425 | ||
74e2afc6 QH |
1426 | static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
1427 | long nr_pages, struct list_head *pages) | |
7a8e76a3 | 1428 | { |
044fa782 | 1429 | struct buffer_page *bpage, *tmp; |
927e56db SRV |
1430 | bool user_thread = current->mm != NULL; |
1431 | gfp_t mflags; | |
9b94a8fb | 1432 | long i; |
3adc54fa | 1433 | |
927e56db SRV |
1434 | /* |
1435 | * Check if the available memory is there first. | |
1436 | * Note, si_mem_available() only gives us a rough estimate of available | |
1437 | * memory. It may not be accurate. But we don't care, we just want | |
1438 | * to prevent doing any allocation when it is obvious that it is | |
1439 | * not going to succeed. | |
1440 | */ | |
2a872fa4 SRV |
1441 | i = si_mem_available(); |
1442 | if (i < nr_pages) | |
1443 | return -ENOMEM; | |
1444 | ||
927e56db SRV |
1445 | /* |
1446 | * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails | |
1447 | * gracefully without invoking oom-killer and the system is not | |
1448 | * destabilized. | |
1449 | */ | |
1450 | mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL; | |
1451 | ||
1452 | /* | |
1453 | * If a user thread allocates too much, and si_mem_available() | |
1454 | * reports there's enough memory, even though there is not. | |
1455 | * Make sure the OOM killer kills this thread. This can happen | |
1456 | * even with RETRY_MAYFAIL because another task may be doing | |
1457 | * an allocation after this task has taken all memory. | |
1458 | * This is the task the OOM killer needs to take out during this | |
1459 | * loop, even if it was triggered by an allocation somewhere else. | |
1460 | */ | |
1461 | if (user_thread) | |
1462 | set_current_oom_origin(); | |
7a8e76a3 | 1463 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 1464 | struct page *page; |
927e56db | 1465 | |
044fa782 | 1466 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
74e2afc6 | 1467 | mflags, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 1468 | if (!bpage) |
e4c2ce82 | 1469 | goto free_pages; |
77ae365e | 1470 | |
74e2afc6 QH |
1471 | rb_check_bpage(cpu_buffer, bpage); |
1472 | ||
438ced17 | 1473 | list_add(&bpage->list, pages); |
77ae365e | 1474 | |
f9b94daa TSV |
1475 | page = alloc_pages_node(cpu_to_node(cpu_buffer->cpu), mflags, |
1476 | cpu_buffer->buffer->subbuf_order); | |
7ea59064 | 1477 | if (!page) |
7a8e76a3 | 1478 | goto free_pages; |
7ea59064 | 1479 | bpage->page = page_address(page); |
f9b94daa | 1480 | bpage->order = cpu_buffer->buffer->subbuf_order; |
044fa782 | 1481 | rb_init_page(bpage->page); |
927e56db SRV |
1482 | |
1483 | if (user_thread && fatal_signal_pending(current)) | |
1484 | goto free_pages; | |
7a8e76a3 | 1485 | } |
927e56db SRV |
1486 | if (user_thread) |
1487 | clear_current_oom_origin(); | |
7a8e76a3 | 1488 | |
438ced17 VN |
1489 | return 0; |
1490 | ||
1491 | free_pages: | |
1492 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1493 | list_del_init(&bpage->list); | |
1494 | free_buffer_page(bpage); | |
1495 | } | |
927e56db SRV |
1496 | if (user_thread) |
1497 | clear_current_oom_origin(); | |
438ced17 VN |
1498 | |
1499 | return -ENOMEM; | |
1500 | } | |
1501 | ||
1502 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
9b94a8fb | 1503 | unsigned long nr_pages) |
438ced17 VN |
1504 | { |
1505 | LIST_HEAD(pages); | |
1506 | ||
1507 | WARN_ON(!nr_pages); | |
1508 | ||
74e2afc6 | 1509 | if (__rb_allocate_pages(cpu_buffer, nr_pages, &pages)) |
438ced17 VN |
1510 | return -ENOMEM; |
1511 | ||
3adc54fa SR |
1512 | /* |
1513 | * The ring buffer page list is a circular list that does not | |
1514 | * start and end with a list head. All page list items point to | |
1515 | * other pages. | |
1516 | */ | |
1517 | cpu_buffer->pages = pages.next; | |
1518 | list_del(&pages); | |
7a8e76a3 | 1519 | |
438ced17 VN |
1520 | cpu_buffer->nr_pages = nr_pages; |
1521 | ||
7a8e76a3 SR |
1522 | rb_check_pages(cpu_buffer); |
1523 | ||
1524 | return 0; | |
7a8e76a3 SR |
1525 | } |
1526 | ||
1527 | static struct ring_buffer_per_cpu * | |
13292494 | 1528 | rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu) |
7a8e76a3 SR |
1529 | { |
1530 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1531 | struct buffer_page *bpage; |
7ea59064 | 1532 | struct page *page; |
7a8e76a3 SR |
1533 | int ret; |
1534 | ||
1535 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1536 | GFP_KERNEL, cpu_to_node(cpu)); | |
1537 | if (!cpu_buffer) | |
1538 | return NULL; | |
1539 | ||
1540 | cpu_buffer->cpu = cpu; | |
1541 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1542 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1543 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1544 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1545 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1546 | init_completion(&cpu_buffer->update_done); |
15693458 | 1547 | init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1548 | init_waitqueue_head(&cpu_buffer->irq_work.waiters); |
1e0d6714 | 1549 | init_waitqueue_head(&cpu_buffer->irq_work.full_waiters); |
7a8e76a3 | 1550 | |
044fa782 | 1551 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1552 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1553 | if (!bpage) |
e4c2ce82 SR |
1554 | goto fail_free_buffer; |
1555 | ||
77ae365e SR |
1556 | rb_check_bpage(cpu_buffer, bpage); |
1557 | ||
044fa782 | 1558 | cpu_buffer->reader_page = bpage; |
f9b94daa TSV |
1559 | |
1560 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, cpu_buffer->buffer->subbuf_order); | |
7ea59064 | 1561 | if (!page) |
e4c2ce82 | 1562 | goto fail_free_reader; |
7ea59064 | 1563 | bpage->page = page_address(page); |
044fa782 | 1564 | rb_init_page(bpage->page); |
e4c2ce82 | 1565 | |
d769041f | 1566 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1567 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1568 | |
438ced17 | 1569 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1570 | if (ret < 0) |
d769041f | 1571 | goto fail_free_reader; |
7a8e76a3 SR |
1572 | |
1573 | cpu_buffer->head_page | |
3adc54fa | 1574 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1575 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1576 | |
77ae365e SR |
1577 | rb_head_page_activate(cpu_buffer); |
1578 | ||
7a8e76a3 SR |
1579 | return cpu_buffer; |
1580 | ||
d769041f SR |
1581 | fail_free_reader: |
1582 | free_buffer_page(cpu_buffer->reader_page); | |
1583 | ||
7a8e76a3 SR |
1584 | fail_free_buffer: |
1585 | kfree(cpu_buffer); | |
1586 | return NULL; | |
1587 | } | |
1588 | ||
1589 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1590 | { | |
3adc54fa | 1591 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1592 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1593 | |
675751bb JB |
1594 | irq_work_sync(&cpu_buffer->irq_work.work); |
1595 | ||
d769041f SR |
1596 | free_buffer_page(cpu_buffer->reader_page); |
1597 | ||
3adc54fa | 1598 | if (head) { |
56f4ca0a DT |
1599 | rb_head_page_deactivate(cpu_buffer); |
1600 | ||
3adc54fa SR |
1601 | list_for_each_entry_safe(bpage, tmp, head, list) { |
1602 | list_del_init(&bpage->list); | |
1603 | free_buffer_page(bpage); | |
1604 | } | |
1605 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1606 | free_buffer_page(bpage); |
7a8e76a3 | 1607 | } |
3adc54fa | 1608 | |
17d80175 SRG |
1609 | free_page((unsigned long)cpu_buffer->free_page); |
1610 | ||
7a8e76a3 SR |
1611 | kfree(cpu_buffer); |
1612 | } | |
1613 | ||
1614 | /** | |
d611851b | 1615 | * __ring_buffer_alloc - allocate a new ring_buffer |
68814b58 | 1616 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 | 1617 | * @flags: attributes to set for the ring buffer. |
59e7cffe | 1618 | * @key: ring buffer reader_lock_key. |
7a8e76a3 SR |
1619 | * |
1620 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1621 | * flag. This flag means that the buffer will overwrite old data | |
1622 | * when the buffer wraps. If this flag is not set, the buffer will | |
1623 | * drop data when the tail hits the head. | |
1624 | */ | |
13292494 | 1625 | struct trace_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1f8a6a10 | 1626 | struct lock_class_key *key) |
7a8e76a3 | 1627 | { |
13292494 | 1628 | struct trace_buffer *buffer; |
9b94a8fb | 1629 | long nr_pages; |
7a8e76a3 | 1630 | int bsize; |
9b94a8fb | 1631 | int cpu; |
b32614c0 | 1632 | int ret; |
7a8e76a3 SR |
1633 | |
1634 | /* keep it in its own cache line */ | |
1635 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1636 | GFP_KERNEL); | |
1637 | if (!buffer) | |
1638 | return NULL; | |
1639 | ||
b18cc3de | 1640 | if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
9e01c1b7 RR |
1641 | goto fail_free_buffer; |
1642 | ||
139f8400 | 1643 | /* Default buffer page size - one system page */ |
f9b94daa | 1644 | buffer->subbuf_order = 0; |
139f8400 TSV |
1645 | buffer->subbuf_size = PAGE_SIZE - BUF_PAGE_HDR_SIZE; |
1646 | ||
1647 | /* Max payload is buffer page size - header (8bytes) */ | |
1648 | buffer->max_data_size = buffer->subbuf_size - (sizeof(u32) * 2); | |
1649 | ||
1650 | nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size); | |
7a8e76a3 | 1651 | buffer->flags = flags; |
37886f6a | 1652 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1653 | buffer->reader_lock_key = key; |
7a8e76a3 | 1654 | |
15693458 | 1655 | init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1656 | init_waitqueue_head(&buffer->irq_work.waiters); |
15693458 | 1657 | |
7a8e76a3 | 1658 | /* need at least two pages */ |
438ced17 VN |
1659 | if (nr_pages < 2) |
1660 | nr_pages = 2; | |
7a8e76a3 | 1661 | |
7a8e76a3 SR |
1662 | buffer->cpus = nr_cpu_ids; |
1663 | ||
1664 | bsize = sizeof(void *) * nr_cpu_ids; | |
1665 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1666 | GFP_KERNEL); | |
1667 | if (!buffer->buffers) | |
9e01c1b7 | 1668 | goto fail_free_cpumask; |
7a8e76a3 | 1669 | |
b32614c0 SAS |
1670 | cpu = raw_smp_processor_id(); |
1671 | cpumask_set_cpu(cpu, buffer->cpumask); | |
1672 | buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu); | |
1673 | if (!buffer->buffers[cpu]) | |
1674 | goto fail_free_buffers; | |
7a8e76a3 | 1675 | |
b32614c0 SAS |
1676 | ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node); |
1677 | if (ret < 0) | |
1678 | goto fail_free_buffers; | |
554f786e | 1679 | |
7a8e76a3 SR |
1680 | mutex_init(&buffer->mutex); |
1681 | ||
1682 | return buffer; | |
1683 | ||
1684 | fail_free_buffers: | |
1685 | for_each_buffer_cpu(buffer, cpu) { | |
1686 | if (buffer->buffers[cpu]) | |
1687 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1688 | } | |
1689 | kfree(buffer->buffers); | |
1690 | ||
9e01c1b7 RR |
1691 | fail_free_cpumask: |
1692 | free_cpumask_var(buffer->cpumask); | |
1693 | ||
7a8e76a3 SR |
1694 | fail_free_buffer: |
1695 | kfree(buffer); | |
1696 | return NULL; | |
1697 | } | |
1f8a6a10 | 1698 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1699 | |
1700 | /** | |
1701 | * ring_buffer_free - free a ring buffer. | |
1702 | * @buffer: the buffer to free. | |
1703 | */ | |
1704 | void | |
13292494 | 1705 | ring_buffer_free(struct trace_buffer *buffer) |
7a8e76a3 SR |
1706 | { |
1707 | int cpu; | |
1708 | ||
b32614c0 | 1709 | cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node); |
554f786e | 1710 | |
675751bb JB |
1711 | irq_work_sync(&buffer->irq_work.work); |
1712 | ||
7a8e76a3 SR |
1713 | for_each_buffer_cpu(buffer, cpu) |
1714 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1715 | ||
bd3f0221 | 1716 | kfree(buffer->buffers); |
9e01c1b7 RR |
1717 | free_cpumask_var(buffer->cpumask); |
1718 | ||
7a8e76a3 SR |
1719 | kfree(buffer); |
1720 | } | |
c4f50183 | 1721 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1722 | |
13292494 | 1723 | void ring_buffer_set_clock(struct trace_buffer *buffer, |
37886f6a SR |
1724 | u64 (*clock)(void)) |
1725 | { | |
1726 | buffer->clock = clock; | |
1727 | } | |
1728 | ||
13292494 | 1729 | void ring_buffer_set_time_stamp_abs(struct trace_buffer *buffer, bool abs) |
00b41452 TZ |
1730 | { |
1731 | buffer->time_stamp_abs = abs; | |
1732 | } | |
1733 | ||
13292494 | 1734 | bool ring_buffer_time_stamp_abs(struct trace_buffer *buffer) |
00b41452 TZ |
1735 | { |
1736 | return buffer->time_stamp_abs; | |
1737 | } | |
1738 | ||
7a8e76a3 SR |
1739 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1740 | ||
83f40318 VN |
1741 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1742 | { | |
1743 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1744 | } | |
1745 | ||
1746 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1747 | { | |
1748 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1749 | } | |
1750 | ||
bc92b956 | 1751 | static bool |
9b94a8fb | 1752 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages) |
7a8e76a3 | 1753 | { |
83f40318 VN |
1754 | struct list_head *tail_page, *to_remove, *next_page; |
1755 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1756 | struct buffer_page *last_page, *first_page; | |
9b94a8fb | 1757 | unsigned long nr_removed; |
83f40318 VN |
1758 | unsigned long head_bit; |
1759 | int page_entries; | |
1760 | ||
1761 | head_bit = 0; | |
7a8e76a3 | 1762 | |
5389f6fa | 1763 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1764 | atomic_inc(&cpu_buffer->record_disabled); |
1765 | /* | |
1766 | * We don't race with the readers since we have acquired the reader | |
1767 | * lock. We also don't race with writers after disabling recording. | |
1768 | * This makes it easy to figure out the first and the last page to be | |
1769 | * removed from the list. We unlink all the pages in between including | |
1770 | * the first and last pages. This is done in a busy loop so that we | |
1771 | * lose the least number of traces. | |
1772 | * The pages are freed after we restart recording and unlock readers. | |
1773 | */ | |
1774 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1775 | |
83f40318 VN |
1776 | /* |
1777 | * tail page might be on reader page, we remove the next page | |
1778 | * from the ring buffer | |
1779 | */ | |
1780 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1781 | tail_page = rb_list_head(tail_page->next); | |
1782 | to_remove = tail_page; | |
1783 | ||
1784 | /* start of pages to remove */ | |
1785 | first_page = list_entry(rb_list_head(to_remove->next), | |
1786 | struct buffer_page, list); | |
1787 | ||
1788 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1789 | to_remove = rb_list_head(to_remove)->next; | |
1790 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1791 | } |
2d093282 ZY |
1792 | /* Read iterators need to reset themselves when some pages removed */ |
1793 | cpu_buffer->pages_removed += nr_removed; | |
7a8e76a3 | 1794 | |
83f40318 | 1795 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1796 | |
83f40318 VN |
1797 | /* |
1798 | * Now we remove all pages between tail_page and next_page. | |
1799 | * Make sure that we have head_bit value preserved for the | |
1800 | * next page | |
1801 | */ | |
1802 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1803 | head_bit); | |
1804 | next_page = rb_list_head(next_page); | |
1805 | next_page->prev = tail_page; | |
1806 | ||
1807 | /* make sure pages points to a valid page in the ring buffer */ | |
1808 | cpu_buffer->pages = next_page; | |
1809 | ||
1810 | /* update head page */ | |
1811 | if (head_bit) | |
1812 | cpu_buffer->head_page = list_entry(next_page, | |
1813 | struct buffer_page, list); | |
1814 | ||
83f40318 VN |
1815 | /* pages are removed, resume tracing and then free the pages */ |
1816 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1817 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1818 | |
1819 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1820 | ||
1821 | /* last buffer page to remove */ | |
1822 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1823 | list); | |
1824 | tmp_iter_page = first_page; | |
1825 | ||
1826 | do { | |
83f36555 VN |
1827 | cond_resched(); |
1828 | ||
83f40318 | 1829 | to_remove_page = tmp_iter_page; |
6689bed3 | 1830 | rb_inc_page(&tmp_iter_page); |
83f40318 VN |
1831 | |
1832 | /* update the counters */ | |
1833 | page_entries = rb_page_entries(to_remove_page); | |
1834 | if (page_entries) { | |
1835 | /* | |
1836 | * If something was added to this page, it was full | |
1837 | * since it is not the tail page. So we deduct the | |
1838 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1839 | * Increment overrun to account for the lost events. |
83f40318 | 1840 | */ |
48fdc72f | 1841 | local_add(page_entries, &cpu_buffer->overrun); |
45d99ea4 | 1842 | local_sub(rb_page_commit(to_remove_page), &cpu_buffer->entries_bytes); |
31029a8b | 1843 | local_inc(&cpu_buffer->pages_lost); |
83f40318 VN |
1844 | } |
1845 | ||
1846 | /* | |
1847 | * We have already removed references to this list item, just | |
1848 | * free up the buffer_page and its page | |
1849 | */ | |
1850 | free_buffer_page(to_remove_page); | |
1851 | nr_removed--; | |
1852 | ||
1853 | } while (to_remove_page != last_page); | |
1854 | ||
1855 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1856 | |
1857 | return nr_removed == 0; | |
7a8e76a3 SR |
1858 | } |
1859 | ||
bc92b956 | 1860 | static bool |
5040b4b7 | 1861 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1862 | { |
5040b4b7 | 1863 | struct list_head *pages = &cpu_buffer->new_pages; |
88ca6a71 | 1864 | unsigned long flags; |
bc92b956 UB |
1865 | bool success; |
1866 | int retries; | |
7a8e76a3 | 1867 | |
88ca6a71 SR |
1868 | /* Can be called at early boot up, where interrupts must not been enabled */ |
1869 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
5040b4b7 VN |
1870 | /* |
1871 | * We are holding the reader lock, so the reader page won't be swapped | |
1872 | * in the ring buffer. Now we are racing with the writer trying to | |
1873 | * move head page and the tail page. | |
1874 | * We are going to adapt the reader page update process where: | |
1875 | * 1. We first splice the start and end of list of new pages between | |
1876 | * the head page and its previous page. | |
1877 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1878 | * start of new pages list. | |
1879 | * 3. Finally, we update the head->prev to the end of new list. | |
1880 | * | |
1881 | * We will try this process 10 times, to make sure that we don't keep | |
1882 | * spinning. | |
1883 | */ | |
1884 | retries = 10; | |
bc92b956 | 1885 | success = false; |
5040b4b7 | 1886 | while (retries--) { |
bdf4fb62 | 1887 | struct list_head *head_page, *prev_page; |
5040b4b7 VN |
1888 | struct list_head *last_page, *first_page; |
1889 | struct list_head *head_page_with_bit; | |
625ed527 | 1890 | struct buffer_page *hpage = rb_set_head_page(cpu_buffer); |
77ae365e | 1891 | |
625ed527 | 1892 | if (!hpage) |
54f7be5b | 1893 | break; |
625ed527 | 1894 | head_page = &hpage->list; |
5040b4b7 VN |
1895 | prev_page = head_page->prev; |
1896 | ||
1897 | first_page = pages->next; | |
1898 | last_page = pages->prev; | |
1899 | ||
1900 | head_page_with_bit = (struct list_head *) | |
1901 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1902 | ||
1903 | last_page->next = head_page_with_bit; | |
1904 | first_page->prev = prev_page; | |
1905 | ||
bdf4fb62 UB |
1906 | /* caution: head_page_with_bit gets updated on cmpxchg failure */ |
1907 | if (try_cmpxchg(&prev_page->next, | |
1908 | &head_page_with_bit, first_page)) { | |
5040b4b7 VN |
1909 | /* |
1910 | * yay, we replaced the page pointer to our new list, | |
1911 | * now, we just have to update to head page's prev | |
1912 | * pointer to point to end of list | |
1913 | */ | |
1914 | head_page->prev = last_page; | |
bc92b956 | 1915 | success = true; |
5040b4b7 VN |
1916 | break; |
1917 | } | |
7a8e76a3 | 1918 | } |
7a8e76a3 | 1919 | |
5040b4b7 VN |
1920 | if (success) |
1921 | INIT_LIST_HEAD(pages); | |
1922 | /* | |
1923 | * If we weren't successful in adding in new pages, warn and stop | |
1924 | * tracing | |
1925 | */ | |
1926 | RB_WARN_ON(cpu_buffer, !success); | |
88ca6a71 | 1927 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
5040b4b7 VN |
1928 | |
1929 | /* free pages if they weren't inserted */ | |
1930 | if (!success) { | |
1931 | struct buffer_page *bpage, *tmp; | |
1932 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1933 | list) { | |
1934 | list_del_init(&bpage->list); | |
1935 | free_buffer_page(bpage); | |
1936 | } | |
1937 | } | |
1938 | return success; | |
7a8e76a3 SR |
1939 | } |
1940 | ||
83f40318 | 1941 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1942 | { |
bc92b956 | 1943 | bool success; |
5040b4b7 | 1944 | |
438ced17 | 1945 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1946 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1947 | else |
5040b4b7 VN |
1948 | success = rb_remove_pages(cpu_buffer, |
1949 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1950 | |
5040b4b7 VN |
1951 | if (success) |
1952 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1953 | } |
1954 | ||
1955 | static void update_pages_handler(struct work_struct *work) | |
1956 | { | |
1957 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1958 | struct ring_buffer_per_cpu, update_pages_work); | |
1959 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1960 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1961 | } |
1962 | ||
7a8e76a3 SR |
1963 | /** |
1964 | * ring_buffer_resize - resize the ring buffer | |
1965 | * @buffer: the buffer to resize. | |
1966 | * @size: the new size. | |
d611851b | 1967 | * @cpu_id: the cpu buffer to resize |
7a8e76a3 | 1968 | * |
139f8400 | 1969 | * Minimum size is 2 * buffer->subbuf_size. |
7a8e76a3 | 1970 | * |
83f40318 | 1971 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1972 | */ |
13292494 | 1973 | int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size, |
438ced17 | 1974 | int cpu_id) |
7a8e76a3 SR |
1975 | { |
1976 | struct ring_buffer_per_cpu *cpu_buffer; | |
9b94a8fb | 1977 | unsigned long nr_pages; |
0a1754b2 | 1978 | int cpu, err; |
7a8e76a3 | 1979 | |
ee51a1de IM |
1980 | /* |
1981 | * Always succeed at resizing a non-existent buffer: | |
1982 | */ | |
1983 | if (!buffer) | |
0a1754b2 | 1984 | return 0; |
ee51a1de | 1985 | |
6a31e1f1 SR |
1986 | /* Make sure the requested buffer exists */ |
1987 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1988 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
0a1754b2 | 1989 | return 0; |
6a31e1f1 | 1990 | |
139f8400 | 1991 | nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size); |
7a8e76a3 SR |
1992 | |
1993 | /* we need a minimum of two pages */ | |
59643d15 SRRH |
1994 | if (nr_pages < 2) |
1995 | nr_pages = 2; | |
7a8e76a3 | 1996 | |
83f40318 | 1997 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1998 | mutex_lock(&buffer->mutex); |
8a96c028 | 1999 | atomic_inc(&buffer->resizing); |
07b8b10e | 2000 | |
438ced17 | 2001 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
07b8b10e SRV |
2002 | /* |
2003 | * Don't succeed if resizing is disabled, as a reader might be | |
2004 | * manipulating the ring buffer and is expecting a sane state while | |
2005 | * this is true. | |
2006 | */ | |
2007 | for_each_buffer_cpu(buffer, cpu) { | |
2008 | cpu_buffer = buffer->buffers[cpu]; | |
2009 | if (atomic_read(&cpu_buffer->resize_disabled)) { | |
2010 | err = -EBUSY; | |
2011 | goto out_err_unlock; | |
2012 | } | |
2013 | } | |
2014 | ||
438ced17 | 2015 | /* calculate the pages to update */ |
7a8e76a3 SR |
2016 | for_each_buffer_cpu(buffer, cpu) { |
2017 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 2018 | |
438ced17 VN |
2019 | cpu_buffer->nr_pages_to_update = nr_pages - |
2020 | cpu_buffer->nr_pages; | |
438ced17 VN |
2021 | /* |
2022 | * nothing more to do for removing pages or no update | |
2023 | */ | |
2024 | if (cpu_buffer->nr_pages_to_update <= 0) | |
2025 | continue; | |
d7ec4bfe | 2026 | /* |
438ced17 VN |
2027 | * to add pages, make sure all new pages can be |
2028 | * allocated without receiving ENOMEM | |
d7ec4bfe | 2029 | */ |
438ced17 | 2030 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
74e2afc6 QH |
2031 | if (__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update, |
2032 | &cpu_buffer->new_pages)) { | |
438ced17 | 2033 | /* not enough memory for new pages */ |
83f40318 VN |
2034 | err = -ENOMEM; |
2035 | goto out_err; | |
2036 | } | |
f6bd2c92 ZY |
2037 | |
2038 | cond_resched(); | |
83f40318 VN |
2039 | } |
2040 | ||
99c37d1a | 2041 | cpus_read_lock(); |
83f40318 VN |
2042 | /* |
2043 | * Fire off all the required work handlers | |
05fdd70d | 2044 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
2045 | * since we can change their buffer sizes without any race. |
2046 | */ | |
2047 | for_each_buffer_cpu(buffer, cpu) { | |
2048 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 2049 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
2050 | continue; |
2051 | ||
021c5b34 CM |
2052 | /* Can't run something on an offline CPU. */ |
2053 | if (!cpu_online(cpu)) { | |
f5eb5588 SRRH |
2054 | rb_update_pages(cpu_buffer); |
2055 | cpu_buffer->nr_pages_to_update = 0; | |
2056 | } else { | |
88ca6a71 SR |
2057 | /* Run directly if possible. */ |
2058 | migrate_disable(); | |
2059 | if (cpu != smp_processor_id()) { | |
2060 | migrate_enable(); | |
2061 | schedule_work_on(cpu, | |
2062 | &cpu_buffer->update_pages_work); | |
2063 | } else { | |
2064 | update_pages_handler(&cpu_buffer->update_pages_work); | |
2065 | migrate_enable(); | |
2066 | } | |
f5eb5588 | 2067 | } |
7a8e76a3 | 2068 | } |
7a8e76a3 | 2069 | |
438ced17 VN |
2070 | /* wait for all the updates to complete */ |
2071 | for_each_buffer_cpu(buffer, cpu) { | |
2072 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 2073 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
2074 | continue; |
2075 | ||
05fdd70d VN |
2076 | if (cpu_online(cpu)) |
2077 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 2078 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 2079 | } |
83f40318 | 2080 | |
99c37d1a | 2081 | cpus_read_unlock(); |
438ced17 VN |
2082 | } else { |
2083 | cpu_buffer = buffer->buffers[cpu_id]; | |
83f40318 | 2084 | |
438ced17 VN |
2085 | if (nr_pages == cpu_buffer->nr_pages) |
2086 | goto out; | |
7a8e76a3 | 2087 | |
07b8b10e SRV |
2088 | /* |
2089 | * Don't succeed if resizing is disabled, as a reader might be | |
2090 | * manipulating the ring buffer and is expecting a sane state while | |
2091 | * this is true. | |
2092 | */ | |
2093 | if (atomic_read(&cpu_buffer->resize_disabled)) { | |
2094 | err = -EBUSY; | |
2095 | goto out_err_unlock; | |
2096 | } | |
2097 | ||
438ced17 VN |
2098 | cpu_buffer->nr_pages_to_update = nr_pages - |
2099 | cpu_buffer->nr_pages; | |
2100 | ||
2101 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
2102 | if (cpu_buffer->nr_pages_to_update > 0 && | |
74e2afc6 QH |
2103 | __rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update, |
2104 | &cpu_buffer->new_pages)) { | |
83f40318 VN |
2105 | err = -ENOMEM; |
2106 | goto out_err; | |
2107 | } | |
438ced17 | 2108 | |
99c37d1a | 2109 | cpus_read_lock(); |
83f40318 | 2110 | |
021c5b34 CM |
2111 | /* Can't run something on an offline CPU. */ |
2112 | if (!cpu_online(cpu_id)) | |
f5eb5588 SRRH |
2113 | rb_update_pages(cpu_buffer); |
2114 | else { | |
88ca6a71 SR |
2115 | /* Run directly if possible. */ |
2116 | migrate_disable(); | |
2117 | if (cpu_id == smp_processor_id()) { | |
2118 | rb_update_pages(cpu_buffer); | |
2119 | migrate_enable(); | |
2120 | } else { | |
2121 | migrate_enable(); | |
2122 | schedule_work_on(cpu_id, | |
2123 | &cpu_buffer->update_pages_work); | |
2124 | wait_for_completion(&cpu_buffer->update_done); | |
2125 | } | |
f5eb5588 | 2126 | } |
83f40318 | 2127 | |
83f40318 | 2128 | cpu_buffer->nr_pages_to_update = 0; |
99c37d1a | 2129 | cpus_read_unlock(); |
438ced17 | 2130 | } |
7a8e76a3 SR |
2131 | |
2132 | out: | |
659f451f SR |
2133 | /* |
2134 | * The ring buffer resize can happen with the ring buffer | |
2135 | * enabled, so that the update disturbs the tracing as little | |
2136 | * as possible. But if the buffer is disabled, we do not need | |
2137 | * to worry about that, and we can take the time to verify | |
2138 | * that the buffer is not corrupt. | |
2139 | */ | |
2140 | if (atomic_read(&buffer->record_disabled)) { | |
2141 | atomic_inc(&buffer->record_disabled); | |
2142 | /* | |
2143 | * Even though the buffer was disabled, we must make sure | |
2144 | * that it is truly disabled before calling rb_check_pages. | |
2145 | * There could have been a race between checking | |
2146 | * record_disable and incrementing it. | |
2147 | */ | |
74401729 | 2148 | synchronize_rcu(); |
659f451f SR |
2149 | for_each_buffer_cpu(buffer, cpu) { |
2150 | cpu_buffer = buffer->buffers[cpu]; | |
2151 | rb_check_pages(cpu_buffer); | |
2152 | } | |
2153 | atomic_dec(&buffer->record_disabled); | |
2154 | } | |
2155 | ||
8a96c028 | 2156 | atomic_dec(&buffer->resizing); |
7a8e76a3 | 2157 | mutex_unlock(&buffer->mutex); |
0a1754b2 | 2158 | return 0; |
7a8e76a3 | 2159 | |
83f40318 | 2160 | out_err: |
438ced17 VN |
2161 | for_each_buffer_cpu(buffer, cpu) { |
2162 | struct buffer_page *bpage, *tmp; | |
83f40318 | 2163 | |
438ced17 | 2164 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 2165 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 2166 | |
438ced17 VN |
2167 | if (list_empty(&cpu_buffer->new_pages)) |
2168 | continue; | |
83f40318 | 2169 | |
438ced17 VN |
2170 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
2171 | list) { | |
2172 | list_del_init(&bpage->list); | |
2173 | free_buffer_page(bpage); | |
2174 | } | |
7a8e76a3 | 2175 | } |
07b8b10e | 2176 | out_err_unlock: |
8a96c028 | 2177 | atomic_dec(&buffer->resizing); |
641d2f63 | 2178 | mutex_unlock(&buffer->mutex); |
83f40318 | 2179 | return err; |
7a8e76a3 | 2180 | } |
c4f50183 | 2181 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 2182 | |
13292494 | 2183 | void ring_buffer_change_overwrite(struct trace_buffer *buffer, int val) |
750912fa DS |
2184 | { |
2185 | mutex_lock(&buffer->mutex); | |
2186 | if (val) | |
2187 | buffer->flags |= RB_FL_OVERWRITE; | |
2188 | else | |
2189 | buffer->flags &= ~RB_FL_OVERWRITE; | |
2190 | mutex_unlock(&buffer->mutex); | |
2191 | } | |
2192 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
2193 | ||
2289d567 | 2194 | static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 2195 | { |
044fa782 | 2196 | return bpage->page->data + index; |
7a8e76a3 SR |
2197 | } |
2198 | ||
2289d567 | 2199 | static __always_inline struct ring_buffer_event * |
d769041f | 2200 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 2201 | { |
6f807acd SR |
2202 | return __rb_page_index(cpu_buffer->reader_page, |
2203 | cpu_buffer->reader_page->read); | |
2204 | } | |
2205 | ||
785888c5 SRV |
2206 | static struct ring_buffer_event * |
2207 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
bf41a158 | 2208 | { |
785888c5 SRV |
2209 | struct ring_buffer_event *event; |
2210 | struct buffer_page *iter_head_page = iter->head_page; | |
2211 | unsigned long commit; | |
2212 | unsigned length; | |
2213 | ||
153368ce SRV |
2214 | if (iter->head != iter->next_event) |
2215 | return iter->event; | |
2216 | ||
785888c5 SRV |
2217 | /* |
2218 | * When the writer goes across pages, it issues a cmpxchg which | |
2219 | * is a mb(), which will synchronize with the rmb here. | |
2220 | * (see rb_tail_page_update() and __rb_reserve_next()) | |
2221 | */ | |
2222 | commit = rb_page_commit(iter_head_page); | |
2223 | smp_rmb(); | |
95a404bd SRG |
2224 | |
2225 | /* An event needs to be at least 8 bytes in size */ | |
2226 | if (iter->head > commit - 8) | |
2227 | goto reset; | |
2228 | ||
785888c5 SRV |
2229 | event = __rb_page_index(iter_head_page, iter->head); |
2230 | length = rb_event_length(event); | |
2231 | ||
2232 | /* | |
2233 | * READ_ONCE() doesn't work on functions and we don't want the | |
2234 | * compiler doing any crazy optimizations with length. | |
2235 | */ | |
2236 | barrier(); | |
2237 | ||
139f8400 | 2238 | if ((iter->head + length) > commit || length > iter->event_size) |
785888c5 SRV |
2239 | /* Writer corrupted the read? */ |
2240 | goto reset; | |
2241 | ||
2242 | memcpy(iter->event, event, length); | |
2243 | /* | |
2244 | * If the page stamp is still the same after this rmb() then the | |
2245 | * event was safely copied without the writer entering the page. | |
2246 | */ | |
2247 | smp_rmb(); | |
2248 | ||
2249 | /* Make sure the page didn't change since we read this */ | |
2250 | if (iter->page_stamp != iter_head_page->page->time_stamp || | |
2251 | commit > rb_page_commit(iter_head_page)) | |
2252 | goto reset; | |
2253 | ||
2254 | iter->next_event = iter->head + length; | |
2255 | return iter->event; | |
2256 | reset: | |
2257 | /* Reset to the beginning */ | |
2258 | iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp; | |
2259 | iter->head = 0; | |
2260 | iter->next_event = 0; | |
c9b7a4a7 | 2261 | iter->missed_events = 1; |
785888c5 | 2262 | return NULL; |
bf41a158 SR |
2263 | } |
2264 | ||
25985edc | 2265 | /* Size is determined by what has been committed */ |
2289d567 | 2266 | static __always_inline unsigned rb_page_size(struct buffer_page *bpage) |
bf41a158 SR |
2267 | { |
2268 | return rb_page_commit(bpage); | |
2269 | } | |
2270 | ||
2289d567 | 2271 | static __always_inline unsigned |
bf41a158 SR |
2272 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) |
2273 | { | |
2274 | return rb_page_commit(cpu_buffer->commit_page); | |
2275 | } | |
2276 | ||
2289d567 | 2277 | static __always_inline unsigned |
3cb30911 | 2278 | rb_event_index(struct ring_buffer_per_cpu *cpu_buffer, struct ring_buffer_event *event) |
bf41a158 SR |
2279 | { |
2280 | unsigned long addr = (unsigned long)event; | |
2281 | ||
3cb30911 SRG |
2282 | addr &= (PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1; |
2283 | ||
2284 | return addr - BUF_PAGE_HDR_SIZE; | |
bf41a158 SR |
2285 | } |
2286 | ||
34a148bf | 2287 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
2288 | { |
2289 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2290 | ||
2291 | /* | |
2292 | * The iterator could be on the reader page (it starts there). | |
2293 | * But the head could have moved, since the reader was | |
2294 | * found. Check for this case and assign the iterator | |
2295 | * to the head page instead of next. | |
2296 | */ | |
2297 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 2298 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f | 2299 | else |
6689bed3 | 2300 | rb_inc_page(&iter->head_page); |
d769041f | 2301 | |
28e3fc56 | 2302 | iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 | 2303 | iter->head = 0; |
785888c5 | 2304 | iter->next_event = 0; |
7a8e76a3 SR |
2305 | } |
2306 | ||
77ae365e SR |
2307 | /* |
2308 | * rb_handle_head_page - writer hit the head page | |
2309 | * | |
2310 | * Returns: +1 to retry page | |
2311 | * 0 to continue | |
2312 | * -1 on error | |
2313 | */ | |
2314 | static int | |
2315 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
2316 | struct buffer_page *tail_page, | |
2317 | struct buffer_page *next_page) | |
2318 | { | |
2319 | struct buffer_page *new_head; | |
2320 | int entries; | |
2321 | int type; | |
2322 | int ret; | |
2323 | ||
2324 | entries = rb_page_entries(next_page); | |
2325 | ||
2326 | /* | |
2327 | * The hard part is here. We need to move the head | |
2328 | * forward, and protect against both readers on | |
2329 | * other CPUs and writers coming in via interrupts. | |
2330 | */ | |
2331 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
2332 | RB_PAGE_HEAD); | |
2333 | ||
2334 | /* | |
2335 | * type can be one of four: | |
2336 | * NORMAL - an interrupt already moved it for us | |
2337 | * HEAD - we are the first to get here. | |
2338 | * UPDATE - we are the interrupt interrupting | |
2339 | * a current move. | |
2340 | * MOVED - a reader on another CPU moved the next | |
2341 | * pointer to its reader page. Give up | |
2342 | * and try again. | |
2343 | */ | |
2344 | ||
2345 | switch (type) { | |
2346 | case RB_PAGE_HEAD: | |
2347 | /* | |
2348 | * We changed the head to UPDATE, thus | |
2349 | * it is our responsibility to update | |
2350 | * the counters. | |
2351 | */ | |
2352 | local_add(entries, &cpu_buffer->overrun); | |
45d99ea4 | 2353 | local_sub(rb_page_commit(next_page), &cpu_buffer->entries_bytes); |
31029a8b | 2354 | local_inc(&cpu_buffer->pages_lost); |
77ae365e SR |
2355 | |
2356 | /* | |
2357 | * The entries will be zeroed out when we move the | |
2358 | * tail page. | |
2359 | */ | |
2360 | ||
2361 | /* still more to do */ | |
2362 | break; | |
2363 | ||
2364 | case RB_PAGE_UPDATE: | |
2365 | /* | |
2366 | * This is an interrupt that interrupt the | |
2367 | * previous update. Still more to do. | |
2368 | */ | |
2369 | break; | |
2370 | case RB_PAGE_NORMAL: | |
2371 | /* | |
2372 | * An interrupt came in before the update | |
2373 | * and processed this for us. | |
2374 | * Nothing left to do. | |
2375 | */ | |
2376 | return 1; | |
2377 | case RB_PAGE_MOVED: | |
2378 | /* | |
2379 | * The reader is on another CPU and just did | |
2380 | * a swap with our next_page. | |
2381 | * Try again. | |
2382 | */ | |
2383 | return 1; | |
2384 | default: | |
2385 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
2386 | return -1; | |
2387 | } | |
2388 | ||
2389 | /* | |
2390 | * Now that we are here, the old head pointer is | |
2391 | * set to UPDATE. This will keep the reader from | |
2392 | * swapping the head page with the reader page. | |
2393 | * The reader (on another CPU) will spin till | |
2394 | * we are finished. | |
2395 | * | |
2396 | * We just need to protect against interrupts | |
2397 | * doing the job. We will set the next pointer | |
2398 | * to HEAD. After that, we set the old pointer | |
2399 | * to NORMAL, but only if it was HEAD before. | |
2400 | * otherwise we are an interrupt, and only | |
2401 | * want the outer most commit to reset it. | |
2402 | */ | |
2403 | new_head = next_page; | |
6689bed3 | 2404 | rb_inc_page(&new_head); |
77ae365e SR |
2405 | |
2406 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
2407 | RB_PAGE_NORMAL); | |
2408 | ||
2409 | /* | |
2410 | * Valid returns are: | |
2411 | * HEAD - an interrupt came in and already set it. | |
2412 | * NORMAL - One of two things: | |
2413 | * 1) We really set it. | |
2414 | * 2) A bunch of interrupts came in and moved | |
2415 | * the page forward again. | |
2416 | */ | |
2417 | switch (ret) { | |
2418 | case RB_PAGE_HEAD: | |
2419 | case RB_PAGE_NORMAL: | |
2420 | /* OK */ | |
2421 | break; | |
2422 | default: | |
2423 | RB_WARN_ON(cpu_buffer, 1); | |
2424 | return -1; | |
2425 | } | |
2426 | ||
2427 | /* | |
2428 | * It is possible that an interrupt came in, | |
2429 | * set the head up, then more interrupts came in | |
2430 | * and moved it again. When we get back here, | |
2431 | * the page would have been set to NORMAL but we | |
2432 | * just set it back to HEAD. | |
2433 | * | |
2434 | * How do you detect this? Well, if that happened | |
2435 | * the tail page would have moved. | |
2436 | */ | |
2437 | if (ret == RB_PAGE_NORMAL) { | |
8573636e SRRH |
2438 | struct buffer_page *buffer_tail_page; |
2439 | ||
2440 | buffer_tail_page = READ_ONCE(cpu_buffer->tail_page); | |
77ae365e SR |
2441 | /* |
2442 | * If the tail had moved passed next, then we need | |
2443 | * to reset the pointer. | |
2444 | */ | |
8573636e SRRH |
2445 | if (buffer_tail_page != tail_page && |
2446 | buffer_tail_page != next_page) | |
77ae365e SR |
2447 | rb_head_page_set_normal(cpu_buffer, new_head, |
2448 | next_page, | |
2449 | RB_PAGE_HEAD); | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * If this was the outer most commit (the one that | |
2454 | * changed the original pointer from HEAD to UPDATE), | |
2455 | * then it is up to us to reset it to NORMAL. | |
2456 | */ | |
2457 | if (type == RB_PAGE_HEAD) { | |
2458 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2459 | tail_page, | |
2460 | RB_PAGE_UPDATE); | |
2461 | if (RB_WARN_ON(cpu_buffer, | |
2462 | ret != RB_PAGE_UPDATE)) | |
2463 | return -1; | |
2464 | } | |
2465 | ||
2466 | return 0; | |
2467 | } | |
2468 | ||
c7b09308 SR |
2469 | static inline void |
2470 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 2471 | unsigned long tail, struct rb_event_info *info) |
c7b09308 | 2472 | { |
139f8400 | 2473 | unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size); |
fcc742ea | 2474 | struct buffer_page *tail_page = info->tail_page; |
c7b09308 | 2475 | struct ring_buffer_event *event; |
fcc742ea | 2476 | unsigned long length = info->length; |
c7b09308 SR |
2477 | |
2478 | /* | |
2479 | * Only the event that crossed the page boundary | |
2480 | * must fill the old tail_page with padding. | |
2481 | */ | |
139f8400 | 2482 | if (tail >= bsize) { |
b3230c8b SR |
2483 | /* |
2484 | * If the page was filled, then we still need | |
2485 | * to update the real_end. Reset it to zero | |
2486 | * and the reader will ignore it. | |
2487 | */ | |
139f8400 | 2488 | if (tail == bsize) |
b3230c8b SR |
2489 | tail_page->real_end = 0; |
2490 | ||
c7b09308 SR |
2491 | local_sub(length, &tail_page->write); |
2492 | return; | |
2493 | } | |
2494 | ||
2495 | event = __rb_page_index(tail_page, tail); | |
2496 | ||
ff0ff84a SR |
2497 | /* |
2498 | * Save the original length to the meta data. | |
2499 | * This will be used by the reader to add lost event | |
2500 | * counter. | |
2501 | */ | |
2502 | tail_page->real_end = tail; | |
2503 | ||
c7b09308 SR |
2504 | /* |
2505 | * If this event is bigger than the minimum size, then | |
2506 | * we need to be careful that we don't subtract the | |
2507 | * write counter enough to allow another writer to slip | |
2508 | * in on this page. | |
2509 | * We put in a discarded commit instead, to make sure | |
45d99ea4 ZY |
2510 | * that this space is not used again, and this space will |
2511 | * not be accounted into 'entries_bytes'. | |
c7b09308 SR |
2512 | * |
2513 | * If we are less than the minimum size, we don't need to | |
2514 | * worry about it. | |
2515 | */ | |
139f8400 | 2516 | if (tail > (bsize - RB_EVNT_MIN_SIZE)) { |
c7b09308 SR |
2517 | /* No room for any events */ |
2518 | ||
2519 | /* Mark the rest of the page with padding */ | |
2520 | rb_event_set_padding(event); | |
2521 | ||
a0fcaaed SRG |
2522 | /* Make sure the padding is visible before the write update */ |
2523 | smp_wmb(); | |
2524 | ||
c7b09308 SR |
2525 | /* Set the write back to the previous setting */ |
2526 | local_sub(length, &tail_page->write); | |
2527 | return; | |
2528 | } | |
2529 | ||
2530 | /* Put in a discarded event */ | |
139f8400 | 2531 | event->array[0] = (bsize - tail) - RB_EVNT_HDR_SIZE; |
c7b09308 SR |
2532 | event->type_len = RINGBUF_TYPE_PADDING; |
2533 | /* time delta must be non zero */ | |
2534 | event->time_delta = 1; | |
c7b09308 | 2535 | |
45d99ea4 | 2536 | /* account for padding bytes */ |
139f8400 | 2537 | local_add(bsize - tail, &cpu_buffer->entries_bytes); |
45d99ea4 | 2538 | |
a0fcaaed SRG |
2539 | /* Make sure the padding is visible before the tail_page->write update */ |
2540 | smp_wmb(); | |
2541 | ||
c7b09308 | 2542 | /* Set write to end of buffer */ |
139f8400 | 2543 | length = (tail + length) - bsize; |
c7b09308 SR |
2544 | local_sub(length, &tail_page->write); |
2545 | } | |
6634ff26 | 2546 | |
4239c38f SRRH |
2547 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer); |
2548 | ||
747e94ae SR |
2549 | /* |
2550 | * This is the slow path, force gcc not to inline it. | |
2551 | */ | |
2552 | static noinline struct ring_buffer_event * | |
6634ff26 | 2553 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
fcc742ea | 2554 | unsigned long tail, struct rb_event_info *info) |
7a8e76a3 | 2555 | { |
fcc742ea | 2556 | struct buffer_page *tail_page = info->tail_page; |
5a50e33c | 2557 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
13292494 | 2558 | struct trace_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2559 | struct buffer_page *next_page; |
2560 | int ret; | |
aa20ae84 SR |
2561 | |
2562 | next_page = tail_page; | |
2563 | ||
6689bed3 | 2564 | rb_inc_page(&next_page); |
aa20ae84 | 2565 | |
aa20ae84 SR |
2566 | /* |
2567 | * If for some reason, we had an interrupt storm that made | |
2568 | * it all the way around the buffer, bail, and warn | |
2569 | * about it. | |
2570 | */ | |
2571 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2572 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2573 | goto out_reset; |
2574 | } | |
2575 | ||
77ae365e SR |
2576 | /* |
2577 | * This is where the fun begins! | |
2578 | * | |
2579 | * We are fighting against races between a reader that | |
2580 | * could be on another CPU trying to swap its reader | |
2581 | * page with the buffer head. | |
2582 | * | |
2583 | * We are also fighting against interrupts coming in and | |
2584 | * moving the head or tail on us as well. | |
2585 | * | |
2586 | * If the next page is the head page then we have filled | |
2587 | * the buffer, unless the commit page is still on the | |
2588 | * reader page. | |
2589 | */ | |
6689bed3 | 2590 | if (rb_is_head_page(next_page, &tail_page->list)) { |
aa20ae84 | 2591 | |
77ae365e SR |
2592 | /* |
2593 | * If the commit is not on the reader page, then | |
2594 | * move the header page. | |
2595 | */ | |
2596 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2597 | /* | |
2598 | * If we are not in overwrite mode, | |
2599 | * this is easy, just stop here. | |
2600 | */ | |
884bfe89 SP |
2601 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2602 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2603 | goto out_reset; |
884bfe89 | 2604 | } |
77ae365e SR |
2605 | |
2606 | ret = rb_handle_head_page(cpu_buffer, | |
2607 | tail_page, | |
2608 | next_page); | |
2609 | if (ret < 0) | |
2610 | goto out_reset; | |
2611 | if (ret) | |
2612 | goto out_again; | |
2613 | } else { | |
2614 | /* | |
2615 | * We need to be careful here too. The | |
2616 | * commit page could still be on the reader | |
2617 | * page. We could have a small buffer, and | |
2618 | * have filled up the buffer with events | |
2619 | * from interrupts and such, and wrapped. | |
2620 | * | |
c6358bac | 2621 | * Note, if the tail page is also on the |
77ae365e SR |
2622 | * reader_page, we let it move out. |
2623 | */ | |
2624 | if (unlikely((cpu_buffer->commit_page != | |
2625 | cpu_buffer->tail_page) && | |
2626 | (cpu_buffer->commit_page == | |
2627 | cpu_buffer->reader_page))) { | |
2628 | local_inc(&cpu_buffer->commit_overrun); | |
2629 | goto out_reset; | |
2630 | } | |
aa20ae84 SR |
2631 | } |
2632 | } | |
2633 | ||
70004986 | 2634 | rb_tail_page_update(cpu_buffer, tail_page, next_page); |
aa20ae84 | 2635 | |
77ae365e | 2636 | out_again: |
aa20ae84 | 2637 | |
fcc742ea | 2638 | rb_reset_tail(cpu_buffer, tail, info); |
aa20ae84 | 2639 | |
4239c38f SRRH |
2640 | /* Commit what we have for now. */ |
2641 | rb_end_commit(cpu_buffer); | |
2642 | /* rb_end_commit() decs committing */ | |
2643 | local_inc(&cpu_buffer->committing); | |
2644 | ||
aa20ae84 SR |
2645 | /* fail and let the caller try again */ |
2646 | return ERR_PTR(-EAGAIN); | |
2647 | ||
45141d46 | 2648 | out_reset: |
6f3b3440 | 2649 | /* reset write */ |
fcc742ea | 2650 | rb_reset_tail(cpu_buffer, tail, info); |
6f3b3440 | 2651 | |
bf41a158 | 2652 | return NULL; |
7a8e76a3 SR |
2653 | } |
2654 | ||
74e87937 SRV |
2655 | /* Slow path */ |
2656 | static struct ring_buffer_event * | |
3cb30911 SRG |
2657 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, |
2658 | struct ring_buffer_event *event, u64 delta, bool abs) | |
9826b273 | 2659 | { |
dc4e2801 TZ |
2660 | if (abs) |
2661 | event->type_len = RINGBUF_TYPE_TIME_STAMP; | |
2662 | else | |
2663 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
9826b273 | 2664 | |
dc4e2801 | 2665 | /* Not the first event on the page, or not delta? */ |
3cb30911 | 2666 | if (abs || rb_event_index(cpu_buffer, event)) { |
d90fd774 SRRH |
2667 | event->time_delta = delta & TS_MASK; |
2668 | event->array[0] = delta >> TS_SHIFT; | |
2669 | } else { | |
2670 | /* nope, just zero it */ | |
2671 | event->time_delta = 0; | |
2672 | event->array[0] = 0; | |
2673 | } | |
a4543a2f | 2674 | |
d90fd774 SRRH |
2675 | return skip_time_extend(event); |
2676 | } | |
a4543a2f | 2677 | |
58fbc3c6 SRV |
2678 | #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK |
2679 | static inline bool sched_clock_stable(void) | |
2680 | { | |
2681 | return true; | |
2682 | } | |
2683 | #endif | |
2684 | ||
74e87937 | 2685 | static void |
58fbc3c6 SRV |
2686 | rb_check_timestamp(struct ring_buffer_per_cpu *cpu_buffer, |
2687 | struct rb_event_info *info) | |
2688 | { | |
2689 | u64 write_stamp; | |
2690 | ||
29ce2451 | 2691 | WARN_ONCE(1, "Delta way too big! %llu ts=%llu before=%llu after=%llu write stamp=%llu\n%s", |
58fbc3c6 SRV |
2692 | (unsigned long long)info->delta, |
2693 | (unsigned long long)info->ts, | |
2694 | (unsigned long long)info->before, | |
2695 | (unsigned long long)info->after, | |
c84897c0 | 2696 | (unsigned long long)({rb_time_read(&cpu_buffer->write_stamp, &write_stamp); write_stamp;}), |
58fbc3c6 SRV |
2697 | sched_clock_stable() ? "" : |
2698 | "If you just came from a suspend/resume,\n" | |
2699 | "please switch to the trace global clock:\n" | |
2455f0e1 | 2700 | " echo global > /sys/kernel/tracing/trace_clock\n" |
58fbc3c6 SRV |
2701 | "or add trace_clock=global to the kernel command line\n"); |
2702 | } | |
2703 | ||
74e87937 SRV |
2704 | static void rb_add_timestamp(struct ring_buffer_per_cpu *cpu_buffer, |
2705 | struct ring_buffer_event **event, | |
2706 | struct rb_event_info *info, | |
2707 | u64 *delta, | |
2708 | unsigned int *length) | |
2709 | { | |
2710 | bool abs = info->add_timestamp & | |
2711 | (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE); | |
2712 | ||
29ce2451 | 2713 | if (unlikely(info->delta > (1ULL << 59))) { |
6695da58 SRG |
2714 | /* |
2715 | * Some timers can use more than 59 bits, and when a timestamp | |
2716 | * is added to the buffer, it will lose those bits. | |
2717 | */ | |
2718 | if (abs && (info->ts & TS_MSB)) { | |
2719 | info->delta &= ABS_TS_MASK; | |
2720 | ||
29ce2451 | 2721 | /* did the clock go backwards */ |
6695da58 | 2722 | } else if (info->before == info->after && info->before > info->ts) { |
29ce2451 SRV |
2723 | /* not interrupted */ |
2724 | static int once; | |
2725 | ||
2726 | /* | |
2727 | * This is possible with a recalibrating of the TSC. | |
2728 | * Do not produce a call stack, but just report it. | |
2729 | */ | |
2730 | if (!once) { | |
2731 | once++; | |
2732 | pr_warn("Ring buffer clock went backwards: %llu -> %llu\n", | |
2733 | info->before, info->ts); | |
2734 | } | |
2735 | } else | |
2736 | rb_check_timestamp(cpu_buffer, info); | |
2737 | if (!abs) | |
2738 | info->delta = 0; | |
2739 | } | |
3cb30911 | 2740 | *event = rb_add_time_stamp(cpu_buffer, *event, info->delta, abs); |
74e87937 SRV |
2741 | *length -= RB_LEN_TIME_EXTEND; |
2742 | *delta = 0; | |
2743 | } | |
2744 | ||
d90fd774 SRRH |
2745 | /** |
2746 | * rb_update_event - update event type and data | |
cfc585a4 | 2747 | * @cpu_buffer: The per cpu buffer of the @event |
d90fd774 | 2748 | * @event: the event to update |
cfc585a4 | 2749 | * @info: The info to update the @event with (contains length and delta) |
d90fd774 | 2750 | * |
cfc585a4 | 2751 | * Update the type and data fields of the @event. The length |
d90fd774 SRRH |
2752 | * is the actual size that is written to the ring buffer, |
2753 | * and with this, we can determine what to place into the | |
2754 | * data field. | |
2755 | */ | |
b7dc42fd | 2756 | static void |
d90fd774 SRRH |
2757 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
2758 | struct ring_buffer_event *event, | |
2759 | struct rb_event_info *info) | |
2760 | { | |
2761 | unsigned length = info->length; | |
2762 | u64 delta = info->delta; | |
8672e494 SRV |
2763 | unsigned int nest = local_read(&cpu_buffer->committing) - 1; |
2764 | ||
a948c69d | 2765 | if (!WARN_ON_ONCE(nest >= MAX_NEST)) |
8672e494 | 2766 | cpu_buffer->event_stamp[nest] = info->ts; |
a4543a2f SRRH |
2767 | |
2768 | /* | |
d90fd774 | 2769 | * If we need to add a timestamp, then we |
6167c205 | 2770 | * add it to the start of the reserved space. |
a4543a2f | 2771 | */ |
74e87937 SRV |
2772 | if (unlikely(info->add_timestamp)) |
2773 | rb_add_timestamp(cpu_buffer, &event, info, &delta, &length); | |
a4543a2f | 2774 | |
d90fd774 SRRH |
2775 | event->time_delta = delta; |
2776 | length -= RB_EVNT_HDR_SIZE; | |
adab66b7 | 2777 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { |
d90fd774 SRRH |
2778 | event->type_len = 0; |
2779 | event->array[0] = length; | |
2780 | } else | |
2781 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
2782 | } | |
2783 | ||
2784 | static unsigned rb_calculate_event_length(unsigned length) | |
2785 | { | |
2786 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2787 | ||
2788 | /* zero length can cause confusions */ | |
2789 | if (!length) | |
2790 | length++; | |
2791 | ||
adab66b7 | 2792 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
d90fd774 SRRH |
2793 | length += sizeof(event.array[0]); |
2794 | ||
2795 | length += RB_EVNT_HDR_SIZE; | |
adab66b7 | 2796 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
d90fd774 SRRH |
2797 | |
2798 | /* | |
2799 | * In case the time delta is larger than the 27 bits for it | |
2800 | * in the header, we need to add a timestamp. If another | |
2801 | * event comes in when trying to discard this one to increase | |
2802 | * the length, then the timestamp will be added in the allocated | |
2803 | * space of this event. If length is bigger than the size needed | |
2804 | * for the TIME_EXTEND, then padding has to be used. The events | |
2805 | * length must be either RB_LEN_TIME_EXTEND, or greater than or equal | |
2806 | * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding. | |
2807 | * As length is a multiple of 4, we only need to worry if it | |
2808 | * is 12 (RB_LEN_TIME_EXTEND + 4). | |
2809 | */ | |
2810 | if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT) | |
2811 | length += RB_ALIGNMENT; | |
2812 | ||
2813 | return length; | |
2814 | } | |
2815 | ||
bc92b956 | 2816 | static inline bool |
d90fd774 SRRH |
2817 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, |
2818 | struct ring_buffer_event *event) | |
2819 | { | |
2820 | unsigned long new_index, old_index; | |
2821 | struct buffer_page *bpage; | |
d90fd774 SRRH |
2822 | unsigned long addr; |
2823 | ||
3cb30911 | 2824 | new_index = rb_event_index(cpu_buffer, event); |
d90fd774 SRRH |
2825 | old_index = new_index + rb_event_ts_length(event); |
2826 | addr = (unsigned long)event; | |
3cb30911 | 2827 | addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1); |
d90fd774 | 2828 | |
8573636e | 2829 | bpage = READ_ONCE(cpu_buffer->tail_page); |
d90fd774 | 2830 | |
083e9f65 SRG |
2831 | /* |
2832 | * Make sure the tail_page is still the same and | |
2833 | * the next write location is the end of this event | |
2834 | */ | |
d90fd774 SRRH |
2835 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { |
2836 | unsigned long write_mask = | |
2837 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
2838 | unsigned long event_length = rb_event_length(event); | |
a389d86f | 2839 | |
b2dd7975 SRG |
2840 | /* |
2841 | * For the before_stamp to be different than the write_stamp | |
2842 | * to make sure that the next event adds an absolute | |
2843 | * value and does not rely on the saved write stamp, which | |
2844 | * is now going to be bogus. | |
083e9f65 SRG |
2845 | * |
2846 | * By setting the before_stamp to zero, the next event | |
2847 | * is not going to use the write_stamp and will instead | |
2848 | * create an absolute timestamp. This means there's no | |
2849 | * reason to update the wirte_stamp! | |
b2dd7975 SRG |
2850 | */ |
2851 | rb_time_set(&cpu_buffer->before_stamp, 0); | |
2852 | ||
a389d86f SRV |
2853 | /* |
2854 | * If an event were to come in now, it would see that the | |
2855 | * write_stamp and the before_stamp are different, and assume | |
2856 | * that this event just added itself before updating | |
2857 | * the write stamp. The interrupting event will fix the | |
083e9f65 | 2858 | * write stamp for us, and use an absolute timestamp. |
a389d86f SRV |
2859 | */ |
2860 | ||
d90fd774 SRRH |
2861 | /* |
2862 | * This is on the tail page. It is possible that | |
2863 | * a write could come in and move the tail page | |
2864 | * and write to the next page. That is fine | |
2865 | * because we just shorten what is on this page. | |
2866 | */ | |
2867 | old_index += write_mask; | |
2868 | new_index += write_mask; | |
00a8478f UB |
2869 | |
2870 | /* caution: old_index gets updated on cmpxchg failure */ | |
2871 | if (local_try_cmpxchg(&bpage->write, &old_index, new_index)) { | |
d90fd774 SRRH |
2872 | /* update counters */ |
2873 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
bc92b956 | 2874 | return true; |
d90fd774 SRRH |
2875 | } |
2876 | } | |
2877 | ||
2878 | /* could not discard */ | |
bc92b956 | 2879 | return false; |
d90fd774 SRRH |
2880 | } |
2881 | ||
2882 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) | |
2883 | { | |
2884 | local_inc(&cpu_buffer->committing); | |
2885 | local_inc(&cpu_buffer->commits); | |
2886 | } | |
2887 | ||
38e11df1 | 2888 | static __always_inline void |
d90fd774 SRRH |
2889 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
2890 | { | |
2891 | unsigned long max_count; | |
2892 | ||
2893 | /* | |
2894 | * We only race with interrupts and NMIs on this CPU. | |
2895 | * If we own the commit event, then we can commit | |
2896 | * all others that interrupted us, since the interruptions | |
2897 | * are in stack format (they finish before they come | |
2898 | * back to us). This allows us to do a simple loop to | |
2899 | * assign the commit to the tail. | |
2900 | */ | |
2901 | again: | |
2902 | max_count = cpu_buffer->nr_pages * 100; | |
2903 | ||
8573636e | 2904 | while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) { |
d90fd774 SRRH |
2905 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
2906 | return; | |
2907 | if (RB_WARN_ON(cpu_buffer, | |
2908 | rb_is_reader_page(cpu_buffer->tail_page))) | |
2909 | return; | |
6455b616 ZY |
2910 | /* |
2911 | * No need for a memory barrier here, as the update | |
2912 | * of the tail_page did it for this page. | |
2913 | */ | |
d90fd774 SRRH |
2914 | local_set(&cpu_buffer->commit_page->page->commit, |
2915 | rb_page_write(cpu_buffer->commit_page)); | |
6689bed3 | 2916 | rb_inc_page(&cpu_buffer->commit_page); |
d90fd774 SRRH |
2917 | /* add barrier to keep gcc from optimizing too much */ |
2918 | barrier(); | |
2919 | } | |
2920 | while (rb_commit_index(cpu_buffer) != | |
2921 | rb_page_write(cpu_buffer->commit_page)) { | |
2922 | ||
6455b616 ZY |
2923 | /* Make sure the readers see the content of what is committed. */ |
2924 | smp_wmb(); | |
d90fd774 SRRH |
2925 | local_set(&cpu_buffer->commit_page->page->commit, |
2926 | rb_page_write(cpu_buffer->commit_page)); | |
2927 | RB_WARN_ON(cpu_buffer, | |
2928 | local_read(&cpu_buffer->commit_page->page->commit) & | |
2929 | ~RB_WRITE_MASK); | |
2930 | barrier(); | |
2931 | } | |
2932 | ||
2933 | /* again, keep gcc from optimizing */ | |
2934 | barrier(); | |
2935 | ||
2936 | /* | |
2937 | * If an interrupt came in just after the first while loop | |
2938 | * and pushed the tail page forward, we will be left with | |
2939 | * a dangling commit that will never go forward. | |
2940 | */ | |
8573636e | 2941 | if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page))) |
d90fd774 SRRH |
2942 | goto again; |
2943 | } | |
2944 | ||
38e11df1 | 2945 | static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
d90fd774 SRRH |
2946 | { |
2947 | unsigned long commits; | |
2948 | ||
2949 | if (RB_WARN_ON(cpu_buffer, | |
2950 | !local_read(&cpu_buffer->committing))) | |
2951 | return; | |
2952 | ||
2953 | again: | |
2954 | commits = local_read(&cpu_buffer->commits); | |
2955 | /* synchronize with interrupts */ | |
2956 | barrier(); | |
2957 | if (local_read(&cpu_buffer->committing) == 1) | |
2958 | rb_set_commit_to_write(cpu_buffer); | |
2959 | ||
2960 | local_dec(&cpu_buffer->committing); | |
2961 | ||
2962 | /* synchronize with interrupts */ | |
2963 | barrier(); | |
2964 | ||
2965 | /* | |
2966 | * Need to account for interrupts coming in between the | |
2967 | * updating of the commit page and the clearing of the | |
2968 | * committing counter. | |
2969 | */ | |
2970 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2971 | !local_read(&cpu_buffer->committing)) { | |
2972 | local_inc(&cpu_buffer->committing); | |
2973 | goto again; | |
2974 | } | |
2975 | } | |
2976 | ||
2977 | static inline void rb_event_discard(struct ring_buffer_event *event) | |
2978 | { | |
dc4e2801 | 2979 | if (extended_time(event)) |
d90fd774 SRRH |
2980 | event = skip_time_extend(event); |
2981 | ||
2982 | /* array[0] holds the actual length for the discarded event */ | |
2983 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2984 | event->type_len = RINGBUF_TYPE_PADDING; | |
2985 | /* time delta must be non zero */ | |
2986 | if (!event->time_delta) | |
2987 | event->time_delta = 1; | |
2988 | } | |
2989 | ||
04aabc32 | 2990 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer) |
d90fd774 SRRH |
2991 | { |
2992 | local_inc(&cpu_buffer->entries); | |
d90fd774 SRRH |
2993 | rb_end_commit(cpu_buffer); |
2994 | } | |
2995 | ||
2996 | static __always_inline void | |
13292494 | 2997 | rb_wakeups(struct trace_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer) |
d90fd774 | 2998 | { |
d90fd774 SRRH |
2999 | if (buffer->irq_work.waiters_pending) { |
3000 | buffer->irq_work.waiters_pending = false; | |
3001 | /* irq_work_queue() supplies it's own memory barriers */ | |
3002 | irq_work_queue(&buffer->irq_work.work); | |
3003 | } | |
3004 | ||
3005 | if (cpu_buffer->irq_work.waiters_pending) { | |
3006 | cpu_buffer->irq_work.waiters_pending = false; | |
3007 | /* irq_work_queue() supplies it's own memory barriers */ | |
3008 | irq_work_queue(&cpu_buffer->irq_work.work); | |
3009 | } | |
3010 | ||
03329f99 SRV |
3011 | if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched)) |
3012 | return; | |
d90fd774 | 3013 | |
03329f99 SRV |
3014 | if (cpu_buffer->reader_page == cpu_buffer->commit_page) |
3015 | return; | |
2c2b0a78 | 3016 | |
03329f99 SRV |
3017 | if (!cpu_buffer->irq_work.full_waiters_pending) |
3018 | return; | |
2c2b0a78 | 3019 | |
03329f99 SRV |
3020 | cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched); |
3021 | ||
42fb0a1e | 3022 | if (!full_hit(buffer, cpu_buffer->cpu, cpu_buffer->shortest_full)) |
03329f99 SRV |
3023 | return; |
3024 | ||
3025 | cpu_buffer->irq_work.wakeup_full = true; | |
3026 | cpu_buffer->irq_work.full_waiters_pending = false; | |
3027 | /* irq_work_queue() supplies it's own memory barriers */ | |
3028 | irq_work_queue(&cpu_buffer->irq_work.work); | |
d90fd774 SRRH |
3029 | } |
3030 | ||
28575c61 SRV |
3031 | #ifdef CONFIG_RING_BUFFER_RECORD_RECURSION |
3032 | # define do_ring_buffer_record_recursion() \ | |
3033 | do_ftrace_record_recursion(_THIS_IP_, _RET_IP_) | |
3034 | #else | |
3035 | # define do_ring_buffer_record_recursion() do { } while (0) | |
3036 | #endif | |
3037 | ||
d90fd774 SRRH |
3038 | /* |
3039 | * The lock and unlock are done within a preempt disable section. | |
3040 | * The current_context per_cpu variable can only be modified | |
3041 | * by the current task between lock and unlock. But it can | |
a0e3a18f SRV |
3042 | * be modified more than once via an interrupt. To pass this |
3043 | * information from the lock to the unlock without having to | |
3044 | * access the 'in_interrupt()' functions again (which do show | |
3045 | * a bit of overhead in something as critical as function tracing, | |
3046 | * we use a bitmask trick. | |
d90fd774 | 3047 | * |
b02414c8 SRV |
3048 | * bit 1 = NMI context |
3049 | * bit 2 = IRQ context | |
3050 | * bit 3 = SoftIRQ context | |
3051 | * bit 4 = normal context. | |
d90fd774 | 3052 | * |
a0e3a18f SRV |
3053 | * This works because this is the order of contexts that can |
3054 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
3055 | * context. | |
3056 | * | |
3057 | * When the context is determined, the corresponding bit is | |
3058 | * checked and set (if it was set, then a recursion of that context | |
3059 | * happened). | |
3060 | * | |
3061 | * On unlock, we need to clear this bit. To do so, just subtract | |
3062 | * 1 from the current_context and AND it to itself. | |
3063 | * | |
3064 | * (binary) | |
3065 | * 101 - 1 = 100 | |
3066 | * 101 & 100 = 100 (clearing bit zero) | |
3067 | * | |
3068 | * 1010 - 1 = 1001 | |
3069 | * 1010 & 1001 = 1000 (clearing bit 1) | |
3070 | * | |
3071 | * The least significant bit can be cleared this way, and it | |
3072 | * just so happens that it is the same bit corresponding to | |
3073 | * the current context. | |
b02414c8 SRV |
3074 | * |
3075 | * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit | |
3076 | * is set when a recursion is detected at the current context, and if | |
3077 | * the TRANSITION bit is already set, it will fail the recursion. | |
3078 | * This is needed because there's a lag between the changing of | |
3079 | * interrupt context and updating the preempt count. In this case, | |
3080 | * a false positive will be found. To handle this, one extra recursion | |
3081 | * is allowed, and this is done by the TRANSITION bit. If the TRANSITION | |
3082 | * bit is already set, then it is considered a recursion and the function | |
3083 | * ends. Otherwise, the TRANSITION bit is set, and that bit is returned. | |
3084 | * | |
3085 | * On the trace_recursive_unlock(), the TRANSITION bit will be the first | |
3086 | * to be cleared. Even if it wasn't the context that set it. That is, | |
3087 | * if an interrupt comes in while NORMAL bit is set and the ring buffer | |
3088 | * is called before preempt_count() is updated, since the check will | |
3089 | * be on the NORMAL bit, the TRANSITION bit will then be set. If an | |
3090 | * NMI then comes in, it will set the NMI bit, but when the NMI code | |
f2cc020d | 3091 | * does the trace_recursive_unlock() it will clear the TRANSITION bit |
b02414c8 SRV |
3092 | * and leave the NMI bit set. But this is fine, because the interrupt |
3093 | * code that set the TRANSITION bit will then clear the NMI bit when it | |
3094 | * calls trace_recursive_unlock(). If another NMI comes in, it will | |
3095 | * set the TRANSITION bit and continue. | |
3096 | * | |
3097 | * Note: The TRANSITION bit only handles a single transition between context. | |
d90fd774 SRRH |
3098 | */ |
3099 | ||
bc92b956 | 3100 | static __always_inline bool |
d90fd774 SRRH |
3101 | trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer) |
3102 | { | |
a0e3a18f | 3103 | unsigned int val = cpu_buffer->current_context; |
91ebe8bc | 3104 | int bit = interrupt_context_level(); |
9b84fadc SRV |
3105 | |
3106 | bit = RB_CTX_NORMAL - bit; | |
a0e3a18f | 3107 | |
b02414c8 SRV |
3108 | if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) { |
3109 | /* | |
3110 | * It is possible that this was called by transitioning | |
3111 | * between interrupt context, and preempt_count() has not | |
3112 | * been updated yet. In this case, use the TRANSITION bit. | |
3113 | */ | |
3114 | bit = RB_CTX_TRANSITION; | |
28575c61 SRV |
3115 | if (val & (1 << (bit + cpu_buffer->nest))) { |
3116 | do_ring_buffer_record_recursion(); | |
bc92b956 | 3117 | return true; |
28575c61 | 3118 | } |
b02414c8 | 3119 | } |
d90fd774 | 3120 | |
8e012066 | 3121 | val |= (1 << (bit + cpu_buffer->nest)); |
a0e3a18f | 3122 | cpu_buffer->current_context = val; |
d90fd774 | 3123 | |
bc92b956 | 3124 | return false; |
d90fd774 SRRH |
3125 | } |
3126 | ||
3127 | static __always_inline void | |
3128 | trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer) | |
3129 | { | |
8e012066 SRV |
3130 | cpu_buffer->current_context &= |
3131 | cpu_buffer->current_context - (1 << cpu_buffer->nest); | |
3132 | } | |
3133 | ||
b02414c8 SRV |
3134 | /* The recursive locking above uses 5 bits */ |
3135 | #define NESTED_BITS 5 | |
8e012066 SRV |
3136 | |
3137 | /** | |
3138 | * ring_buffer_nest_start - Allow to trace while nested | |
3139 | * @buffer: The ring buffer to modify | |
3140 | * | |
6167c205 | 3141 | * The ring buffer has a safety mechanism to prevent recursion. |
8e012066 SRV |
3142 | * But there may be a case where a trace needs to be done while |
3143 | * tracing something else. In this case, calling this function | |
3144 | * will allow this function to nest within a currently active | |
3145 | * ring_buffer_lock_reserve(). | |
3146 | * | |
3147 | * Call this function before calling another ring_buffer_lock_reserve() and | |
3148 | * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit(). | |
3149 | */ | |
13292494 | 3150 | void ring_buffer_nest_start(struct trace_buffer *buffer) |
8e012066 SRV |
3151 | { |
3152 | struct ring_buffer_per_cpu *cpu_buffer; | |
3153 | int cpu; | |
3154 | ||
3155 | /* Enabled by ring_buffer_nest_end() */ | |
3156 | preempt_disable_notrace(); | |
3157 | cpu = raw_smp_processor_id(); | |
3158 | cpu_buffer = buffer->buffers[cpu]; | |
6167c205 | 3159 | /* This is the shift value for the above recursive locking */ |
8e012066 SRV |
3160 | cpu_buffer->nest += NESTED_BITS; |
3161 | } | |
3162 | ||
3163 | /** | |
3164 | * ring_buffer_nest_end - Allow to trace while nested | |
3165 | * @buffer: The ring buffer to modify | |
3166 | * | |
3167 | * Must be called after ring_buffer_nest_start() and after the | |
3168 | * ring_buffer_unlock_commit(). | |
3169 | */ | |
13292494 | 3170 | void ring_buffer_nest_end(struct trace_buffer *buffer) |
8e012066 SRV |
3171 | { |
3172 | struct ring_buffer_per_cpu *cpu_buffer; | |
3173 | int cpu; | |
3174 | ||
3175 | /* disabled by ring_buffer_nest_start() */ | |
3176 | cpu = raw_smp_processor_id(); | |
3177 | cpu_buffer = buffer->buffers[cpu]; | |
6167c205 | 3178 | /* This is the shift value for the above recursive locking */ |
8e012066 SRV |
3179 | cpu_buffer->nest -= NESTED_BITS; |
3180 | preempt_enable_notrace(); | |
d90fd774 SRRH |
3181 | } |
3182 | ||
3183 | /** | |
3184 | * ring_buffer_unlock_commit - commit a reserved | |
3185 | * @buffer: The buffer to commit to | |
d90fd774 SRRH |
3186 | * |
3187 | * This commits the data to the ring buffer, and releases any locks held. | |
3188 | * | |
3189 | * Must be paired with ring_buffer_lock_reserve. | |
3190 | */ | |
04aabc32 | 3191 | int ring_buffer_unlock_commit(struct trace_buffer *buffer) |
d90fd774 SRRH |
3192 | { |
3193 | struct ring_buffer_per_cpu *cpu_buffer; | |
3194 | int cpu = raw_smp_processor_id(); | |
3195 | ||
3196 | cpu_buffer = buffer->buffers[cpu]; | |
3197 | ||
04aabc32 | 3198 | rb_commit(cpu_buffer); |
d90fd774 SRRH |
3199 | |
3200 | rb_wakeups(buffer, cpu_buffer); | |
3201 | ||
3202 | trace_recursive_unlock(cpu_buffer); | |
3203 | ||
3204 | preempt_enable_notrace(); | |
3205 | ||
3206 | return 0; | |
3207 | } | |
3208 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); | |
3209 | ||
5b7be9c7 SRV |
3210 | /* Special value to validate all deltas on a page. */ |
3211 | #define CHECK_FULL_PAGE 1L | |
3212 | ||
3213 | #ifdef CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS | |
d40dbb61 SRG |
3214 | |
3215 | static const char *show_irq_str(int bits) | |
3216 | { | |
3217 | const char *type[] = { | |
3218 | ".", // 0 | |
3219 | "s", // 1 | |
3220 | "h", // 2 | |
3221 | "Hs", // 3 | |
3222 | "n", // 4 | |
3223 | "Ns", // 5 | |
3224 | "Nh", // 6 | |
3225 | "NHs", // 7 | |
3226 | }; | |
3227 | ||
3228 | return type[bits]; | |
3229 | } | |
3230 | ||
3231 | /* Assume this is an trace event */ | |
3232 | static const char *show_flags(struct ring_buffer_event *event) | |
3233 | { | |
3234 | struct trace_entry *entry; | |
3235 | int bits = 0; | |
3236 | ||
3237 | if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry)) | |
3238 | return "X"; | |
3239 | ||
3240 | entry = ring_buffer_event_data(event); | |
3241 | ||
3242 | if (entry->flags & TRACE_FLAG_SOFTIRQ) | |
3243 | bits |= 1; | |
3244 | ||
3245 | if (entry->flags & TRACE_FLAG_HARDIRQ) | |
3246 | bits |= 2; | |
3247 | ||
3248 | if (entry->flags & TRACE_FLAG_NMI) | |
3249 | bits |= 4; | |
3250 | ||
3251 | return show_irq_str(bits); | |
3252 | } | |
3253 | ||
3254 | static const char *show_irq(struct ring_buffer_event *event) | |
3255 | { | |
3256 | struct trace_entry *entry; | |
3257 | ||
3258 | if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry)) | |
3259 | return ""; | |
3260 | ||
3261 | entry = ring_buffer_event_data(event); | |
3262 | if (entry->flags & TRACE_FLAG_IRQS_OFF) | |
3263 | return "d"; | |
3264 | return ""; | |
3265 | } | |
3266 | ||
3267 | static const char *show_interrupt_level(void) | |
3268 | { | |
3269 | unsigned long pc = preempt_count(); | |
3270 | unsigned char level = 0; | |
3271 | ||
3272 | if (pc & SOFTIRQ_OFFSET) | |
3273 | level |= 1; | |
3274 | ||
3275 | if (pc & HARDIRQ_MASK) | |
3276 | level |= 2; | |
3277 | ||
3278 | if (pc & NMI_MASK) | |
3279 | level |= 4; | |
3280 | ||
3281 | return show_irq_str(level); | |
3282 | } | |
3283 | ||
5b7be9c7 SRV |
3284 | static void dump_buffer_page(struct buffer_data_page *bpage, |
3285 | struct rb_event_info *info, | |
3286 | unsigned long tail) | |
3287 | { | |
3288 | struct ring_buffer_event *event; | |
3289 | u64 ts, delta; | |
3290 | int e; | |
3291 | ||
3292 | ts = bpage->time_stamp; | |
3293 | pr_warn(" [%lld] PAGE TIME STAMP\n", ts); | |
3294 | ||
3295 | for (e = 0; e < tail; e += rb_event_length(event)) { | |
3296 | ||
3297 | event = (struct ring_buffer_event *)(bpage->data + e); | |
3298 | ||
3299 | switch (event->type_len) { | |
3300 | ||
3301 | case RINGBUF_TYPE_TIME_EXTEND: | |
e20044f7 | 3302 | delta = rb_event_time_stamp(event); |
5b7be9c7 | 3303 | ts += delta; |
0b9036ef SRG |
3304 | pr_warn(" 0x%x: [%lld] delta:%lld TIME EXTEND\n", |
3305 | e, ts, delta); | |
5b7be9c7 SRV |
3306 | break; |
3307 | ||
3308 | case RINGBUF_TYPE_TIME_STAMP: | |
e20044f7 | 3309 | delta = rb_event_time_stamp(event); |
6695da58 | 3310 | ts = rb_fix_abs_ts(delta, ts); |
0b9036ef SRG |
3311 | pr_warn(" 0x%x: [%lld] absolute:%lld TIME STAMP\n", |
3312 | e, ts, delta); | |
5b7be9c7 SRV |
3313 | break; |
3314 | ||
3315 | case RINGBUF_TYPE_PADDING: | |
3316 | ts += event->time_delta; | |
0b9036ef SRG |
3317 | pr_warn(" 0x%x: [%lld] delta:%d PADDING\n", |
3318 | e, ts, event->time_delta); | |
5b7be9c7 SRV |
3319 | break; |
3320 | ||
3321 | case RINGBUF_TYPE_DATA: | |
3322 | ts += event->time_delta; | |
d40dbb61 SRG |
3323 | pr_warn(" 0x%x: [%lld] delta:%d %s%s\n", |
3324 | e, ts, event->time_delta, | |
3325 | show_flags(event), show_irq(event)); | |
5b7be9c7 SRV |
3326 | break; |
3327 | ||
3328 | default: | |
3329 | break; | |
3330 | } | |
3331 | } | |
0b9036ef | 3332 | pr_warn("expected end:0x%lx last event actually ended at:0x%x\n", tail, e); |
5b7be9c7 SRV |
3333 | } |
3334 | ||
3335 | static DEFINE_PER_CPU(atomic_t, checking); | |
3336 | static atomic_t ts_dump; | |
3337 | ||
f50345b4 SRG |
3338 | #define buffer_warn_return(fmt, ...) \ |
3339 | do { \ | |
3340 | /* If another report is happening, ignore this one */ \ | |
3341 | if (atomic_inc_return(&ts_dump) != 1) { \ | |
3342 | atomic_dec(&ts_dump); \ | |
3343 | goto out; \ | |
3344 | } \ | |
3345 | atomic_inc(&cpu_buffer->record_disabled); \ | |
3346 | pr_warn(fmt, ##__VA_ARGS__); \ | |
3347 | dump_buffer_page(bpage, info, tail); \ | |
3348 | atomic_dec(&ts_dump); \ | |
3349 | /* There's some cases in boot up that this can happen */ \ | |
3350 | if (WARN_ON_ONCE(system_state != SYSTEM_BOOTING)) \ | |
3351 | /* Do not re-enable checking */ \ | |
3352 | return; \ | |
3353 | } while (0) | |
3354 | ||
5b7be9c7 SRV |
3355 | /* |
3356 | * Check if the current event time stamp matches the deltas on | |
3357 | * the buffer page. | |
3358 | */ | |
3359 | static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer, | |
3360 | struct rb_event_info *info, | |
3361 | unsigned long tail) | |
3362 | { | |
3363 | struct ring_buffer_event *event; | |
3364 | struct buffer_data_page *bpage; | |
3365 | u64 ts, delta; | |
3366 | bool full = false; | |
3367 | int e; | |
3368 | ||
3369 | bpage = info->tail_page->page; | |
3370 | ||
3371 | if (tail == CHECK_FULL_PAGE) { | |
3372 | full = true; | |
3373 | tail = local_read(&bpage->commit); | |
3374 | } else if (info->add_timestamp & | |
3375 | (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)) { | |
3376 | /* Ignore events with absolute time stamps */ | |
3377 | return; | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * Do not check the first event (skip possible extends too). | |
3382 | * Also do not check if previous events have not been committed. | |
3383 | */ | |
3384 | if (tail <= 8 || tail > local_read(&bpage->commit)) | |
3385 | return; | |
3386 | ||
3387 | /* | |
083e9f65 | 3388 | * If this interrupted another event, |
5b7be9c7 SRV |
3389 | */ |
3390 | if (atomic_inc_return(this_cpu_ptr(&checking)) != 1) | |
3391 | goto out; | |
3392 | ||
3393 | ts = bpage->time_stamp; | |
3394 | ||
3395 | for (e = 0; e < tail; e += rb_event_length(event)) { | |
3396 | ||
3397 | event = (struct ring_buffer_event *)(bpage->data + e); | |
3398 | ||
3399 | switch (event->type_len) { | |
3400 | ||
3401 | case RINGBUF_TYPE_TIME_EXTEND: | |
e20044f7 | 3402 | delta = rb_event_time_stamp(event); |
5b7be9c7 SRV |
3403 | ts += delta; |
3404 | break; | |
3405 | ||
3406 | case RINGBUF_TYPE_TIME_STAMP: | |
e20044f7 | 3407 | delta = rb_event_time_stamp(event); |
f50345b4 SRG |
3408 | delta = rb_fix_abs_ts(delta, ts); |
3409 | if (delta < ts) { | |
3410 | buffer_warn_return("[CPU: %d]ABSOLUTE TIME WENT BACKWARDS: last ts: %lld absolute ts: %lld\n", | |
3411 | cpu_buffer->cpu, ts, delta); | |
3412 | } | |
3413 | ts = delta; | |
5b7be9c7 SRV |
3414 | break; |
3415 | ||
3416 | case RINGBUF_TYPE_PADDING: | |
3417 | if (event->time_delta == 1) | |
3418 | break; | |
957cdcd9 | 3419 | fallthrough; |
5b7be9c7 SRV |
3420 | case RINGBUF_TYPE_DATA: |
3421 | ts += event->time_delta; | |
3422 | break; | |
3423 | ||
3424 | default: | |
3425 | RB_WARN_ON(cpu_buffer, 1); | |
3426 | } | |
3427 | } | |
3428 | if ((full && ts > info->ts) || | |
3429 | (!full && ts + info->delta != info->ts)) { | |
f50345b4 SRG |
3430 | buffer_warn_return("[CPU: %d]TIME DOES NOT MATCH expected:%lld actual:%lld delta:%lld before:%lld after:%lld%s context:%s\n", |
3431 | cpu_buffer->cpu, | |
3432 | ts + info->delta, info->ts, info->delta, | |
3433 | info->before, info->after, | |
3434 | full ? " (full)" : "", show_interrupt_level()); | |
5b7be9c7 SRV |
3435 | } |
3436 | out: | |
3437 | atomic_dec(this_cpu_ptr(&checking)); | |
3438 | } | |
3439 | #else | |
3440 | static inline void check_buffer(struct ring_buffer_per_cpu *cpu_buffer, | |
3441 | struct rb_event_info *info, | |
3442 | unsigned long tail) | |
3443 | { | |
3444 | } | |
3445 | #endif /* CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS */ | |
3446 | ||
6634ff26 SR |
3447 | static struct ring_buffer_event * |
3448 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 3449 | struct rb_event_info *info) |
6634ff26 | 3450 | { |
6634ff26 | 3451 | struct ring_buffer_event *event; |
fcc742ea | 3452 | struct buffer_page *tail_page; |
a389d86f | 3453 | unsigned long tail, write, w; |
69d1b839 | 3454 | |
8573636e SRRH |
3455 | /* Don't let the compiler play games with cpu_buffer->tail_page */ |
3456 | tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page); | |
a389d86f SRV |
3457 | |
3458 | /*A*/ w = local_read(&tail_page->write) & RB_WRITE_MASK; | |
3459 | barrier(); | |
c84897c0 SRG |
3460 | rb_time_read(&cpu_buffer->before_stamp, &info->before); |
3461 | rb_time_read(&cpu_buffer->write_stamp, &info->after); | |
a389d86f SRV |
3462 | barrier(); |
3463 | info->ts = rb_time_stamp(cpu_buffer->buffer); | |
3464 | ||
58fbc3c6 | 3465 | if ((info->add_timestamp & RB_ADD_STAMP_ABSOLUTE)) { |
a389d86f | 3466 | info->delta = info->ts; |
a389d86f | 3467 | } else { |
58fbc3c6 SRV |
3468 | /* |
3469 | * If interrupting an event time update, we may need an | |
3470 | * absolute timestamp. | |
3471 | * Don't bother if this is the start of a new page (w == 0). | |
3472 | */ | |
b3ae7b67 SRG |
3473 | if (!w) { |
3474 | /* Use the sub-buffer timestamp */ | |
3475 | info->delta = 0; | |
c84897c0 | 3476 | } else if (unlikely(info->before != info->after)) { |
58fbc3c6 SRV |
3477 | info->add_timestamp |= RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND; |
3478 | info->length += RB_LEN_TIME_EXTEND; | |
3479 | } else { | |
3480 | info->delta = info->ts - info->after; | |
3481 | if (unlikely(test_time_stamp(info->delta))) { | |
3482 | info->add_timestamp |= RB_ADD_STAMP_EXTEND; | |
3483 | info->length += RB_LEN_TIME_EXTEND; | |
3484 | } | |
10464b4a | 3485 | } |
7c4b4a51 | 3486 | } |
b7dc42fd | 3487 | |
10464b4a | 3488 | /*B*/ rb_time_set(&cpu_buffer->before_stamp, info->ts); |
a389d86f SRV |
3489 | |
3490 | /*C*/ write = local_add_return(info->length, &tail_page->write); | |
77ae365e SR |
3491 | |
3492 | /* set write to only the index of the write */ | |
3493 | write &= RB_WRITE_MASK; | |
a389d86f | 3494 | |
fcc742ea | 3495 | tail = write - info->length; |
6634ff26 | 3496 | |
a389d86f | 3497 | /* See if we shot pass the end of this buffer page */ |
139f8400 | 3498 | if (unlikely(write > cpu_buffer->buffer->subbuf_size)) { |
9e45e39d | 3499 | check_buffer(cpu_buffer, info, CHECK_FULL_PAGE); |
a389d86f SRV |
3500 | return rb_move_tail(cpu_buffer, tail, info); |
3501 | } | |
3502 | ||
3503 | if (likely(tail == w)) { | |
a389d86f | 3504 | /* Nothing interrupted us between A and C */ |
10464b4a | 3505 | /*D*/ rb_time_set(&cpu_buffer->write_stamp, info->ts); |
dd939425 SRG |
3506 | /* |
3507 | * If something came in between C and D, the write stamp | |
3508 | * may now not be in sync. But that's fine as the before_stamp | |
3509 | * will be different and then next event will just be forced | |
3510 | * to use an absolute timestamp. | |
3511 | */ | |
7c4b4a51 SRV |
3512 | if (likely(!(info->add_timestamp & |
3513 | (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)))) | |
a389d86f | 3514 | /* This did not interrupt any time update */ |
58fbc3c6 | 3515 | info->delta = info->ts - info->after; |
a389d86f | 3516 | else |
82db909e | 3517 | /* Just use full timestamp for interrupting event */ |
a389d86f | 3518 | info->delta = info->ts; |
5b7be9c7 | 3519 | check_buffer(cpu_buffer, info, tail); |
a389d86f SRV |
3520 | } else { |
3521 | u64 ts; | |
3522 | /* SLOW PATH - Interrupted between A and C */ | |
b803d7c6 SRG |
3523 | |
3524 | /* Save the old before_stamp */ | |
c84897c0 | 3525 | rb_time_read(&cpu_buffer->before_stamp, &info->before); |
b803d7c6 SRG |
3526 | |
3527 | /* | |
3528 | * Read a new timestamp and update the before_stamp to make | |
3529 | * the next event after this one force using an absolute | |
3530 | * timestamp. This is in case an interrupt were to come in | |
3531 | * between E and F. | |
3532 | */ | |
a389d86f | 3533 | ts = rb_time_stamp(cpu_buffer->buffer); |
b803d7c6 SRG |
3534 | rb_time_set(&cpu_buffer->before_stamp, ts); |
3535 | ||
3536 | barrier(); | |
c84897c0 | 3537 | /*E*/ rb_time_read(&cpu_buffer->write_stamp, &info->after); |
a389d86f | 3538 | barrier(); |
b803d7c6 SRG |
3539 | /*F*/ if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) && |
3540 | info->after == info->before && info->after < ts) { | |
3541 | /* | |
3542 | * Nothing came after this event between C and F, it is | |
3543 | * safe to use info->after for the delta as it | |
3544 | * matched info->before and is still valid. | |
3545 | */ | |
58fbc3c6 | 3546 | info->delta = ts - info->after; |
a389d86f SRV |
3547 | } else { |
3548 | /* | |
b803d7c6 | 3549 | * Interrupted between C and F: |
a389d86f SRV |
3550 | * Lost the previous events time stamp. Just set the |
3551 | * delta to zero, and this will be the same time as | |
3552 | * the event this event interrupted. And the events that | |
3553 | * came after this will still be correct (as they would | |
3554 | * have built their delta on the previous event. | |
3555 | */ | |
3556 | info->delta = 0; | |
3557 | } | |
8672e494 | 3558 | info->ts = ts; |
7c4b4a51 | 3559 | info->add_timestamp &= ~RB_ADD_STAMP_FORCE; |
a389d86f SRV |
3560 | } |
3561 | ||
6634ff26 | 3562 | /* |
a4543a2f | 3563 | * If this is the first commit on the page, then it has the same |
b7dc42fd | 3564 | * timestamp as the page itself. |
6634ff26 | 3565 | */ |
7c4b4a51 SRV |
3566 | if (unlikely(!tail && !(info->add_timestamp & |
3567 | (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)))) | |
a4543a2f SRRH |
3568 | info->delta = 0; |
3569 | ||
b7dc42fd SRRH |
3570 | /* We reserved something on the buffer */ |
3571 | ||
3572 | event = __rb_page_index(tail_page, tail); | |
a4543a2f SRRH |
3573 | rb_update_event(cpu_buffer, event, info); |
3574 | ||
3575 | local_inc(&tail_page->entries); | |
6634ff26 | 3576 | |
b7dc42fd SRRH |
3577 | /* |
3578 | * If this is the first commit on the page, then update | |
3579 | * its timestamp. | |
3580 | */ | |
75b21c6d | 3581 | if (unlikely(!tail)) |
b7dc42fd SRRH |
3582 | tail_page->page->time_stamp = info->ts; |
3583 | ||
c64e148a | 3584 | /* account for these added bytes */ |
fcc742ea | 3585 | local_add(info->length, &cpu_buffer->entries_bytes); |
c64e148a | 3586 | |
6634ff26 SR |
3587 | return event; |
3588 | } | |
3589 | ||
fa7ffb39 | 3590 | static __always_inline struct ring_buffer_event * |
13292494 | 3591 | rb_reserve_next_event(struct trace_buffer *buffer, |
62f0b3eb | 3592 | struct ring_buffer_per_cpu *cpu_buffer, |
1cd8d735 | 3593 | unsigned long length) |
7a8e76a3 SR |
3594 | { |
3595 | struct ring_buffer_event *event; | |
fcc742ea | 3596 | struct rb_event_info info; |
818e3dd3 | 3597 | int nr_loops = 0; |
58fbc3c6 | 3598 | int add_ts_default; |
7a8e76a3 | 3599 | |
71229230 SRG |
3600 | /* ring buffer does cmpxchg, make sure it is safe in NMI context */ |
3601 | if (!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) && | |
3602 | (unlikely(in_nmi()))) { | |
3603 | return NULL; | |
3604 | } | |
3605 | ||
fa743953 | 3606 | rb_start_commit(cpu_buffer); |
a389d86f | 3607 | /* The commit page can not change after this */ |
fa743953 | 3608 | |
85bac32c | 3609 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
3610 | /* |
3611 | * Due to the ability to swap a cpu buffer from a buffer | |
3612 | * it is possible it was swapped before we committed. | |
3613 | * (committing stops a swap). We check for it here and | |
3614 | * if it happened, we have to fail the write. | |
3615 | */ | |
3616 | barrier(); | |
6aa7de05 | 3617 | if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) { |
62f0b3eb SR |
3618 | local_dec(&cpu_buffer->committing); |
3619 | local_dec(&cpu_buffer->commits); | |
3620 | return NULL; | |
3621 | } | |
85bac32c | 3622 | #endif |
b7dc42fd | 3623 | |
fcc742ea | 3624 | info.length = rb_calculate_event_length(length); |
58fbc3c6 SRV |
3625 | |
3626 | if (ring_buffer_time_stamp_abs(cpu_buffer->buffer)) { | |
3627 | add_ts_default = RB_ADD_STAMP_ABSOLUTE; | |
3628 | info.length += RB_LEN_TIME_EXTEND; | |
139f8400 | 3629 | if (info.length > cpu_buffer->buffer->max_data_size) |
b3ae7b67 | 3630 | goto out_fail; |
58fbc3c6 SRV |
3631 | } else { |
3632 | add_ts_default = RB_ADD_STAMP_NONE; | |
3633 | } | |
3634 | ||
a4543a2f | 3635 | again: |
58fbc3c6 | 3636 | info.add_timestamp = add_ts_default; |
b7dc42fd SRRH |
3637 | info.delta = 0; |
3638 | ||
818e3dd3 SR |
3639 | /* |
3640 | * We allow for interrupts to reenter here and do a trace. | |
3641 | * If one does, it will cause this original code to loop | |
3642 | * back here. Even with heavy interrupts happening, this | |
3643 | * should only happen a few times in a row. If this happens | |
3644 | * 1000 times in a row, there must be either an interrupt | |
3645 | * storm or we have something buggy. | |
3646 | * Bail! | |
3647 | */ | |
3e89c7bb | 3648 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 3649 | goto out_fail; |
818e3dd3 | 3650 | |
fcc742ea SRRH |
3651 | event = __rb_reserve_next(cpu_buffer, &info); |
3652 | ||
bd1b7cd3 | 3653 | if (unlikely(PTR_ERR(event) == -EAGAIN)) { |
58fbc3c6 | 3654 | if (info.add_timestamp & (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND)) |
bd1b7cd3 | 3655 | info.length -= RB_LEN_TIME_EXTEND; |
bf41a158 | 3656 | goto again; |
bd1b7cd3 | 3657 | } |
bf41a158 | 3658 | |
a389d86f SRV |
3659 | if (likely(event)) |
3660 | return event; | |
fa743953 SR |
3661 | out_fail: |
3662 | rb_end_commit(cpu_buffer); | |
3663 | return NULL; | |
7a8e76a3 SR |
3664 | } |
3665 | ||
3666 | /** | |
3667 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
3668 | * @buffer: the ring buffer to reserve from | |
3669 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 | 3670 | * |
6167c205 | 3671 | * Returns a reserved event on the ring buffer to copy directly to. |
7a8e76a3 SR |
3672 | * The user of this interface will need to get the body to write into |
3673 | * and can use the ring_buffer_event_data() interface. | |
3674 | * | |
3675 | * The length is the length of the data needed, not the event length | |
3676 | * which also includes the event header. | |
3677 | * | |
3678 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
3679 | * If NULL is returned, then nothing has been allocated or locked. | |
3680 | */ | |
3681 | struct ring_buffer_event * | |
13292494 | 3682 | ring_buffer_lock_reserve(struct trace_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
3683 | { |
3684 | struct ring_buffer_per_cpu *cpu_buffer; | |
3685 | struct ring_buffer_event *event; | |
5168ae50 | 3686 | int cpu; |
7a8e76a3 | 3687 | |
bf41a158 | 3688 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 3689 | preempt_disable_notrace(); |
bf41a158 | 3690 | |
3205f806 | 3691 | if (unlikely(atomic_read(&buffer->record_disabled))) |
58a09ec6 | 3692 | goto out; |
261842b7 | 3693 | |
7a8e76a3 SR |
3694 | cpu = raw_smp_processor_id(); |
3695 | ||
3205f806 | 3696 | if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask))) |
d769041f | 3697 | goto out; |
7a8e76a3 SR |
3698 | |
3699 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 3700 | |
3205f806 | 3701 | if (unlikely(atomic_read(&cpu_buffer->record_disabled))) |
d769041f | 3702 | goto out; |
7a8e76a3 | 3703 | |
139f8400 | 3704 | if (unlikely(length > buffer->max_data_size)) |
bf41a158 | 3705 | goto out; |
7a8e76a3 | 3706 | |
58a09ec6 SRRH |
3707 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
3708 | goto out; | |
3709 | ||
62f0b3eb | 3710 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 3711 | if (!event) |
58a09ec6 | 3712 | goto out_unlock; |
7a8e76a3 SR |
3713 | |
3714 | return event; | |
3715 | ||
58a09ec6 SRRH |
3716 | out_unlock: |
3717 | trace_recursive_unlock(cpu_buffer); | |
d769041f | 3718 | out: |
5168ae50 | 3719 | preempt_enable_notrace(); |
7a8e76a3 SR |
3720 | return NULL; |
3721 | } | |
c4f50183 | 3722 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 3723 | |
a1863c21 SR |
3724 | /* |
3725 | * Decrement the entries to the page that an event is on. | |
3726 | * The event does not even need to exist, only the pointer | |
3727 | * to the page it is on. This may only be called before the commit | |
3728 | * takes place. | |
3729 | */ | |
3730 | static inline void | |
3731 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
3732 | struct ring_buffer_event *event) | |
3733 | { | |
3734 | unsigned long addr = (unsigned long)event; | |
3735 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
3736 | struct buffer_page *start; | |
3737 | ||
3cb30911 | 3738 | addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1); |
a1863c21 SR |
3739 | |
3740 | /* Do the likely case first */ | |
3741 | if (likely(bpage->page == (void *)addr)) { | |
3742 | local_dec(&bpage->entries); | |
3743 | return; | |
3744 | } | |
3745 | ||
3746 | /* | |
3747 | * Because the commit page may be on the reader page we | |
3748 | * start with the next page and check the end loop there. | |
3749 | */ | |
6689bed3 | 3750 | rb_inc_page(&bpage); |
a1863c21 SR |
3751 | start = bpage; |
3752 | do { | |
3753 | if (bpage->page == (void *)addr) { | |
3754 | local_dec(&bpage->entries); | |
3755 | return; | |
3756 | } | |
6689bed3 | 3757 | rb_inc_page(&bpage); |
a1863c21 SR |
3758 | } while (bpage != start); |
3759 | ||
3760 | /* commit not part of this buffer?? */ | |
3761 | RB_WARN_ON(cpu_buffer, 1); | |
3762 | } | |
3763 | ||
fa1b47dd | 3764 | /** |
88883490 | 3765 | * ring_buffer_discard_commit - discard an event that has not been committed |
fa1b47dd SR |
3766 | * @buffer: the ring buffer |
3767 | * @event: non committed event to discard | |
3768 | * | |
dc892f73 SR |
3769 | * Sometimes an event that is in the ring buffer needs to be ignored. |
3770 | * This function lets the user discard an event in the ring buffer | |
3771 | * and then that event will not be read later. | |
3772 | * | |
6167c205 | 3773 | * This function only works if it is called before the item has been |
dc892f73 | 3774 | * committed. It will try to free the event from the ring buffer |
fa1b47dd SR |
3775 | * if another event has not been added behind it. |
3776 | * | |
3777 | * If another event has been added behind it, it will set the event | |
3778 | * up as discarded, and perform the commit. | |
3779 | * | |
3780 | * If this function is called, do not call ring_buffer_unlock_commit on | |
3781 | * the event. | |
3782 | */ | |
13292494 | 3783 | void ring_buffer_discard_commit(struct trace_buffer *buffer, |
fa1b47dd SR |
3784 | struct ring_buffer_event *event) |
3785 | { | |
3786 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
3787 | int cpu; |
3788 | ||
3789 | /* The event is discarded regardless */ | |
f3b9aae1 | 3790 | rb_event_discard(event); |
fa1b47dd | 3791 | |
fa743953 SR |
3792 | cpu = smp_processor_id(); |
3793 | cpu_buffer = buffer->buffers[cpu]; | |
3794 | ||
fa1b47dd SR |
3795 | /* |
3796 | * This must only be called if the event has not been | |
3797 | * committed yet. Thus we can assume that preemption | |
3798 | * is still disabled. | |
3799 | */ | |
fa743953 | 3800 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 3801 | |
a1863c21 | 3802 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 3803 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 3804 | goto out; |
fa1b47dd | 3805 | |
fa1b47dd | 3806 | out: |
fa743953 | 3807 | rb_end_commit(cpu_buffer); |
fa1b47dd | 3808 | |
58a09ec6 | 3809 | trace_recursive_unlock(cpu_buffer); |
f3b9aae1 | 3810 | |
5168ae50 | 3811 | preempt_enable_notrace(); |
fa1b47dd SR |
3812 | |
3813 | } | |
3814 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
3815 | ||
7a8e76a3 SR |
3816 | /** |
3817 | * ring_buffer_write - write data to the buffer without reserving | |
3818 | * @buffer: The ring buffer to write to. | |
3819 | * @length: The length of the data being written (excluding the event header) | |
3820 | * @data: The data to write to the buffer. | |
3821 | * | |
3822 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
3823 | * one function. If you already have the data to write to the buffer, it | |
3824 | * may be easier to simply call this function. | |
3825 | * | |
3826 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
3827 | * and not the length of the event which would hold the header. | |
3828 | */ | |
13292494 | 3829 | int ring_buffer_write(struct trace_buffer *buffer, |
01e3e710 DS |
3830 | unsigned long length, |
3831 | void *data) | |
7a8e76a3 SR |
3832 | { |
3833 | struct ring_buffer_per_cpu *cpu_buffer; | |
3834 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
3835 | void *body; |
3836 | int ret = -EBUSY; | |
5168ae50 | 3837 | int cpu; |
7a8e76a3 | 3838 | |
5168ae50 | 3839 | preempt_disable_notrace(); |
bf41a158 | 3840 | |
52fbe9cd LJ |
3841 | if (atomic_read(&buffer->record_disabled)) |
3842 | goto out; | |
3843 | ||
7a8e76a3 SR |
3844 | cpu = raw_smp_processor_id(); |
3845 | ||
9e01c1b7 | 3846 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 3847 | goto out; |
7a8e76a3 SR |
3848 | |
3849 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
3850 | |
3851 | if (atomic_read(&cpu_buffer->record_disabled)) | |
3852 | goto out; | |
3853 | ||
139f8400 | 3854 | if (length > buffer->max_data_size) |
be957c44 SR |
3855 | goto out; |
3856 | ||
985e871b SRRH |
3857 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
3858 | goto out; | |
3859 | ||
62f0b3eb | 3860 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 3861 | if (!event) |
985e871b | 3862 | goto out_unlock; |
7a8e76a3 SR |
3863 | |
3864 | body = rb_event_data(event); | |
3865 | ||
3866 | memcpy(body, data, length); | |
3867 | ||
04aabc32 | 3868 | rb_commit(cpu_buffer); |
7a8e76a3 | 3869 | |
15693458 SRRH |
3870 | rb_wakeups(buffer, cpu_buffer); |
3871 | ||
7a8e76a3 | 3872 | ret = 0; |
985e871b SRRH |
3873 | |
3874 | out_unlock: | |
3875 | trace_recursive_unlock(cpu_buffer); | |
3876 | ||
7a8e76a3 | 3877 | out: |
5168ae50 | 3878 | preempt_enable_notrace(); |
7a8e76a3 SR |
3879 | |
3880 | return ret; | |
3881 | } | |
c4f50183 | 3882 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 3883 | |
da58834c | 3884 | static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
3885 | { |
3886 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 3887 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
3888 | struct buffer_page *commit = cpu_buffer->commit_page; |
3889 | ||
77ae365e SR |
3890 | /* In case of error, head will be NULL */ |
3891 | if (unlikely(!head)) | |
da58834c | 3892 | return true; |
77ae365e | 3893 | |
67f0d6d9 HL |
3894 | /* Reader should exhaust content in reader page */ |
3895 | if (reader->read != rb_page_commit(reader)) | |
3896 | return false; | |
3897 | ||
3898 | /* | |
3899 | * If writers are committing on the reader page, knowing all | |
3900 | * committed content has been read, the ring buffer is empty. | |
3901 | */ | |
3902 | if (commit == reader) | |
3903 | return true; | |
3904 | ||
3905 | /* | |
3906 | * If writers are committing on a page other than reader page | |
3907 | * and head page, there should always be content to read. | |
3908 | */ | |
3909 | if (commit != head) | |
3910 | return false; | |
3911 | ||
3912 | /* | |
3913 | * Writers are committing on the head page, we just need | |
3914 | * to care about there're committed data, and the reader will | |
3915 | * swap reader page with head page when it is to read data. | |
3916 | */ | |
3917 | return rb_page_commit(commit) == 0; | |
bf41a158 SR |
3918 | } |
3919 | ||
7a8e76a3 SR |
3920 | /** |
3921 | * ring_buffer_record_disable - stop all writes into the buffer | |
3922 | * @buffer: The ring buffer to stop writes to. | |
3923 | * | |
3924 | * This prevents all writes to the buffer. Any attempt to write | |
3925 | * to the buffer after this will fail and return NULL. | |
3926 | * | |
74401729 | 3927 | * The caller should call synchronize_rcu() after this. |
7a8e76a3 | 3928 | */ |
13292494 | 3929 | void ring_buffer_record_disable(struct trace_buffer *buffer) |
7a8e76a3 SR |
3930 | { |
3931 | atomic_inc(&buffer->record_disabled); | |
3932 | } | |
c4f50183 | 3933 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
3934 | |
3935 | /** | |
3936 | * ring_buffer_record_enable - enable writes to the buffer | |
3937 | * @buffer: The ring buffer to enable writes | |
3938 | * | |
3939 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3940 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 | 3941 | */ |
13292494 | 3942 | void ring_buffer_record_enable(struct trace_buffer *buffer) |
7a8e76a3 SR |
3943 | { |
3944 | atomic_dec(&buffer->record_disabled); | |
3945 | } | |
c4f50183 | 3946 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 3947 | |
499e5470 SR |
3948 | /** |
3949 | * ring_buffer_record_off - stop all writes into the buffer | |
3950 | * @buffer: The ring buffer to stop writes to. | |
3951 | * | |
3952 | * This prevents all writes to the buffer. Any attempt to write | |
3953 | * to the buffer after this will fail and return NULL. | |
3954 | * | |
3955 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 3956 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
3957 | * must be paired with a enable(). |
3958 | */ | |
13292494 | 3959 | void ring_buffer_record_off(struct trace_buffer *buffer) |
499e5470 SR |
3960 | { |
3961 | unsigned int rd; | |
3962 | unsigned int new_rd; | |
3963 | ||
8328e36d | 3964 | rd = atomic_read(&buffer->record_disabled); |
499e5470 | 3965 | do { |
499e5470 | 3966 | new_rd = rd | RB_BUFFER_OFF; |
8328e36d | 3967 | } while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd)); |
499e5470 SR |
3968 | } |
3969 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
3970 | ||
3971 | /** | |
3972 | * ring_buffer_record_on - restart writes into the buffer | |
3973 | * @buffer: The ring buffer to start writes to. | |
3974 | * | |
3975 | * This enables all writes to the buffer that was disabled by | |
3976 | * ring_buffer_record_off(). | |
3977 | * | |
3978 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 3979 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
3980 | * must be paired with a disable(). |
3981 | */ | |
13292494 | 3982 | void ring_buffer_record_on(struct trace_buffer *buffer) |
499e5470 SR |
3983 | { |
3984 | unsigned int rd; | |
3985 | unsigned int new_rd; | |
3986 | ||
8328e36d | 3987 | rd = atomic_read(&buffer->record_disabled); |
499e5470 | 3988 | do { |
499e5470 | 3989 | new_rd = rd & ~RB_BUFFER_OFF; |
8328e36d | 3990 | } while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd)); |
499e5470 SR |
3991 | } |
3992 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
3993 | ||
3994 | /** | |
3995 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
3996 | * @buffer: The ring buffer to see if write is enabled | |
3997 | * | |
3998 | * Returns true if the ring buffer is in a state that it accepts writes. | |
3999 | */ | |
13292494 | 4000 | bool ring_buffer_record_is_on(struct trace_buffer *buffer) |
499e5470 SR |
4001 | { |
4002 | return !atomic_read(&buffer->record_disabled); | |
4003 | } | |
4004 | ||
73c8d894 MH |
4005 | /** |
4006 | * ring_buffer_record_is_set_on - return true if the ring buffer is set writable | |
4007 | * @buffer: The ring buffer to see if write is set enabled | |
4008 | * | |
4009 | * Returns true if the ring buffer is set writable by ring_buffer_record_on(). | |
4010 | * Note that this does NOT mean it is in a writable state. | |
4011 | * | |
4012 | * It may return true when the ring buffer has been disabled by | |
4013 | * ring_buffer_record_disable(), as that is a temporary disabling of | |
4014 | * the ring buffer. | |
4015 | */ | |
13292494 | 4016 | bool ring_buffer_record_is_set_on(struct trace_buffer *buffer) |
73c8d894 MH |
4017 | { |
4018 | return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF); | |
4019 | } | |
4020 | ||
7a8e76a3 SR |
4021 | /** |
4022 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
4023 | * @buffer: The ring buffer to stop writes to. | |
4024 | * @cpu: The CPU buffer to stop | |
4025 | * | |
4026 | * This prevents all writes to the buffer. Any attempt to write | |
4027 | * to the buffer after this will fail and return NULL. | |
4028 | * | |
74401729 | 4029 | * The caller should call synchronize_rcu() after this. |
7a8e76a3 | 4030 | */ |
13292494 | 4031 | void ring_buffer_record_disable_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
4032 | { |
4033 | struct ring_buffer_per_cpu *cpu_buffer; | |
4034 | ||
9e01c1b7 | 4035 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4036 | return; |
7a8e76a3 SR |
4037 | |
4038 | cpu_buffer = buffer->buffers[cpu]; | |
4039 | atomic_inc(&cpu_buffer->record_disabled); | |
4040 | } | |
c4f50183 | 4041 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
4042 | |
4043 | /** | |
4044 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
4045 | * @buffer: The ring buffer to enable writes | |
4046 | * @cpu: The CPU to enable. | |
4047 | * | |
4048 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 4049 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 | 4050 | */ |
13292494 | 4051 | void ring_buffer_record_enable_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
4052 | { |
4053 | struct ring_buffer_per_cpu *cpu_buffer; | |
4054 | ||
9e01c1b7 | 4055 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4056 | return; |
7a8e76a3 SR |
4057 | |
4058 | cpu_buffer = buffer->buffers[cpu]; | |
4059 | atomic_dec(&cpu_buffer->record_disabled); | |
4060 | } | |
c4f50183 | 4061 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 4062 | |
f6195aa0 SR |
4063 | /* |
4064 | * The total entries in the ring buffer is the running counter | |
4065 | * of entries entered into the ring buffer, minus the sum of | |
4066 | * the entries read from the ring buffer and the number of | |
4067 | * entries that were overwritten. | |
4068 | */ | |
4069 | static inline unsigned long | |
4070 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
4071 | { | |
4072 | return local_read(&cpu_buffer->entries) - | |
4073 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
4074 | } | |
4075 | ||
c64e148a VN |
4076 | /** |
4077 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
4078 | * @buffer: The ring buffer | |
4079 | * @cpu: The per CPU buffer to read from. | |
4080 | */ | |
13292494 | 4081 | u64 ring_buffer_oldest_event_ts(struct trace_buffer *buffer, int cpu) |
c64e148a VN |
4082 | { |
4083 | unsigned long flags; | |
4084 | struct ring_buffer_per_cpu *cpu_buffer; | |
4085 | struct buffer_page *bpage; | |
da830e58 | 4086 | u64 ret = 0; |
c64e148a VN |
4087 | |
4088 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4089 | return 0; | |
4090 | ||
4091 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 4092 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
4093 | /* |
4094 | * if the tail is on reader_page, oldest time stamp is on the reader | |
4095 | * page | |
4096 | */ | |
4097 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
4098 | bpage = cpu_buffer->reader_page; | |
4099 | else | |
4100 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
4101 | if (bpage) |
4102 | ret = bpage->page->time_stamp; | |
7115e3fc | 4103 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
4104 | |
4105 | return ret; | |
4106 | } | |
4107 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
4108 | ||
4109 | /** | |
45d99ea4 | 4110 | * ring_buffer_bytes_cpu - get the number of bytes unconsumed in a cpu buffer |
c64e148a VN |
4111 | * @buffer: The ring buffer |
4112 | * @cpu: The per CPU buffer to read from. | |
4113 | */ | |
13292494 | 4114 | unsigned long ring_buffer_bytes_cpu(struct trace_buffer *buffer, int cpu) |
c64e148a VN |
4115 | { |
4116 | struct ring_buffer_per_cpu *cpu_buffer; | |
4117 | unsigned long ret; | |
4118 | ||
4119 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4120 | return 0; | |
4121 | ||
4122 | cpu_buffer = buffer->buffers[cpu]; | |
4123 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
4124 | ||
4125 | return ret; | |
4126 | } | |
4127 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
4128 | ||
7a8e76a3 SR |
4129 | /** |
4130 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
4131 | * @buffer: The ring buffer | |
4132 | * @cpu: The per CPU buffer to get the entries from. | |
4133 | */ | |
13292494 | 4134 | unsigned long ring_buffer_entries_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
4135 | { |
4136 | struct ring_buffer_per_cpu *cpu_buffer; | |
4137 | ||
9e01c1b7 | 4138 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4139 | return 0; |
7a8e76a3 SR |
4140 | |
4141 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 4142 | |
f6195aa0 | 4143 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 4144 | } |
c4f50183 | 4145 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
4146 | |
4147 | /** | |
884bfe89 SP |
4148 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
4149 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
4150 | * @buffer: The ring buffer |
4151 | * @cpu: The per CPU buffer to get the number of overruns from | |
4152 | */ | |
13292494 | 4153 | unsigned long ring_buffer_overrun_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
4154 | { |
4155 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 4156 | unsigned long ret; |
7a8e76a3 | 4157 | |
9e01c1b7 | 4158 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4159 | return 0; |
7a8e76a3 SR |
4160 | |
4161 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 4162 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
4163 | |
4164 | return ret; | |
7a8e76a3 | 4165 | } |
c4f50183 | 4166 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 4167 | |
f0d2c681 | 4168 | /** |
884bfe89 SP |
4169 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
4170 | * commits failing due to the buffer wrapping around while there are uncommitted | |
4171 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
4172 | * @buffer: The ring buffer |
4173 | * @cpu: The per CPU buffer to get the number of overruns from | |
4174 | */ | |
4175 | unsigned long | |
13292494 | 4176 | ring_buffer_commit_overrun_cpu(struct trace_buffer *buffer, int cpu) |
f0d2c681 SR |
4177 | { |
4178 | struct ring_buffer_per_cpu *cpu_buffer; | |
4179 | unsigned long ret; | |
4180 | ||
4181 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4182 | return 0; | |
4183 | ||
4184 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 4185 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
4186 | |
4187 | return ret; | |
4188 | } | |
4189 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
4190 | ||
884bfe89 SP |
4191 | /** |
4192 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
4193 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
4194 | * @buffer: The ring buffer | |
4195 | * @cpu: The per CPU buffer to get the number of overruns from | |
4196 | */ | |
4197 | unsigned long | |
13292494 | 4198 | ring_buffer_dropped_events_cpu(struct trace_buffer *buffer, int cpu) |
884bfe89 SP |
4199 | { |
4200 | struct ring_buffer_per_cpu *cpu_buffer; | |
4201 | unsigned long ret; | |
4202 | ||
4203 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4204 | return 0; | |
4205 | ||
4206 | cpu_buffer = buffer->buffers[cpu]; | |
4207 | ret = local_read(&cpu_buffer->dropped_events); | |
4208 | ||
4209 | return ret; | |
4210 | } | |
4211 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
4212 | ||
ad964704 SRRH |
4213 | /** |
4214 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
4215 | * @buffer: The ring buffer | |
4216 | * @cpu: The per CPU buffer to get the number of events read | |
4217 | */ | |
4218 | unsigned long | |
13292494 | 4219 | ring_buffer_read_events_cpu(struct trace_buffer *buffer, int cpu) |
ad964704 SRRH |
4220 | { |
4221 | struct ring_buffer_per_cpu *cpu_buffer; | |
4222 | ||
4223 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4224 | return 0; | |
4225 | ||
4226 | cpu_buffer = buffer->buffers[cpu]; | |
4227 | return cpu_buffer->read; | |
4228 | } | |
4229 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
4230 | ||
7a8e76a3 SR |
4231 | /** |
4232 | * ring_buffer_entries - get the number of entries in a buffer | |
4233 | * @buffer: The ring buffer | |
4234 | * | |
4235 | * Returns the total number of entries in the ring buffer | |
4236 | * (all CPU entries) | |
4237 | */ | |
13292494 | 4238 | unsigned long ring_buffer_entries(struct trace_buffer *buffer) |
7a8e76a3 SR |
4239 | { |
4240 | struct ring_buffer_per_cpu *cpu_buffer; | |
4241 | unsigned long entries = 0; | |
4242 | int cpu; | |
4243 | ||
4244 | /* if you care about this being correct, lock the buffer */ | |
4245 | for_each_buffer_cpu(buffer, cpu) { | |
4246 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 4247 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
4248 | } |
4249 | ||
4250 | return entries; | |
4251 | } | |
c4f50183 | 4252 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
4253 | |
4254 | /** | |
67b394f7 | 4255 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
4256 | * @buffer: The ring buffer |
4257 | * | |
4258 | * Returns the total number of overruns in the ring buffer | |
4259 | * (all CPU entries) | |
4260 | */ | |
13292494 | 4261 | unsigned long ring_buffer_overruns(struct trace_buffer *buffer) |
7a8e76a3 SR |
4262 | { |
4263 | struct ring_buffer_per_cpu *cpu_buffer; | |
4264 | unsigned long overruns = 0; | |
4265 | int cpu; | |
4266 | ||
4267 | /* if you care about this being correct, lock the buffer */ | |
4268 | for_each_buffer_cpu(buffer, cpu) { | |
4269 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 4270 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
4271 | } |
4272 | ||
4273 | return overruns; | |
4274 | } | |
c4f50183 | 4275 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 4276 | |
642edba5 | 4277 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
4278 | { |
4279 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
4280 | ||
d769041f | 4281 | /* Iterator usage is expected to have record disabled */ |
651e22f2 SRRH |
4282 | iter->head_page = cpu_buffer->reader_page; |
4283 | iter->head = cpu_buffer->reader_page->read; | |
785888c5 | 4284 | iter->next_event = iter->head; |
651e22f2 SRRH |
4285 | |
4286 | iter->cache_reader_page = iter->head_page; | |
24607f11 | 4287 | iter->cache_read = cpu_buffer->read; |
2d093282 | 4288 | iter->cache_pages_removed = cpu_buffer->pages_removed; |
651e22f2 | 4289 | |
28e3fc56 | 4290 | if (iter->head) { |
d769041f | 4291 | iter->read_stamp = cpu_buffer->read_stamp; |
28e3fc56 SRV |
4292 | iter->page_stamp = cpu_buffer->reader_page->page->time_stamp; |
4293 | } else { | |
abc9b56d | 4294 | iter->read_stamp = iter->head_page->page->time_stamp; |
28e3fc56 SRV |
4295 | iter->page_stamp = iter->read_stamp; |
4296 | } | |
642edba5 | 4297 | } |
f83c9d0f | 4298 | |
642edba5 SR |
4299 | /** |
4300 | * ring_buffer_iter_reset - reset an iterator | |
4301 | * @iter: The iterator to reset | |
4302 | * | |
4303 | * Resets the iterator, so that it will start from the beginning | |
4304 | * again. | |
4305 | */ | |
4306 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
4307 | { | |
554f786e | 4308 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
4309 | unsigned long flags; |
4310 | ||
554f786e SR |
4311 | if (!iter) |
4312 | return; | |
4313 | ||
4314 | cpu_buffer = iter->cpu_buffer; | |
4315 | ||
5389f6fa | 4316 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 4317 | rb_iter_reset(iter); |
5389f6fa | 4318 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 4319 | } |
c4f50183 | 4320 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
4321 | |
4322 | /** | |
4323 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
4324 | * @iter: The iterator to check | |
4325 | */ | |
4326 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
4327 | { | |
4328 | struct ring_buffer_per_cpu *cpu_buffer; | |
78f7a45d SRV |
4329 | struct buffer_page *reader; |
4330 | struct buffer_page *head_page; | |
4331 | struct buffer_page *commit_page; | |
ead6ecfd | 4332 | struct buffer_page *curr_commit_page; |
78f7a45d | 4333 | unsigned commit; |
ead6ecfd SRV |
4334 | u64 curr_commit_ts; |
4335 | u64 commit_ts; | |
7a8e76a3 SR |
4336 | |
4337 | cpu_buffer = iter->cpu_buffer; | |
78f7a45d SRV |
4338 | reader = cpu_buffer->reader_page; |
4339 | head_page = cpu_buffer->head_page; | |
4340 | commit_page = cpu_buffer->commit_page; | |
ead6ecfd SRV |
4341 | commit_ts = commit_page->page->time_stamp; |
4342 | ||
4343 | /* | |
4344 | * When the writer goes across pages, it issues a cmpxchg which | |
4345 | * is a mb(), which will synchronize with the rmb here. | |
4346 | * (see rb_tail_page_update()) | |
4347 | */ | |
4348 | smp_rmb(); | |
78f7a45d | 4349 | commit = rb_page_commit(commit_page); |
ead6ecfd SRV |
4350 | /* We want to make sure that the commit page doesn't change */ |
4351 | smp_rmb(); | |
4352 | ||
4353 | /* Make sure commit page didn't change */ | |
4354 | curr_commit_page = READ_ONCE(cpu_buffer->commit_page); | |
4355 | curr_commit_ts = READ_ONCE(curr_commit_page->page->time_stamp); | |
4356 | ||
4357 | /* If the commit page changed, then there's more data */ | |
4358 | if (curr_commit_page != commit_page || | |
4359 | curr_commit_ts != commit_ts) | |
4360 | return 0; | |
78f7a45d | 4361 | |
ead6ecfd | 4362 | /* Still racy, as it may return a false positive, but that's OK */ |
785888c5 | 4363 | return ((iter->head_page == commit_page && iter->head >= commit) || |
78f7a45d SRV |
4364 | (iter->head_page == reader && commit_page == head_page && |
4365 | head_page->read == commit && | |
4366 | iter->head == rb_page_commit(cpu_buffer->reader_page))); | |
7a8e76a3 | 4367 | } |
c4f50183 | 4368 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
4369 | |
4370 | static void | |
4371 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
4372 | struct ring_buffer_event *event) | |
4373 | { | |
4374 | u64 delta; | |
4375 | ||
334d4169 | 4376 | switch (event->type_len) { |
7a8e76a3 SR |
4377 | case RINGBUF_TYPE_PADDING: |
4378 | return; | |
4379 | ||
4380 | case RINGBUF_TYPE_TIME_EXTEND: | |
e20044f7 | 4381 | delta = rb_event_time_stamp(event); |
7a8e76a3 SR |
4382 | cpu_buffer->read_stamp += delta; |
4383 | return; | |
4384 | ||
4385 | case RINGBUF_TYPE_TIME_STAMP: | |
e20044f7 | 4386 | delta = rb_event_time_stamp(event); |
6695da58 | 4387 | delta = rb_fix_abs_ts(delta, cpu_buffer->read_stamp); |
dc4e2801 | 4388 | cpu_buffer->read_stamp = delta; |
7a8e76a3 SR |
4389 | return; |
4390 | ||
4391 | case RINGBUF_TYPE_DATA: | |
4392 | cpu_buffer->read_stamp += event->time_delta; | |
4393 | return; | |
4394 | ||
4395 | default: | |
da4d401a | 4396 | RB_WARN_ON(cpu_buffer, 1); |
7a8e76a3 | 4397 | } |
7a8e76a3 SR |
4398 | } |
4399 | ||
4400 | static void | |
4401 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
4402 | struct ring_buffer_event *event) | |
4403 | { | |
4404 | u64 delta; | |
4405 | ||
334d4169 | 4406 | switch (event->type_len) { |
7a8e76a3 SR |
4407 | case RINGBUF_TYPE_PADDING: |
4408 | return; | |
4409 | ||
4410 | case RINGBUF_TYPE_TIME_EXTEND: | |
e20044f7 | 4411 | delta = rb_event_time_stamp(event); |
7a8e76a3 SR |
4412 | iter->read_stamp += delta; |
4413 | return; | |
4414 | ||
4415 | case RINGBUF_TYPE_TIME_STAMP: | |
e20044f7 | 4416 | delta = rb_event_time_stamp(event); |
6695da58 | 4417 | delta = rb_fix_abs_ts(delta, iter->read_stamp); |
dc4e2801 | 4418 | iter->read_stamp = delta; |
7a8e76a3 SR |
4419 | return; |
4420 | ||
4421 | case RINGBUF_TYPE_DATA: | |
4422 | iter->read_stamp += event->time_delta; | |
4423 | return; | |
4424 | ||
4425 | default: | |
da4d401a | 4426 | RB_WARN_ON(iter->cpu_buffer, 1); |
7a8e76a3 | 4427 | } |
7a8e76a3 SR |
4428 | } |
4429 | ||
d769041f SR |
4430 | static struct buffer_page * |
4431 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 4432 | { |
d769041f | 4433 | struct buffer_page *reader = NULL; |
139f8400 | 4434 | unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size); |
66a8cb95 | 4435 | unsigned long overwrite; |
d769041f | 4436 | unsigned long flags; |
818e3dd3 | 4437 | int nr_loops = 0; |
bc92b956 | 4438 | bool ret; |
d769041f | 4439 | |
3e03fb7f | 4440 | local_irq_save(flags); |
0199c4e6 | 4441 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
4442 | |
4443 | again: | |
818e3dd3 SR |
4444 | /* |
4445 | * This should normally only loop twice. But because the | |
4446 | * start of the reader inserts an empty page, it causes | |
4447 | * a case where we will loop three times. There should be no | |
4448 | * reason to loop four times (that I know of). | |
4449 | */ | |
3e89c7bb | 4450 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
4451 | reader = NULL; |
4452 | goto out; | |
4453 | } | |
4454 | ||
d769041f SR |
4455 | reader = cpu_buffer->reader_page; |
4456 | ||
4457 | /* If there's more to read, return this page */ | |
bf41a158 | 4458 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
4459 | goto out; |
4460 | ||
4461 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
4462 | if (RB_WARN_ON(cpu_buffer, |
4463 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
4464 | goto out; | |
d769041f SR |
4465 | |
4466 | /* check if we caught up to the tail */ | |
4467 | reader = NULL; | |
bf41a158 | 4468 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 4469 | goto out; |
7a8e76a3 | 4470 | |
a5fb8331 SR |
4471 | /* Don't bother swapping if the ring buffer is empty */ |
4472 | if (rb_num_of_entries(cpu_buffer) == 0) | |
4473 | goto out; | |
4474 | ||
7a8e76a3 | 4475 | /* |
d769041f | 4476 | * Reset the reader page to size zero. |
7a8e76a3 | 4477 | */ |
77ae365e SR |
4478 | local_set(&cpu_buffer->reader_page->write, 0); |
4479 | local_set(&cpu_buffer->reader_page->entries, 0); | |
4480 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 4481 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 4482 | |
77ae365e SR |
4483 | spin: |
4484 | /* | |
4485 | * Splice the empty reader page into the list around the head. | |
4486 | */ | |
4487 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
4488 | if (!reader) |
4489 | goto out; | |
0e1ff5d7 | 4490 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 4491 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 4492 | |
3adc54fa SR |
4493 | /* |
4494 | * cpu_buffer->pages just needs to point to the buffer, it | |
4495 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 4496 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
4497 | */ |
4498 | cpu_buffer->pages = reader->list.prev; | |
4499 | ||
77ae365e | 4500 | /* The reader page will be pointing to the new head */ |
6689bed3 | 4501 | rb_set_list_to_head(&cpu_buffer->reader_page->list); |
7a8e76a3 | 4502 | |
66a8cb95 SR |
4503 | /* |
4504 | * We want to make sure we read the overruns after we set up our | |
4505 | * pointers to the next object. The writer side does a | |
4506 | * cmpxchg to cross pages which acts as the mb on the writer | |
4507 | * side. Note, the reader will constantly fail the swap | |
4508 | * while the writer is updating the pointers, so this | |
4509 | * guarantees that the overwrite recorded here is the one we | |
4510 | * want to compare with the last_overrun. | |
4511 | */ | |
4512 | smp_mb(); | |
4513 | overwrite = local_read(&(cpu_buffer->overrun)); | |
4514 | ||
77ae365e SR |
4515 | /* |
4516 | * Here's the tricky part. | |
4517 | * | |
4518 | * We need to move the pointer past the header page. | |
4519 | * But we can only do that if a writer is not currently | |
4520 | * moving it. The page before the header page has the | |
4521 | * flag bit '1' set if it is pointing to the page we want. | |
4522 | * but if the writer is in the process of moving it | |
4523 | * than it will be '2' or already moved '0'. | |
4524 | */ | |
4525 | ||
4526 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
4527 | |
4528 | /* | |
77ae365e | 4529 | * If we did not convert it, then we must try again. |
7a8e76a3 | 4530 | */ |
77ae365e SR |
4531 | if (!ret) |
4532 | goto spin; | |
7a8e76a3 | 4533 | |
77ae365e | 4534 | /* |
2c2b0a78 | 4535 | * Yay! We succeeded in replacing the page. |
77ae365e SR |
4536 | * |
4537 | * Now make the new head point back to the reader page. | |
4538 | */ | |
5ded3dc6 | 4539 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
6689bed3 | 4540 | rb_inc_page(&cpu_buffer->head_page); |
d769041f | 4541 | |
2c2b0a78 SRV |
4542 | local_inc(&cpu_buffer->pages_read); |
4543 | ||
d769041f SR |
4544 | /* Finally update the reader page to the new head */ |
4545 | cpu_buffer->reader_page = reader; | |
b81f472a | 4546 | cpu_buffer->reader_page->read = 0; |
d769041f | 4547 | |
66a8cb95 SR |
4548 | if (overwrite != cpu_buffer->last_overrun) { |
4549 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
4550 | cpu_buffer->last_overrun = overwrite; | |
4551 | } | |
4552 | ||
d769041f SR |
4553 | goto again; |
4554 | ||
4555 | out: | |
b81f472a SRRH |
4556 | /* Update the read_stamp on the first event */ |
4557 | if (reader && reader->read == 0) | |
4558 | cpu_buffer->read_stamp = reader->page->time_stamp; | |
4559 | ||
0199c4e6 | 4560 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 4561 | local_irq_restore(flags); |
d769041f | 4562 | |
a0fcaaed SRG |
4563 | /* |
4564 | * The writer has preempt disable, wait for it. But not forever | |
4565 | * Although, 1 second is pretty much "forever" | |
4566 | */ | |
4567 | #define USECS_WAIT 1000000 | |
4568 | for (nr_loops = 0; nr_loops < USECS_WAIT; nr_loops++) { | |
4569 | /* If the write is past the end of page, a writer is still updating it */ | |
139f8400 | 4570 | if (likely(!reader || rb_page_write(reader) <= bsize)) |
a0fcaaed SRG |
4571 | break; |
4572 | ||
4573 | udelay(1); | |
4574 | ||
4575 | /* Get the latest version of the reader write value */ | |
4576 | smp_rmb(); | |
4577 | } | |
4578 | ||
4579 | /* The writer is not moving forward? Something is wrong */ | |
4580 | if (RB_WARN_ON(cpu_buffer, nr_loops == USECS_WAIT)) | |
4581 | reader = NULL; | |
4582 | ||
4583 | /* | |
4584 | * Make sure we see any padding after the write update | |
6455b616 ZY |
4585 | * (see rb_reset_tail()). |
4586 | * | |
4587 | * In addition, a writer may be writing on the reader page | |
4588 | * if the page has not been fully filled, so the read barrier | |
4589 | * is also needed to make sure we see the content of what is | |
4590 | * committed by the writer (see rb_set_commit_to_write()). | |
a0fcaaed SRG |
4591 | */ |
4592 | smp_rmb(); | |
4593 | ||
4594 | ||
d769041f SR |
4595 | return reader; |
4596 | } | |
4597 | ||
4598 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
4599 | { | |
4600 | struct ring_buffer_event *event; | |
4601 | struct buffer_page *reader; | |
4602 | unsigned length; | |
4603 | ||
4604 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 4605 | |
d769041f | 4606 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
4607 | if (RB_WARN_ON(cpu_buffer, !reader)) |
4608 | return; | |
7a8e76a3 | 4609 | |
d769041f SR |
4610 | event = rb_reader_event(cpu_buffer); |
4611 | ||
a1863c21 | 4612 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 4613 | cpu_buffer->read++; |
d769041f SR |
4614 | |
4615 | rb_update_read_stamp(cpu_buffer, event); | |
4616 | ||
4617 | length = rb_event_length(event); | |
6f807acd | 4618 | cpu_buffer->reader_page->read += length; |
45d99ea4 | 4619 | cpu_buffer->read_bytes += length; |
7a8e76a3 SR |
4620 | } |
4621 | ||
4622 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
4623 | { | |
7a8e76a3 | 4624 | struct ring_buffer_per_cpu *cpu_buffer; |
7a8e76a3 SR |
4625 | |
4626 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 4627 | |
785888c5 SRV |
4628 | /* If head == next_event then we need to jump to the next event */ |
4629 | if (iter->head == iter->next_event) { | |
4630 | /* If the event gets overwritten again, there's nothing to do */ | |
4631 | if (rb_iter_head_event(iter) == NULL) | |
4632 | return; | |
4633 | } | |
4634 | ||
4635 | iter->head = iter->next_event; | |
4636 | ||
7a8e76a3 SR |
4637 | /* |
4638 | * Check if we are at the end of the buffer. | |
4639 | */ | |
785888c5 | 4640 | if (iter->next_event >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
4641 | /* discarded commits can make the page empty */ |
4642 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 4643 | return; |
d769041f | 4644 | rb_inc_iter(iter); |
7a8e76a3 SR |
4645 | return; |
4646 | } | |
4647 | ||
785888c5 | 4648 | rb_update_iter_read_stamp(iter, iter->event); |
7a8e76a3 SR |
4649 | } |
4650 | ||
66a8cb95 SR |
4651 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
4652 | { | |
4653 | return cpu_buffer->lost_events; | |
4654 | } | |
4655 | ||
f83c9d0f | 4656 | static struct ring_buffer_event * |
66a8cb95 SR |
4657 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
4658 | unsigned long *lost_events) | |
7a8e76a3 | 4659 | { |
7a8e76a3 | 4660 | struct ring_buffer_event *event; |
d769041f | 4661 | struct buffer_page *reader; |
818e3dd3 | 4662 | int nr_loops = 0; |
7a8e76a3 | 4663 | |
dc4e2801 TZ |
4664 | if (ts) |
4665 | *ts = 0; | |
7a8e76a3 | 4666 | again: |
818e3dd3 | 4667 | /* |
69d1b839 SR |
4668 | * We repeat when a time extend is encountered. |
4669 | * Since the time extend is always attached to a data event, | |
4670 | * we should never loop more than once. | |
4671 | * (We never hit the following condition more than twice). | |
818e3dd3 | 4672 | */ |
69d1b839 | 4673 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 4674 | return NULL; |
818e3dd3 | 4675 | |
d769041f SR |
4676 | reader = rb_get_reader_page(cpu_buffer); |
4677 | if (!reader) | |
7a8e76a3 SR |
4678 | return NULL; |
4679 | ||
d769041f | 4680 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 4681 | |
334d4169 | 4682 | switch (event->type_len) { |
7a8e76a3 | 4683 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
4684 | if (rb_null_event(event)) |
4685 | RB_WARN_ON(cpu_buffer, 1); | |
4686 | /* | |
4687 | * Because the writer could be discarding every | |
4688 | * event it creates (which would probably be bad) | |
4689 | * if we were to go back to "again" then we may never | |
4690 | * catch up, and will trigger the warn on, or lock | |
4691 | * the box. Return the padding, and we will release | |
4692 | * the current locks, and try again. | |
4693 | */ | |
2d622719 | 4694 | return event; |
7a8e76a3 SR |
4695 | |
4696 | case RINGBUF_TYPE_TIME_EXTEND: | |
4697 | /* Internal data, OK to advance */ | |
d769041f | 4698 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
4699 | goto again; |
4700 | ||
4701 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 | 4702 | if (ts) { |
e20044f7 | 4703 | *ts = rb_event_time_stamp(event); |
6695da58 | 4704 | *ts = rb_fix_abs_ts(*ts, reader->page->time_stamp); |
dc4e2801 TZ |
4705 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
4706 | cpu_buffer->cpu, ts); | |
4707 | } | |
4708 | /* Internal data, OK to advance */ | |
d769041f | 4709 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
4710 | goto again; |
4711 | ||
4712 | case RINGBUF_TYPE_DATA: | |
dc4e2801 | 4713 | if (ts && !(*ts)) { |
7a8e76a3 | 4714 | *ts = cpu_buffer->read_stamp + event->time_delta; |
d8eeb2d3 | 4715 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 4716 | cpu_buffer->cpu, ts); |
7a8e76a3 | 4717 | } |
66a8cb95 SR |
4718 | if (lost_events) |
4719 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
4720 | return event; |
4721 | ||
4722 | default: | |
da4d401a | 4723 | RB_WARN_ON(cpu_buffer, 1); |
7a8e76a3 SR |
4724 | } |
4725 | ||
4726 | return NULL; | |
4727 | } | |
c4f50183 | 4728 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 4729 | |
f83c9d0f SR |
4730 | static struct ring_buffer_event * |
4731 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 | 4732 | { |
13292494 | 4733 | struct trace_buffer *buffer; |
7a8e76a3 SR |
4734 | struct ring_buffer_per_cpu *cpu_buffer; |
4735 | struct ring_buffer_event *event; | |
818e3dd3 | 4736 | int nr_loops = 0; |
7a8e76a3 | 4737 | |
dc4e2801 TZ |
4738 | if (ts) |
4739 | *ts = 0; | |
4740 | ||
7a8e76a3 SR |
4741 | cpu_buffer = iter->cpu_buffer; |
4742 | buffer = cpu_buffer->buffer; | |
4743 | ||
492a74f4 | 4744 | /* |
2d093282 ZY |
4745 | * Check if someone performed a consuming read to the buffer |
4746 | * or removed some pages from the buffer. In these cases, | |
4747 | * iterator was invalidated and we need to reset it. | |
492a74f4 SR |
4748 | */ |
4749 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
2d093282 ZY |
4750 | iter->cache_reader_page != cpu_buffer->reader_page || |
4751 | iter->cache_pages_removed != cpu_buffer->pages_removed)) | |
492a74f4 SR |
4752 | rb_iter_reset(iter); |
4753 | ||
7a8e76a3 | 4754 | again: |
3c05d748 SR |
4755 | if (ring_buffer_iter_empty(iter)) |
4756 | return NULL; | |
4757 | ||
818e3dd3 | 4758 | /* |
3d2353de SRV |
4759 | * As the writer can mess with what the iterator is trying |
4760 | * to read, just give up if we fail to get an event after | |
4761 | * three tries. The iterator is not as reliable when reading | |
4762 | * the ring buffer with an active write as the consumer is. | |
4763 | * Do not warn if the three failures is reached. | |
818e3dd3 | 4764 | */ |
3d2353de | 4765 | if (++nr_loops > 3) |
818e3dd3 | 4766 | return NULL; |
818e3dd3 | 4767 | |
7a8e76a3 SR |
4768 | if (rb_per_cpu_empty(cpu_buffer)) |
4769 | return NULL; | |
4770 | ||
10e83fd0 | 4771 | if (iter->head >= rb_page_size(iter->head_page)) { |
3c05d748 SR |
4772 | rb_inc_iter(iter); |
4773 | goto again; | |
4774 | } | |
4775 | ||
7a8e76a3 | 4776 | event = rb_iter_head_event(iter); |
3d2353de | 4777 | if (!event) |
785888c5 | 4778 | goto again; |
7a8e76a3 | 4779 | |
334d4169 | 4780 | switch (event->type_len) { |
7a8e76a3 | 4781 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
4782 | if (rb_null_event(event)) { |
4783 | rb_inc_iter(iter); | |
4784 | goto again; | |
4785 | } | |
4786 | rb_advance_iter(iter); | |
4787 | return event; | |
7a8e76a3 SR |
4788 | |
4789 | case RINGBUF_TYPE_TIME_EXTEND: | |
4790 | /* Internal data, OK to advance */ | |
4791 | rb_advance_iter(iter); | |
4792 | goto again; | |
4793 | ||
4794 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 | 4795 | if (ts) { |
e20044f7 | 4796 | *ts = rb_event_time_stamp(event); |
6695da58 | 4797 | *ts = rb_fix_abs_ts(*ts, iter->head_page->page->time_stamp); |
dc4e2801 TZ |
4798 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
4799 | cpu_buffer->cpu, ts); | |
4800 | } | |
4801 | /* Internal data, OK to advance */ | |
7a8e76a3 SR |
4802 | rb_advance_iter(iter); |
4803 | goto again; | |
4804 | ||
4805 | case RINGBUF_TYPE_DATA: | |
dc4e2801 | 4806 | if (ts && !(*ts)) { |
7a8e76a3 | 4807 | *ts = iter->read_stamp + event->time_delta; |
37886f6a SR |
4808 | ring_buffer_normalize_time_stamp(buffer, |
4809 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
4810 | } |
4811 | return event; | |
4812 | ||
4813 | default: | |
da4d401a | 4814 | RB_WARN_ON(cpu_buffer, 1); |
7a8e76a3 SR |
4815 | } |
4816 | ||
4817 | return NULL; | |
4818 | } | |
c4f50183 | 4819 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 4820 | |
289a5a25 | 4821 | static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer) |
8d707e8e | 4822 | { |
289a5a25 SRRH |
4823 | if (likely(!in_nmi())) { |
4824 | raw_spin_lock(&cpu_buffer->reader_lock); | |
4825 | return true; | |
4826 | } | |
4827 | ||
8d707e8e SR |
4828 | /* |
4829 | * If an NMI die dumps out the content of the ring buffer | |
289a5a25 SRRH |
4830 | * trylock must be used to prevent a deadlock if the NMI |
4831 | * preempted a task that holds the ring buffer locks. If | |
4832 | * we get the lock then all is fine, if not, then continue | |
4833 | * to do the read, but this can corrupt the ring buffer, | |
4834 | * so it must be permanently disabled from future writes. | |
4835 | * Reading from NMI is a oneshot deal. | |
8d707e8e | 4836 | */ |
289a5a25 SRRH |
4837 | if (raw_spin_trylock(&cpu_buffer->reader_lock)) |
4838 | return true; | |
8d707e8e | 4839 | |
289a5a25 SRRH |
4840 | /* Continue without locking, but disable the ring buffer */ |
4841 | atomic_inc(&cpu_buffer->record_disabled); | |
4842 | return false; | |
4843 | } | |
4844 | ||
4845 | static inline void | |
4846 | rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked) | |
4847 | { | |
4848 | if (likely(locked)) | |
4849 | raw_spin_unlock(&cpu_buffer->reader_lock); | |
8d707e8e SR |
4850 | } |
4851 | ||
f83c9d0f SR |
4852 | /** |
4853 | * ring_buffer_peek - peek at the next event to be read | |
4854 | * @buffer: The ring buffer to read | |
4855 | * @cpu: The cpu to peak at | |
4856 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 4857 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
4858 | * |
4859 | * This will return the event that will be read next, but does | |
4860 | * not consume the data. | |
4861 | */ | |
4862 | struct ring_buffer_event * | |
13292494 | 4863 | ring_buffer_peek(struct trace_buffer *buffer, int cpu, u64 *ts, |
66a8cb95 | 4864 | unsigned long *lost_events) |
f83c9d0f SR |
4865 | { |
4866 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 4867 | struct ring_buffer_event *event; |
f83c9d0f | 4868 | unsigned long flags; |
289a5a25 | 4869 | bool dolock; |
f83c9d0f | 4870 | |
554f786e | 4871 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4872 | return NULL; |
554f786e | 4873 | |
2d622719 | 4874 | again: |
8d707e8e | 4875 | local_irq_save(flags); |
289a5a25 | 4876 | dolock = rb_reader_lock(cpu_buffer); |
66a8cb95 | 4877 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
4878 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
4879 | rb_advance_reader(cpu_buffer); | |
289a5a25 | 4880 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4881 | local_irq_restore(flags); |
f83c9d0f | 4882 | |
1b959e18 | 4883 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4884 | goto again; |
2d622719 | 4885 | |
f83c9d0f SR |
4886 | return event; |
4887 | } | |
4888 | ||
c9b7a4a7 SRV |
4889 | /** ring_buffer_iter_dropped - report if there are dropped events |
4890 | * @iter: The ring buffer iterator | |
4891 | * | |
4892 | * Returns true if there was dropped events since the last peek. | |
4893 | */ | |
4894 | bool ring_buffer_iter_dropped(struct ring_buffer_iter *iter) | |
4895 | { | |
4896 | bool ret = iter->missed_events != 0; | |
4897 | ||
4898 | iter->missed_events = 0; | |
4899 | return ret; | |
4900 | } | |
4901 | EXPORT_SYMBOL_GPL(ring_buffer_iter_dropped); | |
4902 | ||
f83c9d0f SR |
4903 | /** |
4904 | * ring_buffer_iter_peek - peek at the next event to be read | |
4905 | * @iter: The ring buffer iterator | |
4906 | * @ts: The timestamp counter of this event. | |
4907 | * | |
4908 | * This will return the event that will be read next, but does | |
4909 | * not increment the iterator. | |
4910 | */ | |
4911 | struct ring_buffer_event * | |
4912 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
4913 | { | |
4914 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
4915 | struct ring_buffer_event *event; | |
4916 | unsigned long flags; | |
4917 | ||
2d622719 | 4918 | again: |
5389f6fa | 4919 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4920 | event = rb_iter_peek(iter, ts); |
5389f6fa | 4921 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4922 | |
1b959e18 | 4923 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4924 | goto again; |
2d622719 | 4925 | |
f83c9d0f SR |
4926 | return event; |
4927 | } | |
4928 | ||
7a8e76a3 SR |
4929 | /** |
4930 | * ring_buffer_consume - return an event and consume it | |
4931 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
4932 | * @cpu: the cpu to read the buffer from |
4933 | * @ts: a variable to store the timestamp (may be NULL) | |
4934 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
4935 | * |
4936 | * Returns the next event in the ring buffer, and that event is consumed. | |
4937 | * Meaning, that sequential reads will keep returning a different event, | |
4938 | * and eventually empty the ring buffer if the producer is slower. | |
4939 | */ | |
4940 | struct ring_buffer_event * | |
13292494 | 4941 | ring_buffer_consume(struct trace_buffer *buffer, int cpu, u64 *ts, |
66a8cb95 | 4942 | unsigned long *lost_events) |
7a8e76a3 | 4943 | { |
554f786e SR |
4944 | struct ring_buffer_per_cpu *cpu_buffer; |
4945 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 4946 | unsigned long flags; |
289a5a25 | 4947 | bool dolock; |
7a8e76a3 | 4948 | |
2d622719 | 4949 | again: |
554f786e SR |
4950 | /* might be called in atomic */ |
4951 | preempt_disable(); | |
4952 | ||
9e01c1b7 | 4953 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 4954 | goto out; |
7a8e76a3 | 4955 | |
554f786e | 4956 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e | 4957 | local_irq_save(flags); |
289a5a25 | 4958 | dolock = rb_reader_lock(cpu_buffer); |
f83c9d0f | 4959 | |
66a8cb95 SR |
4960 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
4961 | if (event) { | |
4962 | cpu_buffer->lost_events = 0; | |
469535a5 | 4963 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 4964 | } |
7a8e76a3 | 4965 | |
289a5a25 | 4966 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4967 | local_irq_restore(flags); |
f83c9d0f | 4968 | |
554f786e SR |
4969 | out: |
4970 | preempt_enable(); | |
4971 | ||
1b959e18 | 4972 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4973 | goto again; |
2d622719 | 4974 | |
7a8e76a3 SR |
4975 | return event; |
4976 | } | |
c4f50183 | 4977 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
4978 | |
4979 | /** | |
72c9ddfd | 4980 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
4981 | * @buffer: The ring buffer to read from |
4982 | * @cpu: The cpu buffer to iterate over | |
31b265b3 | 4983 | * @flags: gfp flags to use for memory allocation |
7a8e76a3 | 4984 | * |
72c9ddfd DM |
4985 | * This performs the initial preparations necessary to iterate |
4986 | * through the buffer. Memory is allocated, buffer recording | |
4987 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 4988 | * |
6167c205 | 4989 | * Disabling buffer recording prevents the reading from being |
72c9ddfd DM |
4990 | * corrupted. This is not a consuming read, so a producer is not |
4991 | * expected. | |
4992 | * | |
4993 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
d611851b | 4994 | * expected to make at least one call to ring_buffer_read_prepare_sync. |
72c9ddfd DM |
4995 | * Afterwards, ring_buffer_read_start is invoked to get things going |
4996 | * for real. | |
4997 | * | |
d611851b | 4998 | * This overall must be paired with ring_buffer_read_finish. |
7a8e76a3 SR |
4999 | */ |
5000 | struct ring_buffer_iter * | |
13292494 | 5001 | ring_buffer_read_prepare(struct trace_buffer *buffer, int cpu, gfp_t flags) |
7a8e76a3 SR |
5002 | { |
5003 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 5004 | struct ring_buffer_iter *iter; |
7a8e76a3 | 5005 | |
9e01c1b7 | 5006 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 5007 | return NULL; |
7a8e76a3 | 5008 | |
785888c5 | 5009 | iter = kzalloc(sizeof(*iter), flags); |
7a8e76a3 | 5010 | if (!iter) |
8aabee57 | 5011 | return NULL; |
7a8e76a3 | 5012 | |
b0495258 | 5013 | /* Holds the entire event: data and meta data */ |
139f8400 TSV |
5014 | iter->event_size = buffer->subbuf_size; |
5015 | iter->event = kmalloc(iter->event_size, flags); | |
785888c5 SRV |
5016 | if (!iter->event) { |
5017 | kfree(iter); | |
5018 | return NULL; | |
5019 | } | |
5020 | ||
7a8e76a3 SR |
5021 | cpu_buffer = buffer->buffers[cpu]; |
5022 | ||
5023 | iter->cpu_buffer = cpu_buffer; | |
5024 | ||
07b8b10e | 5025 | atomic_inc(&cpu_buffer->resize_disabled); |
72c9ddfd DM |
5026 | |
5027 | return iter; | |
5028 | } | |
5029 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
5030 | ||
5031 | /** | |
5032 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
5033 | * | |
5034 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
5035 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
5036 | * calls on those iterators are allowed. | |
5037 | */ | |
5038 | void | |
5039 | ring_buffer_read_prepare_sync(void) | |
5040 | { | |
74401729 | 5041 | synchronize_rcu(); |
72c9ddfd DM |
5042 | } |
5043 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
5044 | ||
5045 | /** | |
5046 | * ring_buffer_read_start - start a non consuming read of the buffer | |
5047 | * @iter: The iterator returned by ring_buffer_read_prepare | |
5048 | * | |
5049 | * This finalizes the startup of an iteration through the buffer. | |
5050 | * The iterator comes from a call to ring_buffer_read_prepare and | |
5051 | * an intervening ring_buffer_read_prepare_sync must have been | |
5052 | * performed. | |
5053 | * | |
d611851b | 5054 | * Must be paired with ring_buffer_read_finish. |
72c9ddfd DM |
5055 | */ |
5056 | void | |
5057 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
5058 | { | |
5059 | struct ring_buffer_per_cpu *cpu_buffer; | |
5060 | unsigned long flags; | |
5061 | ||
5062 | if (!iter) | |
5063 | return; | |
5064 | ||
5065 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 5066 | |
5389f6fa | 5067 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 5068 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 5069 | rb_iter_reset(iter); |
0199c4e6 | 5070 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 5071 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 5072 | } |
c4f50183 | 5073 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
5074 | |
5075 | /** | |
d611851b | 5076 | * ring_buffer_read_finish - finish reading the iterator of the buffer |
7a8e76a3 SR |
5077 | * @iter: The iterator retrieved by ring_buffer_start |
5078 | * | |
5079 | * This re-enables the recording to the buffer, and frees the | |
5080 | * iterator. | |
5081 | */ | |
5082 | void | |
5083 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
5084 | { | |
5085 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 5086 | unsigned long flags; |
7a8e76a3 | 5087 | |
659f451f SR |
5088 | /* |
5089 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
5090 | * to check the integrity of the ring buffer. |
5091 | * Must prevent readers from trying to read, as the check | |
5092 | * clears the HEAD page and readers require it. | |
659f451f | 5093 | */ |
9366c1ba | 5094 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 5095 | rb_check_pages(cpu_buffer); |
9366c1ba | 5096 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 5097 | |
07b8b10e | 5098 | atomic_dec(&cpu_buffer->resize_disabled); |
785888c5 | 5099 | kfree(iter->event); |
7a8e76a3 SR |
5100 | kfree(iter); |
5101 | } | |
c4f50183 | 5102 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
5103 | |
5104 | /** | |
bc1a72af | 5105 | * ring_buffer_iter_advance - advance the iterator to the next location |
7a8e76a3 | 5106 | * @iter: The ring buffer iterator |
7a8e76a3 | 5107 | * |
bc1a72af SRV |
5108 | * Move the location of the iterator such that the next read will |
5109 | * be the next location of the iterator. | |
7a8e76a3 | 5110 | */ |
bc1a72af | 5111 | void ring_buffer_iter_advance(struct ring_buffer_iter *iter) |
7a8e76a3 | 5112 | { |
f83c9d0f SR |
5113 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
5114 | unsigned long flags; | |
7a8e76a3 | 5115 | |
5389f6fa | 5116 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 5117 | |
7a8e76a3 SR |
5118 | rb_advance_iter(iter); |
5119 | ||
bc1a72af | 5120 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 5121 | } |
bc1a72af | 5122 | EXPORT_SYMBOL_GPL(ring_buffer_iter_advance); |
7a8e76a3 SR |
5123 | |
5124 | /** | |
5125 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
5126 | * @buffer: The ring buffer. | |
59e7cffe | 5127 | * @cpu: The CPU to get ring buffer size from. |
7a8e76a3 | 5128 | */ |
13292494 | 5129 | unsigned long ring_buffer_size(struct trace_buffer *buffer, int cpu) |
7a8e76a3 | 5130 | { |
438ced17 VN |
5131 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
5132 | return 0; | |
5133 | ||
139f8400 | 5134 | return buffer->subbuf_size * buffer->buffers[cpu]->nr_pages; |
7a8e76a3 | 5135 | } |
c4f50183 | 5136 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 | 5137 | |
8ec90be7 SRG |
5138 | /** |
5139 | * ring_buffer_max_event_size - return the max data size of an event | |
5140 | * @buffer: The ring buffer. | |
5141 | * | |
5142 | * Returns the maximum size an event can be. | |
5143 | */ | |
5144 | unsigned long ring_buffer_max_event_size(struct trace_buffer *buffer) | |
5145 | { | |
5146 | /* If abs timestamp is requested, events have a timestamp too */ | |
5147 | if (ring_buffer_time_stamp_abs(buffer)) | |
139f8400 TSV |
5148 | return buffer->max_data_size - RB_LEN_TIME_EXTEND; |
5149 | return buffer->max_data_size; | |
8ec90be7 SRG |
5150 | } |
5151 | EXPORT_SYMBOL_GPL(ring_buffer_max_event_size); | |
5152 | ||
7e42907f ZY |
5153 | static void rb_clear_buffer_page(struct buffer_page *page) |
5154 | { | |
5155 | local_set(&page->write, 0); | |
5156 | local_set(&page->entries, 0); | |
5157 | rb_init_page(page->page); | |
5158 | page->read = 0; | |
5159 | } | |
5160 | ||
7a8e76a3 SR |
5161 | static void |
5162 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
5163 | { | |
7e42907f ZY |
5164 | struct buffer_page *page; |
5165 | ||
77ae365e SR |
5166 | rb_head_page_deactivate(cpu_buffer); |
5167 | ||
7a8e76a3 | 5168 | cpu_buffer->head_page |
3adc54fa | 5169 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
7e42907f ZY |
5170 | rb_clear_buffer_page(cpu_buffer->head_page); |
5171 | list_for_each_entry(page, cpu_buffer->pages, list) { | |
5172 | rb_clear_buffer_page(page); | |
5173 | } | |
bf41a158 SR |
5174 | |
5175 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
5176 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
5177 | ||
5178 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 5179 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
7e42907f | 5180 | rb_clear_buffer_page(cpu_buffer->reader_page); |
7a8e76a3 | 5181 | |
c64e148a | 5182 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 5183 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
5184 | local_set(&cpu_buffer->commit_overrun, 0); |
5185 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 5186 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
5187 | local_set(&cpu_buffer->committing, 0); |
5188 | local_set(&cpu_buffer->commits, 0); | |
2c2b0a78 | 5189 | local_set(&cpu_buffer->pages_touched, 0); |
31029a8b | 5190 | local_set(&cpu_buffer->pages_lost, 0); |
2c2b0a78 | 5191 | local_set(&cpu_buffer->pages_read, 0); |
03329f99 | 5192 | cpu_buffer->last_pages_touch = 0; |
2c2b0a78 | 5193 | cpu_buffer->shortest_full = 0; |
77ae365e | 5194 | cpu_buffer->read = 0; |
c64e148a | 5195 | cpu_buffer->read_bytes = 0; |
69507c06 | 5196 | |
10464b4a SRV |
5197 | rb_time_set(&cpu_buffer->write_stamp, 0); |
5198 | rb_time_set(&cpu_buffer->before_stamp, 0); | |
77ae365e | 5199 | |
8672e494 SRV |
5200 | memset(cpu_buffer->event_stamp, 0, sizeof(cpu_buffer->event_stamp)); |
5201 | ||
66a8cb95 SR |
5202 | cpu_buffer->lost_events = 0; |
5203 | cpu_buffer->last_overrun = 0; | |
5204 | ||
77ae365e | 5205 | rb_head_page_activate(cpu_buffer); |
2d093282 | 5206 | cpu_buffer->pages_removed = 0; |
7a8e76a3 SR |
5207 | } |
5208 | ||
b23d7a5f NP |
5209 | /* Must have disabled the cpu buffer then done a synchronize_rcu */ |
5210 | static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
5211 | { | |
5212 | unsigned long flags; | |
5213 | ||
5214 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
5215 | ||
5216 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) | |
5217 | goto out; | |
5218 | ||
5219 | arch_spin_lock(&cpu_buffer->lock); | |
5220 | ||
5221 | rb_reset_cpu(cpu_buffer); | |
5222 | ||
5223 | arch_spin_unlock(&cpu_buffer->lock); | |
5224 | ||
5225 | out: | |
5226 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
5227 | } | |
5228 | ||
7a8e76a3 SR |
5229 | /** |
5230 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
5231 | * @buffer: The ring buffer to reset a per cpu buffer of | |
5232 | * @cpu: The CPU buffer to be reset | |
5233 | */ | |
13292494 | 5234 | void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
5235 | { |
5236 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 5237 | |
9e01c1b7 | 5238 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 5239 | return; |
7a8e76a3 | 5240 | |
bbeb9746 GK |
5241 | /* prevent another thread from changing buffer sizes */ |
5242 | mutex_lock(&buffer->mutex); | |
5243 | ||
07b8b10e | 5244 | atomic_inc(&cpu_buffer->resize_disabled); |
41ede23e SR |
5245 | atomic_inc(&cpu_buffer->record_disabled); |
5246 | ||
83f40318 | 5247 | /* Make sure all commits have finished */ |
74401729 | 5248 | synchronize_rcu(); |
83f40318 | 5249 | |
b23d7a5f | 5250 | reset_disabled_cpu_buffer(cpu_buffer); |
f83c9d0f | 5251 | |
b23d7a5f NP |
5252 | atomic_dec(&cpu_buffer->record_disabled); |
5253 | atomic_dec(&cpu_buffer->resize_disabled); | |
bbeb9746 GK |
5254 | |
5255 | mutex_unlock(&buffer->mutex); | |
b23d7a5f NP |
5256 | } |
5257 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); | |
41b6a95d | 5258 | |
7c339fb4 TW |
5259 | /* Flag to ensure proper resetting of atomic variables */ |
5260 | #define RESET_BIT (1 << 30) | |
5261 | ||
b23d7a5f | 5262 | /** |
b7085b6f | 5263 | * ring_buffer_reset_online_cpus - reset a ring buffer per CPU buffer |
b23d7a5f | 5264 | * @buffer: The ring buffer to reset a per cpu buffer of |
b23d7a5f NP |
5265 | */ |
5266 | void ring_buffer_reset_online_cpus(struct trace_buffer *buffer) | |
5267 | { | |
5268 | struct ring_buffer_per_cpu *cpu_buffer; | |
5269 | int cpu; | |
7a8e76a3 | 5270 | |
bbeb9746 GK |
5271 | /* prevent another thread from changing buffer sizes */ |
5272 | mutex_lock(&buffer->mutex); | |
5273 | ||
b23d7a5f NP |
5274 | for_each_online_buffer_cpu(buffer, cpu) { |
5275 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 5276 | |
7c339fb4 | 5277 | atomic_add(RESET_BIT, &cpu_buffer->resize_disabled); |
b23d7a5f NP |
5278 | atomic_inc(&cpu_buffer->record_disabled); |
5279 | } | |
f83c9d0f | 5280 | |
b23d7a5f NP |
5281 | /* Make sure all commits have finished */ |
5282 | synchronize_rcu(); | |
41ede23e | 5283 | |
7c339fb4 | 5284 | for_each_buffer_cpu(buffer, cpu) { |
b23d7a5f NP |
5285 | cpu_buffer = buffer->buffers[cpu]; |
5286 | ||
7c339fb4 TW |
5287 | /* |
5288 | * If a CPU came online during the synchronize_rcu(), then | |
5289 | * ignore it. | |
5290 | */ | |
5291 | if (!(atomic_read(&cpu_buffer->resize_disabled) & RESET_BIT)) | |
5292 | continue; | |
5293 | ||
b23d7a5f NP |
5294 | reset_disabled_cpu_buffer(cpu_buffer); |
5295 | ||
5296 | atomic_dec(&cpu_buffer->record_disabled); | |
7c339fb4 | 5297 | atomic_sub(RESET_BIT, &cpu_buffer->resize_disabled); |
b23d7a5f | 5298 | } |
bbeb9746 GK |
5299 | |
5300 | mutex_unlock(&buffer->mutex); | |
7a8e76a3 SR |
5301 | } |
5302 | ||
5303 | /** | |
5304 | * ring_buffer_reset - reset a ring buffer | |
5305 | * @buffer: The ring buffer to reset all cpu buffers | |
5306 | */ | |
13292494 | 5307 | void ring_buffer_reset(struct trace_buffer *buffer) |
7a8e76a3 | 5308 | { |
b23d7a5f | 5309 | struct ring_buffer_per_cpu *cpu_buffer; |
7a8e76a3 SR |
5310 | int cpu; |
5311 | ||
51d15794 SRV |
5312 | /* prevent another thread from changing buffer sizes */ |
5313 | mutex_lock(&buffer->mutex); | |
5314 | ||
b23d7a5f NP |
5315 | for_each_buffer_cpu(buffer, cpu) { |
5316 | cpu_buffer = buffer->buffers[cpu]; | |
5317 | ||
5318 | atomic_inc(&cpu_buffer->resize_disabled); | |
5319 | atomic_inc(&cpu_buffer->record_disabled); | |
5320 | } | |
5321 | ||
5322 | /* Make sure all commits have finished */ | |
5323 | synchronize_rcu(); | |
5324 | ||
5325 | for_each_buffer_cpu(buffer, cpu) { | |
5326 | cpu_buffer = buffer->buffers[cpu]; | |
5327 | ||
5328 | reset_disabled_cpu_buffer(cpu_buffer); | |
5329 | ||
5330 | atomic_dec(&cpu_buffer->record_disabled); | |
5331 | atomic_dec(&cpu_buffer->resize_disabled); | |
5332 | } | |
51d15794 SRV |
5333 | |
5334 | mutex_unlock(&buffer->mutex); | |
7a8e76a3 | 5335 | } |
c4f50183 | 5336 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
5337 | |
5338 | /** | |
b7085b6f | 5339 | * ring_buffer_empty - is the ring buffer empty? |
7a8e76a3 SR |
5340 | * @buffer: The ring buffer to test |
5341 | */ | |
13292494 | 5342 | bool ring_buffer_empty(struct trace_buffer *buffer) |
7a8e76a3 SR |
5343 | { |
5344 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 5345 | unsigned long flags; |
289a5a25 | 5346 | bool dolock; |
bc92b956 | 5347 | bool ret; |
7a8e76a3 SR |
5348 | int cpu; |
5349 | ||
5350 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
5351 | for_each_buffer_cpu(buffer, cpu) { | |
5352 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 5353 | local_irq_save(flags); |
289a5a25 | 5354 | dolock = rb_reader_lock(cpu_buffer); |
d4788207 | 5355 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 5356 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e SR |
5357 | local_irq_restore(flags); |
5358 | ||
d4788207 | 5359 | if (!ret) |
3d4e204d | 5360 | return false; |
7a8e76a3 | 5361 | } |
554f786e | 5362 | |
3d4e204d | 5363 | return true; |
7a8e76a3 | 5364 | } |
c4f50183 | 5365 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
5366 | |
5367 | /** | |
5368 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
5369 | * @buffer: The ring buffer | |
5370 | * @cpu: The CPU buffer to test | |
5371 | */ | |
13292494 | 5372 | bool ring_buffer_empty_cpu(struct trace_buffer *buffer, int cpu) |
7a8e76a3 SR |
5373 | { |
5374 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 5375 | unsigned long flags; |
289a5a25 | 5376 | bool dolock; |
bc92b956 | 5377 | bool ret; |
7a8e76a3 | 5378 | |
9e01c1b7 | 5379 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
3d4e204d | 5380 | return true; |
7a8e76a3 SR |
5381 | |
5382 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 5383 | local_irq_save(flags); |
289a5a25 | 5384 | dolock = rb_reader_lock(cpu_buffer); |
554f786e | 5385 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 5386 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 5387 | local_irq_restore(flags); |
554f786e SR |
5388 | |
5389 | return ret; | |
7a8e76a3 | 5390 | } |
c4f50183 | 5391 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 5392 | |
85bac32c | 5393 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
5394 | /** |
5395 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
5396 | * @buffer_a: One buffer to swap with | |
5397 | * @buffer_b: The other buffer to swap with | |
59e7cffe | 5398 | * @cpu: the CPU of the buffers to swap |
7a8e76a3 SR |
5399 | * |
5400 | * This function is useful for tracers that want to take a "snapshot" | |
5401 | * of a CPU buffer and has another back up buffer lying around. | |
5402 | * it is expected that the tracer handles the cpu buffer not being | |
5403 | * used at the moment. | |
5404 | */ | |
13292494 SRV |
5405 | int ring_buffer_swap_cpu(struct trace_buffer *buffer_a, |
5406 | struct trace_buffer *buffer_b, int cpu) | |
7a8e76a3 SR |
5407 | { |
5408 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
5409 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
5410 | int ret = -EINVAL; |
5411 | ||
9e01c1b7 RR |
5412 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
5413 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 5414 | goto out; |
7a8e76a3 | 5415 | |
438ced17 VN |
5416 | cpu_buffer_a = buffer_a->buffers[cpu]; |
5417 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
5418 | ||
7a8e76a3 | 5419 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 5420 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
5421 | goto out; |
5422 | ||
b81e03a2 SRG |
5423 | if (buffer_a->subbuf_order != buffer_b->subbuf_order) |
5424 | goto out; | |
5425 | ||
554f786e | 5426 | ret = -EAGAIN; |
7a8e76a3 | 5427 | |
97b17efe | 5428 | if (atomic_read(&buffer_a->record_disabled)) |
554f786e | 5429 | goto out; |
97b17efe SR |
5430 | |
5431 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 5432 | goto out; |
97b17efe | 5433 | |
97b17efe | 5434 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 5435 | goto out; |
97b17efe SR |
5436 | |
5437 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 5438 | goto out; |
97b17efe | 5439 | |
7a8e76a3 | 5440 | /* |
74401729 | 5441 | * We can't do a synchronize_rcu here because this |
7a8e76a3 SR |
5442 | * function can be called in atomic context. |
5443 | * Normally this will be called from the same CPU as cpu. | |
5444 | * If not it's up to the caller to protect this. | |
5445 | */ | |
5446 | atomic_inc(&cpu_buffer_a->record_disabled); | |
5447 | atomic_inc(&cpu_buffer_b->record_disabled); | |
5448 | ||
98277991 SR |
5449 | ret = -EBUSY; |
5450 | if (local_read(&cpu_buffer_a->committing)) | |
5451 | goto out_dec; | |
5452 | if (local_read(&cpu_buffer_b->committing)) | |
5453 | goto out_dec; | |
5454 | ||
8a96c028 CL |
5455 | /* |
5456 | * When resize is in progress, we cannot swap it because | |
5457 | * it will mess the state of the cpu buffer. | |
5458 | */ | |
5459 | if (atomic_read(&buffer_a->resizing)) | |
5460 | goto out_dec; | |
5461 | if (atomic_read(&buffer_b->resizing)) | |
5462 | goto out_dec; | |
5463 | ||
7a8e76a3 SR |
5464 | buffer_a->buffers[cpu] = cpu_buffer_b; |
5465 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
5466 | ||
5467 | cpu_buffer_b->buffer = buffer_a; | |
5468 | cpu_buffer_a->buffer = buffer_b; | |
5469 | ||
98277991 SR |
5470 | ret = 0; |
5471 | ||
5472 | out_dec: | |
7a8e76a3 SR |
5473 | atomic_dec(&cpu_buffer_a->record_disabled); |
5474 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 5475 | out: |
554f786e | 5476 | return ret; |
7a8e76a3 | 5477 | } |
c4f50183 | 5478 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 5479 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 5480 | |
8789a9e7 SR |
5481 | /** |
5482 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
5483 | * @buffer: the buffer to allocate for. | |
d611851b | 5484 | * @cpu: the cpu buffer to allocate. |
8789a9e7 SR |
5485 | * |
5486 | * This function is used in conjunction with ring_buffer_read_page. | |
5487 | * When reading a full page from the ring buffer, these functions | |
5488 | * can be used to speed up the process. The calling function should | |
5489 | * allocate a few pages first with this function. Then when it | |
5490 | * needs to get pages from the ring buffer, it passes the result | |
5491 | * of this function into ring_buffer_read_page, which will swap | |
5492 | * the page that was allocated, with the read page of the buffer. | |
5493 | * | |
5494 | * Returns: | |
a7e52ad7 | 5495 | * The page allocated, or ERR_PTR |
8789a9e7 | 5496 | */ |
bce761d7 TSV |
5497 | struct buffer_data_read_page * |
5498 | ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu) | |
8789a9e7 | 5499 | { |
a7e52ad7 | 5500 | struct ring_buffer_per_cpu *cpu_buffer; |
bce761d7 | 5501 | struct buffer_data_read_page *bpage = NULL; |
73a757e6 | 5502 | unsigned long flags; |
7ea59064 | 5503 | struct page *page; |
8789a9e7 | 5504 | |
a7e52ad7 SRV |
5505 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
5506 | return ERR_PTR(-ENODEV); | |
5507 | ||
bce761d7 TSV |
5508 | bpage = kzalloc(sizeof(*bpage), GFP_KERNEL); |
5509 | if (!bpage) | |
5510 | return ERR_PTR(-ENOMEM); | |
5511 | ||
5512 | bpage->order = buffer->subbuf_order; | |
a7e52ad7 | 5513 | cpu_buffer = buffer->buffers[cpu]; |
73a757e6 SRV |
5514 | local_irq_save(flags); |
5515 | arch_spin_lock(&cpu_buffer->lock); | |
5516 | ||
5517 | if (cpu_buffer->free_page) { | |
bce761d7 | 5518 | bpage->data = cpu_buffer->free_page; |
73a757e6 SRV |
5519 | cpu_buffer->free_page = NULL; |
5520 | } | |
5521 | ||
5522 | arch_spin_unlock(&cpu_buffer->lock); | |
5523 | local_irq_restore(flags); | |
5524 | ||
bce761d7 | 5525 | if (bpage->data) |
73a757e6 SRV |
5526 | goto out; |
5527 | ||
f9b94daa TSV |
5528 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL | __GFP_NORETRY, |
5529 | cpu_buffer->buffer->subbuf_order); | |
bce761d7 TSV |
5530 | if (!page) { |
5531 | kfree(bpage); | |
a7e52ad7 | 5532 | return ERR_PTR(-ENOMEM); |
bce761d7 | 5533 | } |
8789a9e7 | 5534 | |
bce761d7 | 5535 | bpage->data = page_address(page); |
8789a9e7 | 5536 | |
73a757e6 | 5537 | out: |
bce761d7 | 5538 | rb_init_page(bpage->data); |
ef7a4a16 | 5539 | |
044fa782 | 5540 | return bpage; |
8789a9e7 | 5541 | } |
d6ce96da | 5542 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
5543 | |
5544 | /** | |
5545 | * ring_buffer_free_read_page - free an allocated read page | |
5546 | * @buffer: the buffer the page was allocate for | |
73a757e6 | 5547 | * @cpu: the cpu buffer the page came from |
bce761d7 | 5548 | * @data_page: the page to free |
8789a9e7 SR |
5549 | * |
5550 | * Free a page allocated from ring_buffer_alloc_read_page. | |
5551 | */ | |
bce761d7 TSV |
5552 | void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, |
5553 | struct buffer_data_read_page *data_page) | |
8789a9e7 | 5554 | { |
3e4272b9 | 5555 | struct ring_buffer_per_cpu *cpu_buffer; |
bce761d7 | 5556 | struct buffer_data_page *bpage = data_page->data; |
ae415fa4 | 5557 | struct page *page = virt_to_page(bpage); |
73a757e6 SRV |
5558 | unsigned long flags; |
5559 | ||
3e4272b9 JJB |
5560 | if (!buffer || !buffer->buffers || !buffer->buffers[cpu]) |
5561 | return; | |
5562 | ||
5563 | cpu_buffer = buffer->buffers[cpu]; | |
5564 | ||
bce761d7 TSV |
5565 | /* |
5566 | * If the page is still in use someplace else, or order of the page | |
5567 | * is different from the subbuffer order of the buffer - | |
5568 | * we can't reuse it | |
5569 | */ | |
5570 | if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order) | |
ae415fa4 SRV |
5571 | goto out; |
5572 | ||
73a757e6 SRV |
5573 | local_irq_save(flags); |
5574 | arch_spin_lock(&cpu_buffer->lock); | |
5575 | ||
5576 | if (!cpu_buffer->free_page) { | |
5577 | cpu_buffer->free_page = bpage; | |
5578 | bpage = NULL; | |
5579 | } | |
5580 | ||
5581 | arch_spin_unlock(&cpu_buffer->lock); | |
5582 | local_irq_restore(flags); | |
5583 | ||
ae415fa4 | 5584 | out: |
bce761d7 TSV |
5585 | free_pages((unsigned long)bpage, data_page->order); |
5586 | kfree(data_page); | |
8789a9e7 | 5587 | } |
d6ce96da | 5588 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
5589 | |
5590 | /** | |
5591 | * ring_buffer_read_page - extract a page from the ring buffer | |
5592 | * @buffer: buffer to extract from | |
5593 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 5594 | * @len: amount to extract |
8789a9e7 SR |
5595 | * @cpu: the cpu of the buffer to extract |
5596 | * @full: should the extraction only happen when the page is full. | |
5597 | * | |
5598 | * This function will pull out a page from the ring buffer and consume it. | |
5599 | * @data_page must be the address of the variable that was returned | |
5600 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
5601 | * to swap with a page in the ring buffer. | |
5602 | * | |
5603 | * for example: | |
d611851b | 5604 | * rpage = ring_buffer_alloc_read_page(buffer, cpu); |
a7e52ad7 SRV |
5605 | * if (IS_ERR(rpage)) |
5606 | * return PTR_ERR(rpage); | |
bce761d7 | 5607 | * ret = ring_buffer_read_page(buffer, rpage, len, cpu, 0); |
667d2412 | 5608 | * if (ret >= 0) |
bce761d7 TSV |
5609 | * process_page(ring_buffer_read_page_data(rpage), ret); |
5610 | * ring_buffer_free_read_page(buffer, cpu, rpage); | |
8789a9e7 SR |
5611 | * |
5612 | * When @full is set, the function will not return true unless | |
5613 | * the writer is off the reader page. | |
5614 | * | |
5615 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
5616 | * The ring buffer can be used anywhere in the kernel and can not | |
5617 | * blindly call wake_up. The layer that uses the ring buffer must be | |
5618 | * responsible for that. | |
5619 | * | |
5620 | * Returns: | |
667d2412 LJ |
5621 | * >=0 if data has been transferred, returns the offset of consumed data. |
5622 | * <0 if no data has been transferred. | |
8789a9e7 | 5623 | */ |
13292494 | 5624 | int ring_buffer_read_page(struct trace_buffer *buffer, |
bce761d7 TSV |
5625 | struct buffer_data_read_page *data_page, |
5626 | size_t len, int cpu, int full) | |
8789a9e7 SR |
5627 | { |
5628 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
5629 | struct ring_buffer_event *event; | |
044fa782 | 5630 | struct buffer_data_page *bpage; |
ef7a4a16 | 5631 | struct buffer_page *reader; |
ff0ff84a | 5632 | unsigned long missed_events; |
8789a9e7 | 5633 | unsigned long flags; |
ef7a4a16 | 5634 | unsigned int commit; |
667d2412 | 5635 | unsigned int read; |
4f3640f8 | 5636 | u64 save_timestamp; |
667d2412 | 5637 | int ret = -1; |
8789a9e7 | 5638 | |
554f786e SR |
5639 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
5640 | goto out; | |
5641 | ||
474d32b6 SR |
5642 | /* |
5643 | * If len is not big enough to hold the page header, then | |
5644 | * we can not copy anything. | |
5645 | */ | |
5646 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 5647 | goto out; |
474d32b6 SR |
5648 | |
5649 | len -= BUF_PAGE_HDR_SIZE; | |
5650 | ||
bce761d7 TSV |
5651 | if (!data_page || !data_page->data) |
5652 | goto out; | |
5653 | if (data_page->order != buffer->subbuf_order) | |
554f786e | 5654 | goto out; |
8789a9e7 | 5655 | |
bce761d7 | 5656 | bpage = data_page->data; |
044fa782 | 5657 | if (!bpage) |
554f786e | 5658 | goto out; |
8789a9e7 | 5659 | |
5389f6fa | 5660 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 5661 | |
ef7a4a16 SR |
5662 | reader = rb_get_reader_page(cpu_buffer); |
5663 | if (!reader) | |
554f786e | 5664 | goto out_unlock; |
8789a9e7 | 5665 | |
ef7a4a16 SR |
5666 | event = rb_reader_event(cpu_buffer); |
5667 | ||
5668 | read = reader->read; | |
5669 | commit = rb_page_commit(reader); | |
667d2412 | 5670 | |
66a8cb95 | 5671 | /* Check if any events were dropped */ |
ff0ff84a | 5672 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 5673 | |
8789a9e7 | 5674 | /* |
474d32b6 SR |
5675 | * If this page has been partially read or |
5676 | * if len is not big enough to read the rest of the page or | |
5677 | * a writer is still on the page, then | |
5678 | * we must copy the data from the page to the buffer. | |
5679 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 5680 | */ |
474d32b6 | 5681 | if (read || (len < (commit - read)) || |
ef7a4a16 | 5682 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 5683 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
5684 | unsigned int rpos = read; |
5685 | unsigned int pos = 0; | |
ef7a4a16 | 5686 | unsigned int size; |
8789a9e7 | 5687 | |
fa8f4a89 SRG |
5688 | /* |
5689 | * If a full page is expected, this can still be returned | |
5690 | * if there's been a previous partial read and the | |
5691 | * rest of the page can be read and the commit page is off | |
5692 | * the reader page. | |
5693 | */ | |
5694 | if (full && | |
5695 | (!read || (len < (commit - read)) || | |
5696 | cpu_buffer->reader_page == cpu_buffer->commit_page)) | |
554f786e | 5697 | goto out_unlock; |
8789a9e7 | 5698 | |
ef7a4a16 SR |
5699 | if (len > (commit - read)) |
5700 | len = (commit - read); | |
5701 | ||
69d1b839 SR |
5702 | /* Always keep the time extend and data together */ |
5703 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
5704 | |
5705 | if (len < size) | |
554f786e | 5706 | goto out_unlock; |
ef7a4a16 | 5707 | |
4f3640f8 SR |
5708 | /* save the current timestamp, since the user will need it */ |
5709 | save_timestamp = cpu_buffer->read_stamp; | |
5710 | ||
ef7a4a16 SR |
5711 | /* Need to copy one event at a time */ |
5712 | do { | |
e1e35927 DS |
5713 | /* We need the size of one event, because |
5714 | * rb_advance_reader only advances by one event, | |
5715 | * whereas rb_event_ts_length may include the size of | |
5716 | * one or two events. | |
5717 | * We have already ensured there's enough space if this | |
5718 | * is a time extend. */ | |
5719 | size = rb_event_length(event); | |
474d32b6 | 5720 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
5721 | |
5722 | len -= size; | |
5723 | ||
5724 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
5725 | rpos = reader->read; |
5726 | pos += size; | |
ef7a4a16 | 5727 | |
18fab912 HY |
5728 | if (rpos >= commit) |
5729 | break; | |
5730 | ||
ef7a4a16 | 5731 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
5732 | /* Always keep the time extend and data together */ |
5733 | size = rb_event_ts_length(event); | |
e1e35927 | 5734 | } while (len >= size); |
667d2412 LJ |
5735 | |
5736 | /* update bpage */ | |
ef7a4a16 | 5737 | local_set(&bpage->commit, pos); |
4f3640f8 | 5738 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 5739 | |
474d32b6 SR |
5740 | /* we copied everything to the beginning */ |
5741 | read = 0; | |
8789a9e7 | 5742 | } else { |
afbab76a | 5743 | /* update the entry counter */ |
77ae365e | 5744 | cpu_buffer->read += rb_page_entries(reader); |
45d99ea4 | 5745 | cpu_buffer->read_bytes += rb_page_commit(reader); |
afbab76a | 5746 | |
8789a9e7 | 5747 | /* swap the pages */ |
044fa782 | 5748 | rb_init_page(bpage); |
ef7a4a16 | 5749 | bpage = reader->page; |
bce761d7 | 5750 | reader->page = data_page->data; |
ef7a4a16 | 5751 | local_set(&reader->write, 0); |
778c55d4 | 5752 | local_set(&reader->entries, 0); |
ef7a4a16 | 5753 | reader->read = 0; |
bce761d7 | 5754 | data_page->data = bpage; |
ff0ff84a SR |
5755 | |
5756 | /* | |
5757 | * Use the real_end for the data size, | |
5758 | * This gives us a chance to store the lost events | |
5759 | * on the page. | |
5760 | */ | |
5761 | if (reader->real_end) | |
5762 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 5763 | } |
667d2412 | 5764 | ret = read; |
8789a9e7 | 5765 | |
66a8cb95 | 5766 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
5767 | |
5768 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
5769 | /* |
5770 | * Set a flag in the commit field if we lost events | |
5771 | */ | |
ff0ff84a | 5772 | if (missed_events) { |
ff0ff84a SR |
5773 | /* If there is room at the end of the page to save the |
5774 | * missed events, then record it there. | |
5775 | */ | |
139f8400 | 5776 | if (buffer->subbuf_size - commit >= sizeof(missed_events)) { |
ff0ff84a SR |
5777 | memcpy(&bpage->data[commit], &missed_events, |
5778 | sizeof(missed_events)); | |
5779 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 5780 | commit += sizeof(missed_events); |
ff0ff84a | 5781 | } |
66a8cb95 | 5782 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 5783 | } |
66a8cb95 | 5784 | |
2711ca23 SR |
5785 | /* |
5786 | * This page may be off to user land. Zero it out here. | |
5787 | */ | |
139f8400 TSV |
5788 | if (commit < buffer->subbuf_size) |
5789 | memset(&bpage->data[commit], 0, buffer->subbuf_size - commit); | |
2711ca23 | 5790 | |
554f786e | 5791 | out_unlock: |
5389f6fa | 5792 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 5793 | |
554f786e | 5794 | out: |
8789a9e7 SR |
5795 | return ret; |
5796 | } | |
d6ce96da | 5797 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 5798 | |
bce761d7 TSV |
5799 | /** |
5800 | * ring_buffer_read_page_data - get pointer to the data in the page. | |
5801 | * @page: the page to get the data from | |
5802 | * | |
5803 | * Returns pointer to the actual data in this page. | |
5804 | */ | |
5805 | void *ring_buffer_read_page_data(struct buffer_data_read_page *page) | |
5806 | { | |
5807 | return page->data; | |
5808 | } | |
5809 | EXPORT_SYMBOL_GPL(ring_buffer_read_page_data); | |
5810 | ||
2808e31e TSV |
5811 | /** |
5812 | * ring_buffer_subbuf_size_get - get size of the sub buffer. | |
5813 | * @buffer: the buffer to get the sub buffer size from | |
5814 | * | |
5815 | * Returns size of the sub buffer, in bytes. | |
5816 | */ | |
5817 | int ring_buffer_subbuf_size_get(struct trace_buffer *buffer) | |
5818 | { | |
5819 | return buffer->subbuf_size + BUF_PAGE_HDR_SIZE; | |
5820 | } | |
5821 | EXPORT_SYMBOL_GPL(ring_buffer_subbuf_size_get); | |
5822 | ||
5823 | /** | |
5824 | * ring_buffer_subbuf_order_get - get order of system sub pages in one buffer page. | |
5825 | * @buffer: The ring_buffer to get the system sub page order from | |
5826 | * | |
5827 | * By default, one ring buffer sub page equals to one system page. This parameter | |
5828 | * is configurable, per ring buffer. The size of the ring buffer sub page can be | |
5829 | * extended, but must be an order of system page size. | |
5830 | * | |
5831 | * Returns the order of buffer sub page size, in system pages: | |
5832 | * 0 means the sub buffer size is 1 system page and so forth. | |
5833 | * In case of an error < 0 is returned. | |
5834 | */ | |
5835 | int ring_buffer_subbuf_order_get(struct trace_buffer *buffer) | |
5836 | { | |
5837 | if (!buffer) | |
5838 | return -EINVAL; | |
5839 | ||
5840 | return buffer->subbuf_order; | |
5841 | } | |
5842 | EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_get); | |
5843 | ||
5844 | /** | |
5845 | * ring_buffer_subbuf_order_set - set the size of ring buffer sub page. | |
5846 | * @buffer: The ring_buffer to set the new page size. | |
5847 | * @order: Order of the system pages in one sub buffer page | |
5848 | * | |
5849 | * By default, one ring buffer pages equals to one system page. This API can be | |
5850 | * used to set new size of the ring buffer page. The size must be order of | |
5851 | * system page size, that's why the input parameter @order is the order of | |
5852 | * system pages that are allocated for one ring buffer page: | |
5853 | * 0 - 1 system page | |
5854 | * 1 - 2 system pages | |
5855 | * 3 - 4 system pages | |
5856 | * ... | |
5857 | * | |
5858 | * Returns 0 on success or < 0 in case of an error. | |
5859 | */ | |
5860 | int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order) | |
5861 | { | |
8e7b58c2 SRG |
5862 | struct ring_buffer_per_cpu *cpu_buffer; |
5863 | struct buffer_page *bpage, *tmp; | |
f9b94daa TSV |
5864 | int old_order, old_size; |
5865 | int nr_pages; | |
2808e31e | 5866 | int psize; |
f9b94daa TSV |
5867 | int err; |
5868 | int cpu; | |
2808e31e TSV |
5869 | |
5870 | if (!buffer || order < 0) | |
5871 | return -EINVAL; | |
5872 | ||
5873 | if (buffer->subbuf_order == order) | |
5874 | return 0; | |
5875 | ||
5876 | psize = (1 << order) * PAGE_SIZE; | |
5877 | if (psize <= BUF_PAGE_HDR_SIZE) | |
5878 | return -EINVAL; | |
5879 | ||
f9b94daa TSV |
5880 | old_order = buffer->subbuf_order; |
5881 | old_size = buffer->subbuf_size; | |
5882 | ||
5883 | /* prevent another thread from changing buffer sizes */ | |
5884 | mutex_lock(&buffer->mutex); | |
5885 | atomic_inc(&buffer->record_disabled); | |
5886 | ||
5887 | /* Make sure all commits have finished */ | |
5888 | synchronize_rcu(); | |
5889 | ||
2808e31e TSV |
5890 | buffer->subbuf_order = order; |
5891 | buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE; | |
5892 | ||
f9b94daa TSV |
5893 | /* Make sure all new buffers are allocated, before deleting the old ones */ |
5894 | for_each_buffer_cpu(buffer, cpu) { | |
8e7b58c2 | 5895 | |
f9b94daa TSV |
5896 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
5897 | continue; | |
5898 | ||
8e7b58c2 SRG |
5899 | cpu_buffer = buffer->buffers[cpu]; |
5900 | ||
353cc219 SRG |
5901 | /* Update the number of pages to match the new size */ |
5902 | nr_pages = old_size * buffer->buffers[cpu]->nr_pages; | |
5903 | nr_pages = DIV_ROUND_UP(nr_pages, buffer->subbuf_size); | |
5904 | ||
8e7b58c2 SRG |
5905 | /* we need a minimum of two pages */ |
5906 | if (nr_pages < 2) | |
5907 | nr_pages = 2; | |
5908 | ||
5909 | cpu_buffer->nr_pages_to_update = nr_pages; | |
5910 | ||
5911 | /* Include the reader page */ | |
5912 | nr_pages++; | |
5913 | ||
5914 | /* Allocate the new size buffer */ | |
5915 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
5916 | if (__rb_allocate_pages(cpu_buffer, nr_pages, | |
5917 | &cpu_buffer->new_pages)) { | |
5918 | /* not enough memory for new pages */ | |
f9b94daa TSV |
5919 | err = -ENOMEM; |
5920 | goto error; | |
5921 | } | |
5922 | } | |
5923 | ||
5924 | for_each_buffer_cpu(buffer, cpu) { | |
8e7b58c2 | 5925 | |
f9b94daa TSV |
5926 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
5927 | continue; | |
5928 | ||
8e7b58c2 SRG |
5929 | cpu_buffer = buffer->buffers[cpu]; |
5930 | ||
5931 | /* Clear the head bit to make the link list normal to read */ | |
5932 | rb_head_page_deactivate(cpu_buffer); | |
5933 | ||
5934 | /* Now walk the list and free all the old sub buffers */ | |
5935 | list_for_each_entry_safe(bpage, tmp, cpu_buffer->pages, list) { | |
5936 | list_del_init(&bpage->list); | |
5937 | free_buffer_page(bpage); | |
5938 | } | |
5939 | /* The above loop stopped an the last page needing to be freed */ | |
5940 | bpage = list_entry(cpu_buffer->pages, struct buffer_page, list); | |
5941 | free_buffer_page(bpage); | |
5942 | ||
5943 | /* Free the current reader page */ | |
5944 | free_buffer_page(cpu_buffer->reader_page); | |
5945 | ||
5946 | /* One page was allocated for the reader page */ | |
5947 | cpu_buffer->reader_page = list_entry(cpu_buffer->new_pages.next, | |
5948 | struct buffer_page, list); | |
5949 | list_del_init(&cpu_buffer->reader_page->list); | |
5950 | ||
5951 | /* The cpu_buffer pages are a link list with no head */ | |
5952 | cpu_buffer->pages = cpu_buffer->new_pages.next; | |
5953 | cpu_buffer->new_pages.next->prev = cpu_buffer->new_pages.prev; | |
5954 | cpu_buffer->new_pages.prev->next = cpu_buffer->new_pages.next; | |
5955 | ||
5956 | /* Clear the new_pages list */ | |
5957 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
5958 | ||
5959 | cpu_buffer->head_page | |
5960 | = list_entry(cpu_buffer->pages, struct buffer_page, list); | |
5961 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; | |
5962 | ||
5963 | cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update; | |
5964 | cpu_buffer->nr_pages_to_update = 0; | |
5965 | ||
5966 | free_pages((unsigned long)cpu_buffer->free_page, old_order); | |
5967 | cpu_buffer->free_page = NULL; | |
5968 | ||
5969 | rb_head_page_activate(cpu_buffer); | |
5970 | ||
5971 | rb_check_pages(cpu_buffer); | |
f9b94daa TSV |
5972 | } |
5973 | ||
5974 | atomic_dec(&buffer->record_disabled); | |
5975 | mutex_unlock(&buffer->mutex); | |
5976 | ||
2808e31e | 5977 | return 0; |
f9b94daa TSV |
5978 | |
5979 | error: | |
5980 | buffer->subbuf_order = old_order; | |
5981 | buffer->subbuf_size = old_size; | |
5982 | ||
5983 | atomic_dec(&buffer->record_disabled); | |
5984 | mutex_unlock(&buffer->mutex); | |
5985 | ||
5986 | for_each_buffer_cpu(buffer, cpu) { | |
8e7b58c2 SRG |
5987 | cpu_buffer = buffer->buffers[cpu]; |
5988 | ||
5989 | if (!cpu_buffer->nr_pages_to_update) | |
f9b94daa | 5990 | continue; |
8e7b58c2 SRG |
5991 | |
5992 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, list) { | |
5993 | list_del_init(&bpage->list); | |
5994 | free_buffer_page(bpage); | |
5995 | } | |
f9b94daa | 5996 | } |
f9b94daa TSV |
5997 | |
5998 | return err; | |
2808e31e TSV |
5999 | } |
6000 | EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_set); | |
6001 | ||
b32614c0 SAS |
6002 | /* |
6003 | * We only allocate new buffers, never free them if the CPU goes down. | |
6004 | * If we were to free the buffer, then the user would lose any trace that was in | |
6005 | * the buffer. | |
6006 | */ | |
6007 | int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node) | |
554f786e | 6008 | { |
13292494 | 6009 | struct trace_buffer *buffer; |
9b94a8fb SRRH |
6010 | long nr_pages_same; |
6011 | int cpu_i; | |
6012 | unsigned long nr_pages; | |
554f786e | 6013 | |
13292494 | 6014 | buffer = container_of(node, struct trace_buffer, node); |
b32614c0 SAS |
6015 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
6016 | return 0; | |
6017 | ||
6018 | nr_pages = 0; | |
6019 | nr_pages_same = 1; | |
6020 | /* check if all cpu sizes are same */ | |
6021 | for_each_buffer_cpu(buffer, cpu_i) { | |
6022 | /* fill in the size from first enabled cpu */ | |
6023 | if (nr_pages == 0) | |
6024 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
6025 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
6026 | nr_pages_same = 0; | |
6027 | break; | |
554f786e | 6028 | } |
554f786e | 6029 | } |
b32614c0 SAS |
6030 | /* allocate minimum pages, user can later expand it */ |
6031 | if (!nr_pages_same) | |
6032 | nr_pages = 2; | |
6033 | buffer->buffers[cpu] = | |
6034 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); | |
6035 | if (!buffer->buffers[cpu]) { | |
6036 | WARN(1, "failed to allocate ring buffer on CPU %u\n", | |
6037 | cpu); | |
6038 | return -ENOMEM; | |
6039 | } | |
6040 | smp_wmb(); | |
6041 | cpumask_set_cpu(cpu, buffer->cpumask); | |
6042 | return 0; | |
554f786e | 6043 | } |
6c43e554 SRRH |
6044 | |
6045 | #ifdef CONFIG_RING_BUFFER_STARTUP_TEST | |
6046 | /* | |
6047 | * This is a basic integrity check of the ring buffer. | |
6048 | * Late in the boot cycle this test will run when configured in. | |
6049 | * It will kick off a thread per CPU that will go into a loop | |
6050 | * writing to the per cpu ring buffer various sizes of data. | |
6051 | * Some of the data will be large items, some small. | |
6052 | * | |
6053 | * Another thread is created that goes into a spin, sending out | |
6054 | * IPIs to the other CPUs to also write into the ring buffer. | |
6055 | * this is to test the nesting ability of the buffer. | |
6056 | * | |
6057 | * Basic stats are recorded and reported. If something in the | |
6058 | * ring buffer should happen that's not expected, a big warning | |
6059 | * is displayed and all ring buffers are disabled. | |
6060 | */ | |
6061 | static struct task_struct *rb_threads[NR_CPUS] __initdata; | |
6062 | ||
6063 | struct rb_test_data { | |
13292494 | 6064 | struct trace_buffer *buffer; |
6c43e554 SRRH |
6065 | unsigned long events; |
6066 | unsigned long bytes_written; | |
6067 | unsigned long bytes_alloc; | |
6068 | unsigned long bytes_dropped; | |
6069 | unsigned long events_nested; | |
6070 | unsigned long bytes_written_nested; | |
6071 | unsigned long bytes_alloc_nested; | |
6072 | unsigned long bytes_dropped_nested; | |
6073 | int min_size_nested; | |
6074 | int max_size_nested; | |
6075 | int max_size; | |
6076 | int min_size; | |
6077 | int cpu; | |
6078 | int cnt; | |
6079 | }; | |
6080 | ||
6081 | static struct rb_test_data rb_data[NR_CPUS] __initdata; | |
6082 | ||
6083 | /* 1 meg per cpu */ | |
6084 | #define RB_TEST_BUFFER_SIZE 1048576 | |
6085 | ||
6086 | static char rb_string[] __initdata = | |
6087 | "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\" | |
6088 | "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890" | |
6089 | "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv"; | |
6090 | ||
6091 | static bool rb_test_started __initdata; | |
6092 | ||
6093 | struct rb_item { | |
6094 | int size; | |
6095 | char str[]; | |
6096 | }; | |
6097 | ||
6098 | static __init int rb_write_something(struct rb_test_data *data, bool nested) | |
6099 | { | |
6100 | struct ring_buffer_event *event; | |
6101 | struct rb_item *item; | |
6102 | bool started; | |
6103 | int event_len; | |
6104 | int size; | |
6105 | int len; | |
6106 | int cnt; | |
6107 | ||
6108 | /* Have nested writes different that what is written */ | |
6109 | cnt = data->cnt + (nested ? 27 : 0); | |
6110 | ||
6111 | /* Multiply cnt by ~e, to make some unique increment */ | |
40ed29b3 | 6112 | size = (cnt * 68 / 25) % (sizeof(rb_string) - 1); |
6c43e554 SRRH |
6113 | |
6114 | len = size + sizeof(struct rb_item); | |
6115 | ||
6116 | started = rb_test_started; | |
6117 | /* read rb_test_started before checking buffer enabled */ | |
6118 | smp_rmb(); | |
6119 | ||
6120 | event = ring_buffer_lock_reserve(data->buffer, len); | |
6121 | if (!event) { | |
6122 | /* Ignore dropped events before test starts. */ | |
6123 | if (started) { | |
6124 | if (nested) | |
6125 | data->bytes_dropped += len; | |
6126 | else | |
6127 | data->bytes_dropped_nested += len; | |
6128 | } | |
6129 | return len; | |
6130 | } | |
6131 | ||
6132 | event_len = ring_buffer_event_length(event); | |
6133 | ||
6134 | if (RB_WARN_ON(data->buffer, event_len < len)) | |
6135 | goto out; | |
6136 | ||
6137 | item = ring_buffer_event_data(event); | |
6138 | item->size = size; | |
6139 | memcpy(item->str, rb_string, size); | |
6140 | ||
6141 | if (nested) { | |
6142 | data->bytes_alloc_nested += event_len; | |
6143 | data->bytes_written_nested += len; | |
6144 | data->events_nested++; | |
6145 | if (!data->min_size_nested || len < data->min_size_nested) | |
6146 | data->min_size_nested = len; | |
6147 | if (len > data->max_size_nested) | |
6148 | data->max_size_nested = len; | |
6149 | } else { | |
6150 | data->bytes_alloc += event_len; | |
6151 | data->bytes_written += len; | |
6152 | data->events++; | |
6153 | if (!data->min_size || len < data->min_size) | |
6154 | data->max_size = len; | |
6155 | if (len > data->max_size) | |
6156 | data->max_size = len; | |
6157 | } | |
6158 | ||
6159 | out: | |
04aabc32 | 6160 | ring_buffer_unlock_commit(data->buffer); |
6c43e554 SRRH |
6161 | |
6162 | return 0; | |
6163 | } | |
6164 | ||
6165 | static __init int rb_test(void *arg) | |
6166 | { | |
6167 | struct rb_test_data *data = arg; | |
6168 | ||
6169 | while (!kthread_should_stop()) { | |
6170 | rb_write_something(data, false); | |
6171 | data->cnt++; | |
6172 | ||
6173 | set_current_state(TASK_INTERRUPTIBLE); | |
6174 | /* Now sleep between a min of 100-300us and a max of 1ms */ | |
6175 | usleep_range(((data->cnt % 3) + 1) * 100, 1000); | |
6176 | } | |
6177 | ||
6178 | return 0; | |
6179 | } | |
6180 | ||
6181 | static __init void rb_ipi(void *ignore) | |
6182 | { | |
6183 | struct rb_test_data *data; | |
6184 | int cpu = smp_processor_id(); | |
6185 | ||
6186 | data = &rb_data[cpu]; | |
6187 | rb_write_something(data, true); | |
6188 | } | |
6189 | ||
6190 | static __init int rb_hammer_test(void *arg) | |
6191 | { | |
6192 | while (!kthread_should_stop()) { | |
6193 | ||
6194 | /* Send an IPI to all cpus to write data! */ | |
6195 | smp_call_function(rb_ipi, NULL, 1); | |
6196 | /* No sleep, but for non preempt, let others run */ | |
6197 | schedule(); | |
6198 | } | |
6199 | ||
6200 | return 0; | |
6201 | } | |
6202 | ||
6203 | static __init int test_ringbuffer(void) | |
6204 | { | |
6205 | struct task_struct *rb_hammer; | |
13292494 | 6206 | struct trace_buffer *buffer; |
6c43e554 SRRH |
6207 | int cpu; |
6208 | int ret = 0; | |
6209 | ||
a356646a | 6210 | if (security_locked_down(LOCKDOWN_TRACEFS)) { |
ee195452 | 6211 | pr_warn("Lockdown is enabled, skipping ring buffer tests\n"); |
a356646a SRV |
6212 | return 0; |
6213 | } | |
6214 | ||
6c43e554 SRRH |
6215 | pr_info("Running ring buffer tests...\n"); |
6216 | ||
6217 | buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE); | |
6218 | if (WARN_ON(!buffer)) | |
6219 | return 0; | |
6220 | ||
6221 | /* Disable buffer so that threads can't write to it yet */ | |
6222 | ring_buffer_record_off(buffer); | |
6223 | ||
6224 | for_each_online_cpu(cpu) { | |
6225 | rb_data[cpu].buffer = buffer; | |
6226 | rb_data[cpu].cpu = cpu; | |
6227 | rb_data[cpu].cnt = cpu; | |
64ed3a04 CH |
6228 | rb_threads[cpu] = kthread_run_on_cpu(rb_test, &rb_data[cpu], |
6229 | cpu, "rbtester/%u"); | |
62277de7 | 6230 | if (WARN_ON(IS_ERR(rb_threads[cpu]))) { |
6c43e554 | 6231 | pr_cont("FAILED\n"); |
62277de7 | 6232 | ret = PTR_ERR(rb_threads[cpu]); |
6c43e554 SRRH |
6233 | goto out_free; |
6234 | } | |
6c43e554 SRRH |
6235 | } |
6236 | ||
6237 | /* Now create the rb hammer! */ | |
6238 | rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer"); | |
62277de7 | 6239 | if (WARN_ON(IS_ERR(rb_hammer))) { |
6c43e554 | 6240 | pr_cont("FAILED\n"); |
62277de7 | 6241 | ret = PTR_ERR(rb_hammer); |
6c43e554 SRRH |
6242 | goto out_free; |
6243 | } | |
6244 | ||
6245 | ring_buffer_record_on(buffer); | |
6246 | /* | |
6247 | * Show buffer is enabled before setting rb_test_started. | |
6248 | * Yes there's a small race window where events could be | |
6249 | * dropped and the thread wont catch it. But when a ring | |
6250 | * buffer gets enabled, there will always be some kind of | |
6251 | * delay before other CPUs see it. Thus, we don't care about | |
6252 | * those dropped events. We care about events dropped after | |
6253 | * the threads see that the buffer is active. | |
6254 | */ | |
6255 | smp_wmb(); | |
6256 | rb_test_started = true; | |
6257 | ||
6258 | set_current_state(TASK_INTERRUPTIBLE); | |
6259 | /* Just run for 10 seconds */; | |
6260 | schedule_timeout(10 * HZ); | |
6261 | ||
6262 | kthread_stop(rb_hammer); | |
6263 | ||
6264 | out_free: | |
6265 | for_each_online_cpu(cpu) { | |
6266 | if (!rb_threads[cpu]) | |
6267 | break; | |
6268 | kthread_stop(rb_threads[cpu]); | |
6269 | } | |
6270 | if (ret) { | |
6271 | ring_buffer_free(buffer); | |
6272 | return ret; | |
6273 | } | |
6274 | ||
6275 | /* Report! */ | |
6276 | pr_info("finished\n"); | |
6277 | for_each_online_cpu(cpu) { | |
6278 | struct ring_buffer_event *event; | |
6279 | struct rb_test_data *data = &rb_data[cpu]; | |
6280 | struct rb_item *item; | |
6281 | unsigned long total_events; | |
6282 | unsigned long total_dropped; | |
6283 | unsigned long total_written; | |
6284 | unsigned long total_alloc; | |
6285 | unsigned long total_read = 0; | |
6286 | unsigned long total_size = 0; | |
6287 | unsigned long total_len = 0; | |
6288 | unsigned long total_lost = 0; | |
6289 | unsigned long lost; | |
6290 | int big_event_size; | |
6291 | int small_event_size; | |
6292 | ||
6293 | ret = -1; | |
6294 | ||
6295 | total_events = data->events + data->events_nested; | |
6296 | total_written = data->bytes_written + data->bytes_written_nested; | |
6297 | total_alloc = data->bytes_alloc + data->bytes_alloc_nested; | |
6298 | total_dropped = data->bytes_dropped + data->bytes_dropped_nested; | |
6299 | ||
6300 | big_event_size = data->max_size + data->max_size_nested; | |
6301 | small_event_size = data->min_size + data->min_size_nested; | |
6302 | ||
6303 | pr_info("CPU %d:\n", cpu); | |
6304 | pr_info(" events: %ld\n", total_events); | |
6305 | pr_info(" dropped bytes: %ld\n", total_dropped); | |
6306 | pr_info(" alloced bytes: %ld\n", total_alloc); | |
6307 | pr_info(" written bytes: %ld\n", total_written); | |
6308 | pr_info(" biggest event: %d\n", big_event_size); | |
6309 | pr_info(" smallest event: %d\n", small_event_size); | |
6310 | ||
6311 | if (RB_WARN_ON(buffer, total_dropped)) | |
6312 | break; | |
6313 | ||
6314 | ret = 0; | |
6315 | ||
6316 | while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) { | |
6317 | total_lost += lost; | |
6318 | item = ring_buffer_event_data(event); | |
6319 | total_len += ring_buffer_event_length(event); | |
6320 | total_size += item->size + sizeof(struct rb_item); | |
6321 | if (memcmp(&item->str[0], rb_string, item->size) != 0) { | |
6322 | pr_info("FAILED!\n"); | |
6323 | pr_info("buffer had: %.*s\n", item->size, item->str); | |
6324 | pr_info("expected: %.*s\n", item->size, rb_string); | |
6325 | RB_WARN_ON(buffer, 1); | |
6326 | ret = -1; | |
6327 | break; | |
6328 | } | |
6329 | total_read++; | |
6330 | } | |
6331 | if (ret) | |
6332 | break; | |
6333 | ||
6334 | ret = -1; | |
6335 | ||
6336 | pr_info(" read events: %ld\n", total_read); | |
6337 | pr_info(" lost events: %ld\n", total_lost); | |
6338 | pr_info(" total events: %ld\n", total_lost + total_read); | |
6339 | pr_info(" recorded len bytes: %ld\n", total_len); | |
6340 | pr_info(" recorded size bytes: %ld\n", total_size); | |
ed888241 | 6341 | if (total_lost) { |
6c43e554 SRRH |
6342 | pr_info(" With dropped events, record len and size may not match\n" |
6343 | " alloced and written from above\n"); | |
ed888241 | 6344 | } else { |
6c43e554 SRRH |
6345 | if (RB_WARN_ON(buffer, total_len != total_alloc || |
6346 | total_size != total_written)) | |
6347 | break; | |
6348 | } | |
6349 | if (RB_WARN_ON(buffer, total_lost + total_read != total_events)) | |
6350 | break; | |
6351 | ||
6352 | ret = 0; | |
6353 | } | |
6354 | if (!ret) | |
6355 | pr_info("Ring buffer PASSED!\n"); | |
6356 | ||
6357 | ring_buffer_free(buffer); | |
6358 | return 0; | |
6359 | } | |
6360 | ||
6361 | late_initcall(test_ringbuffer); | |
6362 | #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */ |