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