• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Generic ring buffer
4  *
5  * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6  */
7 #include <linux/trace_events.h>
8 #include <linux/ring_buffer.h>
9 #include <linux/trace_clock.h>
10 #include <linux/sched/clock.h>
11 #include <linux/trace_seq.h>
12 #include <linux/spinlock.h>
13 #include <linux/irq_work.h>
14 #include <linux/security.h>
15 #include <linux/uaccess.h>
16 #include <linux/hardirq.h>
17 #include <linux/kthread.h>	/* for self test */
18 #include <linux/module.h>
19 #include <linux/percpu.h>
20 #include <linux/mutex.h>
21 #include <linux/delay.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/list.h>
26 #include <linux/cpu.h>
27 #include <linux/oom.h>
28 
29 #include <asm/local.h>
30 
31 static void update_pages_handler(struct work_struct *work);
32 
33 /*
34  * The ring buffer header is special. We must manually up keep it.
35  */
ring_buffer_print_entry_header(struct trace_seq * s)36 int ring_buffer_print_entry_header(struct trace_seq *s)
37 {
38 	trace_seq_puts(s, "# compressed entry header\n");
39 	trace_seq_puts(s, "\ttype_len    :    5 bits\n");
40 	trace_seq_puts(s, "\ttime_delta  :   27 bits\n");
41 	trace_seq_puts(s, "\tarray       :   32 bits\n");
42 	trace_seq_putc(s, '\n');
43 	trace_seq_printf(s, "\tpadding     : type == %d\n",
44 			 RINGBUF_TYPE_PADDING);
45 	trace_seq_printf(s, "\ttime_extend : type == %d\n",
46 			 RINGBUF_TYPE_TIME_EXTEND);
47 	trace_seq_printf(s, "\ttime_stamp : type == %d\n",
48 			 RINGBUF_TYPE_TIME_STAMP);
49 	trace_seq_printf(s, "\tdata max type_len  == %d\n",
50 			 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
51 
52 	return !trace_seq_has_overflowed(s);
53 }
54 
55 /*
56  * The ring buffer is made up of a list of pages. A separate list of pages is
57  * allocated for each CPU. A writer may only write to a buffer that is
58  * associated with the CPU it is currently executing on.  A reader may read
59  * from any per cpu buffer.
60  *
61  * The reader is special. For each per cpu buffer, the reader has its own
62  * reader page. When a reader has read the entire reader page, this reader
63  * page is swapped with another page in the ring buffer.
64  *
65  * Now, as long as the writer is off the reader page, the reader can do what
66  * ever it wants with that page. The writer will never write to that page
67  * again (as long as it is out of the ring buffer).
68  *
69  * Here's some silly ASCII art.
70  *
71  *   +------+
72  *   |reader|          RING BUFFER
73  *   |page  |
74  *   +------+        +---+   +---+   +---+
75  *                   |   |-->|   |-->|   |
76  *                   +---+   +---+   +---+
77  *                     ^               |
78  *                     |               |
79  *                     +---------------+
80  *
81  *
82  *   +------+
83  *   |reader|          RING BUFFER
84  *   |page  |------------------v
85  *   +------+        +---+   +---+   +---+
86  *                   |   |-->|   |-->|   |
87  *                   +---+   +---+   +---+
88  *                     ^               |
89  *                     |               |
90  *                     +---------------+
91  *
92  *
93  *   +------+
94  *   |reader|          RING BUFFER
95  *   |page  |------------------v
96  *   +------+        +---+   +---+   +---+
97  *      ^            |   |-->|   |-->|   |
98  *      |            +---+   +---+   +---+
99  *      |                              |
100  *      |                              |
101  *      +------------------------------+
102  *
103  *
104  *   +------+
105  *   |buffer|          RING BUFFER
106  *   |page  |------------------v
107  *   +------+        +---+   +---+   +---+
108  *      ^            |   |   |   |-->|   |
109  *      |   New      +---+   +---+   +---+
110  *      |  Reader------^               |
111  *      |   page                       |
112  *      +------------------------------+
113  *
114  *
115  * After we make this swap, the reader can hand this page off to the splice
116  * code and be done with it. It can even allocate a new page if it needs to
117  * and swap that into the ring buffer.
118  *
119  * We will be using cmpxchg soon to make all this lockless.
120  *
121  */
122 
123 /* Used for individual buffers (after the counter) */
124 #define RB_BUFFER_OFF		(1 << 20)
125 
126 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
127 
128 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
129 #define RB_ALIGNMENT		4U
130 #define RB_MAX_SMALL_DATA	(RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
131 #define RB_EVNT_MIN_SIZE	8U	/* two 32bit words */
132 #define RB_ALIGN_DATA		__aligned(RB_ALIGNMENT)
133 
134 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
135 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
136 
137 enum {
138 	RB_LEN_TIME_EXTEND = 8,
139 	RB_LEN_TIME_STAMP =  8,
140 };
141 
142 #define skip_time_extend(event) \
143 	((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
144 
145 #define extended_time(event) \
146 	(event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
147 
rb_null_event(struct ring_buffer_event * event)148 static inline int rb_null_event(struct ring_buffer_event *event)
149 {
150 	return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
151 }
152 
rb_event_set_padding(struct ring_buffer_event * event)153 static void rb_event_set_padding(struct ring_buffer_event *event)
154 {
155 	/* padding has a NULL time_delta */
156 	event->type_len = RINGBUF_TYPE_PADDING;
157 	event->time_delta = 0;
158 }
159 
160 static unsigned
rb_event_data_length(struct ring_buffer_event * event)161 rb_event_data_length(struct ring_buffer_event *event)
162 {
163 	unsigned length;
164 
165 	if (event->type_len)
166 		length = event->type_len * RB_ALIGNMENT;
167 	else
168 		length = event->array[0];
169 	return length + RB_EVNT_HDR_SIZE;
170 }
171 
172 /*
173  * Return the length of the given event. Will return
174  * the length of the time extend if the event is a
175  * time extend.
176  */
177 static inline unsigned
rb_event_length(struct ring_buffer_event * event)178 rb_event_length(struct ring_buffer_event *event)
179 {
180 	switch (event->type_len) {
181 	case RINGBUF_TYPE_PADDING:
182 		if (rb_null_event(event))
183 			/* undefined */
184 			return -1;
185 		return  event->array[0] + RB_EVNT_HDR_SIZE;
186 
187 	case RINGBUF_TYPE_TIME_EXTEND:
188 		return RB_LEN_TIME_EXTEND;
189 
190 	case RINGBUF_TYPE_TIME_STAMP:
191 		return RB_LEN_TIME_STAMP;
192 
193 	case RINGBUF_TYPE_DATA:
194 		return rb_event_data_length(event);
195 	default:
196 		BUG();
197 	}
198 	/* not hit */
199 	return 0;
200 }
201 
202 /*
203  * Return total length of time extend and data,
204  *   or just the event length for all other events.
205  */
206 static inline unsigned
rb_event_ts_length(struct ring_buffer_event * event)207 rb_event_ts_length(struct ring_buffer_event *event)
208 {
209 	unsigned len = 0;
210 
211 	if (extended_time(event)) {
212 		/* time extends include the data event after it */
213 		len = RB_LEN_TIME_EXTEND;
214 		event = skip_time_extend(event);
215 	}
216 	return len + rb_event_length(event);
217 }
218 
219 /**
220  * ring_buffer_event_length - return the length of the event
221  * @event: the event to get the length of
222  *
223  * Returns the size of the data load of a data event.
224  * If the event is something other than a data event, it
225  * returns the size of the event itself. With the exception
226  * of a TIME EXTEND, where it still returns the size of the
227  * data load of the data event after it.
228  */
ring_buffer_event_length(struct ring_buffer_event * event)229 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
230 {
231 	unsigned length;
232 
233 	if (extended_time(event))
234 		event = skip_time_extend(event);
235 
236 	length = rb_event_length(event);
237 	if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
238 		return length;
239 	length -= RB_EVNT_HDR_SIZE;
240 	if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
241                 length -= sizeof(event->array[0]);
242 	return length;
243 }
244 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
245 
246 /* inline for ring buffer fast paths */
247 static __always_inline void *
rb_event_data(struct ring_buffer_event * event)248 rb_event_data(struct ring_buffer_event *event)
249 {
250 	if (extended_time(event))
251 		event = skip_time_extend(event);
252 	BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
253 	/* If length is in len field, then array[0] has the data */
254 	if (event->type_len)
255 		return (void *)&event->array[0];
256 	/* Otherwise length is in array[0] and array[1] has the data */
257 	return (void *)&event->array[1];
258 }
259 
260 /**
261  * ring_buffer_event_data - return the data of the event
262  * @event: the event to get the data from
263  */
ring_buffer_event_data(struct ring_buffer_event * event)264 void *ring_buffer_event_data(struct ring_buffer_event *event)
265 {
266 	return rb_event_data(event);
267 }
268 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
269 
270 #define for_each_buffer_cpu(buffer, cpu)		\
271 	for_each_cpu(cpu, buffer->cpumask)
272 
273 #define TS_SHIFT	27
274 #define TS_MASK		((1ULL << TS_SHIFT) - 1)
275 #define TS_DELTA_TEST	(~TS_MASK)
276 
277 /**
278  * ring_buffer_event_time_stamp - return the event's extended timestamp
279  * @event: the event to get the timestamp of
280  *
281  * Returns the extended timestamp associated with a data event.
282  * An extended time_stamp is a 64-bit timestamp represented
283  * internally in a special way that makes the best use of space
284  * contained within a ring buffer event.  This function decodes
285  * it and maps it to a straight u64 value.
286  */
ring_buffer_event_time_stamp(struct ring_buffer_event * event)287 u64 ring_buffer_event_time_stamp(struct ring_buffer_event *event)
288 {
289 	u64 ts;
290 
291 	ts = event->array[0];
292 	ts <<= TS_SHIFT;
293 	ts += event->time_delta;
294 
295 	return ts;
296 }
297 
298 /* Flag when events were overwritten */
299 #define RB_MISSED_EVENTS	(1 << 31)
300 /* Missed count stored at end */
301 #define RB_MISSED_STORED	(1 << 30)
302 
303 #define RB_MISSED_FLAGS		(RB_MISSED_EVENTS|RB_MISSED_STORED)
304 
305 struct buffer_data_page {
306 	u64		 time_stamp;	/* page time stamp */
307 	local_t		 commit;	/* write committed index */
308 	unsigned char	 data[] RB_ALIGN_DATA;	/* data of buffer page */
309 };
310 
311 /*
312  * Note, the buffer_page list must be first. The buffer pages
313  * are allocated in cache lines, which means that each buffer
314  * page will be at the beginning of a cache line, and thus
315  * the least significant bits will be zero. We use this to
316  * add flags in the list struct pointers, to make the ring buffer
317  * lockless.
318  */
319 struct buffer_page {
320 	struct list_head list;		/* list of buffer pages */
321 	local_t		 write;		/* index for next write */
322 	unsigned	 read;		/* index for next read */
323 	local_t		 entries;	/* entries on this page */
324 	unsigned long	 real_end;	/* real end of data */
325 	struct buffer_data_page *page;	/* Actual data page */
326 };
327 
328 /*
329  * The buffer page counters, write and entries, must be reset
330  * atomically when crossing page boundaries. To synchronize this
331  * update, two counters are inserted into the number. One is
332  * the actual counter for the write position or count on the page.
333  *
334  * The other is a counter of updaters. Before an update happens
335  * the update partition of the counter is incremented. This will
336  * allow the updater to update the counter atomically.
337  *
338  * The counter is 20 bits, and the state data is 12.
339  */
340 #define RB_WRITE_MASK		0xfffff
341 #define RB_WRITE_INTCNT		(1 << 20)
342 
rb_init_page(struct buffer_data_page * bpage)343 static void rb_init_page(struct buffer_data_page *bpage)
344 {
345 	local_set(&bpage->commit, 0);
346 }
347 
348 /*
349  * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
350  * this issue out.
351  */
free_buffer_page(struct buffer_page * bpage)352 static void free_buffer_page(struct buffer_page *bpage)
353 {
354 	free_page((unsigned long)bpage->page);
355 	kfree(bpage);
356 }
357 
358 /*
359  * We need to fit the time_stamp delta into 27 bits.
360  */
test_time_stamp(u64 delta)361 static inline int test_time_stamp(u64 delta)
362 {
363 	if (delta & TS_DELTA_TEST)
364 		return 1;
365 	return 0;
366 }
367 
368 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
369 
370 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
371 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
372 
ring_buffer_print_page_header(struct trace_seq * s)373 int ring_buffer_print_page_header(struct trace_seq *s)
374 {
375 	struct buffer_data_page field;
376 
377 	trace_seq_printf(s, "\tfield: u64 timestamp;\t"
378 			 "offset:0;\tsize:%u;\tsigned:%u;\n",
379 			 (unsigned int)sizeof(field.time_stamp),
380 			 (unsigned int)is_signed_type(u64));
381 
382 	trace_seq_printf(s, "\tfield: local_t commit;\t"
383 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
384 			 (unsigned int)offsetof(typeof(field), commit),
385 			 (unsigned int)sizeof(field.commit),
386 			 (unsigned int)is_signed_type(long));
387 
388 	trace_seq_printf(s, "\tfield: int overwrite;\t"
389 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
390 			 (unsigned int)offsetof(typeof(field), commit),
391 			 1,
392 			 (unsigned int)is_signed_type(long));
393 
394 	trace_seq_printf(s, "\tfield: char data;\t"
395 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
396 			 (unsigned int)offsetof(typeof(field), data),
397 			 (unsigned int)BUF_PAGE_SIZE,
398 			 (unsigned int)is_signed_type(char));
399 
400 	return !trace_seq_has_overflowed(s);
401 }
402 
403 struct rb_irq_work {
404 	struct irq_work			work;
405 	wait_queue_head_t		waiters;
406 	wait_queue_head_t		full_waiters;
407 	bool				waiters_pending;
408 	bool				full_waiters_pending;
409 	bool				wakeup_full;
410 };
411 
412 /*
413  * Structure to hold event state and handle nested events.
414  */
415 struct rb_event_info {
416 	u64			ts;
417 	u64			delta;
418 	unsigned long		length;
419 	struct buffer_page	*tail_page;
420 	int			add_timestamp;
421 };
422 
423 /*
424  * Used for which event context the event is in.
425  *  NMI     = 0
426  *  IRQ     = 1
427  *  SOFTIRQ = 2
428  *  NORMAL  = 3
429  *
430  * See trace_recursive_lock() comment below for more details.
431  */
432 enum {
433 	RB_CTX_NMI,
434 	RB_CTX_IRQ,
435 	RB_CTX_SOFTIRQ,
436 	RB_CTX_NORMAL,
437 	RB_CTX_MAX
438 };
439 
440 /*
441  * head_page == tail_page && head == tail then buffer is empty.
442  */
443 struct ring_buffer_per_cpu {
444 	int				cpu;
445 	atomic_t			record_disabled;
446 	struct ring_buffer		*buffer;
447 	raw_spinlock_t			reader_lock;	/* serialize readers */
448 	arch_spinlock_t			lock;
449 	struct lock_class_key		lock_key;
450 	struct buffer_data_page		*free_page;
451 	unsigned long			nr_pages;
452 	unsigned int			current_context;
453 	struct list_head		*pages;
454 	struct buffer_page		*head_page;	/* read from head */
455 	struct buffer_page		*tail_page;	/* write to tail */
456 	struct buffer_page		*commit_page;	/* committed pages */
457 	struct buffer_page		*reader_page;
458 	unsigned long			lost_events;
459 	unsigned long			last_overrun;
460 	unsigned long			nest;
461 	local_t				entries_bytes;
462 	local_t				entries;
463 	local_t				overrun;
464 	local_t				commit_overrun;
465 	local_t				dropped_events;
466 	local_t				committing;
467 	local_t				commits;
468 	local_t				pages_touched;
469 	local_t				pages_read;
470 	long				last_pages_touch;
471 	size_t				shortest_full;
472 	unsigned long			read;
473 	unsigned long			read_bytes;
474 	u64				write_stamp;
475 	u64				read_stamp;
476 	/* ring buffer pages to update, > 0 to add, < 0 to remove */
477 	long				nr_pages_to_update;
478 	struct list_head		new_pages; /* new pages to add */
479 	struct work_struct		update_pages_work;
480 	struct completion		update_done;
481 
482 	struct rb_irq_work		irq_work;
483 };
484 
485 struct ring_buffer {
486 	unsigned			flags;
487 	int				cpus;
488 	atomic_t			record_disabled;
489 	atomic_t			resize_disabled;
490 	cpumask_var_t			cpumask;
491 
492 	struct lock_class_key		*reader_lock_key;
493 
494 	struct mutex			mutex;
495 
496 	struct ring_buffer_per_cpu	**buffers;
497 
498 	struct hlist_node		node;
499 	u64				(*clock)(void);
500 
501 	struct rb_irq_work		irq_work;
502 	bool				time_stamp_abs;
503 };
504 
505 struct ring_buffer_iter {
506 	struct ring_buffer_per_cpu	*cpu_buffer;
507 	unsigned long			head;
508 	struct buffer_page		*head_page;
509 	struct buffer_page		*cache_reader_page;
510 	unsigned long			cache_read;
511 	u64				read_stamp;
512 };
513 
514 /**
515  * ring_buffer_nr_pages - get the number of buffer pages in the ring buffer
516  * @buffer: The ring_buffer to get the number of pages from
517  * @cpu: The cpu of the ring_buffer to get the number of pages from
518  *
519  * Returns the number of pages used by a per_cpu buffer of the ring buffer.
520  */
ring_buffer_nr_pages(struct ring_buffer * buffer,int cpu)521 size_t ring_buffer_nr_pages(struct ring_buffer *buffer, int cpu)
522 {
523 	return buffer->buffers[cpu]->nr_pages;
524 }
525 
526 /**
527  * ring_buffer_nr_pages_dirty - get the number of used pages in the ring buffer
528  * @buffer: The ring_buffer to get the number of pages from
529  * @cpu: The cpu of the ring_buffer to get the number of pages from
530  *
531  * Returns the number of pages that have content in the ring buffer.
532  */
ring_buffer_nr_dirty_pages(struct ring_buffer * buffer,int cpu)533 size_t ring_buffer_nr_dirty_pages(struct ring_buffer *buffer, int cpu)
534 {
535 	size_t read;
536 	size_t cnt;
537 
538 	read = local_read(&buffer->buffers[cpu]->pages_read);
539 	cnt = local_read(&buffer->buffers[cpu]->pages_touched);
540 	/* The reader can read an empty page, but not more than that */
541 	if (cnt < read) {
542 		WARN_ON_ONCE(read > cnt + 1);
543 		return 0;
544 	}
545 
546 	return cnt - read;
547 }
548 
549 /*
550  * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
551  *
552  * Schedules a delayed work to wake up any task that is blocked on the
553  * ring buffer waiters queue.
554  */
rb_wake_up_waiters(struct irq_work * work)555 static void rb_wake_up_waiters(struct irq_work *work)
556 {
557 	struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
558 
559 	wake_up_all(&rbwork->waiters);
560 	if (rbwork->wakeup_full) {
561 		rbwork->wakeup_full = false;
562 		wake_up_all(&rbwork->full_waiters);
563 	}
564 }
565 
566 /**
567  * ring_buffer_wait - wait for input to the ring buffer
568  * @buffer: buffer to wait on
569  * @cpu: the cpu buffer to wait on
570  * @full: wait until a full page is available, if @cpu != RING_BUFFER_ALL_CPUS
571  *
572  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
573  * as data is added to any of the @buffer's cpu buffers. Otherwise
574  * it will wait for data to be added to a specific cpu buffer.
575  */
ring_buffer_wait(struct ring_buffer * buffer,int cpu,int full)576 int ring_buffer_wait(struct ring_buffer *buffer, int cpu, int full)
577 {
578 	struct ring_buffer_per_cpu *uninitialized_var(cpu_buffer);
579 	DEFINE_WAIT(wait);
580 	struct rb_irq_work *work;
581 	int ret = 0;
582 
583 	/*
584 	 * Depending on what the caller is waiting for, either any
585 	 * data in any cpu buffer, or a specific buffer, put the
586 	 * caller on the appropriate wait queue.
587 	 */
588 	if (cpu == RING_BUFFER_ALL_CPUS) {
589 		work = &buffer->irq_work;
590 		/* Full only makes sense on per cpu reads */
591 		full = 0;
592 	} else {
593 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
594 			return -ENODEV;
595 		cpu_buffer = buffer->buffers[cpu];
596 		work = &cpu_buffer->irq_work;
597 	}
598 
599 
600 	while (true) {
601 		if (full)
602 			prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE);
603 		else
604 			prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE);
605 
606 		/*
607 		 * The events can happen in critical sections where
608 		 * checking a work queue can cause deadlocks.
609 		 * After adding a task to the queue, this flag is set
610 		 * only to notify events to try to wake up the queue
611 		 * using irq_work.
612 		 *
613 		 * We don't clear it even if the buffer is no longer
614 		 * empty. The flag only causes the next event to run
615 		 * irq_work to do the work queue wake up. The worse
616 		 * that can happen if we race with !trace_empty() is that
617 		 * an event will cause an irq_work to try to wake up
618 		 * an empty queue.
619 		 *
620 		 * There's no reason to protect this flag either, as
621 		 * the work queue and irq_work logic will do the necessary
622 		 * synchronization for the wake ups. The only thing
623 		 * that is necessary is that the wake up happens after
624 		 * a task has been queued. It's OK for spurious wake ups.
625 		 */
626 		if (full)
627 			work->full_waiters_pending = true;
628 		else
629 			work->waiters_pending = true;
630 
631 		if (signal_pending(current)) {
632 			ret = -EINTR;
633 			break;
634 		}
635 
636 		if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer))
637 			break;
638 
639 		if (cpu != RING_BUFFER_ALL_CPUS &&
640 		    !ring_buffer_empty_cpu(buffer, cpu)) {
641 			unsigned long flags;
642 			bool pagebusy;
643 			size_t nr_pages;
644 			size_t dirty;
645 
646 			if (!full)
647 				break;
648 
649 			raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
650 			pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
651 			nr_pages = cpu_buffer->nr_pages;
652 			dirty = ring_buffer_nr_dirty_pages(buffer, cpu);
653 			if (!cpu_buffer->shortest_full ||
654 			    cpu_buffer->shortest_full < full)
655 				cpu_buffer->shortest_full = full;
656 			raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
657 			if (!pagebusy &&
658 			    (!nr_pages || (dirty * 100) > full * nr_pages))
659 				break;
660 		}
661 
662 		schedule();
663 	}
664 
665 	if (full)
666 		finish_wait(&work->full_waiters, &wait);
667 	else
668 		finish_wait(&work->waiters, &wait);
669 
670 	return ret;
671 }
672 
673 /**
674  * ring_buffer_poll_wait - poll on buffer input
675  * @buffer: buffer to wait on
676  * @cpu: the cpu buffer to wait on
677  * @filp: the file descriptor
678  * @poll_table: The poll descriptor
679  *
680  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
681  * as data is added to any of the @buffer's cpu buffers. Otherwise
682  * it will wait for data to be added to a specific cpu buffer.
683  *
684  * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
685  * zero otherwise.
686  */
ring_buffer_poll_wait(struct ring_buffer * buffer,int cpu,struct file * filp,poll_table * poll_table)687 __poll_t ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu,
688 			  struct file *filp, poll_table *poll_table)
689 {
690 	struct ring_buffer_per_cpu *cpu_buffer;
691 	struct rb_irq_work *work;
692 
693 	if (cpu == RING_BUFFER_ALL_CPUS)
694 		work = &buffer->irq_work;
695 	else {
696 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
697 			return -EINVAL;
698 
699 		cpu_buffer = buffer->buffers[cpu];
700 		work = &cpu_buffer->irq_work;
701 	}
702 
703 	poll_wait(filp, &work->waiters, poll_table);
704 	work->waiters_pending = true;
705 	/*
706 	 * There's a tight race between setting the waiters_pending and
707 	 * checking if the ring buffer is empty.  Once the waiters_pending bit
708 	 * is set, the next event will wake the task up, but we can get stuck
709 	 * if there's only a single event in.
710 	 *
711 	 * FIXME: Ideally, we need a memory barrier on the writer side as well,
712 	 * but adding a memory barrier to all events will cause too much of a
713 	 * performance hit in the fast path.  We only need a memory barrier when
714 	 * the buffer goes from empty to having content.  But as this race is
715 	 * extremely small, and it's not a problem if another event comes in, we
716 	 * will fix it later.
717 	 */
718 	smp_mb();
719 
720 	if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
721 	    (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
722 		return EPOLLIN | EPOLLRDNORM;
723 	return 0;
724 }
725 
726 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
727 #define RB_WARN_ON(b, cond)						\
728 	({								\
729 		int _____ret = unlikely(cond);				\
730 		if (_____ret) {						\
731 			if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
732 				struct ring_buffer_per_cpu *__b =	\
733 					(void *)b;			\
734 				atomic_inc(&__b->buffer->record_disabled); \
735 			} else						\
736 				atomic_inc(&b->record_disabled);	\
737 			WARN_ON(1);					\
738 		}							\
739 		_____ret;						\
740 	})
741 
742 /* Up this if you want to test the TIME_EXTENTS and normalization */
743 #define DEBUG_SHIFT 0
744 
rb_time_stamp(struct ring_buffer * buffer)745 static inline u64 rb_time_stamp(struct ring_buffer *buffer)
746 {
747 	/* shift to debug/test normalization and TIME_EXTENTS */
748 	return buffer->clock() << DEBUG_SHIFT;
749 }
750 
ring_buffer_time_stamp(struct ring_buffer * buffer,int cpu)751 u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
752 {
753 	u64 time;
754 
755 	preempt_disable_notrace();
756 	time = rb_time_stamp(buffer);
757 	preempt_enable_notrace();
758 
759 	return time;
760 }
761 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
762 
ring_buffer_normalize_time_stamp(struct ring_buffer * buffer,int cpu,u64 * ts)763 void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
764 				      int cpu, u64 *ts)
765 {
766 	/* Just stupid testing the normalize function and deltas */
767 	*ts >>= DEBUG_SHIFT;
768 }
769 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
770 
771 /*
772  * Making the ring buffer lockless makes things tricky.
773  * Although writes only happen on the CPU that they are on,
774  * and they only need to worry about interrupts. Reads can
775  * happen on any CPU.
776  *
777  * The reader page is always off the ring buffer, but when the
778  * reader finishes with a page, it needs to swap its page with
779  * a new one from the buffer. The reader needs to take from
780  * the head (writes go to the tail). But if a writer is in overwrite
781  * mode and wraps, it must push the head page forward.
782  *
783  * Here lies the problem.
784  *
785  * The reader must be careful to replace only the head page, and
786  * not another one. As described at the top of the file in the
787  * ASCII art, the reader sets its old page to point to the next
788  * page after head. It then sets the page after head to point to
789  * the old reader page. But if the writer moves the head page
790  * during this operation, the reader could end up with the tail.
791  *
792  * We use cmpxchg to help prevent this race. We also do something
793  * special with the page before head. We set the LSB to 1.
794  *
795  * When the writer must push the page forward, it will clear the
796  * bit that points to the head page, move the head, and then set
797  * the bit that points to the new head page.
798  *
799  * We also don't want an interrupt coming in and moving the head
800  * page on another writer. Thus we use the second LSB to catch
801  * that too. Thus:
802  *
803  * head->list->prev->next        bit 1          bit 0
804  *                              -------        -------
805  * Normal page                     0              0
806  * Points to head page             0              1
807  * New head page                   1              0
808  *
809  * Note we can not trust the prev pointer of the head page, because:
810  *
811  * +----+       +-----+        +-----+
812  * |    |------>|  T  |---X--->|  N  |
813  * |    |<------|     |        |     |
814  * +----+       +-----+        +-----+
815  *   ^                           ^ |
816  *   |          +-----+          | |
817  *   +----------|  R  |----------+ |
818  *              |     |<-----------+
819  *              +-----+
820  *
821  * Key:  ---X-->  HEAD flag set in pointer
822  *         T      Tail page
823  *         R      Reader page
824  *         N      Next page
825  *
826  * (see __rb_reserve_next() to see where this happens)
827  *
828  *  What the above shows is that the reader just swapped out
829  *  the reader page with a page in the buffer, but before it
830  *  could make the new header point back to the new page added
831  *  it was preempted by a writer. The writer moved forward onto
832  *  the new page added by the reader and is about to move forward
833  *  again.
834  *
835  *  You can see, it is legitimate for the previous pointer of
836  *  the head (or any page) not to point back to itself. But only
837  *  temporarily.
838  */
839 
840 #define RB_PAGE_NORMAL		0UL
841 #define RB_PAGE_HEAD		1UL
842 #define RB_PAGE_UPDATE		2UL
843 
844 
845 #define RB_FLAG_MASK		3UL
846 
847 /* PAGE_MOVED is not part of the mask */
848 #define RB_PAGE_MOVED		4UL
849 
850 /*
851  * rb_list_head - remove any bit
852  */
rb_list_head(struct list_head * list)853 static struct list_head *rb_list_head(struct list_head *list)
854 {
855 	unsigned long val = (unsigned long)list;
856 
857 	return (struct list_head *)(val & ~RB_FLAG_MASK);
858 }
859 
860 /*
861  * rb_is_head_page - test if the given page is the head page
862  *
863  * Because the reader may move the head_page pointer, we can
864  * not trust what the head page is (it may be pointing to
865  * the reader page). But if the next page is a header page,
866  * its flags will be non zero.
867  */
868 static inline int
rb_is_head_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * page,struct list_head * list)869 rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer,
870 		struct buffer_page *page, struct list_head *list)
871 {
872 	unsigned long val;
873 
874 	val = (unsigned long)list->next;
875 
876 	if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
877 		return RB_PAGE_MOVED;
878 
879 	return val & RB_FLAG_MASK;
880 }
881 
882 /*
883  * rb_is_reader_page
884  *
885  * The unique thing about the reader page, is that, if the
886  * writer is ever on it, the previous pointer never points
887  * back to the reader page.
888  */
rb_is_reader_page(struct buffer_page * page)889 static bool rb_is_reader_page(struct buffer_page *page)
890 {
891 	struct list_head *list = page->list.prev;
892 
893 	return rb_list_head(list->next) != &page->list;
894 }
895 
896 /*
897  * rb_set_list_to_head - set a list_head to be pointing to head.
898  */
rb_set_list_to_head(struct ring_buffer_per_cpu * cpu_buffer,struct list_head * list)899 static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer,
900 				struct list_head *list)
901 {
902 	unsigned long *ptr;
903 
904 	ptr = (unsigned long *)&list->next;
905 	*ptr |= RB_PAGE_HEAD;
906 	*ptr &= ~RB_PAGE_UPDATE;
907 }
908 
909 /*
910  * rb_head_page_activate - sets up head page
911  */
rb_head_page_activate(struct ring_buffer_per_cpu * cpu_buffer)912 static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
913 {
914 	struct buffer_page *head;
915 
916 	head = cpu_buffer->head_page;
917 	if (!head)
918 		return;
919 
920 	/*
921 	 * Set the previous list pointer to have the HEAD flag.
922 	 */
923 	rb_set_list_to_head(cpu_buffer, head->list.prev);
924 }
925 
rb_list_head_clear(struct list_head * list)926 static void rb_list_head_clear(struct list_head *list)
927 {
928 	unsigned long *ptr = (unsigned long *)&list->next;
929 
930 	*ptr &= ~RB_FLAG_MASK;
931 }
932 
933 /*
934  * rb_head_page_deactivate - clears head page ptr (for free list)
935  */
936 static void
rb_head_page_deactivate(struct ring_buffer_per_cpu * cpu_buffer)937 rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
938 {
939 	struct list_head *hd;
940 
941 	/* Go through the whole list and clear any pointers found. */
942 	rb_list_head_clear(cpu_buffer->pages);
943 
944 	list_for_each(hd, cpu_buffer->pages)
945 		rb_list_head_clear(hd);
946 }
947 
rb_head_page_set(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag,int new_flag)948 static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
949 			    struct buffer_page *head,
950 			    struct buffer_page *prev,
951 			    int old_flag, int new_flag)
952 {
953 	struct list_head *list;
954 	unsigned long val = (unsigned long)&head->list;
955 	unsigned long ret;
956 
957 	list = &prev->list;
958 
959 	val &= ~RB_FLAG_MASK;
960 
961 	ret = cmpxchg((unsigned long *)&list->next,
962 		      val | old_flag, val | new_flag);
963 
964 	/* check if the reader took the page */
965 	if ((ret & ~RB_FLAG_MASK) != val)
966 		return RB_PAGE_MOVED;
967 
968 	return ret & RB_FLAG_MASK;
969 }
970 
rb_head_page_set_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)971 static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
972 				   struct buffer_page *head,
973 				   struct buffer_page *prev,
974 				   int old_flag)
975 {
976 	return rb_head_page_set(cpu_buffer, head, prev,
977 				old_flag, RB_PAGE_UPDATE);
978 }
979 
rb_head_page_set_head(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)980 static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
981 				 struct buffer_page *head,
982 				 struct buffer_page *prev,
983 				 int old_flag)
984 {
985 	return rb_head_page_set(cpu_buffer, head, prev,
986 				old_flag, RB_PAGE_HEAD);
987 }
988 
rb_head_page_set_normal(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)989 static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
990 				   struct buffer_page *head,
991 				   struct buffer_page *prev,
992 				   int old_flag)
993 {
994 	return rb_head_page_set(cpu_buffer, head, prev,
995 				old_flag, RB_PAGE_NORMAL);
996 }
997 
rb_inc_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page ** bpage)998 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
999 			       struct buffer_page **bpage)
1000 {
1001 	struct list_head *p = rb_list_head((*bpage)->list.next);
1002 
1003 	*bpage = list_entry(p, struct buffer_page, list);
1004 }
1005 
1006 static struct buffer_page *
rb_set_head_page(struct ring_buffer_per_cpu * cpu_buffer)1007 rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1008 {
1009 	struct buffer_page *head;
1010 	struct buffer_page *page;
1011 	struct list_head *list;
1012 	int i;
1013 
1014 	if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1015 		return NULL;
1016 
1017 	/* sanity check */
1018 	list = cpu_buffer->pages;
1019 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1020 		return NULL;
1021 
1022 	page = head = cpu_buffer->head_page;
1023 	/*
1024 	 * It is possible that the writer moves the header behind
1025 	 * where we started, and we miss in one loop.
1026 	 * A second loop should grab the header, but we'll do
1027 	 * three loops just because I'm paranoid.
1028 	 */
1029 	for (i = 0; i < 3; i++) {
1030 		do {
1031 			if (rb_is_head_page(cpu_buffer, page, page->list.prev)) {
1032 				cpu_buffer->head_page = page;
1033 				return page;
1034 			}
1035 			rb_inc_page(cpu_buffer, &page);
1036 		} while (page != head);
1037 	}
1038 
1039 	RB_WARN_ON(cpu_buffer, 1);
1040 
1041 	return NULL;
1042 }
1043 
rb_head_page_replace(struct buffer_page * old,struct buffer_page * new)1044 static int rb_head_page_replace(struct buffer_page *old,
1045 				struct buffer_page *new)
1046 {
1047 	unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1048 	unsigned long val;
1049 	unsigned long ret;
1050 
1051 	val = *ptr & ~RB_FLAG_MASK;
1052 	val |= RB_PAGE_HEAD;
1053 
1054 	ret = cmpxchg(ptr, val, (unsigned long)&new->list);
1055 
1056 	return ret == val;
1057 }
1058 
1059 /*
1060  * rb_tail_page_update - move the tail page forward
1061  */
rb_tail_page_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)1062 static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1063 			       struct buffer_page *tail_page,
1064 			       struct buffer_page *next_page)
1065 {
1066 	unsigned long old_entries;
1067 	unsigned long old_write;
1068 
1069 	/*
1070 	 * The tail page now needs to be moved forward.
1071 	 *
1072 	 * We need to reset the tail page, but without messing
1073 	 * with possible erasing of data brought in by interrupts
1074 	 * that have moved the tail page and are currently on it.
1075 	 *
1076 	 * We add a counter to the write field to denote this.
1077 	 */
1078 	old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
1079 	old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
1080 
1081 	local_inc(&cpu_buffer->pages_touched);
1082 	/*
1083 	 * Just make sure we have seen our old_write and synchronize
1084 	 * with any interrupts that come in.
1085 	 */
1086 	barrier();
1087 
1088 	/*
1089 	 * If the tail page is still the same as what we think
1090 	 * it is, then it is up to us to update the tail
1091 	 * pointer.
1092 	 */
1093 	if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1094 		/* Zero the write counter */
1095 		unsigned long val = old_write & ~RB_WRITE_MASK;
1096 		unsigned long eval = old_entries & ~RB_WRITE_MASK;
1097 
1098 		/*
1099 		 * This will only succeed if an interrupt did
1100 		 * not come in and change it. In which case, we
1101 		 * do not want to modify it.
1102 		 *
1103 		 * We add (void) to let the compiler know that we do not care
1104 		 * about the return value of these functions. We use the
1105 		 * cmpxchg to only update if an interrupt did not already
1106 		 * do it for us. If the cmpxchg fails, we don't care.
1107 		 */
1108 		(void)local_cmpxchg(&next_page->write, old_write, val);
1109 		(void)local_cmpxchg(&next_page->entries, old_entries, eval);
1110 
1111 		/*
1112 		 * No need to worry about races with clearing out the commit.
1113 		 * it only can increment when a commit takes place. But that
1114 		 * only happens in the outer most nested commit.
1115 		 */
1116 		local_set(&next_page->page->commit, 0);
1117 
1118 		/* Again, either we update tail_page or an interrupt does */
1119 		(void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page);
1120 	}
1121 }
1122 
rb_check_bpage(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * bpage)1123 static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1124 			  struct buffer_page *bpage)
1125 {
1126 	unsigned long val = (unsigned long)bpage;
1127 
1128 	if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK))
1129 		return 1;
1130 
1131 	return 0;
1132 }
1133 
1134 /**
1135  * rb_check_list - make sure a pointer to a list has the last bits zero
1136  */
rb_check_list(struct ring_buffer_per_cpu * cpu_buffer,struct list_head * list)1137 static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer,
1138 			 struct list_head *list)
1139 {
1140 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev))
1141 		return 1;
1142 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next))
1143 		return 1;
1144 	return 0;
1145 }
1146 
1147 /**
1148  * rb_check_pages - integrity check of buffer pages
1149  * @cpu_buffer: CPU buffer with pages to test
1150  *
1151  * As a safety measure we check to make sure the data pages have not
1152  * been corrupted.
1153  */
rb_check_pages(struct ring_buffer_per_cpu * cpu_buffer)1154 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1155 {
1156 	struct list_head *head = cpu_buffer->pages;
1157 	struct buffer_page *bpage, *tmp;
1158 
1159 	/* Reset the head page if it exists */
1160 	if (cpu_buffer->head_page)
1161 		rb_set_head_page(cpu_buffer);
1162 
1163 	rb_head_page_deactivate(cpu_buffer);
1164 
1165 	if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
1166 		return -1;
1167 	if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
1168 		return -1;
1169 
1170 	if (rb_check_list(cpu_buffer, head))
1171 		return -1;
1172 
1173 	list_for_each_entry_safe(bpage, tmp, head, list) {
1174 		if (RB_WARN_ON(cpu_buffer,
1175 			       bpage->list.next->prev != &bpage->list))
1176 			return -1;
1177 		if (RB_WARN_ON(cpu_buffer,
1178 			       bpage->list.prev->next != &bpage->list))
1179 			return -1;
1180 		if (rb_check_list(cpu_buffer, &bpage->list))
1181 			return -1;
1182 	}
1183 
1184 	rb_head_page_activate(cpu_buffer);
1185 
1186 	return 0;
1187 }
1188 
__rb_allocate_pages(long nr_pages,struct list_head * pages,int cpu)1189 static int __rb_allocate_pages(long nr_pages, struct list_head *pages, int cpu)
1190 {
1191 	struct buffer_page *bpage, *tmp;
1192 	bool user_thread = current->mm != NULL;
1193 	gfp_t mflags;
1194 	long i;
1195 
1196 	/*
1197 	 * Check if the available memory is there first.
1198 	 * Note, si_mem_available() only gives us a rough estimate of available
1199 	 * memory. It may not be accurate. But we don't care, we just want
1200 	 * to prevent doing any allocation when it is obvious that it is
1201 	 * not going to succeed.
1202 	 */
1203 	i = si_mem_available();
1204 	if (i < nr_pages)
1205 		return -ENOMEM;
1206 
1207 	/*
1208 	 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
1209 	 * gracefully without invoking oom-killer and the system is not
1210 	 * destabilized.
1211 	 */
1212 	mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL;
1213 
1214 	/*
1215 	 * If a user thread allocates too much, and si_mem_available()
1216 	 * reports there's enough memory, even though there is not.
1217 	 * Make sure the OOM killer kills this thread. This can happen
1218 	 * even with RETRY_MAYFAIL because another task may be doing
1219 	 * an allocation after this task has taken all memory.
1220 	 * This is the task the OOM killer needs to take out during this
1221 	 * loop, even if it was triggered by an allocation somewhere else.
1222 	 */
1223 	if (user_thread)
1224 		set_current_oom_origin();
1225 	for (i = 0; i < nr_pages; i++) {
1226 		struct page *page;
1227 
1228 		bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1229 				    mflags, cpu_to_node(cpu));
1230 		if (!bpage)
1231 			goto free_pages;
1232 
1233 		list_add(&bpage->list, pages);
1234 
1235 		page = alloc_pages_node(cpu_to_node(cpu), mflags, 0);
1236 		if (!page)
1237 			goto free_pages;
1238 		bpage->page = page_address(page);
1239 		rb_init_page(bpage->page);
1240 
1241 		if (user_thread && fatal_signal_pending(current))
1242 			goto free_pages;
1243 	}
1244 	if (user_thread)
1245 		clear_current_oom_origin();
1246 
1247 	return 0;
1248 
1249 free_pages:
1250 	list_for_each_entry_safe(bpage, tmp, pages, list) {
1251 		list_del_init(&bpage->list);
1252 		free_buffer_page(bpage);
1253 	}
1254 	if (user_thread)
1255 		clear_current_oom_origin();
1256 
1257 	return -ENOMEM;
1258 }
1259 
rb_allocate_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1260 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1261 			     unsigned long nr_pages)
1262 {
1263 	LIST_HEAD(pages);
1264 
1265 	WARN_ON(!nr_pages);
1266 
1267 	if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu))
1268 		return -ENOMEM;
1269 
1270 	/*
1271 	 * The ring buffer page list is a circular list that does not
1272 	 * start and end with a list head. All page list items point to
1273 	 * other pages.
1274 	 */
1275 	cpu_buffer->pages = pages.next;
1276 	list_del(&pages);
1277 
1278 	cpu_buffer->nr_pages = nr_pages;
1279 
1280 	rb_check_pages(cpu_buffer);
1281 
1282 	return 0;
1283 }
1284 
1285 static struct ring_buffer_per_cpu *
rb_allocate_cpu_buffer(struct ring_buffer * buffer,long nr_pages,int cpu)1286 rb_allocate_cpu_buffer(struct ring_buffer *buffer, long nr_pages, int cpu)
1287 {
1288 	struct ring_buffer_per_cpu *cpu_buffer;
1289 	struct buffer_page *bpage;
1290 	struct page *page;
1291 	int ret;
1292 
1293 	cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1294 				  GFP_KERNEL, cpu_to_node(cpu));
1295 	if (!cpu_buffer)
1296 		return NULL;
1297 
1298 	cpu_buffer->cpu = cpu;
1299 	cpu_buffer->buffer = buffer;
1300 	raw_spin_lock_init(&cpu_buffer->reader_lock);
1301 	lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
1302 	cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1303 	INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
1304 	init_completion(&cpu_buffer->update_done);
1305 	init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
1306 	init_waitqueue_head(&cpu_buffer->irq_work.waiters);
1307 	init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
1308 
1309 	bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1310 			    GFP_KERNEL, cpu_to_node(cpu));
1311 	if (!bpage)
1312 		goto fail_free_buffer;
1313 
1314 	rb_check_bpage(cpu_buffer, bpage);
1315 
1316 	cpu_buffer->reader_page = bpage;
1317 	page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1318 	if (!page)
1319 		goto fail_free_reader;
1320 	bpage->page = page_address(page);
1321 	rb_init_page(bpage->page);
1322 
1323 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1324 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
1325 
1326 	ret = rb_allocate_pages(cpu_buffer, nr_pages);
1327 	if (ret < 0)
1328 		goto fail_free_reader;
1329 
1330 	cpu_buffer->head_page
1331 		= list_entry(cpu_buffer->pages, struct buffer_page, list);
1332 	cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
1333 
1334 	rb_head_page_activate(cpu_buffer);
1335 
1336 	return cpu_buffer;
1337 
1338  fail_free_reader:
1339 	free_buffer_page(cpu_buffer->reader_page);
1340 
1341  fail_free_buffer:
1342 	kfree(cpu_buffer);
1343 	return NULL;
1344 }
1345 
rb_free_cpu_buffer(struct ring_buffer_per_cpu * cpu_buffer)1346 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1347 {
1348 	struct list_head *head = cpu_buffer->pages;
1349 	struct buffer_page *bpage, *tmp;
1350 
1351 	free_buffer_page(cpu_buffer->reader_page);
1352 
1353 	rb_head_page_deactivate(cpu_buffer);
1354 
1355 	if (head) {
1356 		list_for_each_entry_safe(bpage, tmp, head, list) {
1357 			list_del_init(&bpage->list);
1358 			free_buffer_page(bpage);
1359 		}
1360 		bpage = list_entry(head, struct buffer_page, list);
1361 		free_buffer_page(bpage);
1362 	}
1363 
1364 	kfree(cpu_buffer);
1365 }
1366 
1367 /**
1368  * __ring_buffer_alloc - allocate a new ring_buffer
1369  * @size: the size in bytes per cpu that is needed.
1370  * @flags: attributes to set for the ring buffer.
1371  *
1372  * Currently the only flag that is available is the RB_FL_OVERWRITE
1373  * flag. This flag means that the buffer will overwrite old data
1374  * when the buffer wraps. If this flag is not set, the buffer will
1375  * drop data when the tail hits the head.
1376  */
__ring_buffer_alloc(unsigned long size,unsigned flags,struct lock_class_key * key)1377 struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1378 					struct lock_class_key *key)
1379 {
1380 	struct ring_buffer *buffer;
1381 	long nr_pages;
1382 	int bsize;
1383 	int cpu;
1384 	int ret;
1385 
1386 	/* keep it in its own cache line */
1387 	buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1388 			 GFP_KERNEL);
1389 	if (!buffer)
1390 		return NULL;
1391 
1392 	if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1393 		goto fail_free_buffer;
1394 
1395 	nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1396 	buffer->flags = flags;
1397 	buffer->clock = trace_clock_local;
1398 	buffer->reader_lock_key = key;
1399 
1400 	init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters);
1401 	init_waitqueue_head(&buffer->irq_work.waiters);
1402 
1403 	/* need at least two pages */
1404 	if (nr_pages < 2)
1405 		nr_pages = 2;
1406 
1407 	buffer->cpus = nr_cpu_ids;
1408 
1409 	bsize = sizeof(void *) * nr_cpu_ids;
1410 	buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1411 				  GFP_KERNEL);
1412 	if (!buffer->buffers)
1413 		goto fail_free_cpumask;
1414 
1415 	cpu = raw_smp_processor_id();
1416 	cpumask_set_cpu(cpu, buffer->cpumask);
1417 	buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
1418 	if (!buffer->buffers[cpu])
1419 		goto fail_free_buffers;
1420 
1421 	ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1422 	if (ret < 0)
1423 		goto fail_free_buffers;
1424 
1425 	mutex_init(&buffer->mutex);
1426 
1427 	return buffer;
1428 
1429  fail_free_buffers:
1430 	for_each_buffer_cpu(buffer, cpu) {
1431 		if (buffer->buffers[cpu])
1432 			rb_free_cpu_buffer(buffer->buffers[cpu]);
1433 	}
1434 	kfree(buffer->buffers);
1435 
1436  fail_free_cpumask:
1437 	free_cpumask_var(buffer->cpumask);
1438 
1439  fail_free_buffer:
1440 	kfree(buffer);
1441 	return NULL;
1442 }
1443 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
1444 
1445 /**
1446  * ring_buffer_free - free a ring buffer.
1447  * @buffer: the buffer to free.
1448  */
1449 void
ring_buffer_free(struct ring_buffer * buffer)1450 ring_buffer_free(struct ring_buffer *buffer)
1451 {
1452 	int cpu;
1453 
1454 	cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1455 
1456 	for_each_buffer_cpu(buffer, cpu)
1457 		rb_free_cpu_buffer(buffer->buffers[cpu]);
1458 
1459 	kfree(buffer->buffers);
1460 	free_cpumask_var(buffer->cpumask);
1461 
1462 	kfree(buffer);
1463 }
1464 EXPORT_SYMBOL_GPL(ring_buffer_free);
1465 
ring_buffer_set_clock(struct ring_buffer * buffer,u64 (* clock)(void))1466 void ring_buffer_set_clock(struct ring_buffer *buffer,
1467 			   u64 (*clock)(void))
1468 {
1469 	buffer->clock = clock;
1470 }
1471 
ring_buffer_set_time_stamp_abs(struct ring_buffer * buffer,bool abs)1472 void ring_buffer_set_time_stamp_abs(struct ring_buffer *buffer, bool abs)
1473 {
1474 	buffer->time_stamp_abs = abs;
1475 }
1476 
ring_buffer_time_stamp_abs(struct ring_buffer * buffer)1477 bool ring_buffer_time_stamp_abs(struct ring_buffer *buffer)
1478 {
1479 	return buffer->time_stamp_abs;
1480 }
1481 
1482 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1483 
rb_page_entries(struct buffer_page * bpage)1484 static inline unsigned long rb_page_entries(struct buffer_page *bpage)
1485 {
1486 	return local_read(&bpage->entries) & RB_WRITE_MASK;
1487 }
1488 
rb_page_write(struct buffer_page * bpage)1489 static inline unsigned long rb_page_write(struct buffer_page *bpage)
1490 {
1491 	return local_read(&bpage->write) & RB_WRITE_MASK;
1492 }
1493 
1494 static int
rb_remove_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1495 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
1496 {
1497 	struct list_head *tail_page, *to_remove, *next_page;
1498 	struct buffer_page *to_remove_page, *tmp_iter_page;
1499 	struct buffer_page *last_page, *first_page;
1500 	unsigned long nr_removed;
1501 	unsigned long head_bit;
1502 	int page_entries;
1503 
1504 	head_bit = 0;
1505 
1506 	raw_spin_lock_irq(&cpu_buffer->reader_lock);
1507 	atomic_inc(&cpu_buffer->record_disabled);
1508 	/*
1509 	 * We don't race with the readers since we have acquired the reader
1510 	 * lock. We also don't race with writers after disabling recording.
1511 	 * This makes it easy to figure out the first and the last page to be
1512 	 * removed from the list. We unlink all the pages in between including
1513 	 * the first and last pages. This is done in a busy loop so that we
1514 	 * lose the least number of traces.
1515 	 * The pages are freed after we restart recording and unlock readers.
1516 	 */
1517 	tail_page = &cpu_buffer->tail_page->list;
1518 
1519 	/*
1520 	 * tail page might be on reader page, we remove the next page
1521 	 * from the ring buffer
1522 	 */
1523 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
1524 		tail_page = rb_list_head(tail_page->next);
1525 	to_remove = tail_page;
1526 
1527 	/* start of pages to remove */
1528 	first_page = list_entry(rb_list_head(to_remove->next),
1529 				struct buffer_page, list);
1530 
1531 	for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
1532 		to_remove = rb_list_head(to_remove)->next;
1533 		head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
1534 	}
1535 
1536 	next_page = rb_list_head(to_remove)->next;
1537 
1538 	/*
1539 	 * Now we remove all pages between tail_page and next_page.
1540 	 * Make sure that we have head_bit value preserved for the
1541 	 * next page
1542 	 */
1543 	tail_page->next = (struct list_head *)((unsigned long)next_page |
1544 						head_bit);
1545 	next_page = rb_list_head(next_page);
1546 	next_page->prev = tail_page;
1547 
1548 	/* make sure pages points to a valid page in the ring buffer */
1549 	cpu_buffer->pages = next_page;
1550 
1551 	/* update head page */
1552 	if (head_bit)
1553 		cpu_buffer->head_page = list_entry(next_page,
1554 						struct buffer_page, list);
1555 
1556 	/*
1557 	 * change read pointer to make sure any read iterators reset
1558 	 * themselves
1559 	 */
1560 	cpu_buffer->read = 0;
1561 
1562 	/* pages are removed, resume tracing and then free the pages */
1563 	atomic_dec(&cpu_buffer->record_disabled);
1564 	raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1565 
1566 	RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
1567 
1568 	/* last buffer page to remove */
1569 	last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
1570 				list);
1571 	tmp_iter_page = first_page;
1572 
1573 	do {
1574 		cond_resched();
1575 
1576 		to_remove_page = tmp_iter_page;
1577 		rb_inc_page(cpu_buffer, &tmp_iter_page);
1578 
1579 		/* update the counters */
1580 		page_entries = rb_page_entries(to_remove_page);
1581 		if (page_entries) {
1582 			/*
1583 			 * If something was added to this page, it was full
1584 			 * since it is not the tail page. So we deduct the
1585 			 * bytes consumed in ring buffer from here.
1586 			 * Increment overrun to account for the lost events.
1587 			 */
1588 			local_add(page_entries, &cpu_buffer->overrun);
1589 			local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
1590 		}
1591 
1592 		/*
1593 		 * We have already removed references to this list item, just
1594 		 * free up the buffer_page and its page
1595 		 */
1596 		free_buffer_page(to_remove_page);
1597 		nr_removed--;
1598 
1599 	} while (to_remove_page != last_page);
1600 
1601 	RB_WARN_ON(cpu_buffer, nr_removed);
1602 
1603 	return nr_removed == 0;
1604 }
1605 
1606 static int
rb_insert_pages(struct ring_buffer_per_cpu * cpu_buffer)1607 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
1608 {
1609 	struct list_head *pages = &cpu_buffer->new_pages;
1610 	int retries, success;
1611 
1612 	raw_spin_lock_irq(&cpu_buffer->reader_lock);
1613 	/*
1614 	 * We are holding the reader lock, so the reader page won't be swapped
1615 	 * in the ring buffer. Now we are racing with the writer trying to
1616 	 * move head page and the tail page.
1617 	 * We are going to adapt the reader page update process where:
1618 	 * 1. We first splice the start and end of list of new pages between
1619 	 *    the head page and its previous page.
1620 	 * 2. We cmpxchg the prev_page->next to point from head page to the
1621 	 *    start of new pages list.
1622 	 * 3. Finally, we update the head->prev to the end of new list.
1623 	 *
1624 	 * We will try this process 10 times, to make sure that we don't keep
1625 	 * spinning.
1626 	 */
1627 	retries = 10;
1628 	success = 0;
1629 	while (retries--) {
1630 		struct list_head *head_page, *prev_page, *r;
1631 		struct list_head *last_page, *first_page;
1632 		struct list_head *head_page_with_bit;
1633 
1634 		head_page = &rb_set_head_page(cpu_buffer)->list;
1635 		if (!head_page)
1636 			break;
1637 		prev_page = head_page->prev;
1638 
1639 		first_page = pages->next;
1640 		last_page  = pages->prev;
1641 
1642 		head_page_with_bit = (struct list_head *)
1643 				     ((unsigned long)head_page | RB_PAGE_HEAD);
1644 
1645 		last_page->next = head_page_with_bit;
1646 		first_page->prev = prev_page;
1647 
1648 		r = cmpxchg(&prev_page->next, head_page_with_bit, first_page);
1649 
1650 		if (r == head_page_with_bit) {
1651 			/*
1652 			 * yay, we replaced the page pointer to our new list,
1653 			 * now, we just have to update to head page's prev
1654 			 * pointer to point to end of list
1655 			 */
1656 			head_page->prev = last_page;
1657 			success = 1;
1658 			break;
1659 		}
1660 	}
1661 
1662 	if (success)
1663 		INIT_LIST_HEAD(pages);
1664 	/*
1665 	 * If we weren't successful in adding in new pages, warn and stop
1666 	 * tracing
1667 	 */
1668 	RB_WARN_ON(cpu_buffer, !success);
1669 	raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1670 
1671 	/* free pages if they weren't inserted */
1672 	if (!success) {
1673 		struct buffer_page *bpage, *tmp;
1674 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1675 					 list) {
1676 			list_del_init(&bpage->list);
1677 			free_buffer_page(bpage);
1678 		}
1679 	}
1680 	return success;
1681 }
1682 
rb_update_pages(struct ring_buffer_per_cpu * cpu_buffer)1683 static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
1684 {
1685 	int success;
1686 
1687 	if (cpu_buffer->nr_pages_to_update > 0)
1688 		success = rb_insert_pages(cpu_buffer);
1689 	else
1690 		success = rb_remove_pages(cpu_buffer,
1691 					-cpu_buffer->nr_pages_to_update);
1692 
1693 	if (success)
1694 		cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
1695 }
1696 
update_pages_handler(struct work_struct * work)1697 static void update_pages_handler(struct work_struct *work)
1698 {
1699 	struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
1700 			struct ring_buffer_per_cpu, update_pages_work);
1701 	rb_update_pages(cpu_buffer);
1702 	complete(&cpu_buffer->update_done);
1703 }
1704 
1705 /**
1706  * ring_buffer_resize - resize the ring buffer
1707  * @buffer: the buffer to resize.
1708  * @size: the new size.
1709  * @cpu_id: the cpu buffer to resize
1710  *
1711  * Minimum size is 2 * BUF_PAGE_SIZE.
1712  *
1713  * Returns 0 on success and < 0 on failure.
1714  */
ring_buffer_resize(struct ring_buffer * buffer,unsigned long size,int cpu_id)1715 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size,
1716 			int cpu_id)
1717 {
1718 	struct ring_buffer_per_cpu *cpu_buffer;
1719 	unsigned long nr_pages;
1720 	int cpu, err = 0;
1721 
1722 	/*
1723 	 * Always succeed at resizing a non-existent buffer:
1724 	 */
1725 	if (!buffer)
1726 		return size;
1727 
1728 	/* Make sure the requested buffer exists */
1729 	if (cpu_id != RING_BUFFER_ALL_CPUS &&
1730 	    !cpumask_test_cpu(cpu_id, buffer->cpumask))
1731 		return size;
1732 
1733 	nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1734 
1735 	/* we need a minimum of two pages */
1736 	if (nr_pages < 2)
1737 		nr_pages = 2;
1738 
1739 	size = nr_pages * BUF_PAGE_SIZE;
1740 
1741 	/*
1742 	 * Don't succeed if resizing is disabled, as a reader might be
1743 	 * manipulating the ring buffer and is expecting a sane state while
1744 	 * this is true.
1745 	 */
1746 	if (atomic_read(&buffer->resize_disabled))
1747 		return -EBUSY;
1748 
1749 	/* prevent another thread from changing buffer sizes */
1750 	mutex_lock(&buffer->mutex);
1751 
1752 	if (cpu_id == RING_BUFFER_ALL_CPUS) {
1753 		/* calculate the pages to update */
1754 		for_each_buffer_cpu(buffer, cpu) {
1755 			cpu_buffer = buffer->buffers[cpu];
1756 
1757 			cpu_buffer->nr_pages_to_update = nr_pages -
1758 							cpu_buffer->nr_pages;
1759 			/*
1760 			 * nothing more to do for removing pages or no update
1761 			 */
1762 			if (cpu_buffer->nr_pages_to_update <= 0)
1763 				continue;
1764 			/*
1765 			 * to add pages, make sure all new pages can be
1766 			 * allocated without receiving ENOMEM
1767 			 */
1768 			INIT_LIST_HEAD(&cpu_buffer->new_pages);
1769 			if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update,
1770 						&cpu_buffer->new_pages, cpu)) {
1771 				/* not enough memory for new pages */
1772 				err = -ENOMEM;
1773 				goto out_err;
1774 			}
1775 		}
1776 
1777 		get_online_cpus();
1778 		/*
1779 		 * Fire off all the required work handlers
1780 		 * We can't schedule on offline CPUs, but it's not necessary
1781 		 * since we can change their buffer sizes without any race.
1782 		 */
1783 		for_each_buffer_cpu(buffer, cpu) {
1784 			cpu_buffer = buffer->buffers[cpu];
1785 			if (!cpu_buffer->nr_pages_to_update)
1786 				continue;
1787 
1788 			/* Can't run something on an offline CPU. */
1789 			if (!cpu_online(cpu)) {
1790 				rb_update_pages(cpu_buffer);
1791 				cpu_buffer->nr_pages_to_update = 0;
1792 			} else {
1793 				schedule_work_on(cpu,
1794 						&cpu_buffer->update_pages_work);
1795 			}
1796 		}
1797 
1798 		/* wait for all the updates to complete */
1799 		for_each_buffer_cpu(buffer, cpu) {
1800 			cpu_buffer = buffer->buffers[cpu];
1801 			if (!cpu_buffer->nr_pages_to_update)
1802 				continue;
1803 
1804 			if (cpu_online(cpu))
1805 				wait_for_completion(&cpu_buffer->update_done);
1806 			cpu_buffer->nr_pages_to_update = 0;
1807 		}
1808 
1809 		put_online_cpus();
1810 	} else {
1811 		/* Make sure this CPU has been initialized */
1812 		if (!cpumask_test_cpu(cpu_id, buffer->cpumask))
1813 			goto out;
1814 
1815 		cpu_buffer = buffer->buffers[cpu_id];
1816 
1817 		if (nr_pages == cpu_buffer->nr_pages)
1818 			goto out;
1819 
1820 		cpu_buffer->nr_pages_to_update = nr_pages -
1821 						cpu_buffer->nr_pages;
1822 
1823 		INIT_LIST_HEAD(&cpu_buffer->new_pages);
1824 		if (cpu_buffer->nr_pages_to_update > 0 &&
1825 			__rb_allocate_pages(cpu_buffer->nr_pages_to_update,
1826 					    &cpu_buffer->new_pages, cpu_id)) {
1827 			err = -ENOMEM;
1828 			goto out_err;
1829 		}
1830 
1831 		get_online_cpus();
1832 
1833 		/* Can't run something on an offline CPU. */
1834 		if (!cpu_online(cpu_id))
1835 			rb_update_pages(cpu_buffer);
1836 		else {
1837 			schedule_work_on(cpu_id,
1838 					 &cpu_buffer->update_pages_work);
1839 			wait_for_completion(&cpu_buffer->update_done);
1840 		}
1841 
1842 		cpu_buffer->nr_pages_to_update = 0;
1843 		put_online_cpus();
1844 	}
1845 
1846  out:
1847 	/*
1848 	 * The ring buffer resize can happen with the ring buffer
1849 	 * enabled, so that the update disturbs the tracing as little
1850 	 * as possible. But if the buffer is disabled, we do not need
1851 	 * to worry about that, and we can take the time to verify
1852 	 * that the buffer is not corrupt.
1853 	 */
1854 	if (atomic_read(&buffer->record_disabled)) {
1855 		atomic_inc(&buffer->record_disabled);
1856 		/*
1857 		 * Even though the buffer was disabled, we must make sure
1858 		 * that it is truly disabled before calling rb_check_pages.
1859 		 * There could have been a race between checking
1860 		 * record_disable and incrementing it.
1861 		 */
1862 		synchronize_rcu();
1863 		for_each_buffer_cpu(buffer, cpu) {
1864 			cpu_buffer = buffer->buffers[cpu];
1865 			rb_check_pages(cpu_buffer);
1866 		}
1867 		atomic_dec(&buffer->record_disabled);
1868 	}
1869 
1870 	mutex_unlock(&buffer->mutex);
1871 	return size;
1872 
1873  out_err:
1874 	for_each_buffer_cpu(buffer, cpu) {
1875 		struct buffer_page *bpage, *tmp;
1876 
1877 		cpu_buffer = buffer->buffers[cpu];
1878 		cpu_buffer->nr_pages_to_update = 0;
1879 
1880 		if (list_empty(&cpu_buffer->new_pages))
1881 			continue;
1882 
1883 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1884 					list) {
1885 			list_del_init(&bpage->list);
1886 			free_buffer_page(bpage);
1887 		}
1888 	}
1889 	mutex_unlock(&buffer->mutex);
1890 	return err;
1891 }
1892 EXPORT_SYMBOL_GPL(ring_buffer_resize);
1893 
ring_buffer_change_overwrite(struct ring_buffer * buffer,int val)1894 void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val)
1895 {
1896 	mutex_lock(&buffer->mutex);
1897 	if (val)
1898 		buffer->flags |= RB_FL_OVERWRITE;
1899 	else
1900 		buffer->flags &= ~RB_FL_OVERWRITE;
1901 	mutex_unlock(&buffer->mutex);
1902 }
1903 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
1904 
__rb_page_index(struct buffer_page * bpage,unsigned index)1905 static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
1906 {
1907 	return bpage->page->data + index;
1908 }
1909 
1910 static __always_inline struct ring_buffer_event *
rb_reader_event(struct ring_buffer_per_cpu * cpu_buffer)1911 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
1912 {
1913 	return __rb_page_index(cpu_buffer->reader_page,
1914 			       cpu_buffer->reader_page->read);
1915 }
1916 
1917 static __always_inline struct ring_buffer_event *
rb_iter_head_event(struct ring_buffer_iter * iter)1918 rb_iter_head_event(struct ring_buffer_iter *iter)
1919 {
1920 	return __rb_page_index(iter->head_page, iter->head);
1921 }
1922 
rb_page_commit(struct buffer_page * bpage)1923 static __always_inline unsigned rb_page_commit(struct buffer_page *bpage)
1924 {
1925 	return local_read(&bpage->page->commit);
1926 }
1927 
1928 /* Size is determined by what has been committed */
rb_page_size(struct buffer_page * bpage)1929 static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
1930 {
1931 	return rb_page_commit(bpage);
1932 }
1933 
1934 static __always_inline unsigned
rb_commit_index(struct ring_buffer_per_cpu * cpu_buffer)1935 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
1936 {
1937 	return rb_page_commit(cpu_buffer->commit_page);
1938 }
1939 
1940 static __always_inline unsigned
rb_event_index(struct ring_buffer_event * event)1941 rb_event_index(struct ring_buffer_event *event)
1942 {
1943 	unsigned long addr = (unsigned long)event;
1944 
1945 	return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
1946 }
1947 
rb_inc_iter(struct ring_buffer_iter * iter)1948 static void rb_inc_iter(struct ring_buffer_iter *iter)
1949 {
1950 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1951 
1952 	/*
1953 	 * The iterator could be on the reader page (it starts there).
1954 	 * But the head could have moved, since the reader was
1955 	 * found. Check for this case and assign the iterator
1956 	 * to the head page instead of next.
1957 	 */
1958 	if (iter->head_page == cpu_buffer->reader_page)
1959 		iter->head_page = rb_set_head_page(cpu_buffer);
1960 	else
1961 		rb_inc_page(cpu_buffer, &iter->head_page);
1962 
1963 	iter->read_stamp = iter->head_page->page->time_stamp;
1964 	iter->head = 0;
1965 }
1966 
1967 /*
1968  * rb_handle_head_page - writer hit the head page
1969  *
1970  * Returns: +1 to retry page
1971  *           0 to continue
1972  *          -1 on error
1973  */
1974 static int
rb_handle_head_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)1975 rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
1976 		    struct buffer_page *tail_page,
1977 		    struct buffer_page *next_page)
1978 {
1979 	struct buffer_page *new_head;
1980 	int entries;
1981 	int type;
1982 	int ret;
1983 
1984 	entries = rb_page_entries(next_page);
1985 
1986 	/*
1987 	 * The hard part is here. We need to move the head
1988 	 * forward, and protect against both readers on
1989 	 * other CPUs and writers coming in via interrupts.
1990 	 */
1991 	type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
1992 				       RB_PAGE_HEAD);
1993 
1994 	/*
1995 	 * type can be one of four:
1996 	 *  NORMAL - an interrupt already moved it for us
1997 	 *  HEAD   - we are the first to get here.
1998 	 *  UPDATE - we are the interrupt interrupting
1999 	 *           a current move.
2000 	 *  MOVED  - a reader on another CPU moved the next
2001 	 *           pointer to its reader page. Give up
2002 	 *           and try again.
2003 	 */
2004 
2005 	switch (type) {
2006 	case RB_PAGE_HEAD:
2007 		/*
2008 		 * We changed the head to UPDATE, thus
2009 		 * it is our responsibility to update
2010 		 * the counters.
2011 		 */
2012 		local_add(entries, &cpu_buffer->overrun);
2013 		local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
2014 
2015 		/*
2016 		 * The entries will be zeroed out when we move the
2017 		 * tail page.
2018 		 */
2019 
2020 		/* still more to do */
2021 		break;
2022 
2023 	case RB_PAGE_UPDATE:
2024 		/*
2025 		 * This is an interrupt that interrupt the
2026 		 * previous update. Still more to do.
2027 		 */
2028 		break;
2029 	case RB_PAGE_NORMAL:
2030 		/*
2031 		 * An interrupt came in before the update
2032 		 * and processed this for us.
2033 		 * Nothing left to do.
2034 		 */
2035 		return 1;
2036 	case RB_PAGE_MOVED:
2037 		/*
2038 		 * The reader is on another CPU and just did
2039 		 * a swap with our next_page.
2040 		 * Try again.
2041 		 */
2042 		return 1;
2043 	default:
2044 		RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
2045 		return -1;
2046 	}
2047 
2048 	/*
2049 	 * Now that we are here, the old head pointer is
2050 	 * set to UPDATE. This will keep the reader from
2051 	 * swapping the head page with the reader page.
2052 	 * The reader (on another CPU) will spin till
2053 	 * we are finished.
2054 	 *
2055 	 * We just need to protect against interrupts
2056 	 * doing the job. We will set the next pointer
2057 	 * to HEAD. After that, we set the old pointer
2058 	 * to NORMAL, but only if it was HEAD before.
2059 	 * otherwise we are an interrupt, and only
2060 	 * want the outer most commit to reset it.
2061 	 */
2062 	new_head = next_page;
2063 	rb_inc_page(cpu_buffer, &new_head);
2064 
2065 	ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
2066 				    RB_PAGE_NORMAL);
2067 
2068 	/*
2069 	 * Valid returns are:
2070 	 *  HEAD   - an interrupt came in and already set it.
2071 	 *  NORMAL - One of two things:
2072 	 *            1) We really set it.
2073 	 *            2) A bunch of interrupts came in and moved
2074 	 *               the page forward again.
2075 	 */
2076 	switch (ret) {
2077 	case RB_PAGE_HEAD:
2078 	case RB_PAGE_NORMAL:
2079 		/* OK */
2080 		break;
2081 	default:
2082 		RB_WARN_ON(cpu_buffer, 1);
2083 		return -1;
2084 	}
2085 
2086 	/*
2087 	 * It is possible that an interrupt came in,
2088 	 * set the head up, then more interrupts came in
2089 	 * and moved it again. When we get back here,
2090 	 * the page would have been set to NORMAL but we
2091 	 * just set it back to HEAD.
2092 	 *
2093 	 * How do you detect this? Well, if that happened
2094 	 * the tail page would have moved.
2095 	 */
2096 	if (ret == RB_PAGE_NORMAL) {
2097 		struct buffer_page *buffer_tail_page;
2098 
2099 		buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
2100 		/*
2101 		 * If the tail had moved passed next, then we need
2102 		 * to reset the pointer.
2103 		 */
2104 		if (buffer_tail_page != tail_page &&
2105 		    buffer_tail_page != next_page)
2106 			rb_head_page_set_normal(cpu_buffer, new_head,
2107 						next_page,
2108 						RB_PAGE_HEAD);
2109 	}
2110 
2111 	/*
2112 	 * If this was the outer most commit (the one that
2113 	 * changed the original pointer from HEAD to UPDATE),
2114 	 * then it is up to us to reset it to NORMAL.
2115 	 */
2116 	if (type == RB_PAGE_HEAD) {
2117 		ret = rb_head_page_set_normal(cpu_buffer, next_page,
2118 					      tail_page,
2119 					      RB_PAGE_UPDATE);
2120 		if (RB_WARN_ON(cpu_buffer,
2121 			       ret != RB_PAGE_UPDATE))
2122 			return -1;
2123 	}
2124 
2125 	return 0;
2126 }
2127 
2128 static inline void
rb_reset_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2129 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
2130 	      unsigned long tail, struct rb_event_info *info)
2131 {
2132 	struct buffer_page *tail_page = info->tail_page;
2133 	struct ring_buffer_event *event;
2134 	unsigned long length = info->length;
2135 
2136 	/*
2137 	 * Only the event that crossed the page boundary
2138 	 * must fill the old tail_page with padding.
2139 	 */
2140 	if (tail >= BUF_PAGE_SIZE) {
2141 		/*
2142 		 * If the page was filled, then we still need
2143 		 * to update the real_end. Reset it to zero
2144 		 * and the reader will ignore it.
2145 		 */
2146 		if (tail == BUF_PAGE_SIZE)
2147 			tail_page->real_end = 0;
2148 
2149 		local_sub(length, &tail_page->write);
2150 		return;
2151 	}
2152 
2153 	event = __rb_page_index(tail_page, tail);
2154 
2155 	/* account for padding bytes */
2156 	local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
2157 
2158 	/*
2159 	 * Save the original length to the meta data.
2160 	 * This will be used by the reader to add lost event
2161 	 * counter.
2162 	 */
2163 	tail_page->real_end = tail;
2164 
2165 	/*
2166 	 * If this event is bigger than the minimum size, then
2167 	 * we need to be careful that we don't subtract the
2168 	 * write counter enough to allow another writer to slip
2169 	 * in on this page.
2170 	 * We put in a discarded commit instead, to make sure
2171 	 * that this space is not used again.
2172 	 *
2173 	 * If we are less than the minimum size, we don't need to
2174 	 * worry about it.
2175 	 */
2176 	if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
2177 		/* No room for any events */
2178 
2179 		/* Mark the rest of the page with padding */
2180 		rb_event_set_padding(event);
2181 
2182 		/* Set the write back to the previous setting */
2183 		local_sub(length, &tail_page->write);
2184 		return;
2185 	}
2186 
2187 	/* Put in a discarded event */
2188 	event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
2189 	event->type_len = RINGBUF_TYPE_PADDING;
2190 	/* time delta must be non zero */
2191 	event->time_delta = 1;
2192 
2193 	/* Set write to end of buffer */
2194 	length = (tail + length) - BUF_PAGE_SIZE;
2195 	local_sub(length, &tail_page->write);
2196 }
2197 
2198 static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
2199 
2200 /*
2201  * This is the slow path, force gcc not to inline it.
2202  */
2203 static noinline struct ring_buffer_event *
rb_move_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2204 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
2205 	     unsigned long tail, struct rb_event_info *info)
2206 {
2207 	struct buffer_page *tail_page = info->tail_page;
2208 	struct buffer_page *commit_page = cpu_buffer->commit_page;
2209 	struct ring_buffer *buffer = cpu_buffer->buffer;
2210 	struct buffer_page *next_page;
2211 	int ret;
2212 
2213 	next_page = tail_page;
2214 
2215 	rb_inc_page(cpu_buffer, &next_page);
2216 
2217 	/*
2218 	 * If for some reason, we had an interrupt storm that made
2219 	 * it all the way around the buffer, bail, and warn
2220 	 * about it.
2221 	 */
2222 	if (unlikely(next_page == commit_page)) {
2223 		local_inc(&cpu_buffer->commit_overrun);
2224 		goto out_reset;
2225 	}
2226 
2227 	/*
2228 	 * This is where the fun begins!
2229 	 *
2230 	 * We are fighting against races between a reader that
2231 	 * could be on another CPU trying to swap its reader
2232 	 * page with the buffer head.
2233 	 *
2234 	 * We are also fighting against interrupts coming in and
2235 	 * moving the head or tail on us as well.
2236 	 *
2237 	 * If the next page is the head page then we have filled
2238 	 * the buffer, unless the commit page is still on the
2239 	 * reader page.
2240 	 */
2241 	if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) {
2242 
2243 		/*
2244 		 * If the commit is not on the reader page, then
2245 		 * move the header page.
2246 		 */
2247 		if (!rb_is_reader_page(cpu_buffer->commit_page)) {
2248 			/*
2249 			 * If we are not in overwrite mode,
2250 			 * this is easy, just stop here.
2251 			 */
2252 			if (!(buffer->flags & RB_FL_OVERWRITE)) {
2253 				local_inc(&cpu_buffer->dropped_events);
2254 				goto out_reset;
2255 			}
2256 
2257 			ret = rb_handle_head_page(cpu_buffer,
2258 						  tail_page,
2259 						  next_page);
2260 			if (ret < 0)
2261 				goto out_reset;
2262 			if (ret)
2263 				goto out_again;
2264 		} else {
2265 			/*
2266 			 * We need to be careful here too. The
2267 			 * commit page could still be on the reader
2268 			 * page. We could have a small buffer, and
2269 			 * have filled up the buffer with events
2270 			 * from interrupts and such, and wrapped.
2271 			 *
2272 			 * Note, if the tail page is also the on the
2273 			 * reader_page, we let it move out.
2274 			 */
2275 			if (unlikely((cpu_buffer->commit_page !=
2276 				      cpu_buffer->tail_page) &&
2277 				     (cpu_buffer->commit_page ==
2278 				      cpu_buffer->reader_page))) {
2279 				local_inc(&cpu_buffer->commit_overrun);
2280 				goto out_reset;
2281 			}
2282 		}
2283 	}
2284 
2285 	rb_tail_page_update(cpu_buffer, tail_page, next_page);
2286 
2287  out_again:
2288 
2289 	rb_reset_tail(cpu_buffer, tail, info);
2290 
2291 	/* Commit what we have for now. */
2292 	rb_end_commit(cpu_buffer);
2293 	/* rb_end_commit() decs committing */
2294 	local_inc(&cpu_buffer->committing);
2295 
2296 	/* fail and let the caller try again */
2297 	return ERR_PTR(-EAGAIN);
2298 
2299  out_reset:
2300 	/* reset write */
2301 	rb_reset_tail(cpu_buffer, tail, info);
2302 
2303 	return NULL;
2304 }
2305 
2306 /* Slow path, do not inline */
2307 static noinline struct ring_buffer_event *
rb_add_time_stamp(struct ring_buffer_event * event,u64 delta,bool abs)2308 rb_add_time_stamp(struct ring_buffer_event *event, u64 delta, bool abs)
2309 {
2310 	if (abs)
2311 		event->type_len = RINGBUF_TYPE_TIME_STAMP;
2312 	else
2313 		event->type_len = RINGBUF_TYPE_TIME_EXTEND;
2314 
2315 	/* Not the first event on the page, or not delta? */
2316 	if (abs || rb_event_index(event)) {
2317 		event->time_delta = delta & TS_MASK;
2318 		event->array[0] = delta >> TS_SHIFT;
2319 	} else {
2320 		/* nope, just zero it */
2321 		event->time_delta = 0;
2322 		event->array[0] = 0;
2323 	}
2324 
2325 	return skip_time_extend(event);
2326 }
2327 
2328 static inline bool rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
2329 				     struct ring_buffer_event *event);
2330 
2331 /**
2332  * rb_update_event - update event type and data
2333  * @event: the event to update
2334  * @type: the type of event
2335  * @length: the size of the event field in the ring buffer
2336  *
2337  * Update the type and data fields of the event. The length
2338  * is the actual size that is written to the ring buffer,
2339  * and with this, we can determine what to place into the
2340  * data field.
2341  */
2342 static void
rb_update_event(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event,struct rb_event_info * info)2343 rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
2344 		struct ring_buffer_event *event,
2345 		struct rb_event_info *info)
2346 {
2347 	unsigned length = info->length;
2348 	u64 delta = info->delta;
2349 
2350 	/* Only a commit updates the timestamp */
2351 	if (unlikely(!rb_event_is_commit(cpu_buffer, event)))
2352 		delta = 0;
2353 
2354 	/*
2355 	 * If we need to add a timestamp, then we
2356 	 * add it to the start of the reserved space.
2357 	 */
2358 	if (unlikely(info->add_timestamp)) {
2359 		bool abs = ring_buffer_time_stamp_abs(cpu_buffer->buffer);
2360 
2361 		event = rb_add_time_stamp(event, info->delta, abs);
2362 		length -= RB_LEN_TIME_EXTEND;
2363 		delta = 0;
2364 	}
2365 
2366 	event->time_delta = delta;
2367 	length -= RB_EVNT_HDR_SIZE;
2368 	if (length > RB_MAX_SMALL_DATA) {
2369 		event->type_len = 0;
2370 		event->array[0] = length;
2371 	} else
2372 		event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
2373 }
2374 
rb_calculate_event_length(unsigned length)2375 static unsigned rb_calculate_event_length(unsigned length)
2376 {
2377 	struct ring_buffer_event event; /* Used only for sizeof array */
2378 
2379 	/* zero length can cause confusions */
2380 	if (!length)
2381 		length++;
2382 
2383 	if (length > RB_MAX_SMALL_DATA)
2384 		length += sizeof(event.array[0]);
2385 
2386 	length += RB_EVNT_HDR_SIZE;
2387 	length = ALIGN(length, RB_ALIGNMENT);
2388 
2389 	/*
2390 	 * In case the time delta is larger than the 27 bits for it
2391 	 * in the header, we need to add a timestamp. If another
2392 	 * event comes in when trying to discard this one to increase
2393 	 * the length, then the timestamp will be added in the allocated
2394 	 * space of this event. If length is bigger than the size needed
2395 	 * for the TIME_EXTEND, then padding has to be used. The events
2396 	 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
2397 	 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
2398 	 * As length is a multiple of 4, we only need to worry if it
2399 	 * is 12 (RB_LEN_TIME_EXTEND + 4).
2400 	 */
2401 	if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
2402 		length += RB_ALIGNMENT;
2403 
2404 	return length;
2405 }
2406 
2407 #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
sched_clock_stable(void)2408 static inline bool sched_clock_stable(void)
2409 {
2410 	return true;
2411 }
2412 #endif
2413 
2414 static inline int
rb_try_to_discard(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2415 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2416 		  struct ring_buffer_event *event)
2417 {
2418 	unsigned long new_index, old_index;
2419 	struct buffer_page *bpage;
2420 	unsigned long index;
2421 	unsigned long addr;
2422 
2423 	new_index = rb_event_index(event);
2424 	old_index = new_index + rb_event_ts_length(event);
2425 	addr = (unsigned long)event;
2426 	addr &= PAGE_MASK;
2427 
2428 	bpage = READ_ONCE(cpu_buffer->tail_page);
2429 
2430 	if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
2431 		unsigned long write_mask =
2432 			local_read(&bpage->write) & ~RB_WRITE_MASK;
2433 		unsigned long event_length = rb_event_length(event);
2434 		/*
2435 		 * This is on the tail page. It is possible that
2436 		 * a write could come in and move the tail page
2437 		 * and write to the next page. That is fine
2438 		 * because we just shorten what is on this page.
2439 		 */
2440 		old_index += write_mask;
2441 		new_index += write_mask;
2442 		index = local_cmpxchg(&bpage->write, old_index, new_index);
2443 		if (index == old_index) {
2444 			/* update counters */
2445 			local_sub(event_length, &cpu_buffer->entries_bytes);
2446 			return 1;
2447 		}
2448 	}
2449 
2450 	/* could not discard */
2451 	return 0;
2452 }
2453 
rb_start_commit(struct ring_buffer_per_cpu * cpu_buffer)2454 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
2455 {
2456 	local_inc(&cpu_buffer->committing);
2457 	local_inc(&cpu_buffer->commits);
2458 }
2459 
2460 static __always_inline void
rb_set_commit_to_write(struct ring_buffer_per_cpu * cpu_buffer)2461 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
2462 {
2463 	unsigned long max_count;
2464 
2465 	/*
2466 	 * We only race with interrupts and NMIs on this CPU.
2467 	 * If we own the commit event, then we can commit
2468 	 * all others that interrupted us, since the interruptions
2469 	 * are in stack format (they finish before they come
2470 	 * back to us). This allows us to do a simple loop to
2471 	 * assign the commit to the tail.
2472 	 */
2473  again:
2474 	max_count = cpu_buffer->nr_pages * 100;
2475 
2476 	while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
2477 		if (RB_WARN_ON(cpu_buffer, !(--max_count)))
2478 			return;
2479 		if (RB_WARN_ON(cpu_buffer,
2480 			       rb_is_reader_page(cpu_buffer->tail_page)))
2481 			return;
2482 		local_set(&cpu_buffer->commit_page->page->commit,
2483 			  rb_page_write(cpu_buffer->commit_page));
2484 		rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
2485 		/* Only update the write stamp if the page has an event */
2486 		if (rb_page_write(cpu_buffer->commit_page))
2487 			cpu_buffer->write_stamp =
2488 				cpu_buffer->commit_page->page->time_stamp;
2489 		/* add barrier to keep gcc from optimizing too much */
2490 		barrier();
2491 	}
2492 	while (rb_commit_index(cpu_buffer) !=
2493 	       rb_page_write(cpu_buffer->commit_page)) {
2494 
2495 		local_set(&cpu_buffer->commit_page->page->commit,
2496 			  rb_page_write(cpu_buffer->commit_page));
2497 		RB_WARN_ON(cpu_buffer,
2498 			   local_read(&cpu_buffer->commit_page->page->commit) &
2499 			   ~RB_WRITE_MASK);
2500 		barrier();
2501 	}
2502 
2503 	/* again, keep gcc from optimizing */
2504 	barrier();
2505 
2506 	/*
2507 	 * If an interrupt came in just after the first while loop
2508 	 * and pushed the tail page forward, we will be left with
2509 	 * a dangling commit that will never go forward.
2510 	 */
2511 	if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
2512 		goto again;
2513 }
2514 
rb_end_commit(struct ring_buffer_per_cpu * cpu_buffer)2515 static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
2516 {
2517 	unsigned long commits;
2518 
2519 	if (RB_WARN_ON(cpu_buffer,
2520 		       !local_read(&cpu_buffer->committing)))
2521 		return;
2522 
2523  again:
2524 	commits = local_read(&cpu_buffer->commits);
2525 	/* synchronize with interrupts */
2526 	barrier();
2527 	if (local_read(&cpu_buffer->committing) == 1)
2528 		rb_set_commit_to_write(cpu_buffer);
2529 
2530 	local_dec(&cpu_buffer->committing);
2531 
2532 	/* synchronize with interrupts */
2533 	barrier();
2534 
2535 	/*
2536 	 * Need to account for interrupts coming in between the
2537 	 * updating of the commit page and the clearing of the
2538 	 * committing counter.
2539 	 */
2540 	if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
2541 	    !local_read(&cpu_buffer->committing)) {
2542 		local_inc(&cpu_buffer->committing);
2543 		goto again;
2544 	}
2545 }
2546 
rb_event_discard(struct ring_buffer_event * event)2547 static inline void rb_event_discard(struct ring_buffer_event *event)
2548 {
2549 	if (extended_time(event))
2550 		event = skip_time_extend(event);
2551 
2552 	/* array[0] holds the actual length for the discarded event */
2553 	event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
2554 	event->type_len = RINGBUF_TYPE_PADDING;
2555 	/* time delta must be non zero */
2556 	if (!event->time_delta)
2557 		event->time_delta = 1;
2558 }
2559 
2560 static __always_inline bool
rb_event_is_commit(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2561 rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
2562 		   struct ring_buffer_event *event)
2563 {
2564 	unsigned long addr = (unsigned long)event;
2565 	unsigned long index;
2566 
2567 	index = rb_event_index(event);
2568 	addr &= PAGE_MASK;
2569 
2570 	return cpu_buffer->commit_page->page == (void *)addr &&
2571 		rb_commit_index(cpu_buffer) == index;
2572 }
2573 
2574 static __always_inline void
rb_update_write_stamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2575 rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2576 		      struct ring_buffer_event *event)
2577 {
2578 	u64 delta;
2579 
2580 	/*
2581 	 * The event first in the commit queue updates the
2582 	 * time stamp.
2583 	 */
2584 	if (rb_event_is_commit(cpu_buffer, event)) {
2585 		/*
2586 		 * A commit event that is first on a page
2587 		 * updates the write timestamp with the page stamp
2588 		 */
2589 		if (!rb_event_index(event))
2590 			cpu_buffer->write_stamp =
2591 				cpu_buffer->commit_page->page->time_stamp;
2592 		else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
2593 			delta = ring_buffer_event_time_stamp(event);
2594 			cpu_buffer->write_stamp += delta;
2595 		} else if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
2596 			delta = ring_buffer_event_time_stamp(event);
2597 			cpu_buffer->write_stamp = delta;
2598 		} else
2599 			cpu_buffer->write_stamp += event->time_delta;
2600 	}
2601 }
2602 
rb_commit(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2603 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
2604 		      struct ring_buffer_event *event)
2605 {
2606 	local_inc(&cpu_buffer->entries);
2607 	rb_update_write_stamp(cpu_buffer, event);
2608 	rb_end_commit(cpu_buffer);
2609 }
2610 
2611 static __always_inline void
rb_wakeups(struct ring_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer)2612 rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
2613 {
2614 	size_t nr_pages;
2615 	size_t dirty;
2616 	size_t full;
2617 
2618 	if (buffer->irq_work.waiters_pending) {
2619 		buffer->irq_work.waiters_pending = false;
2620 		/* irq_work_queue() supplies it's own memory barriers */
2621 		irq_work_queue(&buffer->irq_work.work);
2622 	}
2623 
2624 	if (cpu_buffer->irq_work.waiters_pending) {
2625 		cpu_buffer->irq_work.waiters_pending = false;
2626 		/* irq_work_queue() supplies it's own memory barriers */
2627 		irq_work_queue(&cpu_buffer->irq_work.work);
2628 	}
2629 
2630 	if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
2631 		return;
2632 
2633 	if (cpu_buffer->reader_page == cpu_buffer->commit_page)
2634 		return;
2635 
2636 	if (!cpu_buffer->irq_work.full_waiters_pending)
2637 		return;
2638 
2639 	cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
2640 
2641 	full = cpu_buffer->shortest_full;
2642 	nr_pages = cpu_buffer->nr_pages;
2643 	dirty = ring_buffer_nr_dirty_pages(buffer, cpu_buffer->cpu);
2644 	if (full && nr_pages && (dirty * 100) <= full * nr_pages)
2645 		return;
2646 
2647 	cpu_buffer->irq_work.wakeup_full = true;
2648 	cpu_buffer->irq_work.full_waiters_pending = false;
2649 	/* irq_work_queue() supplies it's own memory barriers */
2650 	irq_work_queue(&cpu_buffer->irq_work.work);
2651 }
2652 
2653 /*
2654  * The lock and unlock are done within a preempt disable section.
2655  * The current_context per_cpu variable can only be modified
2656  * by the current task between lock and unlock. But it can
2657  * be modified more than once via an interrupt. To pass this
2658  * information from the lock to the unlock without having to
2659  * access the 'in_interrupt()' functions again (which do show
2660  * a bit of overhead in something as critical as function tracing,
2661  * we use a bitmask trick.
2662  *
2663  *  bit 0 =  NMI context
2664  *  bit 1 =  IRQ context
2665  *  bit 2 =  SoftIRQ context
2666  *  bit 3 =  normal context.
2667  *
2668  * This works because this is the order of contexts that can
2669  * preempt other contexts. A SoftIRQ never preempts an IRQ
2670  * context.
2671  *
2672  * When the context is determined, the corresponding bit is
2673  * checked and set (if it was set, then a recursion of that context
2674  * happened).
2675  *
2676  * On unlock, we need to clear this bit. To do so, just subtract
2677  * 1 from the current_context and AND it to itself.
2678  *
2679  * (binary)
2680  *  101 - 1 = 100
2681  *  101 & 100 = 100 (clearing bit zero)
2682  *
2683  *  1010 - 1 = 1001
2684  *  1010 & 1001 = 1000 (clearing bit 1)
2685  *
2686  * The least significant bit can be cleared this way, and it
2687  * just so happens that it is the same bit corresponding to
2688  * the current context.
2689  */
2690 
2691 static __always_inline int
trace_recursive_lock(struct ring_buffer_per_cpu * cpu_buffer)2692 trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
2693 {
2694 	unsigned int val = cpu_buffer->current_context;
2695 	unsigned long pc = preempt_count();
2696 	int bit;
2697 
2698 	if (!(pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET)))
2699 		bit = RB_CTX_NORMAL;
2700 	else
2701 		bit = pc & NMI_MASK ? RB_CTX_NMI :
2702 			pc & HARDIRQ_MASK ? RB_CTX_IRQ : RB_CTX_SOFTIRQ;
2703 
2704 	if (unlikely(val & (1 << (bit + cpu_buffer->nest))))
2705 		return 1;
2706 
2707 	val |= (1 << (bit + cpu_buffer->nest));
2708 	cpu_buffer->current_context = val;
2709 
2710 	return 0;
2711 }
2712 
2713 static __always_inline void
trace_recursive_unlock(struct ring_buffer_per_cpu * cpu_buffer)2714 trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
2715 {
2716 	cpu_buffer->current_context &=
2717 		cpu_buffer->current_context - (1 << cpu_buffer->nest);
2718 }
2719 
2720 /* The recursive locking above uses 4 bits */
2721 #define NESTED_BITS 4
2722 
2723 /**
2724  * ring_buffer_nest_start - Allow to trace while nested
2725  * @buffer: The ring buffer to modify
2726  *
2727  * The ring buffer has a safety mechanism to prevent recursion.
2728  * But there may be a case where a trace needs to be done while
2729  * tracing something else. In this case, calling this function
2730  * will allow this function to nest within a currently active
2731  * ring_buffer_lock_reserve().
2732  *
2733  * Call this function before calling another ring_buffer_lock_reserve() and
2734  * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
2735  */
ring_buffer_nest_start(struct ring_buffer * buffer)2736 void ring_buffer_nest_start(struct ring_buffer *buffer)
2737 {
2738 	struct ring_buffer_per_cpu *cpu_buffer;
2739 	int cpu;
2740 
2741 	/* Enabled by ring_buffer_nest_end() */
2742 	preempt_disable_notrace();
2743 	cpu = raw_smp_processor_id();
2744 	cpu_buffer = buffer->buffers[cpu];
2745 	/* This is the shift value for the above recursive locking */
2746 	cpu_buffer->nest += NESTED_BITS;
2747 }
2748 
2749 /**
2750  * ring_buffer_nest_end - Allow to trace while nested
2751  * @buffer: The ring buffer to modify
2752  *
2753  * Must be called after ring_buffer_nest_start() and after the
2754  * ring_buffer_unlock_commit().
2755  */
ring_buffer_nest_end(struct ring_buffer * buffer)2756 void ring_buffer_nest_end(struct ring_buffer *buffer)
2757 {
2758 	struct ring_buffer_per_cpu *cpu_buffer;
2759 	int cpu;
2760 
2761 	/* disabled by ring_buffer_nest_start() */
2762 	cpu = raw_smp_processor_id();
2763 	cpu_buffer = buffer->buffers[cpu];
2764 	/* This is the shift value for the above recursive locking */
2765 	cpu_buffer->nest -= NESTED_BITS;
2766 	preempt_enable_notrace();
2767 }
2768 
2769 /**
2770  * ring_buffer_unlock_commit - commit a reserved
2771  * @buffer: The buffer to commit to
2772  * @event: The event pointer to commit.
2773  *
2774  * This commits the data to the ring buffer, and releases any locks held.
2775  *
2776  * Must be paired with ring_buffer_lock_reserve.
2777  */
ring_buffer_unlock_commit(struct ring_buffer * buffer,struct ring_buffer_event * event)2778 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
2779 			      struct ring_buffer_event *event)
2780 {
2781 	struct ring_buffer_per_cpu *cpu_buffer;
2782 	int cpu = raw_smp_processor_id();
2783 
2784 	cpu_buffer = buffer->buffers[cpu];
2785 
2786 	rb_commit(cpu_buffer, event);
2787 
2788 	rb_wakeups(buffer, cpu_buffer);
2789 
2790 	trace_recursive_unlock(cpu_buffer);
2791 
2792 	preempt_enable_notrace();
2793 
2794 	return 0;
2795 }
2796 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
2797 
2798 static noinline void
rb_handle_timestamp(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)2799 rb_handle_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2800 		    struct rb_event_info *info)
2801 {
2802 	WARN_ONCE(info->delta > (1ULL << 59),
2803 		  KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
2804 		  (unsigned long long)info->delta,
2805 		  (unsigned long long)info->ts,
2806 		  (unsigned long long)cpu_buffer->write_stamp,
2807 		  sched_clock_stable() ? "" :
2808 		  "If you just came from a suspend/resume,\n"
2809 		  "please switch to the trace global clock:\n"
2810 		  "  echo global > /sys/kernel/debug/tracing/trace_clock\n"
2811 		  "or add trace_clock=global to the kernel command line\n");
2812 	info->add_timestamp = 1;
2813 }
2814 
2815 static struct ring_buffer_event *
__rb_reserve_next(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)2816 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
2817 		  struct rb_event_info *info)
2818 {
2819 	struct ring_buffer_event *event;
2820 	struct buffer_page *tail_page;
2821 	unsigned long tail, write;
2822 
2823 	/*
2824 	 * If the time delta since the last event is too big to
2825 	 * hold in the time field of the event, then we append a
2826 	 * TIME EXTEND event ahead of the data event.
2827 	 */
2828 	if (unlikely(info->add_timestamp))
2829 		info->length += RB_LEN_TIME_EXTEND;
2830 
2831 	/* Don't let the compiler play games with cpu_buffer->tail_page */
2832 	tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
2833 	write = local_add_return(info->length, &tail_page->write);
2834 
2835 	/* set write to only the index of the write */
2836 	write &= RB_WRITE_MASK;
2837 	tail = write - info->length;
2838 
2839 	/*
2840 	 * If this is the first commit on the page, then it has the same
2841 	 * timestamp as the page itself.
2842 	 */
2843 	if (!tail && !ring_buffer_time_stamp_abs(cpu_buffer->buffer))
2844 		info->delta = 0;
2845 
2846 	/* See if we shot pass the end of this buffer page */
2847 	if (unlikely(write > BUF_PAGE_SIZE))
2848 		return rb_move_tail(cpu_buffer, tail, info);
2849 
2850 	/* We reserved something on the buffer */
2851 
2852 	event = __rb_page_index(tail_page, tail);
2853 	rb_update_event(cpu_buffer, event, info);
2854 
2855 	local_inc(&tail_page->entries);
2856 
2857 	/*
2858 	 * If this is the first commit on the page, then update
2859 	 * its timestamp.
2860 	 */
2861 	if (!tail)
2862 		tail_page->page->time_stamp = info->ts;
2863 
2864 	/* account for these added bytes */
2865 	local_add(info->length, &cpu_buffer->entries_bytes);
2866 
2867 	return event;
2868 }
2869 
2870 static __always_inline struct ring_buffer_event *
rb_reserve_next_event(struct ring_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer,unsigned long length)2871 rb_reserve_next_event(struct ring_buffer *buffer,
2872 		      struct ring_buffer_per_cpu *cpu_buffer,
2873 		      unsigned long length)
2874 {
2875 	struct ring_buffer_event *event;
2876 	struct rb_event_info info;
2877 	int nr_loops = 0;
2878 	u64 diff;
2879 
2880 	rb_start_commit(cpu_buffer);
2881 
2882 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2883 	/*
2884 	 * Due to the ability to swap a cpu buffer from a buffer
2885 	 * it is possible it was swapped before we committed.
2886 	 * (committing stops a swap). We check for it here and
2887 	 * if it happened, we have to fail the write.
2888 	 */
2889 	barrier();
2890 	if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
2891 		local_dec(&cpu_buffer->committing);
2892 		local_dec(&cpu_buffer->commits);
2893 		return NULL;
2894 	}
2895 #endif
2896 
2897 	info.length = rb_calculate_event_length(length);
2898  again:
2899 	info.add_timestamp = 0;
2900 	info.delta = 0;
2901 
2902 	/*
2903 	 * We allow for interrupts to reenter here and do a trace.
2904 	 * If one does, it will cause this original code to loop
2905 	 * back here. Even with heavy interrupts happening, this
2906 	 * should only happen a few times in a row. If this happens
2907 	 * 1000 times in a row, there must be either an interrupt
2908 	 * storm or we have something buggy.
2909 	 * Bail!
2910 	 */
2911 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
2912 		goto out_fail;
2913 
2914 	info.ts = rb_time_stamp(cpu_buffer->buffer);
2915 	diff = info.ts - cpu_buffer->write_stamp;
2916 
2917 	/* make sure this diff is calculated here */
2918 	barrier();
2919 
2920 	if (ring_buffer_time_stamp_abs(buffer)) {
2921 		info.delta = info.ts;
2922 		rb_handle_timestamp(cpu_buffer, &info);
2923 	} else /* Did the write stamp get updated already? */
2924 		if (likely(info.ts >= cpu_buffer->write_stamp)) {
2925 		info.delta = diff;
2926 		if (unlikely(test_time_stamp(info.delta)))
2927 			rb_handle_timestamp(cpu_buffer, &info);
2928 	}
2929 
2930 	event = __rb_reserve_next(cpu_buffer, &info);
2931 
2932 	if (unlikely(PTR_ERR(event) == -EAGAIN)) {
2933 		if (info.add_timestamp)
2934 			info.length -= RB_LEN_TIME_EXTEND;
2935 		goto again;
2936 	}
2937 
2938 	if (!event)
2939 		goto out_fail;
2940 
2941 	return event;
2942 
2943  out_fail:
2944 	rb_end_commit(cpu_buffer);
2945 	return NULL;
2946 }
2947 
2948 /**
2949  * ring_buffer_lock_reserve - reserve a part of the buffer
2950  * @buffer: the ring buffer to reserve from
2951  * @length: the length of the data to reserve (excluding event header)
2952  *
2953  * Returns a reserved event on the ring buffer to copy directly to.
2954  * The user of this interface will need to get the body to write into
2955  * and can use the ring_buffer_event_data() interface.
2956  *
2957  * The length is the length of the data needed, not the event length
2958  * which also includes the event header.
2959  *
2960  * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2961  * If NULL is returned, then nothing has been allocated or locked.
2962  */
2963 struct ring_buffer_event *
ring_buffer_lock_reserve(struct ring_buffer * buffer,unsigned long length)2964 ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
2965 {
2966 	struct ring_buffer_per_cpu *cpu_buffer;
2967 	struct ring_buffer_event *event;
2968 	int cpu;
2969 
2970 	/* If we are tracing schedule, we don't want to recurse */
2971 	preempt_disable_notrace();
2972 
2973 	if (unlikely(atomic_read(&buffer->record_disabled)))
2974 		goto out;
2975 
2976 	cpu = raw_smp_processor_id();
2977 
2978 	if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
2979 		goto out;
2980 
2981 	cpu_buffer = buffer->buffers[cpu];
2982 
2983 	if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
2984 		goto out;
2985 
2986 	if (unlikely(length > BUF_MAX_DATA_SIZE))
2987 		goto out;
2988 
2989 	if (unlikely(trace_recursive_lock(cpu_buffer)))
2990 		goto out;
2991 
2992 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
2993 	if (!event)
2994 		goto out_unlock;
2995 
2996 	return event;
2997 
2998  out_unlock:
2999 	trace_recursive_unlock(cpu_buffer);
3000  out:
3001 	preempt_enable_notrace();
3002 	return NULL;
3003 }
3004 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
3005 
3006 /*
3007  * Decrement the entries to the page that an event is on.
3008  * The event does not even need to exist, only the pointer
3009  * to the page it is on. This may only be called before the commit
3010  * takes place.
3011  */
3012 static inline void
rb_decrement_entry(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3013 rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
3014 		   struct ring_buffer_event *event)
3015 {
3016 	unsigned long addr = (unsigned long)event;
3017 	struct buffer_page *bpage = cpu_buffer->commit_page;
3018 	struct buffer_page *start;
3019 
3020 	addr &= PAGE_MASK;
3021 
3022 	/* Do the likely case first */
3023 	if (likely(bpage->page == (void *)addr)) {
3024 		local_dec(&bpage->entries);
3025 		return;
3026 	}
3027 
3028 	/*
3029 	 * Because the commit page may be on the reader page we
3030 	 * start with the next page and check the end loop there.
3031 	 */
3032 	rb_inc_page(cpu_buffer, &bpage);
3033 	start = bpage;
3034 	do {
3035 		if (bpage->page == (void *)addr) {
3036 			local_dec(&bpage->entries);
3037 			return;
3038 		}
3039 		rb_inc_page(cpu_buffer, &bpage);
3040 	} while (bpage != start);
3041 
3042 	/* commit not part of this buffer?? */
3043 	RB_WARN_ON(cpu_buffer, 1);
3044 }
3045 
3046 /**
3047  * ring_buffer_commit_discard - discard an event that has not been committed
3048  * @buffer: the ring buffer
3049  * @event: non committed event to discard
3050  *
3051  * Sometimes an event that is in the ring buffer needs to be ignored.
3052  * This function lets the user discard an event in the ring buffer
3053  * and then that event will not be read later.
3054  *
3055  * This function only works if it is called before the item has been
3056  * committed. It will try to free the event from the ring buffer
3057  * if another event has not been added behind it.
3058  *
3059  * If another event has been added behind it, it will set the event
3060  * up as discarded, and perform the commit.
3061  *
3062  * If this function is called, do not call ring_buffer_unlock_commit on
3063  * the event.
3064  */
ring_buffer_discard_commit(struct ring_buffer * buffer,struct ring_buffer_event * event)3065 void ring_buffer_discard_commit(struct ring_buffer *buffer,
3066 				struct ring_buffer_event *event)
3067 {
3068 	struct ring_buffer_per_cpu *cpu_buffer;
3069 	int cpu;
3070 
3071 	/* The event is discarded regardless */
3072 	rb_event_discard(event);
3073 
3074 	cpu = smp_processor_id();
3075 	cpu_buffer = buffer->buffers[cpu];
3076 
3077 	/*
3078 	 * This must only be called if the event has not been
3079 	 * committed yet. Thus we can assume that preemption
3080 	 * is still disabled.
3081 	 */
3082 	RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
3083 
3084 	rb_decrement_entry(cpu_buffer, event);
3085 	if (rb_try_to_discard(cpu_buffer, event))
3086 		goto out;
3087 
3088 	/*
3089 	 * The commit is still visible by the reader, so we
3090 	 * must still update the timestamp.
3091 	 */
3092 	rb_update_write_stamp(cpu_buffer, event);
3093  out:
3094 	rb_end_commit(cpu_buffer);
3095 
3096 	trace_recursive_unlock(cpu_buffer);
3097 
3098 	preempt_enable_notrace();
3099 
3100 }
3101 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
3102 
3103 /**
3104  * ring_buffer_write - write data to the buffer without reserving
3105  * @buffer: The ring buffer to write to.
3106  * @length: The length of the data being written (excluding the event header)
3107  * @data: The data to write to the buffer.
3108  *
3109  * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
3110  * one function. If you already have the data to write to the buffer, it
3111  * may be easier to simply call this function.
3112  *
3113  * Note, like ring_buffer_lock_reserve, the length is the length of the data
3114  * and not the length of the event which would hold the header.
3115  */
ring_buffer_write(struct ring_buffer * buffer,unsigned long length,void * data)3116 int ring_buffer_write(struct ring_buffer *buffer,
3117 		      unsigned long length,
3118 		      void *data)
3119 {
3120 	struct ring_buffer_per_cpu *cpu_buffer;
3121 	struct ring_buffer_event *event;
3122 	void *body;
3123 	int ret = -EBUSY;
3124 	int cpu;
3125 
3126 	preempt_disable_notrace();
3127 
3128 	if (atomic_read(&buffer->record_disabled))
3129 		goto out;
3130 
3131 	cpu = raw_smp_processor_id();
3132 
3133 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3134 		goto out;
3135 
3136 	cpu_buffer = buffer->buffers[cpu];
3137 
3138 	if (atomic_read(&cpu_buffer->record_disabled))
3139 		goto out;
3140 
3141 	if (length > BUF_MAX_DATA_SIZE)
3142 		goto out;
3143 
3144 	if (unlikely(trace_recursive_lock(cpu_buffer)))
3145 		goto out;
3146 
3147 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
3148 	if (!event)
3149 		goto out_unlock;
3150 
3151 	body = rb_event_data(event);
3152 
3153 	memcpy(body, data, length);
3154 
3155 	rb_commit(cpu_buffer, event);
3156 
3157 	rb_wakeups(buffer, cpu_buffer);
3158 
3159 	ret = 0;
3160 
3161  out_unlock:
3162 	trace_recursive_unlock(cpu_buffer);
3163 
3164  out:
3165 	preempt_enable_notrace();
3166 
3167 	return ret;
3168 }
3169 EXPORT_SYMBOL_GPL(ring_buffer_write);
3170 
rb_per_cpu_empty(struct ring_buffer_per_cpu * cpu_buffer)3171 static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
3172 {
3173 	struct buffer_page *reader = cpu_buffer->reader_page;
3174 	struct buffer_page *head = rb_set_head_page(cpu_buffer);
3175 	struct buffer_page *commit = cpu_buffer->commit_page;
3176 
3177 	/* In case of error, head will be NULL */
3178 	if (unlikely(!head))
3179 		return true;
3180 
3181 	return reader->read == rb_page_commit(reader) &&
3182 		(commit == reader ||
3183 		 (commit == head &&
3184 		  head->read == rb_page_commit(commit)));
3185 }
3186 
3187 /**
3188  * ring_buffer_record_disable - stop all writes into the buffer
3189  * @buffer: The ring buffer to stop writes to.
3190  *
3191  * This prevents all writes to the buffer. Any attempt to write
3192  * to the buffer after this will fail and return NULL.
3193  *
3194  * The caller should call synchronize_rcu() after this.
3195  */
ring_buffer_record_disable(struct ring_buffer * buffer)3196 void ring_buffer_record_disable(struct ring_buffer *buffer)
3197 {
3198 	atomic_inc(&buffer->record_disabled);
3199 }
3200 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
3201 
3202 /**
3203  * ring_buffer_record_enable - enable writes to the buffer
3204  * @buffer: The ring buffer to enable writes
3205  *
3206  * Note, multiple disables will need the same number of enables
3207  * to truly enable the writing (much like preempt_disable).
3208  */
ring_buffer_record_enable(struct ring_buffer * buffer)3209 void ring_buffer_record_enable(struct ring_buffer *buffer)
3210 {
3211 	atomic_dec(&buffer->record_disabled);
3212 }
3213 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
3214 
3215 /**
3216  * ring_buffer_record_off - stop all writes into the buffer
3217  * @buffer: The ring buffer to stop writes to.
3218  *
3219  * This prevents all writes to the buffer. Any attempt to write
3220  * to the buffer after this will fail and return NULL.
3221  *
3222  * This is different than ring_buffer_record_disable() as
3223  * it works like an on/off switch, where as the disable() version
3224  * must be paired with a enable().
3225  */
ring_buffer_record_off(struct ring_buffer * buffer)3226 void ring_buffer_record_off(struct ring_buffer *buffer)
3227 {
3228 	unsigned int rd;
3229 	unsigned int new_rd;
3230 
3231 	do {
3232 		rd = atomic_read(&buffer->record_disabled);
3233 		new_rd = rd | RB_BUFFER_OFF;
3234 	} while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
3235 }
3236 EXPORT_SYMBOL_GPL(ring_buffer_record_off);
3237 
3238 /**
3239  * ring_buffer_record_on - restart writes into the buffer
3240  * @buffer: The ring buffer to start writes to.
3241  *
3242  * This enables all writes to the buffer that was disabled by
3243  * ring_buffer_record_off().
3244  *
3245  * This is different than ring_buffer_record_enable() as
3246  * it works like an on/off switch, where as the enable() version
3247  * must be paired with a disable().
3248  */
ring_buffer_record_on(struct ring_buffer * buffer)3249 void ring_buffer_record_on(struct ring_buffer *buffer)
3250 {
3251 	unsigned int rd;
3252 	unsigned int new_rd;
3253 
3254 	do {
3255 		rd = atomic_read(&buffer->record_disabled);
3256 		new_rd = rd & ~RB_BUFFER_OFF;
3257 	} while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
3258 }
3259 EXPORT_SYMBOL_GPL(ring_buffer_record_on);
3260 
3261 /**
3262  * ring_buffer_record_is_on - return true if the ring buffer can write
3263  * @buffer: The ring buffer to see if write is enabled
3264  *
3265  * Returns true if the ring buffer is in a state that it accepts writes.
3266  */
ring_buffer_record_is_on(struct ring_buffer * buffer)3267 bool ring_buffer_record_is_on(struct ring_buffer *buffer)
3268 {
3269 	return !atomic_read(&buffer->record_disabled);
3270 }
3271 
3272 /**
3273  * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
3274  * @buffer: The ring buffer to see if write is set enabled
3275  *
3276  * Returns true if the ring buffer is set writable by ring_buffer_record_on().
3277  * Note that this does NOT mean it is in a writable state.
3278  *
3279  * It may return true when the ring buffer has been disabled by
3280  * ring_buffer_record_disable(), as that is a temporary disabling of
3281  * the ring buffer.
3282  */
ring_buffer_record_is_set_on(struct ring_buffer * buffer)3283 bool ring_buffer_record_is_set_on(struct ring_buffer *buffer)
3284 {
3285 	return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF);
3286 }
3287 
3288 /**
3289  * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
3290  * @buffer: The ring buffer to stop writes to.
3291  * @cpu: The CPU buffer to stop
3292  *
3293  * This prevents all writes to the buffer. Any attempt to write
3294  * to the buffer after this will fail and return NULL.
3295  *
3296  * The caller should call synchronize_rcu() after this.
3297  */
ring_buffer_record_disable_cpu(struct ring_buffer * buffer,int cpu)3298 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
3299 {
3300 	struct ring_buffer_per_cpu *cpu_buffer;
3301 
3302 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3303 		return;
3304 
3305 	cpu_buffer = buffer->buffers[cpu];
3306 	atomic_inc(&cpu_buffer->record_disabled);
3307 }
3308 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
3309 
3310 /**
3311  * ring_buffer_record_enable_cpu - enable writes to the buffer
3312  * @buffer: The ring buffer to enable writes
3313  * @cpu: The CPU to enable.
3314  *
3315  * Note, multiple disables will need the same number of enables
3316  * to truly enable the writing (much like preempt_disable).
3317  */
ring_buffer_record_enable_cpu(struct ring_buffer * buffer,int cpu)3318 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
3319 {
3320 	struct ring_buffer_per_cpu *cpu_buffer;
3321 
3322 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3323 		return;
3324 
3325 	cpu_buffer = buffer->buffers[cpu];
3326 	atomic_dec(&cpu_buffer->record_disabled);
3327 }
3328 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
3329 
3330 /*
3331  * The total entries in the ring buffer is the running counter
3332  * of entries entered into the ring buffer, minus the sum of
3333  * the entries read from the ring buffer and the number of
3334  * entries that were overwritten.
3335  */
3336 static inline unsigned long
rb_num_of_entries(struct ring_buffer_per_cpu * cpu_buffer)3337 rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
3338 {
3339 	return local_read(&cpu_buffer->entries) -
3340 		(local_read(&cpu_buffer->overrun) + cpu_buffer->read);
3341 }
3342 
3343 /**
3344  * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
3345  * @buffer: The ring buffer
3346  * @cpu: The per CPU buffer to read from.
3347  */
ring_buffer_oldest_event_ts(struct ring_buffer * buffer,int cpu)3348 u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu)
3349 {
3350 	unsigned long flags;
3351 	struct ring_buffer_per_cpu *cpu_buffer;
3352 	struct buffer_page *bpage;
3353 	u64 ret = 0;
3354 
3355 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3356 		return 0;
3357 
3358 	cpu_buffer = buffer->buffers[cpu];
3359 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3360 	/*
3361 	 * if the tail is on reader_page, oldest time stamp is on the reader
3362 	 * page
3363 	 */
3364 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
3365 		bpage = cpu_buffer->reader_page;
3366 	else
3367 		bpage = rb_set_head_page(cpu_buffer);
3368 	if (bpage)
3369 		ret = bpage->page->time_stamp;
3370 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3371 
3372 	return ret;
3373 }
3374 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
3375 
3376 /**
3377  * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
3378  * @buffer: The ring buffer
3379  * @cpu: The per CPU buffer to read from.
3380  */
ring_buffer_bytes_cpu(struct ring_buffer * buffer,int cpu)3381 unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu)
3382 {
3383 	struct ring_buffer_per_cpu *cpu_buffer;
3384 	unsigned long ret;
3385 
3386 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3387 		return 0;
3388 
3389 	cpu_buffer = buffer->buffers[cpu];
3390 	ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
3391 
3392 	return ret;
3393 }
3394 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
3395 
3396 /**
3397  * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
3398  * @buffer: The ring buffer
3399  * @cpu: The per CPU buffer to get the entries from.
3400  */
ring_buffer_entries_cpu(struct ring_buffer * buffer,int cpu)3401 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
3402 {
3403 	struct ring_buffer_per_cpu *cpu_buffer;
3404 
3405 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3406 		return 0;
3407 
3408 	cpu_buffer = buffer->buffers[cpu];
3409 
3410 	return rb_num_of_entries(cpu_buffer);
3411 }
3412 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
3413 
3414 /**
3415  * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
3416  * buffer wrapping around (only if RB_FL_OVERWRITE is on).
3417  * @buffer: The ring buffer
3418  * @cpu: The per CPU buffer to get the number of overruns from
3419  */
ring_buffer_overrun_cpu(struct ring_buffer * buffer,int cpu)3420 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
3421 {
3422 	struct ring_buffer_per_cpu *cpu_buffer;
3423 	unsigned long ret;
3424 
3425 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3426 		return 0;
3427 
3428 	cpu_buffer = buffer->buffers[cpu];
3429 	ret = local_read(&cpu_buffer->overrun);
3430 
3431 	return ret;
3432 }
3433 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
3434 
3435 /**
3436  * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
3437  * commits failing due to the buffer wrapping around while there are uncommitted
3438  * events, such as during an interrupt storm.
3439  * @buffer: The ring buffer
3440  * @cpu: The per CPU buffer to get the number of overruns from
3441  */
3442 unsigned long
ring_buffer_commit_overrun_cpu(struct ring_buffer * buffer,int cpu)3443 ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
3444 {
3445 	struct ring_buffer_per_cpu *cpu_buffer;
3446 	unsigned long ret;
3447 
3448 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3449 		return 0;
3450 
3451 	cpu_buffer = buffer->buffers[cpu];
3452 	ret = local_read(&cpu_buffer->commit_overrun);
3453 
3454 	return ret;
3455 }
3456 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
3457 
3458 /**
3459  * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
3460  * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
3461  * @buffer: The ring buffer
3462  * @cpu: The per CPU buffer to get the number of overruns from
3463  */
3464 unsigned long
ring_buffer_dropped_events_cpu(struct ring_buffer * buffer,int cpu)3465 ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu)
3466 {
3467 	struct ring_buffer_per_cpu *cpu_buffer;
3468 	unsigned long ret;
3469 
3470 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3471 		return 0;
3472 
3473 	cpu_buffer = buffer->buffers[cpu];
3474 	ret = local_read(&cpu_buffer->dropped_events);
3475 
3476 	return ret;
3477 }
3478 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
3479 
3480 /**
3481  * ring_buffer_read_events_cpu - get the number of events successfully read
3482  * @buffer: The ring buffer
3483  * @cpu: The per CPU buffer to get the number of events read
3484  */
3485 unsigned long
ring_buffer_read_events_cpu(struct ring_buffer * buffer,int cpu)3486 ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu)
3487 {
3488 	struct ring_buffer_per_cpu *cpu_buffer;
3489 
3490 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3491 		return 0;
3492 
3493 	cpu_buffer = buffer->buffers[cpu];
3494 	return cpu_buffer->read;
3495 }
3496 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
3497 
3498 /**
3499  * ring_buffer_entries - get the number of entries in a buffer
3500  * @buffer: The ring buffer
3501  *
3502  * Returns the total number of entries in the ring buffer
3503  * (all CPU entries)
3504  */
ring_buffer_entries(struct ring_buffer * buffer)3505 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
3506 {
3507 	struct ring_buffer_per_cpu *cpu_buffer;
3508 	unsigned long entries = 0;
3509 	int cpu;
3510 
3511 	/* if you care about this being correct, lock the buffer */
3512 	for_each_buffer_cpu(buffer, cpu) {
3513 		cpu_buffer = buffer->buffers[cpu];
3514 		entries += rb_num_of_entries(cpu_buffer);
3515 	}
3516 
3517 	return entries;
3518 }
3519 EXPORT_SYMBOL_GPL(ring_buffer_entries);
3520 
3521 /**
3522  * ring_buffer_overruns - get the number of overruns in buffer
3523  * @buffer: The ring buffer
3524  *
3525  * Returns the total number of overruns in the ring buffer
3526  * (all CPU entries)
3527  */
ring_buffer_overruns(struct ring_buffer * buffer)3528 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
3529 {
3530 	struct ring_buffer_per_cpu *cpu_buffer;
3531 	unsigned long overruns = 0;
3532 	int cpu;
3533 
3534 	/* if you care about this being correct, lock the buffer */
3535 	for_each_buffer_cpu(buffer, cpu) {
3536 		cpu_buffer = buffer->buffers[cpu];
3537 		overruns += local_read(&cpu_buffer->overrun);
3538 	}
3539 
3540 	return overruns;
3541 }
3542 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
3543 
rb_iter_reset(struct ring_buffer_iter * iter)3544 static void rb_iter_reset(struct ring_buffer_iter *iter)
3545 {
3546 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3547 
3548 	/* Iterator usage is expected to have record disabled */
3549 	iter->head_page = cpu_buffer->reader_page;
3550 	iter->head = cpu_buffer->reader_page->read;
3551 
3552 	iter->cache_reader_page = iter->head_page;
3553 	iter->cache_read = cpu_buffer->read;
3554 
3555 	if (iter->head)
3556 		iter->read_stamp = cpu_buffer->read_stamp;
3557 	else
3558 		iter->read_stamp = iter->head_page->page->time_stamp;
3559 }
3560 
3561 /**
3562  * ring_buffer_iter_reset - reset an iterator
3563  * @iter: The iterator to reset
3564  *
3565  * Resets the iterator, so that it will start from the beginning
3566  * again.
3567  */
ring_buffer_iter_reset(struct ring_buffer_iter * iter)3568 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
3569 {
3570 	struct ring_buffer_per_cpu *cpu_buffer;
3571 	unsigned long flags;
3572 
3573 	if (!iter)
3574 		return;
3575 
3576 	cpu_buffer = iter->cpu_buffer;
3577 
3578 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3579 	rb_iter_reset(iter);
3580 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3581 }
3582 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
3583 
3584 /**
3585  * ring_buffer_iter_empty - check if an iterator has no more to read
3586  * @iter: The iterator to check
3587  */
ring_buffer_iter_empty(struct ring_buffer_iter * iter)3588 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
3589 {
3590 	struct ring_buffer_per_cpu *cpu_buffer;
3591 	struct buffer_page *reader;
3592 	struct buffer_page *head_page;
3593 	struct buffer_page *commit_page;
3594 	unsigned commit;
3595 
3596 	cpu_buffer = iter->cpu_buffer;
3597 
3598 	/* Remember, trace recording is off when iterator is in use */
3599 	reader = cpu_buffer->reader_page;
3600 	head_page = cpu_buffer->head_page;
3601 	commit_page = cpu_buffer->commit_page;
3602 	commit = rb_page_commit(commit_page);
3603 
3604 	return ((iter->head_page == commit_page && iter->head == commit) ||
3605 		(iter->head_page == reader && commit_page == head_page &&
3606 		 head_page->read == commit &&
3607 		 iter->head == rb_page_commit(cpu_buffer->reader_page)));
3608 }
3609 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
3610 
3611 static void
rb_update_read_stamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3612 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
3613 		     struct ring_buffer_event *event)
3614 {
3615 	u64 delta;
3616 
3617 	switch (event->type_len) {
3618 	case RINGBUF_TYPE_PADDING:
3619 		return;
3620 
3621 	case RINGBUF_TYPE_TIME_EXTEND:
3622 		delta = ring_buffer_event_time_stamp(event);
3623 		cpu_buffer->read_stamp += delta;
3624 		return;
3625 
3626 	case RINGBUF_TYPE_TIME_STAMP:
3627 		delta = ring_buffer_event_time_stamp(event);
3628 		cpu_buffer->read_stamp = delta;
3629 		return;
3630 
3631 	case RINGBUF_TYPE_DATA:
3632 		cpu_buffer->read_stamp += event->time_delta;
3633 		return;
3634 
3635 	default:
3636 		BUG();
3637 	}
3638 	return;
3639 }
3640 
3641 static void
rb_update_iter_read_stamp(struct ring_buffer_iter * iter,struct ring_buffer_event * event)3642 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
3643 			  struct ring_buffer_event *event)
3644 {
3645 	u64 delta;
3646 
3647 	switch (event->type_len) {
3648 	case RINGBUF_TYPE_PADDING:
3649 		return;
3650 
3651 	case RINGBUF_TYPE_TIME_EXTEND:
3652 		delta = ring_buffer_event_time_stamp(event);
3653 		iter->read_stamp += delta;
3654 		return;
3655 
3656 	case RINGBUF_TYPE_TIME_STAMP:
3657 		delta = ring_buffer_event_time_stamp(event);
3658 		iter->read_stamp = delta;
3659 		return;
3660 
3661 	case RINGBUF_TYPE_DATA:
3662 		iter->read_stamp += event->time_delta;
3663 		return;
3664 
3665 	default:
3666 		BUG();
3667 	}
3668 	return;
3669 }
3670 
3671 static struct buffer_page *
rb_get_reader_page(struct ring_buffer_per_cpu * cpu_buffer)3672 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
3673 {
3674 	struct buffer_page *reader = NULL;
3675 	unsigned long overwrite;
3676 	unsigned long flags;
3677 	int nr_loops = 0;
3678 	int ret;
3679 
3680 	local_irq_save(flags);
3681 	arch_spin_lock(&cpu_buffer->lock);
3682 
3683  again:
3684 	/*
3685 	 * This should normally only loop twice. But because the
3686 	 * start of the reader inserts an empty page, it causes
3687 	 * a case where we will loop three times. There should be no
3688 	 * reason to loop four times (that I know of).
3689 	 */
3690 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
3691 		reader = NULL;
3692 		goto out;
3693 	}
3694 
3695 	reader = cpu_buffer->reader_page;
3696 
3697 	/* If there's more to read, return this page */
3698 	if (cpu_buffer->reader_page->read < rb_page_size(reader))
3699 		goto out;
3700 
3701 	/* Never should we have an index greater than the size */
3702 	if (RB_WARN_ON(cpu_buffer,
3703 		       cpu_buffer->reader_page->read > rb_page_size(reader)))
3704 		goto out;
3705 
3706 	/* check if we caught up to the tail */
3707 	reader = NULL;
3708 	if (cpu_buffer->commit_page == cpu_buffer->reader_page)
3709 		goto out;
3710 
3711 	/* Don't bother swapping if the ring buffer is empty */
3712 	if (rb_num_of_entries(cpu_buffer) == 0)
3713 		goto out;
3714 
3715 	/*
3716 	 * Reset the reader page to size zero.
3717 	 */
3718 	local_set(&cpu_buffer->reader_page->write, 0);
3719 	local_set(&cpu_buffer->reader_page->entries, 0);
3720 	local_set(&cpu_buffer->reader_page->page->commit, 0);
3721 	cpu_buffer->reader_page->real_end = 0;
3722 
3723  spin:
3724 	/*
3725 	 * Splice the empty reader page into the list around the head.
3726 	 */
3727 	reader = rb_set_head_page(cpu_buffer);
3728 	if (!reader)
3729 		goto out;
3730 	cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
3731 	cpu_buffer->reader_page->list.prev = reader->list.prev;
3732 
3733 	/*
3734 	 * cpu_buffer->pages just needs to point to the buffer, it
3735 	 *  has no specific buffer page to point to. Lets move it out
3736 	 *  of our way so we don't accidentally swap it.
3737 	 */
3738 	cpu_buffer->pages = reader->list.prev;
3739 
3740 	/* The reader page will be pointing to the new head */
3741 	rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list);
3742 
3743 	/*
3744 	 * We want to make sure we read the overruns after we set up our
3745 	 * pointers to the next object. The writer side does a
3746 	 * cmpxchg to cross pages which acts as the mb on the writer
3747 	 * side. Note, the reader will constantly fail the swap
3748 	 * while the writer is updating the pointers, so this
3749 	 * guarantees that the overwrite recorded here is the one we
3750 	 * want to compare with the last_overrun.
3751 	 */
3752 	smp_mb();
3753 	overwrite = local_read(&(cpu_buffer->overrun));
3754 
3755 	/*
3756 	 * Here's the tricky part.
3757 	 *
3758 	 * We need to move the pointer past the header page.
3759 	 * But we can only do that if a writer is not currently
3760 	 * moving it. The page before the header page has the
3761 	 * flag bit '1' set if it is pointing to the page we want.
3762 	 * but if the writer is in the process of moving it
3763 	 * than it will be '2' or already moved '0'.
3764 	 */
3765 
3766 	ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
3767 
3768 	/*
3769 	 * If we did not convert it, then we must try again.
3770 	 */
3771 	if (!ret)
3772 		goto spin;
3773 
3774 	/*
3775 	 * Yay! We succeeded in replacing the page.
3776 	 *
3777 	 * Now make the new head point back to the reader page.
3778 	 */
3779 	rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
3780 	rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
3781 
3782 	local_inc(&cpu_buffer->pages_read);
3783 
3784 	/* Finally update the reader page to the new head */
3785 	cpu_buffer->reader_page = reader;
3786 	cpu_buffer->reader_page->read = 0;
3787 
3788 	if (overwrite != cpu_buffer->last_overrun) {
3789 		cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
3790 		cpu_buffer->last_overrun = overwrite;
3791 	}
3792 
3793 	goto again;
3794 
3795  out:
3796 	/* Update the read_stamp on the first event */
3797 	if (reader && reader->read == 0)
3798 		cpu_buffer->read_stamp = reader->page->time_stamp;
3799 
3800 	arch_spin_unlock(&cpu_buffer->lock);
3801 	local_irq_restore(flags);
3802 
3803 	return reader;
3804 }
3805 
rb_advance_reader(struct ring_buffer_per_cpu * cpu_buffer)3806 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
3807 {
3808 	struct ring_buffer_event *event;
3809 	struct buffer_page *reader;
3810 	unsigned length;
3811 
3812 	reader = rb_get_reader_page(cpu_buffer);
3813 
3814 	/* This function should not be called when buffer is empty */
3815 	if (RB_WARN_ON(cpu_buffer, !reader))
3816 		return;
3817 
3818 	event = rb_reader_event(cpu_buffer);
3819 
3820 	if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
3821 		cpu_buffer->read++;
3822 
3823 	rb_update_read_stamp(cpu_buffer, event);
3824 
3825 	length = rb_event_length(event);
3826 	cpu_buffer->reader_page->read += length;
3827 }
3828 
rb_advance_iter(struct ring_buffer_iter * iter)3829 static void rb_advance_iter(struct ring_buffer_iter *iter)
3830 {
3831 	struct ring_buffer_per_cpu *cpu_buffer;
3832 	struct ring_buffer_event *event;
3833 	unsigned length;
3834 
3835 	cpu_buffer = iter->cpu_buffer;
3836 
3837 	/*
3838 	 * Check if we are at the end of the buffer.
3839 	 */
3840 	if (iter->head >= rb_page_size(iter->head_page)) {
3841 		/* discarded commits can make the page empty */
3842 		if (iter->head_page == cpu_buffer->commit_page)
3843 			return;
3844 		rb_inc_iter(iter);
3845 		return;
3846 	}
3847 
3848 	event = rb_iter_head_event(iter);
3849 
3850 	length = rb_event_length(event);
3851 
3852 	/*
3853 	 * This should not be called to advance the header if we are
3854 	 * at the tail of the buffer.
3855 	 */
3856 	if (RB_WARN_ON(cpu_buffer,
3857 		       (iter->head_page == cpu_buffer->commit_page) &&
3858 		       (iter->head + length > rb_commit_index(cpu_buffer))))
3859 		return;
3860 
3861 	rb_update_iter_read_stamp(iter, event);
3862 
3863 	iter->head += length;
3864 
3865 	/* check for end of page padding */
3866 	if ((iter->head >= rb_page_size(iter->head_page)) &&
3867 	    (iter->head_page != cpu_buffer->commit_page))
3868 		rb_inc_iter(iter);
3869 }
3870 
rb_lost_events(struct ring_buffer_per_cpu * cpu_buffer)3871 static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
3872 {
3873 	return cpu_buffer->lost_events;
3874 }
3875 
3876 static struct ring_buffer_event *
rb_buffer_peek(struct ring_buffer_per_cpu * cpu_buffer,u64 * ts,unsigned long * lost_events)3877 rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
3878 	       unsigned long *lost_events)
3879 {
3880 	struct ring_buffer_event *event;
3881 	struct buffer_page *reader;
3882 	int nr_loops = 0;
3883 
3884 	if (ts)
3885 		*ts = 0;
3886  again:
3887 	/*
3888 	 * We repeat when a time extend is encountered.
3889 	 * Since the time extend is always attached to a data event,
3890 	 * we should never loop more than once.
3891 	 * (We never hit the following condition more than twice).
3892 	 */
3893 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
3894 		return NULL;
3895 
3896 	reader = rb_get_reader_page(cpu_buffer);
3897 	if (!reader)
3898 		return NULL;
3899 
3900 	event = rb_reader_event(cpu_buffer);
3901 
3902 	switch (event->type_len) {
3903 	case RINGBUF_TYPE_PADDING:
3904 		if (rb_null_event(event))
3905 			RB_WARN_ON(cpu_buffer, 1);
3906 		/*
3907 		 * Because the writer could be discarding every
3908 		 * event it creates (which would probably be bad)
3909 		 * if we were to go back to "again" then we may never
3910 		 * catch up, and will trigger the warn on, or lock
3911 		 * the box. Return the padding, and we will release
3912 		 * the current locks, and try again.
3913 		 */
3914 		return event;
3915 
3916 	case RINGBUF_TYPE_TIME_EXTEND:
3917 		/* Internal data, OK to advance */
3918 		rb_advance_reader(cpu_buffer);
3919 		goto again;
3920 
3921 	case RINGBUF_TYPE_TIME_STAMP:
3922 		if (ts) {
3923 			*ts = ring_buffer_event_time_stamp(event);
3924 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
3925 							 cpu_buffer->cpu, ts);
3926 		}
3927 		/* Internal data, OK to advance */
3928 		rb_advance_reader(cpu_buffer);
3929 		goto again;
3930 
3931 	case RINGBUF_TYPE_DATA:
3932 		if (ts && !(*ts)) {
3933 			*ts = cpu_buffer->read_stamp + event->time_delta;
3934 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
3935 							 cpu_buffer->cpu, ts);
3936 		}
3937 		if (lost_events)
3938 			*lost_events = rb_lost_events(cpu_buffer);
3939 		return event;
3940 
3941 	default:
3942 		BUG();
3943 	}
3944 
3945 	return NULL;
3946 }
3947 EXPORT_SYMBOL_GPL(ring_buffer_peek);
3948 
3949 static struct ring_buffer_event *
rb_iter_peek(struct ring_buffer_iter * iter,u64 * ts)3950 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
3951 {
3952 	struct ring_buffer *buffer;
3953 	struct ring_buffer_per_cpu *cpu_buffer;
3954 	struct ring_buffer_event *event;
3955 	int nr_loops = 0;
3956 
3957 	if (ts)
3958 		*ts = 0;
3959 
3960 	cpu_buffer = iter->cpu_buffer;
3961 	buffer = cpu_buffer->buffer;
3962 
3963 	/*
3964 	 * Check if someone performed a consuming read to
3965 	 * the buffer. A consuming read invalidates the iterator
3966 	 * and we need to reset the iterator in this case.
3967 	 */
3968 	if (unlikely(iter->cache_read != cpu_buffer->read ||
3969 		     iter->cache_reader_page != cpu_buffer->reader_page))
3970 		rb_iter_reset(iter);
3971 
3972  again:
3973 	if (ring_buffer_iter_empty(iter))
3974 		return NULL;
3975 
3976 	/*
3977 	 * We repeat when a time extend is encountered or we hit
3978 	 * the end of the page. Since the time extend is always attached
3979 	 * to a data event, we should never loop more than three times.
3980 	 * Once for going to next page, once on time extend, and
3981 	 * finally once to get the event.
3982 	 * (We never hit the following condition more than thrice).
3983 	 */
3984 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3))
3985 		return NULL;
3986 
3987 	if (rb_per_cpu_empty(cpu_buffer))
3988 		return NULL;
3989 
3990 	if (iter->head >= rb_page_size(iter->head_page)) {
3991 		rb_inc_iter(iter);
3992 		goto again;
3993 	}
3994 
3995 	event = rb_iter_head_event(iter);
3996 
3997 	switch (event->type_len) {
3998 	case RINGBUF_TYPE_PADDING:
3999 		if (rb_null_event(event)) {
4000 			rb_inc_iter(iter);
4001 			goto again;
4002 		}
4003 		rb_advance_iter(iter);
4004 		return event;
4005 
4006 	case RINGBUF_TYPE_TIME_EXTEND:
4007 		/* Internal data, OK to advance */
4008 		rb_advance_iter(iter);
4009 		goto again;
4010 
4011 	case RINGBUF_TYPE_TIME_STAMP:
4012 		if (ts) {
4013 			*ts = ring_buffer_event_time_stamp(event);
4014 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4015 							 cpu_buffer->cpu, ts);
4016 		}
4017 		/* Internal data, OK to advance */
4018 		rb_advance_iter(iter);
4019 		goto again;
4020 
4021 	case RINGBUF_TYPE_DATA:
4022 		if (ts && !(*ts)) {
4023 			*ts = iter->read_stamp + event->time_delta;
4024 			ring_buffer_normalize_time_stamp(buffer,
4025 							 cpu_buffer->cpu, ts);
4026 		}
4027 		return event;
4028 
4029 	default:
4030 		BUG();
4031 	}
4032 
4033 	return NULL;
4034 }
4035 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
4036 
rb_reader_lock(struct ring_buffer_per_cpu * cpu_buffer)4037 static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
4038 {
4039 	if (likely(!in_nmi())) {
4040 		raw_spin_lock(&cpu_buffer->reader_lock);
4041 		return true;
4042 	}
4043 
4044 	/*
4045 	 * If an NMI die dumps out the content of the ring buffer
4046 	 * trylock must be used to prevent a deadlock if the NMI
4047 	 * preempted a task that holds the ring buffer locks. If
4048 	 * we get the lock then all is fine, if not, then continue
4049 	 * to do the read, but this can corrupt the ring buffer,
4050 	 * so it must be permanently disabled from future writes.
4051 	 * Reading from NMI is a oneshot deal.
4052 	 */
4053 	if (raw_spin_trylock(&cpu_buffer->reader_lock))
4054 		return true;
4055 
4056 	/* Continue without locking, but disable the ring buffer */
4057 	atomic_inc(&cpu_buffer->record_disabled);
4058 	return false;
4059 }
4060 
4061 static inline void
rb_reader_unlock(struct ring_buffer_per_cpu * cpu_buffer,bool locked)4062 rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
4063 {
4064 	if (likely(locked))
4065 		raw_spin_unlock(&cpu_buffer->reader_lock);
4066 	return;
4067 }
4068 
4069 /**
4070  * ring_buffer_peek - peek at the next event to be read
4071  * @buffer: The ring buffer to read
4072  * @cpu: The cpu to peak at
4073  * @ts: The timestamp counter of this event.
4074  * @lost_events: a variable to store if events were lost (may be NULL)
4075  *
4076  * This will return the event that will be read next, but does
4077  * not consume the data.
4078  */
4079 struct ring_buffer_event *
ring_buffer_peek(struct ring_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4080 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts,
4081 		 unsigned long *lost_events)
4082 {
4083 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4084 	struct ring_buffer_event *event;
4085 	unsigned long flags;
4086 	bool dolock;
4087 
4088 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4089 		return NULL;
4090 
4091  again:
4092 	local_irq_save(flags);
4093 	dolock = rb_reader_lock(cpu_buffer);
4094 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4095 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4096 		rb_advance_reader(cpu_buffer);
4097 	rb_reader_unlock(cpu_buffer, dolock);
4098 	local_irq_restore(flags);
4099 
4100 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4101 		goto again;
4102 
4103 	return event;
4104 }
4105 
4106 /**
4107  * ring_buffer_iter_peek - peek at the next event to be read
4108  * @iter: The ring buffer iterator
4109  * @ts: The timestamp counter of this event.
4110  *
4111  * This will return the event that will be read next, but does
4112  * not increment the iterator.
4113  */
4114 struct ring_buffer_event *
ring_buffer_iter_peek(struct ring_buffer_iter * iter,u64 * ts)4115 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
4116 {
4117 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4118 	struct ring_buffer_event *event;
4119 	unsigned long flags;
4120 
4121  again:
4122 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4123 	event = rb_iter_peek(iter, ts);
4124 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4125 
4126 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4127 		goto again;
4128 
4129 	return event;
4130 }
4131 
4132 /**
4133  * ring_buffer_consume - return an event and consume it
4134  * @buffer: The ring buffer to get the next event from
4135  * @cpu: the cpu to read the buffer from
4136  * @ts: a variable to store the timestamp (may be NULL)
4137  * @lost_events: a variable to store if events were lost (may be NULL)
4138  *
4139  * Returns the next event in the ring buffer, and that event is consumed.
4140  * Meaning, that sequential reads will keep returning a different event,
4141  * and eventually empty the ring buffer if the producer is slower.
4142  */
4143 struct ring_buffer_event *
ring_buffer_consume(struct ring_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4144 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts,
4145 		    unsigned long *lost_events)
4146 {
4147 	struct ring_buffer_per_cpu *cpu_buffer;
4148 	struct ring_buffer_event *event = NULL;
4149 	unsigned long flags;
4150 	bool dolock;
4151 
4152  again:
4153 	/* might be called in atomic */
4154 	preempt_disable();
4155 
4156 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4157 		goto out;
4158 
4159 	cpu_buffer = buffer->buffers[cpu];
4160 	local_irq_save(flags);
4161 	dolock = rb_reader_lock(cpu_buffer);
4162 
4163 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4164 	if (event) {
4165 		cpu_buffer->lost_events = 0;
4166 		rb_advance_reader(cpu_buffer);
4167 	}
4168 
4169 	rb_reader_unlock(cpu_buffer, dolock);
4170 	local_irq_restore(flags);
4171 
4172  out:
4173 	preempt_enable();
4174 
4175 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4176 		goto again;
4177 
4178 	return event;
4179 }
4180 EXPORT_SYMBOL_GPL(ring_buffer_consume);
4181 
4182 /**
4183  * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
4184  * @buffer: The ring buffer to read from
4185  * @cpu: The cpu buffer to iterate over
4186  * @flags: gfp flags to use for memory allocation
4187  *
4188  * This performs the initial preparations necessary to iterate
4189  * through the buffer.  Memory is allocated, buffer recording
4190  * is disabled, and the iterator pointer is returned to the caller.
4191  *
4192  * Disabling buffer recording prevents the reading from being
4193  * corrupted. This is not a consuming read, so a producer is not
4194  * expected.
4195  *
4196  * After a sequence of ring_buffer_read_prepare calls, the user is
4197  * expected to make at least one call to ring_buffer_read_prepare_sync.
4198  * Afterwards, ring_buffer_read_start is invoked to get things going
4199  * for real.
4200  *
4201  * This overall must be paired with ring_buffer_read_finish.
4202  */
4203 struct ring_buffer_iter *
ring_buffer_read_prepare(struct ring_buffer * buffer,int cpu,gfp_t flags)4204 ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu, gfp_t flags)
4205 {
4206 	struct ring_buffer_per_cpu *cpu_buffer;
4207 	struct ring_buffer_iter *iter;
4208 
4209 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4210 		return NULL;
4211 
4212 	iter = kmalloc(sizeof(*iter), flags);
4213 	if (!iter)
4214 		return NULL;
4215 
4216 	cpu_buffer = buffer->buffers[cpu];
4217 
4218 	iter->cpu_buffer = cpu_buffer;
4219 
4220 	atomic_inc(&buffer->resize_disabled);
4221 	atomic_inc(&cpu_buffer->record_disabled);
4222 
4223 	return iter;
4224 }
4225 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
4226 
4227 /**
4228  * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
4229  *
4230  * All previously invoked ring_buffer_read_prepare calls to prepare
4231  * iterators will be synchronized.  Afterwards, read_buffer_read_start
4232  * calls on those iterators are allowed.
4233  */
4234 void
ring_buffer_read_prepare_sync(void)4235 ring_buffer_read_prepare_sync(void)
4236 {
4237 	synchronize_rcu();
4238 }
4239 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
4240 
4241 /**
4242  * ring_buffer_read_start - start a non consuming read of the buffer
4243  * @iter: The iterator returned by ring_buffer_read_prepare
4244  *
4245  * This finalizes the startup of an iteration through the buffer.
4246  * The iterator comes from a call to ring_buffer_read_prepare and
4247  * an intervening ring_buffer_read_prepare_sync must have been
4248  * performed.
4249  *
4250  * Must be paired with ring_buffer_read_finish.
4251  */
4252 void
ring_buffer_read_start(struct ring_buffer_iter * iter)4253 ring_buffer_read_start(struct ring_buffer_iter *iter)
4254 {
4255 	struct ring_buffer_per_cpu *cpu_buffer;
4256 	unsigned long flags;
4257 
4258 	if (!iter)
4259 		return;
4260 
4261 	cpu_buffer = iter->cpu_buffer;
4262 
4263 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4264 	arch_spin_lock(&cpu_buffer->lock);
4265 	rb_iter_reset(iter);
4266 	arch_spin_unlock(&cpu_buffer->lock);
4267 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4268 }
4269 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
4270 
4271 /**
4272  * ring_buffer_read_finish - finish reading the iterator of the buffer
4273  * @iter: The iterator retrieved by ring_buffer_start
4274  *
4275  * This re-enables the recording to the buffer, and frees the
4276  * iterator.
4277  */
4278 void
ring_buffer_read_finish(struct ring_buffer_iter * iter)4279 ring_buffer_read_finish(struct ring_buffer_iter *iter)
4280 {
4281 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4282 	unsigned long flags;
4283 
4284 	/*
4285 	 * Ring buffer is disabled from recording, here's a good place
4286 	 * to check the integrity of the ring buffer.
4287 	 * Must prevent readers from trying to read, as the check
4288 	 * clears the HEAD page and readers require it.
4289 	 */
4290 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4291 	rb_check_pages(cpu_buffer);
4292 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4293 
4294 	atomic_dec(&cpu_buffer->record_disabled);
4295 	atomic_dec(&cpu_buffer->buffer->resize_disabled);
4296 	kfree(iter);
4297 }
4298 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
4299 
4300 /**
4301  * ring_buffer_read - read the next item in the ring buffer by the iterator
4302  * @iter: The ring buffer iterator
4303  * @ts: The time stamp of the event read.
4304  *
4305  * This reads the next event in the ring buffer and increments the iterator.
4306  */
4307 struct ring_buffer_event *
ring_buffer_read(struct ring_buffer_iter * iter,u64 * ts)4308 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
4309 {
4310 	struct ring_buffer_event *event;
4311 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4312 	unsigned long flags;
4313 
4314 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4315  again:
4316 	event = rb_iter_peek(iter, ts);
4317 	if (!event)
4318 		goto out;
4319 
4320 	if (event->type_len == RINGBUF_TYPE_PADDING)
4321 		goto again;
4322 
4323 	rb_advance_iter(iter);
4324  out:
4325 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4326 
4327 	return event;
4328 }
4329 EXPORT_SYMBOL_GPL(ring_buffer_read);
4330 
4331 /**
4332  * ring_buffer_size - return the size of the ring buffer (in bytes)
4333  * @buffer: The ring buffer.
4334  */
ring_buffer_size(struct ring_buffer * buffer,int cpu)4335 unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu)
4336 {
4337 	/*
4338 	 * Earlier, this method returned
4339 	 *	BUF_PAGE_SIZE * buffer->nr_pages
4340 	 * Since the nr_pages field is now removed, we have converted this to
4341 	 * return the per cpu buffer value.
4342 	 */
4343 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4344 		return 0;
4345 
4346 	return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages;
4347 }
4348 EXPORT_SYMBOL_GPL(ring_buffer_size);
4349 
4350 static void
rb_reset_cpu(struct ring_buffer_per_cpu * cpu_buffer)4351 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
4352 {
4353 	rb_head_page_deactivate(cpu_buffer);
4354 
4355 	cpu_buffer->head_page
4356 		= list_entry(cpu_buffer->pages, struct buffer_page, list);
4357 	local_set(&cpu_buffer->head_page->write, 0);
4358 	local_set(&cpu_buffer->head_page->entries, 0);
4359 	local_set(&cpu_buffer->head_page->page->commit, 0);
4360 
4361 	cpu_buffer->head_page->read = 0;
4362 
4363 	cpu_buffer->tail_page = cpu_buffer->head_page;
4364 	cpu_buffer->commit_page = cpu_buffer->head_page;
4365 
4366 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
4367 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
4368 	local_set(&cpu_buffer->reader_page->write, 0);
4369 	local_set(&cpu_buffer->reader_page->entries, 0);
4370 	local_set(&cpu_buffer->reader_page->page->commit, 0);
4371 	cpu_buffer->reader_page->read = 0;
4372 
4373 	local_set(&cpu_buffer->entries_bytes, 0);
4374 	local_set(&cpu_buffer->overrun, 0);
4375 	local_set(&cpu_buffer->commit_overrun, 0);
4376 	local_set(&cpu_buffer->dropped_events, 0);
4377 	local_set(&cpu_buffer->entries, 0);
4378 	local_set(&cpu_buffer->committing, 0);
4379 	local_set(&cpu_buffer->commits, 0);
4380 	local_set(&cpu_buffer->pages_touched, 0);
4381 	local_set(&cpu_buffer->pages_read, 0);
4382 	cpu_buffer->last_pages_touch = 0;
4383 	cpu_buffer->shortest_full = 0;
4384 	cpu_buffer->read = 0;
4385 	cpu_buffer->read_bytes = 0;
4386 
4387 	cpu_buffer->write_stamp = 0;
4388 	cpu_buffer->read_stamp = 0;
4389 
4390 	cpu_buffer->lost_events = 0;
4391 	cpu_buffer->last_overrun = 0;
4392 
4393 	rb_head_page_activate(cpu_buffer);
4394 }
4395 
4396 /**
4397  * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
4398  * @buffer: The ring buffer to reset a per cpu buffer of
4399  * @cpu: The CPU buffer to be reset
4400  */
ring_buffer_reset_cpu(struct ring_buffer * buffer,int cpu)4401 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
4402 {
4403 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4404 	unsigned long flags;
4405 
4406 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4407 		return;
4408 
4409 	atomic_inc(&buffer->resize_disabled);
4410 	atomic_inc(&cpu_buffer->record_disabled);
4411 
4412 	/* Make sure all commits have finished */
4413 	synchronize_rcu();
4414 
4415 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4416 
4417 	if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
4418 		goto out;
4419 
4420 	arch_spin_lock(&cpu_buffer->lock);
4421 
4422 	rb_reset_cpu(cpu_buffer);
4423 
4424 	arch_spin_unlock(&cpu_buffer->lock);
4425 
4426  out:
4427 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4428 
4429 	atomic_dec(&cpu_buffer->record_disabled);
4430 	atomic_dec(&buffer->resize_disabled);
4431 }
4432 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
4433 
4434 /**
4435  * ring_buffer_reset - reset a ring buffer
4436  * @buffer: The ring buffer to reset all cpu buffers
4437  */
ring_buffer_reset(struct ring_buffer * buffer)4438 void ring_buffer_reset(struct ring_buffer *buffer)
4439 {
4440 	int cpu;
4441 
4442 	for_each_buffer_cpu(buffer, cpu)
4443 		ring_buffer_reset_cpu(buffer, cpu);
4444 }
4445 EXPORT_SYMBOL_GPL(ring_buffer_reset);
4446 
4447 /**
4448  * rind_buffer_empty - is the ring buffer empty?
4449  * @buffer: The ring buffer to test
4450  */
ring_buffer_empty(struct ring_buffer * buffer)4451 bool ring_buffer_empty(struct ring_buffer *buffer)
4452 {
4453 	struct ring_buffer_per_cpu *cpu_buffer;
4454 	unsigned long flags;
4455 	bool dolock;
4456 	int cpu;
4457 	int ret;
4458 
4459 	/* yes this is racy, but if you don't like the race, lock the buffer */
4460 	for_each_buffer_cpu(buffer, cpu) {
4461 		cpu_buffer = buffer->buffers[cpu];
4462 		local_irq_save(flags);
4463 		dolock = rb_reader_lock(cpu_buffer);
4464 		ret = rb_per_cpu_empty(cpu_buffer);
4465 		rb_reader_unlock(cpu_buffer, dolock);
4466 		local_irq_restore(flags);
4467 
4468 		if (!ret)
4469 			return false;
4470 	}
4471 
4472 	return true;
4473 }
4474 EXPORT_SYMBOL_GPL(ring_buffer_empty);
4475 
4476 /**
4477  * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
4478  * @buffer: The ring buffer
4479  * @cpu: The CPU buffer to test
4480  */
ring_buffer_empty_cpu(struct ring_buffer * buffer,int cpu)4481 bool ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
4482 {
4483 	struct ring_buffer_per_cpu *cpu_buffer;
4484 	unsigned long flags;
4485 	bool dolock;
4486 	int ret;
4487 
4488 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4489 		return true;
4490 
4491 	cpu_buffer = buffer->buffers[cpu];
4492 	local_irq_save(flags);
4493 	dolock = rb_reader_lock(cpu_buffer);
4494 	ret = rb_per_cpu_empty(cpu_buffer);
4495 	rb_reader_unlock(cpu_buffer, dolock);
4496 	local_irq_restore(flags);
4497 
4498 	return ret;
4499 }
4500 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
4501 
4502 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4503 /**
4504  * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
4505  * @buffer_a: One buffer to swap with
4506  * @buffer_b: The other buffer to swap with
4507  *
4508  * This function is useful for tracers that want to take a "snapshot"
4509  * of a CPU buffer and has another back up buffer lying around.
4510  * it is expected that the tracer handles the cpu buffer not being
4511  * used at the moment.
4512  */
ring_buffer_swap_cpu(struct ring_buffer * buffer_a,struct ring_buffer * buffer_b,int cpu)4513 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
4514 			 struct ring_buffer *buffer_b, int cpu)
4515 {
4516 	struct ring_buffer_per_cpu *cpu_buffer_a;
4517 	struct ring_buffer_per_cpu *cpu_buffer_b;
4518 	int ret = -EINVAL;
4519 
4520 	if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
4521 	    !cpumask_test_cpu(cpu, buffer_b->cpumask))
4522 		goto out;
4523 
4524 	cpu_buffer_a = buffer_a->buffers[cpu];
4525 	cpu_buffer_b = buffer_b->buffers[cpu];
4526 
4527 	/* At least make sure the two buffers are somewhat the same */
4528 	if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
4529 		goto out;
4530 
4531 	ret = -EAGAIN;
4532 
4533 	if (atomic_read(&buffer_a->record_disabled))
4534 		goto out;
4535 
4536 	if (atomic_read(&buffer_b->record_disabled))
4537 		goto out;
4538 
4539 	if (atomic_read(&cpu_buffer_a->record_disabled))
4540 		goto out;
4541 
4542 	if (atomic_read(&cpu_buffer_b->record_disabled))
4543 		goto out;
4544 
4545 	/*
4546 	 * We can't do a synchronize_rcu here because this
4547 	 * function can be called in atomic context.
4548 	 * Normally this will be called from the same CPU as cpu.
4549 	 * If not it's up to the caller to protect this.
4550 	 */
4551 	atomic_inc(&cpu_buffer_a->record_disabled);
4552 	atomic_inc(&cpu_buffer_b->record_disabled);
4553 
4554 	ret = -EBUSY;
4555 	if (local_read(&cpu_buffer_a->committing))
4556 		goto out_dec;
4557 	if (local_read(&cpu_buffer_b->committing))
4558 		goto out_dec;
4559 
4560 	buffer_a->buffers[cpu] = cpu_buffer_b;
4561 	buffer_b->buffers[cpu] = cpu_buffer_a;
4562 
4563 	cpu_buffer_b->buffer = buffer_a;
4564 	cpu_buffer_a->buffer = buffer_b;
4565 
4566 	ret = 0;
4567 
4568 out_dec:
4569 	atomic_dec(&cpu_buffer_a->record_disabled);
4570 	atomic_dec(&cpu_buffer_b->record_disabled);
4571 out:
4572 	return ret;
4573 }
4574 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
4575 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
4576 
4577 /**
4578  * ring_buffer_alloc_read_page - allocate a page to read from buffer
4579  * @buffer: the buffer to allocate for.
4580  * @cpu: the cpu buffer to allocate.
4581  *
4582  * This function is used in conjunction with ring_buffer_read_page.
4583  * When reading a full page from the ring buffer, these functions
4584  * can be used to speed up the process. The calling function should
4585  * allocate a few pages first with this function. Then when it
4586  * needs to get pages from the ring buffer, it passes the result
4587  * of this function into ring_buffer_read_page, which will swap
4588  * the page that was allocated, with the read page of the buffer.
4589  *
4590  * Returns:
4591  *  The page allocated, or ERR_PTR
4592  */
ring_buffer_alloc_read_page(struct ring_buffer * buffer,int cpu)4593 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu)
4594 {
4595 	struct ring_buffer_per_cpu *cpu_buffer;
4596 	struct buffer_data_page *bpage = NULL;
4597 	unsigned long flags;
4598 	struct page *page;
4599 
4600 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4601 		return ERR_PTR(-ENODEV);
4602 
4603 	cpu_buffer = buffer->buffers[cpu];
4604 	local_irq_save(flags);
4605 	arch_spin_lock(&cpu_buffer->lock);
4606 
4607 	if (cpu_buffer->free_page) {
4608 		bpage = cpu_buffer->free_page;
4609 		cpu_buffer->free_page = NULL;
4610 	}
4611 
4612 	arch_spin_unlock(&cpu_buffer->lock);
4613 	local_irq_restore(flags);
4614 
4615 	if (bpage)
4616 		goto out;
4617 
4618 	page = alloc_pages_node(cpu_to_node(cpu),
4619 				GFP_KERNEL | __GFP_NORETRY, 0);
4620 	if (!page)
4621 		return ERR_PTR(-ENOMEM);
4622 
4623 	bpage = page_address(page);
4624 
4625  out:
4626 	rb_init_page(bpage);
4627 
4628 	return bpage;
4629 }
4630 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
4631 
4632 /**
4633  * ring_buffer_free_read_page - free an allocated read page
4634  * @buffer: the buffer the page was allocate for
4635  * @cpu: the cpu buffer the page came from
4636  * @data: the page to free
4637  *
4638  * Free a page allocated from ring_buffer_alloc_read_page.
4639  */
ring_buffer_free_read_page(struct ring_buffer * buffer,int cpu,void * data)4640 void ring_buffer_free_read_page(struct ring_buffer *buffer, int cpu, void *data)
4641 {
4642 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4643 	struct buffer_data_page *bpage = data;
4644 	struct page *page = virt_to_page(bpage);
4645 	unsigned long flags;
4646 
4647 	/* If the page is still in use someplace else, we can't reuse it */
4648 	if (page_ref_count(page) > 1)
4649 		goto out;
4650 
4651 	local_irq_save(flags);
4652 	arch_spin_lock(&cpu_buffer->lock);
4653 
4654 	if (!cpu_buffer->free_page) {
4655 		cpu_buffer->free_page = bpage;
4656 		bpage = NULL;
4657 	}
4658 
4659 	arch_spin_unlock(&cpu_buffer->lock);
4660 	local_irq_restore(flags);
4661 
4662  out:
4663 	free_page((unsigned long)bpage);
4664 }
4665 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
4666 
4667 /**
4668  * ring_buffer_read_page - extract a page from the ring buffer
4669  * @buffer: buffer to extract from
4670  * @data_page: the page to use allocated from ring_buffer_alloc_read_page
4671  * @len: amount to extract
4672  * @cpu: the cpu of the buffer to extract
4673  * @full: should the extraction only happen when the page is full.
4674  *
4675  * This function will pull out a page from the ring buffer and consume it.
4676  * @data_page must be the address of the variable that was returned
4677  * from ring_buffer_alloc_read_page. This is because the page might be used
4678  * to swap with a page in the ring buffer.
4679  *
4680  * for example:
4681  *	rpage = ring_buffer_alloc_read_page(buffer, cpu);
4682  *	if (IS_ERR(rpage))
4683  *		return PTR_ERR(rpage);
4684  *	ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
4685  *	if (ret >= 0)
4686  *		process_page(rpage, ret);
4687  *
4688  * When @full is set, the function will not return true unless
4689  * the writer is off the reader page.
4690  *
4691  * Note: it is up to the calling functions to handle sleeps and wakeups.
4692  *  The ring buffer can be used anywhere in the kernel and can not
4693  *  blindly call wake_up. The layer that uses the ring buffer must be
4694  *  responsible for that.
4695  *
4696  * Returns:
4697  *  >=0 if data has been transferred, returns the offset of consumed data.
4698  *  <0 if no data has been transferred.
4699  */
ring_buffer_read_page(struct ring_buffer * buffer,void ** data_page,size_t len,int cpu,int full)4700 int ring_buffer_read_page(struct ring_buffer *buffer,
4701 			  void **data_page, size_t len, int cpu, int full)
4702 {
4703 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4704 	struct ring_buffer_event *event;
4705 	struct buffer_data_page *bpage;
4706 	struct buffer_page *reader;
4707 	unsigned long missed_events;
4708 	unsigned long flags;
4709 	unsigned int commit;
4710 	unsigned int read;
4711 	u64 save_timestamp;
4712 	int ret = -1;
4713 
4714 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4715 		goto out;
4716 
4717 	/*
4718 	 * If len is not big enough to hold the page header, then
4719 	 * we can not copy anything.
4720 	 */
4721 	if (len <= BUF_PAGE_HDR_SIZE)
4722 		goto out;
4723 
4724 	len -= BUF_PAGE_HDR_SIZE;
4725 
4726 	if (!data_page)
4727 		goto out;
4728 
4729 	bpage = *data_page;
4730 	if (!bpage)
4731 		goto out;
4732 
4733 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4734 
4735 	reader = rb_get_reader_page(cpu_buffer);
4736 	if (!reader)
4737 		goto out_unlock;
4738 
4739 	event = rb_reader_event(cpu_buffer);
4740 
4741 	read = reader->read;
4742 	commit = rb_page_commit(reader);
4743 
4744 	/* Check if any events were dropped */
4745 	missed_events = cpu_buffer->lost_events;
4746 
4747 	/*
4748 	 * If this page has been partially read or
4749 	 * if len is not big enough to read the rest of the page or
4750 	 * a writer is still on the page, then
4751 	 * we must copy the data from the page to the buffer.
4752 	 * Otherwise, we can simply swap the page with the one passed in.
4753 	 */
4754 	if (read || (len < (commit - read)) ||
4755 	    cpu_buffer->reader_page == cpu_buffer->commit_page) {
4756 		struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
4757 		unsigned int rpos = read;
4758 		unsigned int pos = 0;
4759 		unsigned int size;
4760 
4761 		if (full)
4762 			goto out_unlock;
4763 
4764 		if (len > (commit - read))
4765 			len = (commit - read);
4766 
4767 		/* Always keep the time extend and data together */
4768 		size = rb_event_ts_length(event);
4769 
4770 		if (len < size)
4771 			goto out_unlock;
4772 
4773 		/* save the current timestamp, since the user will need it */
4774 		save_timestamp = cpu_buffer->read_stamp;
4775 
4776 		/* Need to copy one event at a time */
4777 		do {
4778 			/* We need the size of one event, because
4779 			 * rb_advance_reader only advances by one event,
4780 			 * whereas rb_event_ts_length may include the size of
4781 			 * one or two events.
4782 			 * We have already ensured there's enough space if this
4783 			 * is a time extend. */
4784 			size = rb_event_length(event);
4785 			memcpy(bpage->data + pos, rpage->data + rpos, size);
4786 
4787 			len -= size;
4788 
4789 			rb_advance_reader(cpu_buffer);
4790 			rpos = reader->read;
4791 			pos += size;
4792 
4793 			if (rpos >= commit)
4794 				break;
4795 
4796 			event = rb_reader_event(cpu_buffer);
4797 			/* Always keep the time extend and data together */
4798 			size = rb_event_ts_length(event);
4799 		} while (len >= size);
4800 
4801 		/* update bpage */
4802 		local_set(&bpage->commit, pos);
4803 		bpage->time_stamp = save_timestamp;
4804 
4805 		/* we copied everything to the beginning */
4806 		read = 0;
4807 	} else {
4808 		/* update the entry counter */
4809 		cpu_buffer->read += rb_page_entries(reader);
4810 		cpu_buffer->read_bytes += BUF_PAGE_SIZE;
4811 
4812 		/* swap the pages */
4813 		rb_init_page(bpage);
4814 		bpage = reader->page;
4815 		reader->page = *data_page;
4816 		local_set(&reader->write, 0);
4817 		local_set(&reader->entries, 0);
4818 		reader->read = 0;
4819 		*data_page = bpage;
4820 
4821 		/*
4822 		 * Use the real_end for the data size,
4823 		 * This gives us a chance to store the lost events
4824 		 * on the page.
4825 		 */
4826 		if (reader->real_end)
4827 			local_set(&bpage->commit, reader->real_end);
4828 	}
4829 	ret = read;
4830 
4831 	cpu_buffer->lost_events = 0;
4832 
4833 	commit = local_read(&bpage->commit);
4834 	/*
4835 	 * Set a flag in the commit field if we lost events
4836 	 */
4837 	if (missed_events) {
4838 		/* If there is room at the end of the page to save the
4839 		 * missed events, then record it there.
4840 		 */
4841 		if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
4842 			memcpy(&bpage->data[commit], &missed_events,
4843 			       sizeof(missed_events));
4844 			local_add(RB_MISSED_STORED, &bpage->commit);
4845 			commit += sizeof(missed_events);
4846 		}
4847 		local_add(RB_MISSED_EVENTS, &bpage->commit);
4848 	}
4849 
4850 	/*
4851 	 * This page may be off to user land. Zero it out here.
4852 	 */
4853 	if (commit < BUF_PAGE_SIZE)
4854 		memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
4855 
4856  out_unlock:
4857 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4858 
4859  out:
4860 	return ret;
4861 }
4862 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
4863 
4864 /*
4865  * We only allocate new buffers, never free them if the CPU goes down.
4866  * If we were to free the buffer, then the user would lose any trace that was in
4867  * the buffer.
4868  */
trace_rb_cpu_prepare(unsigned int cpu,struct hlist_node * node)4869 int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
4870 {
4871 	struct ring_buffer *buffer;
4872 	long nr_pages_same;
4873 	int cpu_i;
4874 	unsigned long nr_pages;
4875 
4876 	buffer = container_of(node, struct ring_buffer, node);
4877 	if (cpumask_test_cpu(cpu, buffer->cpumask))
4878 		return 0;
4879 
4880 	nr_pages = 0;
4881 	nr_pages_same = 1;
4882 	/* check if all cpu sizes are same */
4883 	for_each_buffer_cpu(buffer, cpu_i) {
4884 		/* fill in the size from first enabled cpu */
4885 		if (nr_pages == 0)
4886 			nr_pages = buffer->buffers[cpu_i]->nr_pages;
4887 		if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
4888 			nr_pages_same = 0;
4889 			break;
4890 		}
4891 	}
4892 	/* allocate minimum pages, user can later expand it */
4893 	if (!nr_pages_same)
4894 		nr_pages = 2;
4895 	buffer->buffers[cpu] =
4896 		rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
4897 	if (!buffer->buffers[cpu]) {
4898 		WARN(1, "failed to allocate ring buffer on CPU %u\n",
4899 		     cpu);
4900 		return -ENOMEM;
4901 	}
4902 	smp_wmb();
4903 	cpumask_set_cpu(cpu, buffer->cpumask);
4904 	return 0;
4905 }
4906 
4907 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
4908 /*
4909  * This is a basic integrity check of the ring buffer.
4910  * Late in the boot cycle this test will run when configured in.
4911  * It will kick off a thread per CPU that will go into a loop
4912  * writing to the per cpu ring buffer various sizes of data.
4913  * Some of the data will be large items, some small.
4914  *
4915  * Another thread is created that goes into a spin, sending out
4916  * IPIs to the other CPUs to also write into the ring buffer.
4917  * this is to test the nesting ability of the buffer.
4918  *
4919  * Basic stats are recorded and reported. If something in the
4920  * ring buffer should happen that's not expected, a big warning
4921  * is displayed and all ring buffers are disabled.
4922  */
4923 static struct task_struct *rb_threads[NR_CPUS] __initdata;
4924 
4925 struct rb_test_data {
4926 	struct ring_buffer	*buffer;
4927 	unsigned long		events;
4928 	unsigned long		bytes_written;
4929 	unsigned long		bytes_alloc;
4930 	unsigned long		bytes_dropped;
4931 	unsigned long		events_nested;
4932 	unsigned long		bytes_written_nested;
4933 	unsigned long		bytes_alloc_nested;
4934 	unsigned long		bytes_dropped_nested;
4935 	int			min_size_nested;
4936 	int			max_size_nested;
4937 	int			max_size;
4938 	int			min_size;
4939 	int			cpu;
4940 	int			cnt;
4941 };
4942 
4943 static struct rb_test_data rb_data[NR_CPUS] __initdata;
4944 
4945 /* 1 meg per cpu */
4946 #define RB_TEST_BUFFER_SIZE	1048576
4947 
4948 static char rb_string[] __initdata =
4949 	"abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
4950 	"?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
4951 	"!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
4952 
4953 static bool rb_test_started __initdata;
4954 
4955 struct rb_item {
4956 	int size;
4957 	char str[];
4958 };
4959 
rb_write_something(struct rb_test_data * data,bool nested)4960 static __init int rb_write_something(struct rb_test_data *data, bool nested)
4961 {
4962 	struct ring_buffer_event *event;
4963 	struct rb_item *item;
4964 	bool started;
4965 	int event_len;
4966 	int size;
4967 	int len;
4968 	int cnt;
4969 
4970 	/* Have nested writes different that what is written */
4971 	cnt = data->cnt + (nested ? 27 : 0);
4972 
4973 	/* Multiply cnt by ~e, to make some unique increment */
4974 	size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
4975 
4976 	len = size + sizeof(struct rb_item);
4977 
4978 	started = rb_test_started;
4979 	/* read rb_test_started before checking buffer enabled */
4980 	smp_rmb();
4981 
4982 	event = ring_buffer_lock_reserve(data->buffer, len);
4983 	if (!event) {
4984 		/* Ignore dropped events before test starts. */
4985 		if (started) {
4986 			if (nested)
4987 				data->bytes_dropped += len;
4988 			else
4989 				data->bytes_dropped_nested += len;
4990 		}
4991 		return len;
4992 	}
4993 
4994 	event_len = ring_buffer_event_length(event);
4995 
4996 	if (RB_WARN_ON(data->buffer, event_len < len))
4997 		goto out;
4998 
4999 	item = ring_buffer_event_data(event);
5000 	item->size = size;
5001 	memcpy(item->str, rb_string, size);
5002 
5003 	if (nested) {
5004 		data->bytes_alloc_nested += event_len;
5005 		data->bytes_written_nested += len;
5006 		data->events_nested++;
5007 		if (!data->min_size_nested || len < data->min_size_nested)
5008 			data->min_size_nested = len;
5009 		if (len > data->max_size_nested)
5010 			data->max_size_nested = len;
5011 	} else {
5012 		data->bytes_alloc += event_len;
5013 		data->bytes_written += len;
5014 		data->events++;
5015 		if (!data->min_size || len < data->min_size)
5016 			data->max_size = len;
5017 		if (len > data->max_size)
5018 			data->max_size = len;
5019 	}
5020 
5021  out:
5022 	ring_buffer_unlock_commit(data->buffer, event);
5023 
5024 	return 0;
5025 }
5026 
rb_test(void * arg)5027 static __init int rb_test(void *arg)
5028 {
5029 	struct rb_test_data *data = arg;
5030 
5031 	while (!kthread_should_stop()) {
5032 		rb_write_something(data, false);
5033 		data->cnt++;
5034 
5035 		set_current_state(TASK_INTERRUPTIBLE);
5036 		/* Now sleep between a min of 100-300us and a max of 1ms */
5037 		usleep_range(((data->cnt % 3) + 1) * 100, 1000);
5038 	}
5039 
5040 	return 0;
5041 }
5042 
rb_ipi(void * ignore)5043 static __init void rb_ipi(void *ignore)
5044 {
5045 	struct rb_test_data *data;
5046 	int cpu = smp_processor_id();
5047 
5048 	data = &rb_data[cpu];
5049 	rb_write_something(data, true);
5050 }
5051 
rb_hammer_test(void * arg)5052 static __init int rb_hammer_test(void *arg)
5053 {
5054 	while (!kthread_should_stop()) {
5055 
5056 		/* Send an IPI to all cpus to write data! */
5057 		smp_call_function(rb_ipi, NULL, 1);
5058 		/* No sleep, but for non preempt, let others run */
5059 		schedule();
5060 	}
5061 
5062 	return 0;
5063 }
5064 
test_ringbuffer(void)5065 static __init int test_ringbuffer(void)
5066 {
5067 	struct task_struct *rb_hammer;
5068 	struct ring_buffer *buffer;
5069 	int cpu;
5070 	int ret = 0;
5071 
5072 	if (security_locked_down(LOCKDOWN_TRACEFS)) {
5073 		pr_warning("Lockdown is enabled, skipping ring buffer tests\n");
5074 		return 0;
5075 	}
5076 
5077 	pr_info("Running ring buffer tests...\n");
5078 
5079 	buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
5080 	if (WARN_ON(!buffer))
5081 		return 0;
5082 
5083 	/* Disable buffer so that threads can't write to it yet */
5084 	ring_buffer_record_off(buffer);
5085 
5086 	for_each_online_cpu(cpu) {
5087 		rb_data[cpu].buffer = buffer;
5088 		rb_data[cpu].cpu = cpu;
5089 		rb_data[cpu].cnt = cpu;
5090 		rb_threads[cpu] = kthread_create(rb_test, &rb_data[cpu],
5091 						 "rbtester/%d", cpu);
5092 		if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
5093 			pr_cont("FAILED\n");
5094 			ret = PTR_ERR(rb_threads[cpu]);
5095 			goto out_free;
5096 		}
5097 
5098 		kthread_bind(rb_threads[cpu], cpu);
5099  		wake_up_process(rb_threads[cpu]);
5100 	}
5101 
5102 	/* Now create the rb hammer! */
5103 	rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
5104 	if (WARN_ON(IS_ERR(rb_hammer))) {
5105 		pr_cont("FAILED\n");
5106 		ret = PTR_ERR(rb_hammer);
5107 		goto out_free;
5108 	}
5109 
5110 	ring_buffer_record_on(buffer);
5111 	/*
5112 	 * Show buffer is enabled before setting rb_test_started.
5113 	 * Yes there's a small race window where events could be
5114 	 * dropped and the thread wont catch it. But when a ring
5115 	 * buffer gets enabled, there will always be some kind of
5116 	 * delay before other CPUs see it. Thus, we don't care about
5117 	 * those dropped events. We care about events dropped after
5118 	 * the threads see that the buffer is active.
5119 	 */
5120 	smp_wmb();
5121 	rb_test_started = true;
5122 
5123 	set_current_state(TASK_INTERRUPTIBLE);
5124 	/* Just run for 10 seconds */;
5125 	schedule_timeout(10 * HZ);
5126 
5127 	kthread_stop(rb_hammer);
5128 
5129  out_free:
5130 	for_each_online_cpu(cpu) {
5131 		if (!rb_threads[cpu])
5132 			break;
5133 		kthread_stop(rb_threads[cpu]);
5134 	}
5135 	if (ret) {
5136 		ring_buffer_free(buffer);
5137 		return ret;
5138 	}
5139 
5140 	/* Report! */
5141 	pr_info("finished\n");
5142 	for_each_online_cpu(cpu) {
5143 		struct ring_buffer_event *event;
5144 		struct rb_test_data *data = &rb_data[cpu];
5145 		struct rb_item *item;
5146 		unsigned long total_events;
5147 		unsigned long total_dropped;
5148 		unsigned long total_written;
5149 		unsigned long total_alloc;
5150 		unsigned long total_read = 0;
5151 		unsigned long total_size = 0;
5152 		unsigned long total_len = 0;
5153 		unsigned long total_lost = 0;
5154 		unsigned long lost;
5155 		int big_event_size;
5156 		int small_event_size;
5157 
5158 		ret = -1;
5159 
5160 		total_events = data->events + data->events_nested;
5161 		total_written = data->bytes_written + data->bytes_written_nested;
5162 		total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
5163 		total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
5164 
5165 		big_event_size = data->max_size + data->max_size_nested;
5166 		small_event_size = data->min_size + data->min_size_nested;
5167 
5168 		pr_info("CPU %d:\n", cpu);
5169 		pr_info("              events:    %ld\n", total_events);
5170 		pr_info("       dropped bytes:    %ld\n", total_dropped);
5171 		pr_info("       alloced bytes:    %ld\n", total_alloc);
5172 		pr_info("       written bytes:    %ld\n", total_written);
5173 		pr_info("       biggest event:    %d\n", big_event_size);
5174 		pr_info("      smallest event:    %d\n", small_event_size);
5175 
5176 		if (RB_WARN_ON(buffer, total_dropped))
5177 			break;
5178 
5179 		ret = 0;
5180 
5181 		while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
5182 			total_lost += lost;
5183 			item = ring_buffer_event_data(event);
5184 			total_len += ring_buffer_event_length(event);
5185 			total_size += item->size + sizeof(struct rb_item);
5186 			if (memcmp(&item->str[0], rb_string, item->size) != 0) {
5187 				pr_info("FAILED!\n");
5188 				pr_info("buffer had: %.*s\n", item->size, item->str);
5189 				pr_info("expected:   %.*s\n", item->size, rb_string);
5190 				RB_WARN_ON(buffer, 1);
5191 				ret = -1;
5192 				break;
5193 			}
5194 			total_read++;
5195 		}
5196 		if (ret)
5197 			break;
5198 
5199 		ret = -1;
5200 
5201 		pr_info("         read events:   %ld\n", total_read);
5202 		pr_info("         lost events:   %ld\n", total_lost);
5203 		pr_info("        total events:   %ld\n", total_lost + total_read);
5204 		pr_info("  recorded len bytes:   %ld\n", total_len);
5205 		pr_info(" recorded size bytes:   %ld\n", total_size);
5206 		if (total_lost)
5207 			pr_info(" With dropped events, record len and size may not match\n"
5208 				" alloced and written from above\n");
5209 		if (!total_lost) {
5210 			if (RB_WARN_ON(buffer, total_len != total_alloc ||
5211 				       total_size != total_written))
5212 				break;
5213 		}
5214 		if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
5215 			break;
5216 
5217 		ret = 0;
5218 	}
5219 	if (!ret)
5220 		pr_info("Ring buffer PASSED!\n");
5221 
5222 	ring_buffer_free(buffer);
5223 	return 0;
5224 }
5225 
5226 late_initcall(test_ringbuffer);
5227 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */
5228