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_recursion.h>
8 #include <linux/trace_events.h>
9 #include <linux/ring_buffer.h>
10 #include <linux/trace_clock.h>
11 #include <linux/sched/clock.h>
12 #include <linux/trace_seq.h>
13 #include <linux/spinlock.h>
14 #include <linux/irq_work.h>
15 #include <linux/security.h>
16 #include <linux/uaccess.h>
17 #include <linux/hardirq.h>
18 #include <linux/kthread.h>	/* for self test */
19 #include <linux/module.h>
20 #include <linux/percpu.h>
21 #include <linux/mutex.h>
22 #include <linux/delay.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/hash.h>
26 #include <linux/list.h>
27 #include <linux/cpu.h>
28 #include <linux/oom.h>
29 
30 #include <asm/local.h>
31 
32 /*
33  * The "absolute" timestamp in the buffer is only 59 bits.
34  * If a clock has the 5 MSBs set, it needs to be saved and
35  * reinserted.
36  */
37 #define TS_MSB		(0xf8ULL << 56)
38 #define ABS_TS_MASK	(~TS_MSB)
39 
40 static void update_pages_handler(struct work_struct *work);
41 
42 /*
43  * The ring buffer header is special. We must manually up keep it.
44  */
ring_buffer_print_entry_header(struct trace_seq * s)45 int ring_buffer_print_entry_header(struct trace_seq *s)
46 {
47 	trace_seq_puts(s, "# compressed entry header\n");
48 	trace_seq_puts(s, "\ttype_len    :    5 bits\n");
49 	trace_seq_puts(s, "\ttime_delta  :   27 bits\n");
50 	trace_seq_puts(s, "\tarray       :   32 bits\n");
51 	trace_seq_putc(s, '\n');
52 	trace_seq_printf(s, "\tpadding     : type == %d\n",
53 			 RINGBUF_TYPE_PADDING);
54 	trace_seq_printf(s, "\ttime_extend : type == %d\n",
55 			 RINGBUF_TYPE_TIME_EXTEND);
56 	trace_seq_printf(s, "\ttime_stamp : type == %d\n",
57 			 RINGBUF_TYPE_TIME_STAMP);
58 	trace_seq_printf(s, "\tdata max type_len  == %d\n",
59 			 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
60 
61 	return !trace_seq_has_overflowed(s);
62 }
63 
64 /*
65  * The ring buffer is made up of a list of pages. A separate list of pages is
66  * allocated for each CPU. A writer may only write to a buffer that is
67  * associated with the CPU it is currently executing on.  A reader may read
68  * from any per cpu buffer.
69  *
70  * The reader is special. For each per cpu buffer, the reader has its own
71  * reader page. When a reader has read the entire reader page, this reader
72  * page is swapped with another page in the ring buffer.
73  *
74  * Now, as long as the writer is off the reader page, the reader can do what
75  * ever it wants with that page. The writer will never write to that page
76  * again (as long as it is out of the ring buffer).
77  *
78  * Here's some silly ASCII art.
79  *
80  *   +------+
81  *   |reader|          RING BUFFER
82  *   |page  |
83  *   +------+        +---+   +---+   +---+
84  *                   |   |-->|   |-->|   |
85  *                   +---+   +---+   +---+
86  *                     ^               |
87  *                     |               |
88  *                     +---------------+
89  *
90  *
91  *   +------+
92  *   |reader|          RING BUFFER
93  *   |page  |------------------v
94  *   +------+        +---+   +---+   +---+
95  *                   |   |-->|   |-->|   |
96  *                   +---+   +---+   +---+
97  *                     ^               |
98  *                     |               |
99  *                     +---------------+
100  *
101  *
102  *   +------+
103  *   |reader|          RING BUFFER
104  *   |page  |------------------v
105  *   +------+        +---+   +---+   +---+
106  *      ^            |   |-->|   |-->|   |
107  *      |            +---+   +---+   +---+
108  *      |                              |
109  *      |                              |
110  *      +------------------------------+
111  *
112  *
113  *   +------+
114  *   |buffer|          RING BUFFER
115  *   |page  |------------------v
116  *   +------+        +---+   +---+   +---+
117  *      ^            |   |   |   |-->|   |
118  *      |   New      +---+   +---+   +---+
119  *      |  Reader------^               |
120  *      |   page                       |
121  *      +------------------------------+
122  *
123  *
124  * After we make this swap, the reader can hand this page off to the splice
125  * code and be done with it. It can even allocate a new page if it needs to
126  * and swap that into the ring buffer.
127  *
128  * We will be using cmpxchg soon to make all this lockless.
129  *
130  */
131 
132 /* Used for individual buffers (after the counter) */
133 #define RB_BUFFER_OFF		(1 << 20)
134 
135 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
136 
137 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
138 #define RB_ALIGNMENT		4U
139 #define RB_MAX_SMALL_DATA	(RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
140 #define RB_EVNT_MIN_SIZE	8U	/* two 32bit words */
141 
142 #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
143 # define RB_FORCE_8BYTE_ALIGNMENT	0
144 # define RB_ARCH_ALIGNMENT		RB_ALIGNMENT
145 #else
146 # define RB_FORCE_8BYTE_ALIGNMENT	1
147 # define RB_ARCH_ALIGNMENT		8U
148 #endif
149 
150 #define RB_ALIGN_DATA		__aligned(RB_ARCH_ALIGNMENT)
151 
152 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
153 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
154 
155 enum {
156 	RB_LEN_TIME_EXTEND = 8,
157 	RB_LEN_TIME_STAMP =  8,
158 };
159 
160 #define skip_time_extend(event) \
161 	((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
162 
163 #define extended_time(event) \
164 	(event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
165 
rb_null_event(struct ring_buffer_event * event)166 static inline int rb_null_event(struct ring_buffer_event *event)
167 {
168 	return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
169 }
170 
rb_event_set_padding(struct ring_buffer_event * event)171 static void rb_event_set_padding(struct ring_buffer_event *event)
172 {
173 	/* padding has a NULL time_delta */
174 	event->type_len = RINGBUF_TYPE_PADDING;
175 	event->time_delta = 0;
176 }
177 
178 static unsigned
rb_event_data_length(struct ring_buffer_event * event)179 rb_event_data_length(struct ring_buffer_event *event)
180 {
181 	unsigned length;
182 
183 	if (event->type_len)
184 		length = event->type_len * RB_ALIGNMENT;
185 	else
186 		length = event->array[0];
187 	return length + RB_EVNT_HDR_SIZE;
188 }
189 
190 /*
191  * Return the length of the given event. Will return
192  * the length of the time extend if the event is a
193  * time extend.
194  */
195 static inline unsigned
rb_event_length(struct ring_buffer_event * event)196 rb_event_length(struct ring_buffer_event *event)
197 {
198 	switch (event->type_len) {
199 	case RINGBUF_TYPE_PADDING:
200 		if (rb_null_event(event))
201 			/* undefined */
202 			return -1;
203 		return  event->array[0] + RB_EVNT_HDR_SIZE;
204 
205 	case RINGBUF_TYPE_TIME_EXTEND:
206 		return RB_LEN_TIME_EXTEND;
207 
208 	case RINGBUF_TYPE_TIME_STAMP:
209 		return RB_LEN_TIME_STAMP;
210 
211 	case RINGBUF_TYPE_DATA:
212 		return rb_event_data_length(event);
213 	default:
214 		WARN_ON_ONCE(1);
215 	}
216 	/* not hit */
217 	return 0;
218 }
219 
220 /*
221  * Return total length of time extend and data,
222  *   or just the event length for all other events.
223  */
224 static inline unsigned
rb_event_ts_length(struct ring_buffer_event * event)225 rb_event_ts_length(struct ring_buffer_event *event)
226 {
227 	unsigned len = 0;
228 
229 	if (extended_time(event)) {
230 		/* time extends include the data event after it */
231 		len = RB_LEN_TIME_EXTEND;
232 		event = skip_time_extend(event);
233 	}
234 	return len + rb_event_length(event);
235 }
236 
237 /**
238  * ring_buffer_event_length - return the length of the event
239  * @event: the event to get the length of
240  *
241  * Returns the size of the data load of a data event.
242  * If the event is something other than a data event, it
243  * returns the size of the event itself. With the exception
244  * of a TIME EXTEND, where it still returns the size of the
245  * data load of the data event after it.
246  */
ring_buffer_event_length(struct ring_buffer_event * event)247 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
248 {
249 	unsigned length;
250 
251 	if (extended_time(event))
252 		event = skip_time_extend(event);
253 
254 	length = rb_event_length(event);
255 	if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
256 		return length;
257 	length -= RB_EVNT_HDR_SIZE;
258 	if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
259                 length -= sizeof(event->array[0]);
260 	return length;
261 }
262 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
263 
264 /* inline for ring buffer fast paths */
265 static __always_inline void *
rb_event_data(struct ring_buffer_event * event)266 rb_event_data(struct ring_buffer_event *event)
267 {
268 	if (extended_time(event))
269 		event = skip_time_extend(event);
270 	WARN_ON_ONCE(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
271 	/* If length is in len field, then array[0] has the data */
272 	if (event->type_len)
273 		return (void *)&event->array[0];
274 	/* Otherwise length is in array[0] and array[1] has the data */
275 	return (void *)&event->array[1];
276 }
277 
278 /**
279  * ring_buffer_event_data - return the data of the event
280  * @event: the event to get the data from
281  */
ring_buffer_event_data(struct ring_buffer_event * event)282 void *ring_buffer_event_data(struct ring_buffer_event *event)
283 {
284 	return rb_event_data(event);
285 }
286 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
287 
288 #define for_each_buffer_cpu(buffer, cpu)		\
289 	for_each_cpu(cpu, buffer->cpumask)
290 
291 #define for_each_online_buffer_cpu(buffer, cpu)		\
292 	for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask)
293 
294 #define TS_SHIFT	27
295 #define TS_MASK		((1ULL << TS_SHIFT) - 1)
296 #define TS_DELTA_TEST	(~TS_MASK)
297 
rb_event_time_stamp(struct ring_buffer_event * event)298 static u64 rb_event_time_stamp(struct ring_buffer_event *event)
299 {
300 	u64 ts;
301 
302 	ts = event->array[0];
303 	ts <<= TS_SHIFT;
304 	ts += event->time_delta;
305 
306 	return ts;
307 }
308 
309 /* Flag when events were overwritten */
310 #define RB_MISSED_EVENTS	(1 << 31)
311 /* Missed count stored at end */
312 #define RB_MISSED_STORED	(1 << 30)
313 
314 struct buffer_data_page {
315 	u64		 time_stamp;	/* page time stamp */
316 	local_t		 commit;	/* write committed index */
317 	unsigned char	 data[] RB_ALIGN_DATA;	/* data of buffer page */
318 };
319 
320 /*
321  * Note, the buffer_page list must be first. The buffer pages
322  * are allocated in cache lines, which means that each buffer
323  * page will be at the beginning of a cache line, and thus
324  * the least significant bits will be zero. We use this to
325  * add flags in the list struct pointers, to make the ring buffer
326  * lockless.
327  */
328 struct buffer_page {
329 	struct list_head list;		/* list of buffer pages */
330 	local_t		 write;		/* index for next write */
331 	unsigned	 read;		/* index for next read */
332 	local_t		 entries;	/* entries on this page */
333 	unsigned long	 real_end;	/* real end of data */
334 	struct buffer_data_page *page;	/* Actual data page */
335 };
336 
337 /*
338  * The buffer page counters, write and entries, must be reset
339  * atomically when crossing page boundaries. To synchronize this
340  * update, two counters are inserted into the number. One is
341  * the actual counter for the write position or count on the page.
342  *
343  * The other is a counter of updaters. Before an update happens
344  * the update partition of the counter is incremented. This will
345  * allow the updater to update the counter atomically.
346  *
347  * The counter is 20 bits, and the state data is 12.
348  */
349 #define RB_WRITE_MASK		0xfffff
350 #define RB_WRITE_INTCNT		(1 << 20)
351 
rb_init_page(struct buffer_data_page * bpage)352 static void rb_init_page(struct buffer_data_page *bpage)
353 {
354 	local_set(&bpage->commit, 0);
355 }
356 
357 /*
358  * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
359  * this issue out.
360  */
free_buffer_page(struct buffer_page * bpage)361 static void free_buffer_page(struct buffer_page *bpage)
362 {
363 	free_page((unsigned long)bpage->page);
364 	kfree(bpage);
365 }
366 
367 /*
368  * We need to fit the time_stamp delta into 27 bits.
369  */
test_time_stamp(u64 delta)370 static inline int test_time_stamp(u64 delta)
371 {
372 	if (delta & TS_DELTA_TEST)
373 		return 1;
374 	return 0;
375 }
376 
377 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
378 
379 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
380 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
381 
ring_buffer_print_page_header(struct trace_seq * s)382 int ring_buffer_print_page_header(struct trace_seq *s)
383 {
384 	struct buffer_data_page field;
385 
386 	trace_seq_printf(s, "\tfield: u64 timestamp;\t"
387 			 "offset:0;\tsize:%u;\tsigned:%u;\n",
388 			 (unsigned int)sizeof(field.time_stamp),
389 			 (unsigned int)is_signed_type(u64));
390 
391 	trace_seq_printf(s, "\tfield: local_t commit;\t"
392 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
393 			 (unsigned int)offsetof(typeof(field), commit),
394 			 (unsigned int)sizeof(field.commit),
395 			 (unsigned int)is_signed_type(long));
396 
397 	trace_seq_printf(s, "\tfield: int overwrite;\t"
398 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
399 			 (unsigned int)offsetof(typeof(field), commit),
400 			 1,
401 			 (unsigned int)is_signed_type(long));
402 
403 	trace_seq_printf(s, "\tfield: char data;\t"
404 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
405 			 (unsigned int)offsetof(typeof(field), data),
406 			 (unsigned int)BUF_PAGE_SIZE,
407 			 (unsigned int)is_signed_type(char));
408 
409 	return !trace_seq_has_overflowed(s);
410 }
411 
412 struct rb_irq_work {
413 	struct irq_work			work;
414 	wait_queue_head_t		waiters;
415 	wait_queue_head_t		full_waiters;
416 	bool				waiters_pending;
417 	bool				full_waiters_pending;
418 	bool				wakeup_full;
419 };
420 
421 /*
422  * Structure to hold event state and handle nested events.
423  */
424 struct rb_event_info {
425 	u64			ts;
426 	u64			delta;
427 	u64			before;
428 	u64			after;
429 	unsigned long		length;
430 	struct buffer_page	*tail_page;
431 	int			add_timestamp;
432 };
433 
434 /*
435  * Used for the add_timestamp
436  *  NONE
437  *  EXTEND - wants a time extend
438  *  ABSOLUTE - the buffer requests all events to have absolute time stamps
439  *  FORCE - force a full time stamp.
440  */
441 enum {
442 	RB_ADD_STAMP_NONE		= 0,
443 	RB_ADD_STAMP_EXTEND		= BIT(1),
444 	RB_ADD_STAMP_ABSOLUTE		= BIT(2),
445 	RB_ADD_STAMP_FORCE		= BIT(3)
446 };
447 /*
448  * Used for which event context the event is in.
449  *  TRANSITION = 0
450  *  NMI     = 1
451  *  IRQ     = 2
452  *  SOFTIRQ = 3
453  *  NORMAL  = 4
454  *
455  * See trace_recursive_lock() comment below for more details.
456  */
457 enum {
458 	RB_CTX_TRANSITION,
459 	RB_CTX_NMI,
460 	RB_CTX_IRQ,
461 	RB_CTX_SOFTIRQ,
462 	RB_CTX_NORMAL,
463 	RB_CTX_MAX
464 };
465 
466 #if BITS_PER_LONG == 32
467 #define RB_TIME_32
468 #endif
469 
470 /* To test on 64 bit machines */
471 //#define RB_TIME_32
472 
473 #ifdef RB_TIME_32
474 
475 struct rb_time_struct {
476 	local_t		cnt;
477 	local_t		top;
478 	local_t		bottom;
479 	local_t		msb;
480 };
481 #else
482 #include <asm/local64.h>
483 struct rb_time_struct {
484 	local64_t	time;
485 };
486 #endif
487 typedef struct rb_time_struct rb_time_t;
488 
489 #define MAX_NEST	5
490 
491 /*
492  * head_page == tail_page && head == tail then buffer is empty.
493  */
494 struct ring_buffer_per_cpu {
495 	int				cpu;
496 	atomic_t			record_disabled;
497 	atomic_t			resize_disabled;
498 	struct trace_buffer	*buffer;
499 	raw_spinlock_t			reader_lock;	/* serialize readers */
500 	arch_spinlock_t			lock;
501 	struct lock_class_key		lock_key;
502 	struct buffer_data_page		*free_page;
503 	unsigned long			nr_pages;
504 	unsigned int			current_context;
505 	struct list_head		*pages;
506 	struct buffer_page		*head_page;	/* read from head */
507 	struct buffer_page		*tail_page;	/* write to tail */
508 	struct buffer_page		*commit_page;	/* committed pages */
509 	struct buffer_page		*reader_page;
510 	unsigned long			lost_events;
511 	unsigned long			last_overrun;
512 	unsigned long			nest;
513 	local_t				entries_bytes;
514 	local_t				entries;
515 	local_t				overrun;
516 	local_t				commit_overrun;
517 	local_t				dropped_events;
518 	local_t				committing;
519 	local_t				commits;
520 	local_t				pages_touched;
521 	local_t				pages_read;
522 	long				last_pages_touch;
523 	size_t				shortest_full;
524 	unsigned long			read;
525 	unsigned long			read_bytes;
526 	rb_time_t			write_stamp;
527 	rb_time_t			before_stamp;
528 	u64				event_stamp[MAX_NEST];
529 	u64				read_stamp;
530 	/* ring buffer pages to update, > 0 to add, < 0 to remove */
531 	long				nr_pages_to_update;
532 	struct list_head		new_pages; /* new pages to add */
533 	struct work_struct		update_pages_work;
534 	struct completion		update_done;
535 
536 	struct rb_irq_work		irq_work;
537 };
538 
539 struct trace_buffer {
540 	unsigned			flags;
541 	int				cpus;
542 	atomic_t			record_disabled;
543 	cpumask_var_t			cpumask;
544 
545 	struct lock_class_key		*reader_lock_key;
546 
547 	struct mutex			mutex;
548 
549 	struct ring_buffer_per_cpu	**buffers;
550 
551 	struct hlist_node		node;
552 	u64				(*clock)(void);
553 
554 	struct rb_irq_work		irq_work;
555 	bool				time_stamp_abs;
556 };
557 
558 struct ring_buffer_iter {
559 	struct ring_buffer_per_cpu	*cpu_buffer;
560 	unsigned long			head;
561 	unsigned long			next_event;
562 	struct buffer_page		*head_page;
563 	struct buffer_page		*cache_reader_page;
564 	unsigned long			cache_read;
565 	u64				read_stamp;
566 	u64				page_stamp;
567 	struct ring_buffer_event	*event;
568 	int				missed_events;
569 };
570 
571 #ifdef RB_TIME_32
572 
573 /*
574  * On 32 bit machines, local64_t is very expensive. As the ring
575  * buffer doesn't need all the features of a true 64 bit atomic,
576  * on 32 bit, it uses these functions (64 still uses local64_t).
577  *
578  * For the ring buffer, 64 bit required operations for the time is
579  * the following:
580  *
581  *  - Reads may fail if it interrupted a modification of the time stamp.
582  *      It will succeed if it did not interrupt another write even if
583  *      the read itself is interrupted by a write.
584  *      It returns whether it was successful or not.
585  *
586  *  - Writes always succeed and will overwrite other writes and writes
587  *      that were done by events interrupting the current write.
588  *
589  *  - A write followed by a read of the same time stamp will always succeed,
590  *      but may not contain the same value.
591  *
592  *  - A cmpxchg will fail if it interrupted another write or cmpxchg.
593  *      Other than that, it acts like a normal cmpxchg.
594  *
595  * The 60 bit time stamp is broken up by 30 bits in a top and bottom half
596  *  (bottom being the least significant 30 bits of the 60 bit time stamp).
597  *
598  * The two most significant bits of each half holds a 2 bit counter (0-3).
599  * Each update will increment this counter by one.
600  * When reading the top and bottom, if the two counter bits match then the
601  *  top and bottom together make a valid 60 bit number.
602  */
603 #define RB_TIME_SHIFT	30
604 #define RB_TIME_VAL_MASK ((1 << RB_TIME_SHIFT) - 1)
605 #define RB_TIME_MSB_SHIFT	 60
606 
rb_time_cnt(unsigned long val)607 static inline int rb_time_cnt(unsigned long val)
608 {
609 	return (val >> RB_TIME_SHIFT) & 3;
610 }
611 
rb_time_val(unsigned long top,unsigned long bottom)612 static inline u64 rb_time_val(unsigned long top, unsigned long bottom)
613 {
614 	u64 val;
615 
616 	val = top & RB_TIME_VAL_MASK;
617 	val <<= RB_TIME_SHIFT;
618 	val |= bottom & RB_TIME_VAL_MASK;
619 
620 	return val;
621 }
622 
__rb_time_read(rb_time_t * t,u64 * ret,unsigned long * cnt)623 static inline bool __rb_time_read(rb_time_t *t, u64 *ret, unsigned long *cnt)
624 {
625 	unsigned long top, bottom, msb;
626 	unsigned long c;
627 
628 	/*
629 	 * If the read is interrupted by a write, then the cnt will
630 	 * be different. Loop until both top and bottom have been read
631 	 * without interruption.
632 	 */
633 	do {
634 		c = local_read(&t->cnt);
635 		top = local_read(&t->top);
636 		bottom = local_read(&t->bottom);
637 		msb = local_read(&t->msb);
638 	} while (c != local_read(&t->cnt));
639 
640 	*cnt = rb_time_cnt(top);
641 
642 	/* If top and bottom counts don't match, this interrupted a write */
643 	if (*cnt != rb_time_cnt(bottom))
644 		return false;
645 
646 	/* The shift to msb will lose its cnt bits */
647 	*ret = rb_time_val(top, bottom) | ((u64)msb << RB_TIME_MSB_SHIFT);
648 	return true;
649 }
650 
rb_time_read(rb_time_t * t,u64 * ret)651 static bool rb_time_read(rb_time_t *t, u64 *ret)
652 {
653 	unsigned long cnt;
654 
655 	return __rb_time_read(t, ret, &cnt);
656 }
657 
rb_time_val_cnt(unsigned long val,unsigned long cnt)658 static inline unsigned long rb_time_val_cnt(unsigned long val, unsigned long cnt)
659 {
660 	return (val & RB_TIME_VAL_MASK) | ((cnt & 3) << RB_TIME_SHIFT);
661 }
662 
rb_time_split(u64 val,unsigned long * top,unsigned long * bottom,unsigned long * msb)663 static inline void rb_time_split(u64 val, unsigned long *top, unsigned long *bottom,
664 				 unsigned long *msb)
665 {
666 	*top = (unsigned long)((val >> RB_TIME_SHIFT) & RB_TIME_VAL_MASK);
667 	*bottom = (unsigned long)(val & RB_TIME_VAL_MASK);
668 	*msb = (unsigned long)(val >> RB_TIME_MSB_SHIFT);
669 }
670 
rb_time_val_set(local_t * t,unsigned long val,unsigned long cnt)671 static inline void rb_time_val_set(local_t *t, unsigned long val, unsigned long cnt)
672 {
673 	val = rb_time_val_cnt(val, cnt);
674 	local_set(t, val);
675 }
676 
rb_time_set(rb_time_t * t,u64 val)677 static void rb_time_set(rb_time_t *t, u64 val)
678 {
679 	unsigned long cnt, top, bottom, msb;
680 
681 	rb_time_split(val, &top, &bottom, &msb);
682 
683 	/* Writes always succeed with a valid number even if it gets interrupted. */
684 	do {
685 		cnt = local_inc_return(&t->cnt);
686 		rb_time_val_set(&t->top, top, cnt);
687 		rb_time_val_set(&t->bottom, bottom, cnt);
688 		rb_time_val_set(&t->msb, val >> RB_TIME_MSB_SHIFT, cnt);
689 	} while (cnt != local_read(&t->cnt));
690 }
691 
692 static inline bool
rb_time_read_cmpxchg(local_t * l,unsigned long expect,unsigned long set)693 rb_time_read_cmpxchg(local_t *l, unsigned long expect, unsigned long set)
694 {
695 	unsigned long ret;
696 
697 	ret = local_cmpxchg(l, expect, set);
698 	return ret == expect;
699 }
700 
rb_time_cmpxchg(rb_time_t * t,u64 expect,u64 set)701 static int rb_time_cmpxchg(rb_time_t *t, u64 expect, u64 set)
702 {
703 	unsigned long cnt, top, bottom, msb;
704 	unsigned long cnt2, top2, bottom2, msb2;
705 	u64 val;
706 
707 	/* The cmpxchg always fails if it interrupted an update */
708 	 if (!__rb_time_read(t, &val, &cnt2))
709 		 return false;
710 
711 	 if (val != expect)
712 		 return false;
713 
714 	 cnt = local_read(&t->cnt);
715 	 if ((cnt & 3) != cnt2)
716 		 return false;
717 
718 	 cnt2 = cnt + 1;
719 
720 	 rb_time_split(val, &top, &bottom, &msb);
721 	 top = rb_time_val_cnt(top, cnt);
722 	 bottom = rb_time_val_cnt(bottom, cnt);
723 
724 	 rb_time_split(set, &top2, &bottom2, &msb2);
725 	 top2 = rb_time_val_cnt(top2, cnt2);
726 	 bottom2 = rb_time_val_cnt(bottom2, cnt2);
727 
728 	if (!rb_time_read_cmpxchg(&t->cnt, cnt, cnt2))
729 		return false;
730 	if (!rb_time_read_cmpxchg(&t->msb, msb, msb2))
731 		return false;
732 	if (!rb_time_read_cmpxchg(&t->top, top, top2))
733 		return false;
734 	if (!rb_time_read_cmpxchg(&t->bottom, bottom, bottom2))
735 		return false;
736 	return true;
737 }
738 
739 #else /* 64 bits */
740 
741 /* local64_t always succeeds */
742 
rb_time_read(rb_time_t * t,u64 * ret)743 static inline bool rb_time_read(rb_time_t *t, u64 *ret)
744 {
745 	*ret = local64_read(&t->time);
746 	return true;
747 }
rb_time_set(rb_time_t * t,u64 val)748 static void rb_time_set(rb_time_t *t, u64 val)
749 {
750 	local64_set(&t->time, val);
751 }
752 
rb_time_cmpxchg(rb_time_t * t,u64 expect,u64 set)753 static bool rb_time_cmpxchg(rb_time_t *t, u64 expect, u64 set)
754 {
755 	u64 val;
756 	val = local64_cmpxchg(&t->time, expect, set);
757 	return val == expect;
758 }
759 #endif
760 
761 /*
762  * Enable this to make sure that the event passed to
763  * ring_buffer_event_time_stamp() is not committed and also
764  * is on the buffer that it passed in.
765  */
766 //#define RB_VERIFY_EVENT
767 #ifdef RB_VERIFY_EVENT
768 static struct list_head *rb_list_head(struct list_head *list);
verify_event(struct ring_buffer_per_cpu * cpu_buffer,void * event)769 static void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
770 			 void *event)
771 {
772 	struct buffer_page *page = cpu_buffer->commit_page;
773 	struct buffer_page *tail_page = READ_ONCE(cpu_buffer->tail_page);
774 	struct list_head *next;
775 	long commit, write;
776 	unsigned long addr = (unsigned long)event;
777 	bool done = false;
778 	int stop = 0;
779 
780 	/* Make sure the event exists and is not committed yet */
781 	do {
782 		if (page == tail_page || WARN_ON_ONCE(stop++ > 100))
783 			done = true;
784 		commit = local_read(&page->page->commit);
785 		write = local_read(&page->write);
786 		if (addr >= (unsigned long)&page->page->data[commit] &&
787 		    addr < (unsigned long)&page->page->data[write])
788 			return;
789 
790 		next = rb_list_head(page->list.next);
791 		page = list_entry(next, struct buffer_page, list);
792 	} while (!done);
793 	WARN_ON_ONCE(1);
794 }
795 #else
verify_event(struct ring_buffer_per_cpu * cpu_buffer,void * event)796 static inline void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
797 			 void *event)
798 {
799 }
800 #endif
801 
802 /*
803  * The absolute time stamp drops the 5 MSBs and some clocks may
804  * require them. The rb_fix_abs_ts() will take a previous full
805  * time stamp, and add the 5 MSB of that time stamp on to the
806  * saved absolute time stamp. Then they are compared in case of
807  * the unlikely event that the latest time stamp incremented
808  * the 5 MSB.
809  */
rb_fix_abs_ts(u64 abs,u64 save_ts)810 static inline u64 rb_fix_abs_ts(u64 abs, u64 save_ts)
811 {
812 	if (save_ts & TS_MSB) {
813 		abs |= save_ts & TS_MSB;
814 		/* Check for overflow */
815 		if (unlikely(abs < save_ts))
816 			abs += 1ULL << 59;
817 	}
818 	return abs;
819 }
820 
821 static inline u64 rb_time_stamp(struct trace_buffer *buffer);
822 
823 /**
824  * ring_buffer_event_time_stamp - return the event's current time stamp
825  * @buffer: The buffer that the event is on
826  * @event: the event to get the time stamp of
827  *
828  * Note, this must be called after @event is reserved, and before it is
829  * committed to the ring buffer. And must be called from the same
830  * context where the event was reserved (normal, softirq, irq, etc).
831  *
832  * Returns the time stamp associated with the current event.
833  * If the event has an extended time stamp, then that is used as
834  * the time stamp to return.
835  * In the highly unlikely case that the event was nested more than
836  * the max nesting, then the write_stamp of the buffer is returned,
837  * otherwise  current time is returned, but that really neither of
838  * the last two cases should ever happen.
839  */
ring_buffer_event_time_stamp(struct trace_buffer * buffer,struct ring_buffer_event * event)840 u64 ring_buffer_event_time_stamp(struct trace_buffer *buffer,
841 				 struct ring_buffer_event *event)
842 {
843 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[smp_processor_id()];
844 	unsigned int nest;
845 	u64 ts;
846 
847 	/* If the event includes an absolute time, then just use that */
848 	if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
849 		ts = rb_event_time_stamp(event);
850 		return rb_fix_abs_ts(ts, cpu_buffer->tail_page->page->time_stamp);
851 	}
852 
853 	nest = local_read(&cpu_buffer->committing);
854 	verify_event(cpu_buffer, event);
855 	if (WARN_ON_ONCE(!nest))
856 		goto fail;
857 
858 	/* Read the current saved nesting level time stamp */
859 	if (likely(--nest < MAX_NEST))
860 		return cpu_buffer->event_stamp[nest];
861 
862 	/* Shouldn't happen, warn if it does */
863 	WARN_ONCE(1, "nest (%d) greater than max", nest);
864 
865  fail:
866 	/* Can only fail on 32 bit */
867 	if (!rb_time_read(&cpu_buffer->write_stamp, &ts))
868 		/* Screw it, just read the current time */
869 		ts = rb_time_stamp(cpu_buffer->buffer);
870 
871 	return ts;
872 }
873 
874 /**
875  * ring_buffer_nr_pages - get the number of buffer pages in the ring buffer
876  * @buffer: The ring_buffer to get the number of pages from
877  * @cpu: The cpu of the ring_buffer to get the number of pages from
878  *
879  * Returns the number of pages used by a per_cpu buffer of the ring buffer.
880  */
ring_buffer_nr_pages(struct trace_buffer * buffer,int cpu)881 size_t ring_buffer_nr_pages(struct trace_buffer *buffer, int cpu)
882 {
883 	return buffer->buffers[cpu]->nr_pages;
884 }
885 
886 /**
887  * ring_buffer_nr_pages_dirty - get the number of used pages in the ring buffer
888  * @buffer: The ring_buffer to get the number of pages from
889  * @cpu: The cpu of the ring_buffer to get the number of pages from
890  *
891  * Returns the number of pages that have content in the ring buffer.
892  */
ring_buffer_nr_dirty_pages(struct trace_buffer * buffer,int cpu)893 size_t ring_buffer_nr_dirty_pages(struct trace_buffer *buffer, int cpu)
894 {
895 	size_t read;
896 	size_t cnt;
897 
898 	read = local_read(&buffer->buffers[cpu]->pages_read);
899 	cnt = local_read(&buffer->buffers[cpu]->pages_touched);
900 	/* The reader can read an empty page, but not more than that */
901 	if (cnt < read) {
902 		WARN_ON_ONCE(read > cnt + 1);
903 		return 0;
904 	}
905 
906 	return cnt - read;
907 }
908 
909 /*
910  * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
911  *
912  * Schedules a delayed work to wake up any task that is blocked on the
913  * ring buffer waiters queue.
914  */
rb_wake_up_waiters(struct irq_work * work)915 static void rb_wake_up_waiters(struct irq_work *work)
916 {
917 	struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
918 
919 	wake_up_all(&rbwork->waiters);
920 	if (rbwork->wakeup_full) {
921 		rbwork->wakeup_full = false;
922 		wake_up_all(&rbwork->full_waiters);
923 	}
924 }
925 
926 /**
927  * ring_buffer_wait - wait for input to the ring buffer
928  * @buffer: buffer to wait on
929  * @cpu: the cpu buffer to wait on
930  * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
931  *
932  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
933  * as data is added to any of the @buffer's cpu buffers. Otherwise
934  * it will wait for data to be added to a specific cpu buffer.
935  */
ring_buffer_wait(struct trace_buffer * buffer,int cpu,int full)936 int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full)
937 {
938 	struct ring_buffer_per_cpu *cpu_buffer;
939 	DEFINE_WAIT(wait);
940 	struct rb_irq_work *work;
941 	int ret = 0;
942 
943 	/*
944 	 * Depending on what the caller is waiting for, either any
945 	 * data in any cpu buffer, or a specific buffer, put the
946 	 * caller on the appropriate wait queue.
947 	 */
948 	if (cpu == RING_BUFFER_ALL_CPUS) {
949 		work = &buffer->irq_work;
950 		/* Full only makes sense on per cpu reads */
951 		full = 0;
952 	} else {
953 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
954 			return -ENODEV;
955 		cpu_buffer = buffer->buffers[cpu];
956 		work = &cpu_buffer->irq_work;
957 	}
958 
959 
960 	while (true) {
961 		if (full)
962 			prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE);
963 		else
964 			prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE);
965 
966 		/*
967 		 * The events can happen in critical sections where
968 		 * checking a work queue can cause deadlocks.
969 		 * After adding a task to the queue, this flag is set
970 		 * only to notify events to try to wake up the queue
971 		 * using irq_work.
972 		 *
973 		 * We don't clear it even if the buffer is no longer
974 		 * empty. The flag only causes the next event to run
975 		 * irq_work to do the work queue wake up. The worse
976 		 * that can happen if we race with !trace_empty() is that
977 		 * an event will cause an irq_work to try to wake up
978 		 * an empty queue.
979 		 *
980 		 * There's no reason to protect this flag either, as
981 		 * the work queue and irq_work logic will do the necessary
982 		 * synchronization for the wake ups. The only thing
983 		 * that is necessary is that the wake up happens after
984 		 * a task has been queued. It's OK for spurious wake ups.
985 		 */
986 		if (full)
987 			work->full_waiters_pending = true;
988 		else
989 			work->waiters_pending = true;
990 
991 		if (signal_pending(current)) {
992 			ret = -EINTR;
993 			break;
994 		}
995 
996 		if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer))
997 			break;
998 
999 		if (cpu != RING_BUFFER_ALL_CPUS &&
1000 		    !ring_buffer_empty_cpu(buffer, cpu)) {
1001 			unsigned long flags;
1002 			bool pagebusy;
1003 			size_t nr_pages;
1004 			size_t dirty;
1005 
1006 			if (!full)
1007 				break;
1008 
1009 			raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1010 			pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
1011 			nr_pages = cpu_buffer->nr_pages;
1012 			dirty = ring_buffer_nr_dirty_pages(buffer, cpu);
1013 			if (!cpu_buffer->shortest_full ||
1014 			    cpu_buffer->shortest_full < full)
1015 				cpu_buffer->shortest_full = full;
1016 			raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1017 			if (!pagebusy &&
1018 			    (!nr_pages || (dirty * 100) > full * nr_pages))
1019 				break;
1020 		}
1021 
1022 		schedule();
1023 	}
1024 
1025 	if (full)
1026 		finish_wait(&work->full_waiters, &wait);
1027 	else
1028 		finish_wait(&work->waiters, &wait);
1029 
1030 	return ret;
1031 }
1032 
1033 /**
1034  * ring_buffer_poll_wait - poll on buffer input
1035  * @buffer: buffer to wait on
1036  * @cpu: the cpu buffer to wait on
1037  * @filp: the file descriptor
1038  * @poll_table: The poll descriptor
1039  *
1040  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
1041  * as data is added to any of the @buffer's cpu buffers. Otherwise
1042  * it will wait for data to be added to a specific cpu buffer.
1043  *
1044  * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
1045  * zero otherwise.
1046  */
ring_buffer_poll_wait(struct trace_buffer * buffer,int cpu,struct file * filp,poll_table * poll_table)1047 __poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu,
1048 			  struct file *filp, poll_table *poll_table)
1049 {
1050 	struct ring_buffer_per_cpu *cpu_buffer;
1051 	struct rb_irq_work *work;
1052 
1053 	if (cpu == RING_BUFFER_ALL_CPUS)
1054 		work = &buffer->irq_work;
1055 	else {
1056 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
1057 			return -EINVAL;
1058 
1059 		cpu_buffer = buffer->buffers[cpu];
1060 		work = &cpu_buffer->irq_work;
1061 	}
1062 
1063 	poll_wait(filp, &work->waiters, poll_table);
1064 	work->waiters_pending = true;
1065 	/*
1066 	 * There's a tight race between setting the waiters_pending and
1067 	 * checking if the ring buffer is empty.  Once the waiters_pending bit
1068 	 * is set, the next event will wake the task up, but we can get stuck
1069 	 * if there's only a single event in.
1070 	 *
1071 	 * FIXME: Ideally, we need a memory barrier on the writer side as well,
1072 	 * but adding a memory barrier to all events will cause too much of a
1073 	 * performance hit in the fast path.  We only need a memory barrier when
1074 	 * the buffer goes from empty to having content.  But as this race is
1075 	 * extremely small, and it's not a problem if another event comes in, we
1076 	 * will fix it later.
1077 	 */
1078 	smp_mb();
1079 
1080 	if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
1081 	    (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
1082 		return EPOLLIN | EPOLLRDNORM;
1083 	return 0;
1084 }
1085 
1086 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
1087 #define RB_WARN_ON(b, cond)						\
1088 	({								\
1089 		int _____ret = unlikely(cond);				\
1090 		if (_____ret) {						\
1091 			if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
1092 				struct ring_buffer_per_cpu *__b =	\
1093 					(void *)b;			\
1094 				atomic_inc(&__b->buffer->record_disabled); \
1095 			} else						\
1096 				atomic_inc(&b->record_disabled);	\
1097 			WARN_ON(1);					\
1098 		}							\
1099 		_____ret;						\
1100 	})
1101 
1102 /* Up this if you want to test the TIME_EXTENTS and normalization */
1103 #define DEBUG_SHIFT 0
1104 
rb_time_stamp(struct trace_buffer * buffer)1105 static inline u64 rb_time_stamp(struct trace_buffer *buffer)
1106 {
1107 	u64 ts;
1108 
1109 	/* Skip retpolines :-( */
1110 	if (IS_ENABLED(CONFIG_RETPOLINE) && likely(buffer->clock == trace_clock_local))
1111 		ts = trace_clock_local();
1112 	else
1113 		ts = buffer->clock();
1114 
1115 	/* shift to debug/test normalization and TIME_EXTENTS */
1116 	return ts << DEBUG_SHIFT;
1117 }
1118 
ring_buffer_time_stamp(struct trace_buffer * buffer)1119 u64 ring_buffer_time_stamp(struct trace_buffer *buffer)
1120 {
1121 	u64 time;
1122 
1123 	preempt_disable_notrace();
1124 	time = rb_time_stamp(buffer);
1125 	preempt_enable_notrace();
1126 
1127 	return time;
1128 }
1129 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
1130 
ring_buffer_normalize_time_stamp(struct trace_buffer * buffer,int cpu,u64 * ts)1131 void ring_buffer_normalize_time_stamp(struct trace_buffer *buffer,
1132 				      int cpu, u64 *ts)
1133 {
1134 	/* Just stupid testing the normalize function and deltas */
1135 	*ts >>= DEBUG_SHIFT;
1136 }
1137 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
1138 
1139 /*
1140  * Making the ring buffer lockless makes things tricky.
1141  * Although writes only happen on the CPU that they are on,
1142  * and they only need to worry about interrupts. Reads can
1143  * happen on any CPU.
1144  *
1145  * The reader page is always off the ring buffer, but when the
1146  * reader finishes with a page, it needs to swap its page with
1147  * a new one from the buffer. The reader needs to take from
1148  * the head (writes go to the tail). But if a writer is in overwrite
1149  * mode and wraps, it must push the head page forward.
1150  *
1151  * Here lies the problem.
1152  *
1153  * The reader must be careful to replace only the head page, and
1154  * not another one. As described at the top of the file in the
1155  * ASCII art, the reader sets its old page to point to the next
1156  * page after head. It then sets the page after head to point to
1157  * the old reader page. But if the writer moves the head page
1158  * during this operation, the reader could end up with the tail.
1159  *
1160  * We use cmpxchg to help prevent this race. We also do something
1161  * special with the page before head. We set the LSB to 1.
1162  *
1163  * When the writer must push the page forward, it will clear the
1164  * bit that points to the head page, move the head, and then set
1165  * the bit that points to the new head page.
1166  *
1167  * We also don't want an interrupt coming in and moving the head
1168  * page on another writer. Thus we use the second LSB to catch
1169  * that too. Thus:
1170  *
1171  * head->list->prev->next        bit 1          bit 0
1172  *                              -------        -------
1173  * Normal page                     0              0
1174  * Points to head page             0              1
1175  * New head page                   1              0
1176  *
1177  * Note we can not trust the prev pointer of the head page, because:
1178  *
1179  * +----+       +-----+        +-----+
1180  * |    |------>|  T  |---X--->|  N  |
1181  * |    |<------|     |        |     |
1182  * +----+       +-----+        +-----+
1183  *   ^                           ^ |
1184  *   |          +-----+          | |
1185  *   +----------|  R  |----------+ |
1186  *              |     |<-----------+
1187  *              +-----+
1188  *
1189  * Key:  ---X-->  HEAD flag set in pointer
1190  *         T      Tail page
1191  *         R      Reader page
1192  *         N      Next page
1193  *
1194  * (see __rb_reserve_next() to see where this happens)
1195  *
1196  *  What the above shows is that the reader just swapped out
1197  *  the reader page with a page in the buffer, but before it
1198  *  could make the new header point back to the new page added
1199  *  it was preempted by a writer. The writer moved forward onto
1200  *  the new page added by the reader and is about to move forward
1201  *  again.
1202  *
1203  *  You can see, it is legitimate for the previous pointer of
1204  *  the head (or any page) not to point back to itself. But only
1205  *  temporarily.
1206  */
1207 
1208 #define RB_PAGE_NORMAL		0UL
1209 #define RB_PAGE_HEAD		1UL
1210 #define RB_PAGE_UPDATE		2UL
1211 
1212 
1213 #define RB_FLAG_MASK		3UL
1214 
1215 /* PAGE_MOVED is not part of the mask */
1216 #define RB_PAGE_MOVED		4UL
1217 
1218 /*
1219  * rb_list_head - remove any bit
1220  */
rb_list_head(struct list_head * list)1221 static struct list_head *rb_list_head(struct list_head *list)
1222 {
1223 	unsigned long val = (unsigned long)list;
1224 
1225 	return (struct list_head *)(val & ~RB_FLAG_MASK);
1226 }
1227 
1228 /*
1229  * rb_is_head_page - test if the given page is the head page
1230  *
1231  * Because the reader may move the head_page pointer, we can
1232  * not trust what the head page is (it may be pointing to
1233  * the reader page). But if the next page is a header page,
1234  * its flags will be non zero.
1235  */
1236 static inline int
rb_is_head_page(struct buffer_page * page,struct list_head * list)1237 rb_is_head_page(struct buffer_page *page, struct list_head *list)
1238 {
1239 	unsigned long val;
1240 
1241 	val = (unsigned long)list->next;
1242 
1243 	if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
1244 		return RB_PAGE_MOVED;
1245 
1246 	return val & RB_FLAG_MASK;
1247 }
1248 
1249 /*
1250  * rb_is_reader_page
1251  *
1252  * The unique thing about the reader page, is that, if the
1253  * writer is ever on it, the previous pointer never points
1254  * back to the reader page.
1255  */
rb_is_reader_page(struct buffer_page * page)1256 static bool rb_is_reader_page(struct buffer_page *page)
1257 {
1258 	struct list_head *list = page->list.prev;
1259 
1260 	return rb_list_head(list->next) != &page->list;
1261 }
1262 
1263 /*
1264  * rb_set_list_to_head - set a list_head to be pointing to head.
1265  */
rb_set_list_to_head(struct list_head * list)1266 static void rb_set_list_to_head(struct list_head *list)
1267 {
1268 	unsigned long *ptr;
1269 
1270 	ptr = (unsigned long *)&list->next;
1271 	*ptr |= RB_PAGE_HEAD;
1272 	*ptr &= ~RB_PAGE_UPDATE;
1273 }
1274 
1275 /*
1276  * rb_head_page_activate - sets up head page
1277  */
rb_head_page_activate(struct ring_buffer_per_cpu * cpu_buffer)1278 static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
1279 {
1280 	struct buffer_page *head;
1281 
1282 	head = cpu_buffer->head_page;
1283 	if (!head)
1284 		return;
1285 
1286 	/*
1287 	 * Set the previous list pointer to have the HEAD flag.
1288 	 */
1289 	rb_set_list_to_head(head->list.prev);
1290 }
1291 
rb_list_head_clear(struct list_head * list)1292 static void rb_list_head_clear(struct list_head *list)
1293 {
1294 	unsigned long *ptr = (unsigned long *)&list->next;
1295 
1296 	*ptr &= ~RB_FLAG_MASK;
1297 }
1298 
1299 /*
1300  * rb_head_page_deactivate - clears head page ptr (for free list)
1301  */
1302 static void
rb_head_page_deactivate(struct ring_buffer_per_cpu * cpu_buffer)1303 rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
1304 {
1305 	struct list_head *hd;
1306 
1307 	/* Go through the whole list and clear any pointers found. */
1308 	rb_list_head_clear(cpu_buffer->pages);
1309 
1310 	list_for_each(hd, cpu_buffer->pages)
1311 		rb_list_head_clear(hd);
1312 }
1313 
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)1314 static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
1315 			    struct buffer_page *head,
1316 			    struct buffer_page *prev,
1317 			    int old_flag, int new_flag)
1318 {
1319 	struct list_head *list;
1320 	unsigned long val = (unsigned long)&head->list;
1321 	unsigned long ret;
1322 
1323 	list = &prev->list;
1324 
1325 	val &= ~RB_FLAG_MASK;
1326 
1327 	ret = cmpxchg((unsigned long *)&list->next,
1328 		      val | old_flag, val | new_flag);
1329 
1330 	/* check if the reader took the page */
1331 	if ((ret & ~RB_FLAG_MASK) != val)
1332 		return RB_PAGE_MOVED;
1333 
1334 	return ret & RB_FLAG_MASK;
1335 }
1336 
rb_head_page_set_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)1337 static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
1338 				   struct buffer_page *head,
1339 				   struct buffer_page *prev,
1340 				   int old_flag)
1341 {
1342 	return rb_head_page_set(cpu_buffer, head, prev,
1343 				old_flag, RB_PAGE_UPDATE);
1344 }
1345 
rb_head_page_set_head(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)1346 static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
1347 				 struct buffer_page *head,
1348 				 struct buffer_page *prev,
1349 				 int old_flag)
1350 {
1351 	return rb_head_page_set(cpu_buffer, head, prev,
1352 				old_flag, RB_PAGE_HEAD);
1353 }
1354 
rb_head_page_set_normal(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)1355 static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
1356 				   struct buffer_page *head,
1357 				   struct buffer_page *prev,
1358 				   int old_flag)
1359 {
1360 	return rb_head_page_set(cpu_buffer, head, prev,
1361 				old_flag, RB_PAGE_NORMAL);
1362 }
1363 
rb_inc_page(struct buffer_page ** bpage)1364 static inline void rb_inc_page(struct buffer_page **bpage)
1365 {
1366 	struct list_head *p = rb_list_head((*bpage)->list.next);
1367 
1368 	*bpage = list_entry(p, struct buffer_page, list);
1369 }
1370 
1371 static struct buffer_page *
rb_set_head_page(struct ring_buffer_per_cpu * cpu_buffer)1372 rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1373 {
1374 	struct buffer_page *head;
1375 	struct buffer_page *page;
1376 	struct list_head *list;
1377 	int i;
1378 
1379 	if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1380 		return NULL;
1381 
1382 	/* sanity check */
1383 	list = cpu_buffer->pages;
1384 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1385 		return NULL;
1386 
1387 	page = head = cpu_buffer->head_page;
1388 	/*
1389 	 * It is possible that the writer moves the header behind
1390 	 * where we started, and we miss in one loop.
1391 	 * A second loop should grab the header, but we'll do
1392 	 * three loops just because I'm paranoid.
1393 	 */
1394 	for (i = 0; i < 3; i++) {
1395 		do {
1396 			if (rb_is_head_page(page, page->list.prev)) {
1397 				cpu_buffer->head_page = page;
1398 				return page;
1399 			}
1400 			rb_inc_page(&page);
1401 		} while (page != head);
1402 	}
1403 
1404 	RB_WARN_ON(cpu_buffer, 1);
1405 
1406 	return NULL;
1407 }
1408 
rb_head_page_replace(struct buffer_page * old,struct buffer_page * new)1409 static int rb_head_page_replace(struct buffer_page *old,
1410 				struct buffer_page *new)
1411 {
1412 	unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1413 	unsigned long val;
1414 	unsigned long ret;
1415 
1416 	val = *ptr & ~RB_FLAG_MASK;
1417 	val |= RB_PAGE_HEAD;
1418 
1419 	ret = cmpxchg(ptr, val, (unsigned long)&new->list);
1420 
1421 	return ret == val;
1422 }
1423 
1424 /*
1425  * rb_tail_page_update - move the tail page forward
1426  */
rb_tail_page_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)1427 static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1428 			       struct buffer_page *tail_page,
1429 			       struct buffer_page *next_page)
1430 {
1431 	unsigned long old_entries;
1432 	unsigned long old_write;
1433 
1434 	/*
1435 	 * The tail page now needs to be moved forward.
1436 	 *
1437 	 * We need to reset the tail page, but without messing
1438 	 * with possible erasing of data brought in by interrupts
1439 	 * that have moved the tail page and are currently on it.
1440 	 *
1441 	 * We add a counter to the write field to denote this.
1442 	 */
1443 	old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
1444 	old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
1445 
1446 	local_inc(&cpu_buffer->pages_touched);
1447 	/*
1448 	 * Just make sure we have seen our old_write and synchronize
1449 	 * with any interrupts that come in.
1450 	 */
1451 	barrier();
1452 
1453 	/*
1454 	 * If the tail page is still the same as what we think
1455 	 * it is, then it is up to us to update the tail
1456 	 * pointer.
1457 	 */
1458 	if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1459 		/* Zero the write counter */
1460 		unsigned long val = old_write & ~RB_WRITE_MASK;
1461 		unsigned long eval = old_entries & ~RB_WRITE_MASK;
1462 
1463 		/*
1464 		 * This will only succeed if an interrupt did
1465 		 * not come in and change it. In which case, we
1466 		 * do not want to modify it.
1467 		 *
1468 		 * We add (void) to let the compiler know that we do not care
1469 		 * about the return value of these functions. We use the
1470 		 * cmpxchg to only update if an interrupt did not already
1471 		 * do it for us. If the cmpxchg fails, we don't care.
1472 		 */
1473 		(void)local_cmpxchg(&next_page->write, old_write, val);
1474 		(void)local_cmpxchg(&next_page->entries, old_entries, eval);
1475 
1476 		/*
1477 		 * No need to worry about races with clearing out the commit.
1478 		 * it only can increment when a commit takes place. But that
1479 		 * only happens in the outer most nested commit.
1480 		 */
1481 		local_set(&next_page->page->commit, 0);
1482 
1483 		/* Again, either we update tail_page or an interrupt does */
1484 		(void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page);
1485 	}
1486 }
1487 
rb_check_bpage(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * bpage)1488 static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1489 			  struct buffer_page *bpage)
1490 {
1491 	unsigned long val = (unsigned long)bpage;
1492 
1493 	if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK))
1494 		return 1;
1495 
1496 	return 0;
1497 }
1498 
1499 /**
1500  * rb_check_list - make sure a pointer to a list has the last bits zero
1501  */
rb_check_list(struct ring_buffer_per_cpu * cpu_buffer,struct list_head * list)1502 static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer,
1503 			 struct list_head *list)
1504 {
1505 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev))
1506 		return 1;
1507 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next))
1508 		return 1;
1509 	return 0;
1510 }
1511 
1512 /**
1513  * rb_check_pages - integrity check of buffer pages
1514  * @cpu_buffer: CPU buffer with pages to test
1515  *
1516  * As a safety measure we check to make sure the data pages have not
1517  * been corrupted.
1518  */
rb_check_pages(struct ring_buffer_per_cpu * cpu_buffer)1519 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1520 {
1521 	struct list_head *head = cpu_buffer->pages;
1522 	struct buffer_page *bpage, *tmp;
1523 
1524 	/* Reset the head page if it exists */
1525 	if (cpu_buffer->head_page)
1526 		rb_set_head_page(cpu_buffer);
1527 
1528 	rb_head_page_deactivate(cpu_buffer);
1529 
1530 	if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
1531 		return -1;
1532 	if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
1533 		return -1;
1534 
1535 	if (rb_check_list(cpu_buffer, head))
1536 		return -1;
1537 
1538 	list_for_each_entry_safe(bpage, tmp, head, list) {
1539 		if (RB_WARN_ON(cpu_buffer,
1540 			       bpage->list.next->prev != &bpage->list))
1541 			return -1;
1542 		if (RB_WARN_ON(cpu_buffer,
1543 			       bpage->list.prev->next != &bpage->list))
1544 			return -1;
1545 		if (rb_check_list(cpu_buffer, &bpage->list))
1546 			return -1;
1547 	}
1548 
1549 	rb_head_page_activate(cpu_buffer);
1550 
1551 	return 0;
1552 }
1553 
__rb_allocate_pages(struct ring_buffer_per_cpu * cpu_buffer,long nr_pages,struct list_head * pages)1554 static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1555 		long nr_pages, struct list_head *pages)
1556 {
1557 	struct buffer_page *bpage, *tmp;
1558 	bool user_thread = current->mm != NULL;
1559 	gfp_t mflags;
1560 	long i;
1561 
1562 	/*
1563 	 * Check if the available memory is there first.
1564 	 * Note, si_mem_available() only gives us a rough estimate of available
1565 	 * memory. It may not be accurate. But we don't care, we just want
1566 	 * to prevent doing any allocation when it is obvious that it is
1567 	 * not going to succeed.
1568 	 */
1569 	i = si_mem_available();
1570 	if (i < nr_pages)
1571 		return -ENOMEM;
1572 
1573 	/*
1574 	 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
1575 	 * gracefully without invoking oom-killer and the system is not
1576 	 * destabilized.
1577 	 */
1578 	mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL;
1579 
1580 	/*
1581 	 * If a user thread allocates too much, and si_mem_available()
1582 	 * reports there's enough memory, even though there is not.
1583 	 * Make sure the OOM killer kills this thread. This can happen
1584 	 * even with RETRY_MAYFAIL because another task may be doing
1585 	 * an allocation after this task has taken all memory.
1586 	 * This is the task the OOM killer needs to take out during this
1587 	 * loop, even if it was triggered by an allocation somewhere else.
1588 	 */
1589 	if (user_thread)
1590 		set_current_oom_origin();
1591 	for (i = 0; i < nr_pages; i++) {
1592 		struct page *page;
1593 
1594 		bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1595 				    mflags, cpu_to_node(cpu_buffer->cpu));
1596 		if (!bpage)
1597 			goto free_pages;
1598 
1599 		rb_check_bpage(cpu_buffer, bpage);
1600 
1601 		list_add(&bpage->list, pages);
1602 
1603 		page = alloc_pages_node(cpu_to_node(cpu_buffer->cpu), mflags, 0);
1604 		if (!page)
1605 			goto free_pages;
1606 		bpage->page = page_address(page);
1607 		rb_init_page(bpage->page);
1608 
1609 		if (user_thread && fatal_signal_pending(current))
1610 			goto free_pages;
1611 	}
1612 	if (user_thread)
1613 		clear_current_oom_origin();
1614 
1615 	return 0;
1616 
1617 free_pages:
1618 	list_for_each_entry_safe(bpage, tmp, pages, list) {
1619 		list_del_init(&bpage->list);
1620 		free_buffer_page(bpage);
1621 	}
1622 	if (user_thread)
1623 		clear_current_oom_origin();
1624 
1625 	return -ENOMEM;
1626 }
1627 
rb_allocate_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1628 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1629 			     unsigned long nr_pages)
1630 {
1631 	LIST_HEAD(pages);
1632 
1633 	WARN_ON(!nr_pages);
1634 
1635 	if (__rb_allocate_pages(cpu_buffer, nr_pages, &pages))
1636 		return -ENOMEM;
1637 
1638 	/*
1639 	 * The ring buffer page list is a circular list that does not
1640 	 * start and end with a list head. All page list items point to
1641 	 * other pages.
1642 	 */
1643 	cpu_buffer->pages = pages.next;
1644 	list_del(&pages);
1645 
1646 	cpu_buffer->nr_pages = nr_pages;
1647 
1648 	rb_check_pages(cpu_buffer);
1649 
1650 	return 0;
1651 }
1652 
1653 static struct ring_buffer_per_cpu *
rb_allocate_cpu_buffer(struct trace_buffer * buffer,long nr_pages,int cpu)1654 rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu)
1655 {
1656 	struct ring_buffer_per_cpu *cpu_buffer;
1657 	struct buffer_page *bpage;
1658 	struct page *page;
1659 	int ret;
1660 
1661 	cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1662 				  GFP_KERNEL, cpu_to_node(cpu));
1663 	if (!cpu_buffer)
1664 		return NULL;
1665 
1666 	cpu_buffer->cpu = cpu;
1667 	cpu_buffer->buffer = buffer;
1668 	raw_spin_lock_init(&cpu_buffer->reader_lock);
1669 	lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
1670 	cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1671 	INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
1672 	init_completion(&cpu_buffer->update_done);
1673 	init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
1674 	init_waitqueue_head(&cpu_buffer->irq_work.waiters);
1675 	init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
1676 
1677 	bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1678 			    GFP_KERNEL, cpu_to_node(cpu));
1679 	if (!bpage)
1680 		goto fail_free_buffer;
1681 
1682 	rb_check_bpage(cpu_buffer, bpage);
1683 
1684 	cpu_buffer->reader_page = bpage;
1685 	page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1686 	if (!page)
1687 		goto fail_free_reader;
1688 	bpage->page = page_address(page);
1689 	rb_init_page(bpage->page);
1690 
1691 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1692 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
1693 
1694 	ret = rb_allocate_pages(cpu_buffer, nr_pages);
1695 	if (ret < 0)
1696 		goto fail_free_reader;
1697 
1698 	cpu_buffer->head_page
1699 		= list_entry(cpu_buffer->pages, struct buffer_page, list);
1700 	cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
1701 
1702 	rb_head_page_activate(cpu_buffer);
1703 
1704 	return cpu_buffer;
1705 
1706  fail_free_reader:
1707 	free_buffer_page(cpu_buffer->reader_page);
1708 
1709  fail_free_buffer:
1710 	kfree(cpu_buffer);
1711 	return NULL;
1712 }
1713 
rb_free_cpu_buffer(struct ring_buffer_per_cpu * cpu_buffer)1714 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1715 {
1716 	struct list_head *head = cpu_buffer->pages;
1717 	struct buffer_page *bpage, *tmp;
1718 
1719 	free_buffer_page(cpu_buffer->reader_page);
1720 
1721 	rb_head_page_deactivate(cpu_buffer);
1722 
1723 	if (head) {
1724 		list_for_each_entry_safe(bpage, tmp, head, list) {
1725 			list_del_init(&bpage->list);
1726 			free_buffer_page(bpage);
1727 		}
1728 		bpage = list_entry(head, struct buffer_page, list);
1729 		free_buffer_page(bpage);
1730 	}
1731 
1732 	kfree(cpu_buffer);
1733 }
1734 
1735 /**
1736  * __ring_buffer_alloc - allocate a new ring_buffer
1737  * @size: the size in bytes per cpu that is needed.
1738  * @flags: attributes to set for the ring buffer.
1739  * @key: ring buffer reader_lock_key.
1740  *
1741  * Currently the only flag that is available is the RB_FL_OVERWRITE
1742  * flag. This flag means that the buffer will overwrite old data
1743  * when the buffer wraps. If this flag is not set, the buffer will
1744  * drop data when the tail hits the head.
1745  */
__ring_buffer_alloc(unsigned long size,unsigned flags,struct lock_class_key * key)1746 struct trace_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1747 					struct lock_class_key *key)
1748 {
1749 	struct trace_buffer *buffer;
1750 	long nr_pages;
1751 	int bsize;
1752 	int cpu;
1753 	int ret;
1754 
1755 	/* keep it in its own cache line */
1756 	buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1757 			 GFP_KERNEL);
1758 	if (!buffer)
1759 		return NULL;
1760 
1761 	if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1762 		goto fail_free_buffer;
1763 
1764 	nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1765 	buffer->flags = flags;
1766 	buffer->clock = trace_clock_local;
1767 	buffer->reader_lock_key = key;
1768 
1769 	init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters);
1770 	init_waitqueue_head(&buffer->irq_work.waiters);
1771 
1772 	/* need at least two pages */
1773 	if (nr_pages < 2)
1774 		nr_pages = 2;
1775 
1776 	buffer->cpus = nr_cpu_ids;
1777 
1778 	bsize = sizeof(void *) * nr_cpu_ids;
1779 	buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1780 				  GFP_KERNEL);
1781 	if (!buffer->buffers)
1782 		goto fail_free_cpumask;
1783 
1784 	cpu = raw_smp_processor_id();
1785 	cpumask_set_cpu(cpu, buffer->cpumask);
1786 	buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
1787 	if (!buffer->buffers[cpu])
1788 		goto fail_free_buffers;
1789 
1790 	ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1791 	if (ret < 0)
1792 		goto fail_free_buffers;
1793 
1794 	mutex_init(&buffer->mutex);
1795 
1796 	return buffer;
1797 
1798  fail_free_buffers:
1799 	for_each_buffer_cpu(buffer, cpu) {
1800 		if (buffer->buffers[cpu])
1801 			rb_free_cpu_buffer(buffer->buffers[cpu]);
1802 	}
1803 	kfree(buffer->buffers);
1804 
1805  fail_free_cpumask:
1806 	free_cpumask_var(buffer->cpumask);
1807 
1808  fail_free_buffer:
1809 	kfree(buffer);
1810 	return NULL;
1811 }
1812 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
1813 
1814 /**
1815  * ring_buffer_free - free a ring buffer.
1816  * @buffer: the buffer to free.
1817  */
1818 void
ring_buffer_free(struct trace_buffer * buffer)1819 ring_buffer_free(struct trace_buffer *buffer)
1820 {
1821 	int cpu;
1822 
1823 	cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1824 
1825 	for_each_buffer_cpu(buffer, cpu)
1826 		rb_free_cpu_buffer(buffer->buffers[cpu]);
1827 
1828 	kfree(buffer->buffers);
1829 	free_cpumask_var(buffer->cpumask);
1830 
1831 	kfree(buffer);
1832 }
1833 EXPORT_SYMBOL_GPL(ring_buffer_free);
1834 
ring_buffer_set_clock(struct trace_buffer * buffer,u64 (* clock)(void))1835 void ring_buffer_set_clock(struct trace_buffer *buffer,
1836 			   u64 (*clock)(void))
1837 {
1838 	buffer->clock = clock;
1839 }
1840 
ring_buffer_set_time_stamp_abs(struct trace_buffer * buffer,bool abs)1841 void ring_buffer_set_time_stamp_abs(struct trace_buffer *buffer, bool abs)
1842 {
1843 	buffer->time_stamp_abs = abs;
1844 }
1845 
ring_buffer_time_stamp_abs(struct trace_buffer * buffer)1846 bool ring_buffer_time_stamp_abs(struct trace_buffer *buffer)
1847 {
1848 	return buffer->time_stamp_abs;
1849 }
1850 
1851 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1852 
rb_page_entries(struct buffer_page * bpage)1853 static inline unsigned long rb_page_entries(struct buffer_page *bpage)
1854 {
1855 	return local_read(&bpage->entries) & RB_WRITE_MASK;
1856 }
1857 
rb_page_write(struct buffer_page * bpage)1858 static inline unsigned long rb_page_write(struct buffer_page *bpage)
1859 {
1860 	return local_read(&bpage->write) & RB_WRITE_MASK;
1861 }
1862 
1863 static int
rb_remove_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1864 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
1865 {
1866 	struct list_head *tail_page, *to_remove, *next_page;
1867 	struct buffer_page *to_remove_page, *tmp_iter_page;
1868 	struct buffer_page *last_page, *first_page;
1869 	unsigned long nr_removed;
1870 	unsigned long head_bit;
1871 	int page_entries;
1872 
1873 	head_bit = 0;
1874 
1875 	raw_spin_lock_irq(&cpu_buffer->reader_lock);
1876 	atomic_inc(&cpu_buffer->record_disabled);
1877 	/*
1878 	 * We don't race with the readers since we have acquired the reader
1879 	 * lock. We also don't race with writers after disabling recording.
1880 	 * This makes it easy to figure out the first and the last page to be
1881 	 * removed from the list. We unlink all the pages in between including
1882 	 * the first and last pages. This is done in a busy loop so that we
1883 	 * lose the least number of traces.
1884 	 * The pages are freed after we restart recording and unlock readers.
1885 	 */
1886 	tail_page = &cpu_buffer->tail_page->list;
1887 
1888 	/*
1889 	 * tail page might be on reader page, we remove the next page
1890 	 * from the ring buffer
1891 	 */
1892 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
1893 		tail_page = rb_list_head(tail_page->next);
1894 	to_remove = tail_page;
1895 
1896 	/* start of pages to remove */
1897 	first_page = list_entry(rb_list_head(to_remove->next),
1898 				struct buffer_page, list);
1899 
1900 	for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
1901 		to_remove = rb_list_head(to_remove)->next;
1902 		head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
1903 	}
1904 
1905 	next_page = rb_list_head(to_remove)->next;
1906 
1907 	/*
1908 	 * Now we remove all pages between tail_page and next_page.
1909 	 * Make sure that we have head_bit value preserved for the
1910 	 * next page
1911 	 */
1912 	tail_page->next = (struct list_head *)((unsigned long)next_page |
1913 						head_bit);
1914 	next_page = rb_list_head(next_page);
1915 	next_page->prev = tail_page;
1916 
1917 	/* make sure pages points to a valid page in the ring buffer */
1918 	cpu_buffer->pages = next_page;
1919 
1920 	/* update head page */
1921 	if (head_bit)
1922 		cpu_buffer->head_page = list_entry(next_page,
1923 						struct buffer_page, list);
1924 
1925 	/*
1926 	 * change read pointer to make sure any read iterators reset
1927 	 * themselves
1928 	 */
1929 	cpu_buffer->read = 0;
1930 
1931 	/* pages are removed, resume tracing and then free the pages */
1932 	atomic_dec(&cpu_buffer->record_disabled);
1933 	raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1934 
1935 	RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
1936 
1937 	/* last buffer page to remove */
1938 	last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
1939 				list);
1940 	tmp_iter_page = first_page;
1941 
1942 	do {
1943 		cond_resched();
1944 
1945 		to_remove_page = tmp_iter_page;
1946 		rb_inc_page(&tmp_iter_page);
1947 
1948 		/* update the counters */
1949 		page_entries = rb_page_entries(to_remove_page);
1950 		if (page_entries) {
1951 			/*
1952 			 * If something was added to this page, it was full
1953 			 * since it is not the tail page. So we deduct the
1954 			 * bytes consumed in ring buffer from here.
1955 			 * Increment overrun to account for the lost events.
1956 			 */
1957 			local_add(page_entries, &cpu_buffer->overrun);
1958 			local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
1959 		}
1960 
1961 		/*
1962 		 * We have already removed references to this list item, just
1963 		 * free up the buffer_page and its page
1964 		 */
1965 		free_buffer_page(to_remove_page);
1966 		nr_removed--;
1967 
1968 	} while (to_remove_page != last_page);
1969 
1970 	RB_WARN_ON(cpu_buffer, nr_removed);
1971 
1972 	return nr_removed == 0;
1973 }
1974 
1975 static int
rb_insert_pages(struct ring_buffer_per_cpu * cpu_buffer)1976 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
1977 {
1978 	struct list_head *pages = &cpu_buffer->new_pages;
1979 	int retries, success;
1980 
1981 	raw_spin_lock_irq(&cpu_buffer->reader_lock);
1982 	/*
1983 	 * We are holding the reader lock, so the reader page won't be swapped
1984 	 * in the ring buffer. Now we are racing with the writer trying to
1985 	 * move head page and the tail page.
1986 	 * We are going to adapt the reader page update process where:
1987 	 * 1. We first splice the start and end of list of new pages between
1988 	 *    the head page and its previous page.
1989 	 * 2. We cmpxchg the prev_page->next to point from head page to the
1990 	 *    start of new pages list.
1991 	 * 3. Finally, we update the head->prev to the end of new list.
1992 	 *
1993 	 * We will try this process 10 times, to make sure that we don't keep
1994 	 * spinning.
1995 	 */
1996 	retries = 10;
1997 	success = 0;
1998 	while (retries--) {
1999 		struct list_head *head_page, *prev_page, *r;
2000 		struct list_head *last_page, *first_page;
2001 		struct list_head *head_page_with_bit;
2002 
2003 		head_page = &rb_set_head_page(cpu_buffer)->list;
2004 		if (!head_page)
2005 			break;
2006 		prev_page = head_page->prev;
2007 
2008 		first_page = pages->next;
2009 		last_page  = pages->prev;
2010 
2011 		head_page_with_bit = (struct list_head *)
2012 				     ((unsigned long)head_page | RB_PAGE_HEAD);
2013 
2014 		last_page->next = head_page_with_bit;
2015 		first_page->prev = prev_page;
2016 
2017 		r = cmpxchg(&prev_page->next, head_page_with_bit, first_page);
2018 
2019 		if (r == head_page_with_bit) {
2020 			/*
2021 			 * yay, we replaced the page pointer to our new list,
2022 			 * now, we just have to update to head page's prev
2023 			 * pointer to point to end of list
2024 			 */
2025 			head_page->prev = last_page;
2026 			success = 1;
2027 			break;
2028 		}
2029 	}
2030 
2031 	if (success)
2032 		INIT_LIST_HEAD(pages);
2033 	/*
2034 	 * If we weren't successful in adding in new pages, warn and stop
2035 	 * tracing
2036 	 */
2037 	RB_WARN_ON(cpu_buffer, !success);
2038 	raw_spin_unlock_irq(&cpu_buffer->reader_lock);
2039 
2040 	/* free pages if they weren't inserted */
2041 	if (!success) {
2042 		struct buffer_page *bpage, *tmp;
2043 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2044 					 list) {
2045 			list_del_init(&bpage->list);
2046 			free_buffer_page(bpage);
2047 		}
2048 	}
2049 	return success;
2050 }
2051 
rb_update_pages(struct ring_buffer_per_cpu * cpu_buffer)2052 static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
2053 {
2054 	int success;
2055 
2056 	if (cpu_buffer->nr_pages_to_update > 0)
2057 		success = rb_insert_pages(cpu_buffer);
2058 	else
2059 		success = rb_remove_pages(cpu_buffer,
2060 					-cpu_buffer->nr_pages_to_update);
2061 
2062 	if (success)
2063 		cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
2064 }
2065 
update_pages_handler(struct work_struct * work)2066 static void update_pages_handler(struct work_struct *work)
2067 {
2068 	struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
2069 			struct ring_buffer_per_cpu, update_pages_work);
2070 	rb_update_pages(cpu_buffer);
2071 	complete(&cpu_buffer->update_done);
2072 }
2073 
2074 /**
2075  * ring_buffer_resize - resize the ring buffer
2076  * @buffer: the buffer to resize.
2077  * @size: the new size.
2078  * @cpu_id: the cpu buffer to resize
2079  *
2080  * Minimum size is 2 * BUF_PAGE_SIZE.
2081  *
2082  * Returns 0 on success and < 0 on failure.
2083  */
ring_buffer_resize(struct trace_buffer * buffer,unsigned long size,int cpu_id)2084 int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size,
2085 			int cpu_id)
2086 {
2087 	struct ring_buffer_per_cpu *cpu_buffer;
2088 	unsigned long nr_pages;
2089 	int cpu, err;
2090 
2091 	/*
2092 	 * Always succeed at resizing a non-existent buffer:
2093 	 */
2094 	if (!buffer)
2095 		return 0;
2096 
2097 	/* Make sure the requested buffer exists */
2098 	if (cpu_id != RING_BUFFER_ALL_CPUS &&
2099 	    !cpumask_test_cpu(cpu_id, buffer->cpumask))
2100 		return 0;
2101 
2102 	nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
2103 
2104 	/* we need a minimum of two pages */
2105 	if (nr_pages < 2)
2106 		nr_pages = 2;
2107 
2108 	/* prevent another thread from changing buffer sizes */
2109 	mutex_lock(&buffer->mutex);
2110 
2111 
2112 	if (cpu_id == RING_BUFFER_ALL_CPUS) {
2113 		/*
2114 		 * Don't succeed if resizing is disabled, as a reader might be
2115 		 * manipulating the ring buffer and is expecting a sane state while
2116 		 * this is true.
2117 		 */
2118 		for_each_buffer_cpu(buffer, cpu) {
2119 			cpu_buffer = buffer->buffers[cpu];
2120 			if (atomic_read(&cpu_buffer->resize_disabled)) {
2121 				err = -EBUSY;
2122 				goto out_err_unlock;
2123 			}
2124 		}
2125 
2126 		/* calculate the pages to update */
2127 		for_each_buffer_cpu(buffer, cpu) {
2128 			cpu_buffer = buffer->buffers[cpu];
2129 
2130 			cpu_buffer->nr_pages_to_update = nr_pages -
2131 							cpu_buffer->nr_pages;
2132 			/*
2133 			 * nothing more to do for removing pages or no update
2134 			 */
2135 			if (cpu_buffer->nr_pages_to_update <= 0)
2136 				continue;
2137 			/*
2138 			 * to add pages, make sure all new pages can be
2139 			 * allocated without receiving ENOMEM
2140 			 */
2141 			INIT_LIST_HEAD(&cpu_buffer->new_pages);
2142 			if (__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
2143 						&cpu_buffer->new_pages)) {
2144 				/* not enough memory for new pages */
2145 				err = -ENOMEM;
2146 				goto out_err;
2147 			}
2148 		}
2149 
2150 		cpus_read_lock();
2151 		/*
2152 		 * Fire off all the required work handlers
2153 		 * We can't schedule on offline CPUs, but it's not necessary
2154 		 * since we can change their buffer sizes without any race.
2155 		 */
2156 		for_each_buffer_cpu(buffer, cpu) {
2157 			cpu_buffer = buffer->buffers[cpu];
2158 			if (!cpu_buffer->nr_pages_to_update)
2159 				continue;
2160 
2161 			/* Can't run something on an offline CPU. */
2162 			if (!cpu_online(cpu)) {
2163 				rb_update_pages(cpu_buffer);
2164 				cpu_buffer->nr_pages_to_update = 0;
2165 			} else {
2166 				schedule_work_on(cpu,
2167 						&cpu_buffer->update_pages_work);
2168 			}
2169 		}
2170 
2171 		/* wait for all the updates to complete */
2172 		for_each_buffer_cpu(buffer, cpu) {
2173 			cpu_buffer = buffer->buffers[cpu];
2174 			if (!cpu_buffer->nr_pages_to_update)
2175 				continue;
2176 
2177 			if (cpu_online(cpu))
2178 				wait_for_completion(&cpu_buffer->update_done);
2179 			cpu_buffer->nr_pages_to_update = 0;
2180 		}
2181 
2182 		cpus_read_unlock();
2183 	} else {
2184 		cpu_buffer = buffer->buffers[cpu_id];
2185 
2186 		if (nr_pages == cpu_buffer->nr_pages)
2187 			goto out;
2188 
2189 		/*
2190 		 * Don't succeed if resizing is disabled, as a reader might be
2191 		 * manipulating the ring buffer and is expecting a sane state while
2192 		 * this is true.
2193 		 */
2194 		if (atomic_read(&cpu_buffer->resize_disabled)) {
2195 			err = -EBUSY;
2196 			goto out_err_unlock;
2197 		}
2198 
2199 		cpu_buffer->nr_pages_to_update = nr_pages -
2200 						cpu_buffer->nr_pages;
2201 
2202 		INIT_LIST_HEAD(&cpu_buffer->new_pages);
2203 		if (cpu_buffer->nr_pages_to_update > 0 &&
2204 			__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
2205 					    &cpu_buffer->new_pages)) {
2206 			err = -ENOMEM;
2207 			goto out_err;
2208 		}
2209 
2210 		cpus_read_lock();
2211 
2212 		/* Can't run something on an offline CPU. */
2213 		if (!cpu_online(cpu_id))
2214 			rb_update_pages(cpu_buffer);
2215 		else {
2216 			schedule_work_on(cpu_id,
2217 					 &cpu_buffer->update_pages_work);
2218 			wait_for_completion(&cpu_buffer->update_done);
2219 		}
2220 
2221 		cpu_buffer->nr_pages_to_update = 0;
2222 		cpus_read_unlock();
2223 	}
2224 
2225  out:
2226 	/*
2227 	 * The ring buffer resize can happen with the ring buffer
2228 	 * enabled, so that the update disturbs the tracing as little
2229 	 * as possible. But if the buffer is disabled, we do not need
2230 	 * to worry about that, and we can take the time to verify
2231 	 * that the buffer is not corrupt.
2232 	 */
2233 	if (atomic_read(&buffer->record_disabled)) {
2234 		atomic_inc(&buffer->record_disabled);
2235 		/*
2236 		 * Even though the buffer was disabled, we must make sure
2237 		 * that it is truly disabled before calling rb_check_pages.
2238 		 * There could have been a race between checking
2239 		 * record_disable and incrementing it.
2240 		 */
2241 		synchronize_rcu();
2242 		for_each_buffer_cpu(buffer, cpu) {
2243 			cpu_buffer = buffer->buffers[cpu];
2244 			rb_check_pages(cpu_buffer);
2245 		}
2246 		atomic_dec(&buffer->record_disabled);
2247 	}
2248 
2249 	mutex_unlock(&buffer->mutex);
2250 	return 0;
2251 
2252  out_err:
2253 	for_each_buffer_cpu(buffer, cpu) {
2254 		struct buffer_page *bpage, *tmp;
2255 
2256 		cpu_buffer = buffer->buffers[cpu];
2257 		cpu_buffer->nr_pages_to_update = 0;
2258 
2259 		if (list_empty(&cpu_buffer->new_pages))
2260 			continue;
2261 
2262 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2263 					list) {
2264 			list_del_init(&bpage->list);
2265 			free_buffer_page(bpage);
2266 		}
2267 	}
2268  out_err_unlock:
2269 	mutex_unlock(&buffer->mutex);
2270 	return err;
2271 }
2272 EXPORT_SYMBOL_GPL(ring_buffer_resize);
2273 
ring_buffer_change_overwrite(struct trace_buffer * buffer,int val)2274 void ring_buffer_change_overwrite(struct trace_buffer *buffer, int val)
2275 {
2276 	mutex_lock(&buffer->mutex);
2277 	if (val)
2278 		buffer->flags |= RB_FL_OVERWRITE;
2279 	else
2280 		buffer->flags &= ~RB_FL_OVERWRITE;
2281 	mutex_unlock(&buffer->mutex);
2282 }
2283 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
2284 
__rb_page_index(struct buffer_page * bpage,unsigned index)2285 static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
2286 {
2287 	return bpage->page->data + index;
2288 }
2289 
2290 static __always_inline struct ring_buffer_event *
rb_reader_event(struct ring_buffer_per_cpu * cpu_buffer)2291 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
2292 {
2293 	return __rb_page_index(cpu_buffer->reader_page,
2294 			       cpu_buffer->reader_page->read);
2295 }
2296 
rb_page_commit(struct buffer_page * bpage)2297 static __always_inline unsigned rb_page_commit(struct buffer_page *bpage)
2298 {
2299 	return local_read(&bpage->page->commit);
2300 }
2301 
2302 static struct ring_buffer_event *
rb_iter_head_event(struct ring_buffer_iter * iter)2303 rb_iter_head_event(struct ring_buffer_iter *iter)
2304 {
2305 	struct ring_buffer_event *event;
2306 	struct buffer_page *iter_head_page = iter->head_page;
2307 	unsigned long commit;
2308 	unsigned length;
2309 
2310 	if (iter->head != iter->next_event)
2311 		return iter->event;
2312 
2313 	/*
2314 	 * When the writer goes across pages, it issues a cmpxchg which
2315 	 * is a mb(), which will synchronize with the rmb here.
2316 	 * (see rb_tail_page_update() and __rb_reserve_next())
2317 	 */
2318 	commit = rb_page_commit(iter_head_page);
2319 	smp_rmb();
2320 	event = __rb_page_index(iter_head_page, iter->head);
2321 	length = rb_event_length(event);
2322 
2323 	/*
2324 	 * READ_ONCE() doesn't work on functions and we don't want the
2325 	 * compiler doing any crazy optimizations with length.
2326 	 */
2327 	barrier();
2328 
2329 	if ((iter->head + length) > commit || length > BUF_MAX_DATA_SIZE)
2330 		/* Writer corrupted the read? */
2331 		goto reset;
2332 
2333 	memcpy(iter->event, event, length);
2334 	/*
2335 	 * If the page stamp is still the same after this rmb() then the
2336 	 * event was safely copied without the writer entering the page.
2337 	 */
2338 	smp_rmb();
2339 
2340 	/* Make sure the page didn't change since we read this */
2341 	if (iter->page_stamp != iter_head_page->page->time_stamp ||
2342 	    commit > rb_page_commit(iter_head_page))
2343 		goto reset;
2344 
2345 	iter->next_event = iter->head + length;
2346 	return iter->event;
2347  reset:
2348 	/* Reset to the beginning */
2349 	iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
2350 	iter->head = 0;
2351 	iter->next_event = 0;
2352 	iter->missed_events = 1;
2353 	return NULL;
2354 }
2355 
2356 /* Size is determined by what has been committed */
rb_page_size(struct buffer_page * bpage)2357 static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
2358 {
2359 	return rb_page_commit(bpage);
2360 }
2361 
2362 static __always_inline unsigned
rb_commit_index(struct ring_buffer_per_cpu * cpu_buffer)2363 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
2364 {
2365 	return rb_page_commit(cpu_buffer->commit_page);
2366 }
2367 
2368 static __always_inline unsigned
rb_event_index(struct ring_buffer_event * event)2369 rb_event_index(struct ring_buffer_event *event)
2370 {
2371 	unsigned long addr = (unsigned long)event;
2372 
2373 	return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
2374 }
2375 
rb_inc_iter(struct ring_buffer_iter * iter)2376 static void rb_inc_iter(struct ring_buffer_iter *iter)
2377 {
2378 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2379 
2380 	/*
2381 	 * The iterator could be on the reader page (it starts there).
2382 	 * But the head could have moved, since the reader was
2383 	 * found. Check for this case and assign the iterator
2384 	 * to the head page instead of next.
2385 	 */
2386 	if (iter->head_page == cpu_buffer->reader_page)
2387 		iter->head_page = rb_set_head_page(cpu_buffer);
2388 	else
2389 		rb_inc_page(&iter->head_page);
2390 
2391 	iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
2392 	iter->head = 0;
2393 	iter->next_event = 0;
2394 }
2395 
2396 /*
2397  * rb_handle_head_page - writer hit the head page
2398  *
2399  * Returns: +1 to retry page
2400  *           0 to continue
2401  *          -1 on error
2402  */
2403 static int
rb_handle_head_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)2404 rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
2405 		    struct buffer_page *tail_page,
2406 		    struct buffer_page *next_page)
2407 {
2408 	struct buffer_page *new_head;
2409 	int entries;
2410 	int type;
2411 	int ret;
2412 
2413 	entries = rb_page_entries(next_page);
2414 
2415 	/*
2416 	 * The hard part is here. We need to move the head
2417 	 * forward, and protect against both readers on
2418 	 * other CPUs and writers coming in via interrupts.
2419 	 */
2420 	type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
2421 				       RB_PAGE_HEAD);
2422 
2423 	/*
2424 	 * type can be one of four:
2425 	 *  NORMAL - an interrupt already moved it for us
2426 	 *  HEAD   - we are the first to get here.
2427 	 *  UPDATE - we are the interrupt interrupting
2428 	 *           a current move.
2429 	 *  MOVED  - a reader on another CPU moved the next
2430 	 *           pointer to its reader page. Give up
2431 	 *           and try again.
2432 	 */
2433 
2434 	switch (type) {
2435 	case RB_PAGE_HEAD:
2436 		/*
2437 		 * We changed the head to UPDATE, thus
2438 		 * it is our responsibility to update
2439 		 * the counters.
2440 		 */
2441 		local_add(entries, &cpu_buffer->overrun);
2442 		local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
2443 
2444 		/*
2445 		 * The entries will be zeroed out when we move the
2446 		 * tail page.
2447 		 */
2448 
2449 		/* still more to do */
2450 		break;
2451 
2452 	case RB_PAGE_UPDATE:
2453 		/*
2454 		 * This is an interrupt that interrupt the
2455 		 * previous update. Still more to do.
2456 		 */
2457 		break;
2458 	case RB_PAGE_NORMAL:
2459 		/*
2460 		 * An interrupt came in before the update
2461 		 * and processed this for us.
2462 		 * Nothing left to do.
2463 		 */
2464 		return 1;
2465 	case RB_PAGE_MOVED:
2466 		/*
2467 		 * The reader is on another CPU and just did
2468 		 * a swap with our next_page.
2469 		 * Try again.
2470 		 */
2471 		return 1;
2472 	default:
2473 		RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
2474 		return -1;
2475 	}
2476 
2477 	/*
2478 	 * Now that we are here, the old head pointer is
2479 	 * set to UPDATE. This will keep the reader from
2480 	 * swapping the head page with the reader page.
2481 	 * The reader (on another CPU) will spin till
2482 	 * we are finished.
2483 	 *
2484 	 * We just need to protect against interrupts
2485 	 * doing the job. We will set the next pointer
2486 	 * to HEAD. After that, we set the old pointer
2487 	 * to NORMAL, but only if it was HEAD before.
2488 	 * otherwise we are an interrupt, and only
2489 	 * want the outer most commit to reset it.
2490 	 */
2491 	new_head = next_page;
2492 	rb_inc_page(&new_head);
2493 
2494 	ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
2495 				    RB_PAGE_NORMAL);
2496 
2497 	/*
2498 	 * Valid returns are:
2499 	 *  HEAD   - an interrupt came in and already set it.
2500 	 *  NORMAL - One of two things:
2501 	 *            1) We really set it.
2502 	 *            2) A bunch of interrupts came in and moved
2503 	 *               the page forward again.
2504 	 */
2505 	switch (ret) {
2506 	case RB_PAGE_HEAD:
2507 	case RB_PAGE_NORMAL:
2508 		/* OK */
2509 		break;
2510 	default:
2511 		RB_WARN_ON(cpu_buffer, 1);
2512 		return -1;
2513 	}
2514 
2515 	/*
2516 	 * It is possible that an interrupt came in,
2517 	 * set the head up, then more interrupts came in
2518 	 * and moved it again. When we get back here,
2519 	 * the page would have been set to NORMAL but we
2520 	 * just set it back to HEAD.
2521 	 *
2522 	 * How do you detect this? Well, if that happened
2523 	 * the tail page would have moved.
2524 	 */
2525 	if (ret == RB_PAGE_NORMAL) {
2526 		struct buffer_page *buffer_tail_page;
2527 
2528 		buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
2529 		/*
2530 		 * If the tail had moved passed next, then we need
2531 		 * to reset the pointer.
2532 		 */
2533 		if (buffer_tail_page != tail_page &&
2534 		    buffer_tail_page != next_page)
2535 			rb_head_page_set_normal(cpu_buffer, new_head,
2536 						next_page,
2537 						RB_PAGE_HEAD);
2538 	}
2539 
2540 	/*
2541 	 * If this was the outer most commit (the one that
2542 	 * changed the original pointer from HEAD to UPDATE),
2543 	 * then it is up to us to reset it to NORMAL.
2544 	 */
2545 	if (type == RB_PAGE_HEAD) {
2546 		ret = rb_head_page_set_normal(cpu_buffer, next_page,
2547 					      tail_page,
2548 					      RB_PAGE_UPDATE);
2549 		if (RB_WARN_ON(cpu_buffer,
2550 			       ret != RB_PAGE_UPDATE))
2551 			return -1;
2552 	}
2553 
2554 	return 0;
2555 }
2556 
2557 static inline void
rb_reset_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2558 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
2559 	      unsigned long tail, struct rb_event_info *info)
2560 {
2561 	struct buffer_page *tail_page = info->tail_page;
2562 	struct ring_buffer_event *event;
2563 	unsigned long length = info->length;
2564 
2565 	/*
2566 	 * Only the event that crossed the page boundary
2567 	 * must fill the old tail_page with padding.
2568 	 */
2569 	if (tail >= BUF_PAGE_SIZE) {
2570 		/*
2571 		 * If the page was filled, then we still need
2572 		 * to update the real_end. Reset it to zero
2573 		 * and the reader will ignore it.
2574 		 */
2575 		if (tail == BUF_PAGE_SIZE)
2576 			tail_page->real_end = 0;
2577 
2578 		local_sub(length, &tail_page->write);
2579 		return;
2580 	}
2581 
2582 	event = __rb_page_index(tail_page, tail);
2583 
2584 	/* account for padding bytes */
2585 	local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
2586 
2587 	/*
2588 	 * Save the original length to the meta data.
2589 	 * This will be used by the reader to add lost event
2590 	 * counter.
2591 	 */
2592 	tail_page->real_end = tail;
2593 
2594 	/*
2595 	 * If this event is bigger than the minimum size, then
2596 	 * we need to be careful that we don't subtract the
2597 	 * write counter enough to allow another writer to slip
2598 	 * in on this page.
2599 	 * We put in a discarded commit instead, to make sure
2600 	 * that this space is not used again.
2601 	 *
2602 	 * If we are less than the minimum size, we don't need to
2603 	 * worry about it.
2604 	 */
2605 	if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
2606 		/* No room for any events */
2607 
2608 		/* Mark the rest of the page with padding */
2609 		rb_event_set_padding(event);
2610 
2611 		/* Set the write back to the previous setting */
2612 		local_sub(length, &tail_page->write);
2613 		return;
2614 	}
2615 
2616 	/* Put in a discarded event */
2617 	event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
2618 	event->type_len = RINGBUF_TYPE_PADDING;
2619 	/* time delta must be non zero */
2620 	event->time_delta = 1;
2621 
2622 	/* Set write to end of buffer */
2623 	length = (tail + length) - BUF_PAGE_SIZE;
2624 	local_sub(length, &tail_page->write);
2625 }
2626 
2627 static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
2628 
2629 /*
2630  * This is the slow path, force gcc not to inline it.
2631  */
2632 static noinline struct ring_buffer_event *
rb_move_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2633 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
2634 	     unsigned long tail, struct rb_event_info *info)
2635 {
2636 	struct buffer_page *tail_page = info->tail_page;
2637 	struct buffer_page *commit_page = cpu_buffer->commit_page;
2638 	struct trace_buffer *buffer = cpu_buffer->buffer;
2639 	struct buffer_page *next_page;
2640 	int ret;
2641 
2642 	next_page = tail_page;
2643 
2644 	rb_inc_page(&next_page);
2645 
2646 	/*
2647 	 * If for some reason, we had an interrupt storm that made
2648 	 * it all the way around the buffer, bail, and warn
2649 	 * about it.
2650 	 */
2651 	if (unlikely(next_page == commit_page)) {
2652 		local_inc(&cpu_buffer->commit_overrun);
2653 		goto out_reset;
2654 	}
2655 
2656 	/*
2657 	 * This is where the fun begins!
2658 	 *
2659 	 * We are fighting against races between a reader that
2660 	 * could be on another CPU trying to swap its reader
2661 	 * page with the buffer head.
2662 	 *
2663 	 * We are also fighting against interrupts coming in and
2664 	 * moving the head or tail on us as well.
2665 	 *
2666 	 * If the next page is the head page then we have filled
2667 	 * the buffer, unless the commit page is still on the
2668 	 * reader page.
2669 	 */
2670 	if (rb_is_head_page(next_page, &tail_page->list)) {
2671 
2672 		/*
2673 		 * If the commit is not on the reader page, then
2674 		 * move the header page.
2675 		 */
2676 		if (!rb_is_reader_page(cpu_buffer->commit_page)) {
2677 			/*
2678 			 * If we are not in overwrite mode,
2679 			 * this is easy, just stop here.
2680 			 */
2681 			if (!(buffer->flags & RB_FL_OVERWRITE)) {
2682 				local_inc(&cpu_buffer->dropped_events);
2683 				goto out_reset;
2684 			}
2685 
2686 			ret = rb_handle_head_page(cpu_buffer,
2687 						  tail_page,
2688 						  next_page);
2689 			if (ret < 0)
2690 				goto out_reset;
2691 			if (ret)
2692 				goto out_again;
2693 		} else {
2694 			/*
2695 			 * We need to be careful here too. The
2696 			 * commit page could still be on the reader
2697 			 * page. We could have a small buffer, and
2698 			 * have filled up the buffer with events
2699 			 * from interrupts and such, and wrapped.
2700 			 *
2701 			 * Note, if the tail page is also on the
2702 			 * reader_page, we let it move out.
2703 			 */
2704 			if (unlikely((cpu_buffer->commit_page !=
2705 				      cpu_buffer->tail_page) &&
2706 				     (cpu_buffer->commit_page ==
2707 				      cpu_buffer->reader_page))) {
2708 				local_inc(&cpu_buffer->commit_overrun);
2709 				goto out_reset;
2710 			}
2711 		}
2712 	}
2713 
2714 	rb_tail_page_update(cpu_buffer, tail_page, next_page);
2715 
2716  out_again:
2717 
2718 	rb_reset_tail(cpu_buffer, tail, info);
2719 
2720 	/* Commit what we have for now. */
2721 	rb_end_commit(cpu_buffer);
2722 	/* rb_end_commit() decs committing */
2723 	local_inc(&cpu_buffer->committing);
2724 
2725 	/* fail and let the caller try again */
2726 	return ERR_PTR(-EAGAIN);
2727 
2728  out_reset:
2729 	/* reset write */
2730 	rb_reset_tail(cpu_buffer, tail, info);
2731 
2732 	return NULL;
2733 }
2734 
2735 /* Slow path */
2736 static struct ring_buffer_event *
rb_add_time_stamp(struct ring_buffer_event * event,u64 delta,bool abs)2737 rb_add_time_stamp(struct ring_buffer_event *event, u64 delta, bool abs)
2738 {
2739 	if (abs)
2740 		event->type_len = RINGBUF_TYPE_TIME_STAMP;
2741 	else
2742 		event->type_len = RINGBUF_TYPE_TIME_EXTEND;
2743 
2744 	/* Not the first event on the page, or not delta? */
2745 	if (abs || rb_event_index(event)) {
2746 		event->time_delta = delta & TS_MASK;
2747 		event->array[0] = delta >> TS_SHIFT;
2748 	} else {
2749 		/* nope, just zero it */
2750 		event->time_delta = 0;
2751 		event->array[0] = 0;
2752 	}
2753 
2754 	return skip_time_extend(event);
2755 }
2756 
2757 #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
sched_clock_stable(void)2758 static inline bool sched_clock_stable(void)
2759 {
2760 	return true;
2761 }
2762 #endif
2763 
2764 static void
rb_check_timestamp(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)2765 rb_check_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2766 		   struct rb_event_info *info)
2767 {
2768 	u64 write_stamp;
2769 
2770 	WARN_ONCE(1, "Delta way too big! %llu ts=%llu before=%llu after=%llu write stamp=%llu\n%s",
2771 		  (unsigned long long)info->delta,
2772 		  (unsigned long long)info->ts,
2773 		  (unsigned long long)info->before,
2774 		  (unsigned long long)info->after,
2775 		  (unsigned long long)(rb_time_read(&cpu_buffer->write_stamp, &write_stamp) ? write_stamp : 0),
2776 		  sched_clock_stable() ? "" :
2777 		  "If you just came from a suspend/resume,\n"
2778 		  "please switch to the trace global clock:\n"
2779 		  "  echo global > /sys/kernel/debug/tracing/trace_clock\n"
2780 		  "or add trace_clock=global to the kernel command line\n");
2781 }
2782 
rb_add_timestamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event ** event,struct rb_event_info * info,u64 * delta,unsigned int * length)2783 static void rb_add_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2784 				      struct ring_buffer_event **event,
2785 				      struct rb_event_info *info,
2786 				      u64 *delta,
2787 				      unsigned int *length)
2788 {
2789 	bool abs = info->add_timestamp &
2790 		(RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE);
2791 
2792 	if (unlikely(info->delta > (1ULL << 59))) {
2793 		/*
2794 		 * Some timers can use more than 59 bits, and when a timestamp
2795 		 * is added to the buffer, it will lose those bits.
2796 		 */
2797 		if (abs && (info->ts & TS_MSB)) {
2798 			info->delta &= ABS_TS_MASK;
2799 
2800 		/* did the clock go backwards */
2801 		} else if (info->before == info->after && info->before > info->ts) {
2802 			/* not interrupted */
2803 			static int once;
2804 
2805 			/*
2806 			 * This is possible with a recalibrating of the TSC.
2807 			 * Do not produce a call stack, but just report it.
2808 			 */
2809 			if (!once) {
2810 				once++;
2811 				pr_warn("Ring buffer clock went backwards: %llu -> %llu\n",
2812 					info->before, info->ts);
2813 			}
2814 		} else
2815 			rb_check_timestamp(cpu_buffer, info);
2816 		if (!abs)
2817 			info->delta = 0;
2818 	}
2819 	*event = rb_add_time_stamp(*event, info->delta, abs);
2820 	*length -= RB_LEN_TIME_EXTEND;
2821 	*delta = 0;
2822 }
2823 
2824 /**
2825  * rb_update_event - update event type and data
2826  * @cpu_buffer: The per cpu buffer of the @event
2827  * @event: the event to update
2828  * @info: The info to update the @event with (contains length and delta)
2829  *
2830  * Update the type and data fields of the @event. The length
2831  * is the actual size that is written to the ring buffer,
2832  * and with this, we can determine what to place into the
2833  * data field.
2834  */
2835 static void
rb_update_event(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event,struct rb_event_info * info)2836 rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
2837 		struct ring_buffer_event *event,
2838 		struct rb_event_info *info)
2839 {
2840 	unsigned length = info->length;
2841 	u64 delta = info->delta;
2842 	unsigned int nest = local_read(&cpu_buffer->committing) - 1;
2843 
2844 	if (!WARN_ON_ONCE(nest >= MAX_NEST))
2845 		cpu_buffer->event_stamp[nest] = info->ts;
2846 
2847 	/*
2848 	 * If we need to add a timestamp, then we
2849 	 * add it to the start of the reserved space.
2850 	 */
2851 	if (unlikely(info->add_timestamp))
2852 		rb_add_timestamp(cpu_buffer, &event, info, &delta, &length);
2853 
2854 	event->time_delta = delta;
2855 	length -= RB_EVNT_HDR_SIZE;
2856 	if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
2857 		event->type_len = 0;
2858 		event->array[0] = length;
2859 	} else
2860 		event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
2861 }
2862 
rb_calculate_event_length(unsigned length)2863 static unsigned rb_calculate_event_length(unsigned length)
2864 {
2865 	struct ring_buffer_event event; /* Used only for sizeof array */
2866 
2867 	/* zero length can cause confusions */
2868 	if (!length)
2869 		length++;
2870 
2871 	if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
2872 		length += sizeof(event.array[0]);
2873 
2874 	length += RB_EVNT_HDR_SIZE;
2875 	length = ALIGN(length, RB_ARCH_ALIGNMENT);
2876 
2877 	/*
2878 	 * In case the time delta is larger than the 27 bits for it
2879 	 * in the header, we need to add a timestamp. If another
2880 	 * event comes in when trying to discard this one to increase
2881 	 * the length, then the timestamp will be added in the allocated
2882 	 * space of this event. If length is bigger than the size needed
2883 	 * for the TIME_EXTEND, then padding has to be used. The events
2884 	 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
2885 	 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
2886 	 * As length is a multiple of 4, we only need to worry if it
2887 	 * is 12 (RB_LEN_TIME_EXTEND + 4).
2888 	 */
2889 	if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
2890 		length += RB_ALIGNMENT;
2891 
2892 	return length;
2893 }
2894 
rb_time_delta(struct ring_buffer_event * event)2895 static u64 rb_time_delta(struct ring_buffer_event *event)
2896 {
2897 	switch (event->type_len) {
2898 	case RINGBUF_TYPE_PADDING:
2899 		return 0;
2900 
2901 	case RINGBUF_TYPE_TIME_EXTEND:
2902 		return rb_event_time_stamp(event);
2903 
2904 	case RINGBUF_TYPE_TIME_STAMP:
2905 		return 0;
2906 
2907 	case RINGBUF_TYPE_DATA:
2908 		return event->time_delta;
2909 	default:
2910 		return 0;
2911 	}
2912 }
2913 
2914 static inline int
rb_try_to_discard(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2915 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2916 		  struct ring_buffer_event *event)
2917 {
2918 	unsigned long new_index, old_index;
2919 	struct buffer_page *bpage;
2920 	unsigned long index;
2921 	unsigned long addr;
2922 	u64 write_stamp;
2923 	u64 delta;
2924 
2925 	new_index = rb_event_index(event);
2926 	old_index = new_index + rb_event_ts_length(event);
2927 	addr = (unsigned long)event;
2928 	addr &= PAGE_MASK;
2929 
2930 	bpage = READ_ONCE(cpu_buffer->tail_page);
2931 
2932 	delta = rb_time_delta(event);
2933 
2934 	if (!rb_time_read(&cpu_buffer->write_stamp, &write_stamp))
2935 		return 0;
2936 
2937 	/* Make sure the write stamp is read before testing the location */
2938 	barrier();
2939 
2940 	if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
2941 		unsigned long write_mask =
2942 			local_read(&bpage->write) & ~RB_WRITE_MASK;
2943 		unsigned long event_length = rb_event_length(event);
2944 
2945 		/* Something came in, can't discard */
2946 		if (!rb_time_cmpxchg(&cpu_buffer->write_stamp,
2947 				       write_stamp, write_stamp - delta))
2948 			return 0;
2949 
2950 		/*
2951 		 * It's possible that the event time delta is zero
2952 		 * (has the same time stamp as the previous event)
2953 		 * in which case write_stamp and before_stamp could
2954 		 * be the same. In such a case, force before_stamp
2955 		 * to be different than write_stamp. It doesn't
2956 		 * matter what it is, as long as its different.
2957 		 */
2958 		if (!delta)
2959 			rb_time_set(&cpu_buffer->before_stamp, 0);
2960 
2961 		/*
2962 		 * If an event were to come in now, it would see that the
2963 		 * write_stamp and the before_stamp are different, and assume
2964 		 * that this event just added itself before updating
2965 		 * the write stamp. The interrupting event will fix the
2966 		 * write stamp for us, and use the before stamp as its delta.
2967 		 */
2968 
2969 		/*
2970 		 * This is on the tail page. It is possible that
2971 		 * a write could come in and move the tail page
2972 		 * and write to the next page. That is fine
2973 		 * because we just shorten what is on this page.
2974 		 */
2975 		old_index += write_mask;
2976 		new_index += write_mask;
2977 		index = local_cmpxchg(&bpage->write, old_index, new_index);
2978 		if (index == old_index) {
2979 			/* update counters */
2980 			local_sub(event_length, &cpu_buffer->entries_bytes);
2981 			return 1;
2982 		}
2983 	}
2984 
2985 	/* could not discard */
2986 	return 0;
2987 }
2988 
rb_start_commit(struct ring_buffer_per_cpu * cpu_buffer)2989 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
2990 {
2991 	local_inc(&cpu_buffer->committing);
2992 	local_inc(&cpu_buffer->commits);
2993 }
2994 
2995 static __always_inline void
rb_set_commit_to_write(struct ring_buffer_per_cpu * cpu_buffer)2996 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
2997 {
2998 	unsigned long max_count;
2999 
3000 	/*
3001 	 * We only race with interrupts and NMIs on this CPU.
3002 	 * If we own the commit event, then we can commit
3003 	 * all others that interrupted us, since the interruptions
3004 	 * are in stack format (they finish before they come
3005 	 * back to us). This allows us to do a simple loop to
3006 	 * assign the commit to the tail.
3007 	 */
3008  again:
3009 	max_count = cpu_buffer->nr_pages * 100;
3010 
3011 	while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
3012 		if (RB_WARN_ON(cpu_buffer, !(--max_count)))
3013 			return;
3014 		if (RB_WARN_ON(cpu_buffer,
3015 			       rb_is_reader_page(cpu_buffer->tail_page)))
3016 			return;
3017 		local_set(&cpu_buffer->commit_page->page->commit,
3018 			  rb_page_write(cpu_buffer->commit_page));
3019 		rb_inc_page(&cpu_buffer->commit_page);
3020 		/* add barrier to keep gcc from optimizing too much */
3021 		barrier();
3022 	}
3023 	while (rb_commit_index(cpu_buffer) !=
3024 	       rb_page_write(cpu_buffer->commit_page)) {
3025 
3026 		local_set(&cpu_buffer->commit_page->page->commit,
3027 			  rb_page_write(cpu_buffer->commit_page));
3028 		RB_WARN_ON(cpu_buffer,
3029 			   local_read(&cpu_buffer->commit_page->page->commit) &
3030 			   ~RB_WRITE_MASK);
3031 		barrier();
3032 	}
3033 
3034 	/* again, keep gcc from optimizing */
3035 	barrier();
3036 
3037 	/*
3038 	 * If an interrupt came in just after the first while loop
3039 	 * and pushed the tail page forward, we will be left with
3040 	 * a dangling commit that will never go forward.
3041 	 */
3042 	if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
3043 		goto again;
3044 }
3045 
rb_end_commit(struct ring_buffer_per_cpu * cpu_buffer)3046 static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
3047 {
3048 	unsigned long commits;
3049 
3050 	if (RB_WARN_ON(cpu_buffer,
3051 		       !local_read(&cpu_buffer->committing)))
3052 		return;
3053 
3054  again:
3055 	commits = local_read(&cpu_buffer->commits);
3056 	/* synchronize with interrupts */
3057 	barrier();
3058 	if (local_read(&cpu_buffer->committing) == 1)
3059 		rb_set_commit_to_write(cpu_buffer);
3060 
3061 	local_dec(&cpu_buffer->committing);
3062 
3063 	/* synchronize with interrupts */
3064 	barrier();
3065 
3066 	/*
3067 	 * Need to account for interrupts coming in between the
3068 	 * updating of the commit page and the clearing of the
3069 	 * committing counter.
3070 	 */
3071 	if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
3072 	    !local_read(&cpu_buffer->committing)) {
3073 		local_inc(&cpu_buffer->committing);
3074 		goto again;
3075 	}
3076 }
3077 
rb_event_discard(struct ring_buffer_event * event)3078 static inline void rb_event_discard(struct ring_buffer_event *event)
3079 {
3080 	if (extended_time(event))
3081 		event = skip_time_extend(event);
3082 
3083 	/* array[0] holds the actual length for the discarded event */
3084 	event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
3085 	event->type_len = RINGBUF_TYPE_PADDING;
3086 	/* time delta must be non zero */
3087 	if (!event->time_delta)
3088 		event->time_delta = 1;
3089 }
3090 
rb_commit(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3091 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
3092 		      struct ring_buffer_event *event)
3093 {
3094 	local_inc(&cpu_buffer->entries);
3095 	rb_end_commit(cpu_buffer);
3096 }
3097 
3098 static __always_inline void
rb_wakeups(struct trace_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer)3099 rb_wakeups(struct trace_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
3100 {
3101 	size_t nr_pages;
3102 	size_t dirty;
3103 	size_t full;
3104 
3105 	if (buffer->irq_work.waiters_pending) {
3106 		buffer->irq_work.waiters_pending = false;
3107 		/* irq_work_queue() supplies it's own memory barriers */
3108 		irq_work_queue(&buffer->irq_work.work);
3109 	}
3110 
3111 	if (cpu_buffer->irq_work.waiters_pending) {
3112 		cpu_buffer->irq_work.waiters_pending = false;
3113 		/* irq_work_queue() supplies it's own memory barriers */
3114 		irq_work_queue(&cpu_buffer->irq_work.work);
3115 	}
3116 
3117 	if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
3118 		return;
3119 
3120 	if (cpu_buffer->reader_page == cpu_buffer->commit_page)
3121 		return;
3122 
3123 	if (!cpu_buffer->irq_work.full_waiters_pending)
3124 		return;
3125 
3126 	cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
3127 
3128 	full = cpu_buffer->shortest_full;
3129 	nr_pages = cpu_buffer->nr_pages;
3130 	dirty = ring_buffer_nr_dirty_pages(buffer, cpu_buffer->cpu);
3131 	if (full && nr_pages && (dirty * 100) <= full * nr_pages)
3132 		return;
3133 
3134 	cpu_buffer->irq_work.wakeup_full = true;
3135 	cpu_buffer->irq_work.full_waiters_pending = false;
3136 	/* irq_work_queue() supplies it's own memory barriers */
3137 	irq_work_queue(&cpu_buffer->irq_work.work);
3138 }
3139 
3140 #ifdef CONFIG_RING_BUFFER_RECORD_RECURSION
3141 # define do_ring_buffer_record_recursion()	\
3142 	do_ftrace_record_recursion(_THIS_IP_, _RET_IP_)
3143 #else
3144 # define do_ring_buffer_record_recursion() do { } while (0)
3145 #endif
3146 
3147 /*
3148  * The lock and unlock are done within a preempt disable section.
3149  * The current_context per_cpu variable can only be modified
3150  * by the current task between lock and unlock. But it can
3151  * be modified more than once via an interrupt. To pass this
3152  * information from the lock to the unlock without having to
3153  * access the 'in_interrupt()' functions again (which do show
3154  * a bit of overhead in something as critical as function tracing,
3155  * we use a bitmask trick.
3156  *
3157  *  bit 1 =  NMI context
3158  *  bit 2 =  IRQ context
3159  *  bit 3 =  SoftIRQ context
3160  *  bit 4 =  normal context.
3161  *
3162  * This works because this is the order of contexts that can
3163  * preempt other contexts. A SoftIRQ never preempts an IRQ
3164  * context.
3165  *
3166  * When the context is determined, the corresponding bit is
3167  * checked and set (if it was set, then a recursion of that context
3168  * happened).
3169  *
3170  * On unlock, we need to clear this bit. To do so, just subtract
3171  * 1 from the current_context and AND it to itself.
3172  *
3173  * (binary)
3174  *  101 - 1 = 100
3175  *  101 & 100 = 100 (clearing bit zero)
3176  *
3177  *  1010 - 1 = 1001
3178  *  1010 & 1001 = 1000 (clearing bit 1)
3179  *
3180  * The least significant bit can be cleared this way, and it
3181  * just so happens that it is the same bit corresponding to
3182  * the current context.
3183  *
3184  * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
3185  * is set when a recursion is detected at the current context, and if
3186  * the TRANSITION bit is already set, it will fail the recursion.
3187  * This is needed because there's a lag between the changing of
3188  * interrupt context and updating the preempt count. In this case,
3189  * a false positive will be found. To handle this, one extra recursion
3190  * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
3191  * bit is already set, then it is considered a recursion and the function
3192  * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
3193  *
3194  * On the trace_recursive_unlock(), the TRANSITION bit will be the first
3195  * to be cleared. Even if it wasn't the context that set it. That is,
3196  * if an interrupt comes in while NORMAL bit is set and the ring buffer
3197  * is called before preempt_count() is updated, since the check will
3198  * be on the NORMAL bit, the TRANSITION bit will then be set. If an
3199  * NMI then comes in, it will set the NMI bit, but when the NMI code
3200  * does the trace_recursive_unlock() it will clear the TRANSITION bit
3201  * and leave the NMI bit set. But this is fine, because the interrupt
3202  * code that set the TRANSITION bit will then clear the NMI bit when it
3203  * calls trace_recursive_unlock(). If another NMI comes in, it will
3204  * set the TRANSITION bit and continue.
3205  *
3206  * Note: The TRANSITION bit only handles a single transition between context.
3207  */
3208 
3209 static __always_inline int
trace_recursive_lock(struct ring_buffer_per_cpu * cpu_buffer)3210 trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
3211 {
3212 	unsigned int val = cpu_buffer->current_context;
3213 	int bit = interrupt_context_level();
3214 
3215 	bit = RB_CTX_NORMAL - bit;
3216 
3217 	if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) {
3218 		/*
3219 		 * It is possible that this was called by transitioning
3220 		 * between interrupt context, and preempt_count() has not
3221 		 * been updated yet. In this case, use the TRANSITION bit.
3222 		 */
3223 		bit = RB_CTX_TRANSITION;
3224 		if (val & (1 << (bit + cpu_buffer->nest))) {
3225 			do_ring_buffer_record_recursion();
3226 			return 1;
3227 		}
3228 	}
3229 
3230 	val |= (1 << (bit + cpu_buffer->nest));
3231 	cpu_buffer->current_context = val;
3232 
3233 	return 0;
3234 }
3235 
3236 static __always_inline void
trace_recursive_unlock(struct ring_buffer_per_cpu * cpu_buffer)3237 trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
3238 {
3239 	cpu_buffer->current_context &=
3240 		cpu_buffer->current_context - (1 << cpu_buffer->nest);
3241 }
3242 
3243 /* The recursive locking above uses 5 bits */
3244 #define NESTED_BITS 5
3245 
3246 /**
3247  * ring_buffer_nest_start - Allow to trace while nested
3248  * @buffer: The ring buffer to modify
3249  *
3250  * The ring buffer has a safety mechanism to prevent recursion.
3251  * But there may be a case where a trace needs to be done while
3252  * tracing something else. In this case, calling this function
3253  * will allow this function to nest within a currently active
3254  * ring_buffer_lock_reserve().
3255  *
3256  * Call this function before calling another ring_buffer_lock_reserve() and
3257  * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
3258  */
ring_buffer_nest_start(struct trace_buffer * buffer)3259 void ring_buffer_nest_start(struct trace_buffer *buffer)
3260 {
3261 	struct ring_buffer_per_cpu *cpu_buffer;
3262 	int cpu;
3263 
3264 	/* Enabled by ring_buffer_nest_end() */
3265 	preempt_disable_notrace();
3266 	cpu = raw_smp_processor_id();
3267 	cpu_buffer = buffer->buffers[cpu];
3268 	/* This is the shift value for the above recursive locking */
3269 	cpu_buffer->nest += NESTED_BITS;
3270 }
3271 
3272 /**
3273  * ring_buffer_nest_end - Allow to trace while nested
3274  * @buffer: The ring buffer to modify
3275  *
3276  * Must be called after ring_buffer_nest_start() and after the
3277  * ring_buffer_unlock_commit().
3278  */
ring_buffer_nest_end(struct trace_buffer * buffer)3279 void ring_buffer_nest_end(struct trace_buffer *buffer)
3280 {
3281 	struct ring_buffer_per_cpu *cpu_buffer;
3282 	int cpu;
3283 
3284 	/* disabled by ring_buffer_nest_start() */
3285 	cpu = raw_smp_processor_id();
3286 	cpu_buffer = buffer->buffers[cpu];
3287 	/* This is the shift value for the above recursive locking */
3288 	cpu_buffer->nest -= NESTED_BITS;
3289 	preempt_enable_notrace();
3290 }
3291 
3292 /**
3293  * ring_buffer_unlock_commit - commit a reserved
3294  * @buffer: The buffer to commit to
3295  * @event: The event pointer to commit.
3296  *
3297  * This commits the data to the ring buffer, and releases any locks held.
3298  *
3299  * Must be paired with ring_buffer_lock_reserve.
3300  */
ring_buffer_unlock_commit(struct trace_buffer * buffer,struct ring_buffer_event * event)3301 int ring_buffer_unlock_commit(struct trace_buffer *buffer,
3302 			      struct ring_buffer_event *event)
3303 {
3304 	struct ring_buffer_per_cpu *cpu_buffer;
3305 	int cpu = raw_smp_processor_id();
3306 
3307 	cpu_buffer = buffer->buffers[cpu];
3308 
3309 	rb_commit(cpu_buffer, event);
3310 
3311 	rb_wakeups(buffer, cpu_buffer);
3312 
3313 	trace_recursive_unlock(cpu_buffer);
3314 
3315 	preempt_enable_notrace();
3316 
3317 	return 0;
3318 }
3319 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
3320 
3321 /* Special value to validate all deltas on a page. */
3322 #define CHECK_FULL_PAGE		1L
3323 
3324 #ifdef CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS
dump_buffer_page(struct buffer_data_page * bpage,struct rb_event_info * info,unsigned long tail)3325 static void dump_buffer_page(struct buffer_data_page *bpage,
3326 			     struct rb_event_info *info,
3327 			     unsigned long tail)
3328 {
3329 	struct ring_buffer_event *event;
3330 	u64 ts, delta;
3331 	int e;
3332 
3333 	ts = bpage->time_stamp;
3334 	pr_warn("  [%lld] PAGE TIME STAMP\n", ts);
3335 
3336 	for (e = 0; e < tail; e += rb_event_length(event)) {
3337 
3338 		event = (struct ring_buffer_event *)(bpage->data + e);
3339 
3340 		switch (event->type_len) {
3341 
3342 		case RINGBUF_TYPE_TIME_EXTEND:
3343 			delta = rb_event_time_stamp(event);
3344 			ts += delta;
3345 			pr_warn("  [%lld] delta:%lld TIME EXTEND\n", ts, delta);
3346 			break;
3347 
3348 		case RINGBUF_TYPE_TIME_STAMP:
3349 			delta = rb_event_time_stamp(event);
3350 			ts = rb_fix_abs_ts(delta, ts);
3351 			pr_warn("  [%lld] absolute:%lld TIME STAMP\n", ts, delta);
3352 			break;
3353 
3354 		case RINGBUF_TYPE_PADDING:
3355 			ts += event->time_delta;
3356 			pr_warn("  [%lld] delta:%d PADDING\n", ts, event->time_delta);
3357 			break;
3358 
3359 		case RINGBUF_TYPE_DATA:
3360 			ts += event->time_delta;
3361 			pr_warn("  [%lld] delta:%d\n", ts, event->time_delta);
3362 			break;
3363 
3364 		default:
3365 			break;
3366 		}
3367 	}
3368 }
3369 
3370 static DEFINE_PER_CPU(atomic_t, checking);
3371 static atomic_t ts_dump;
3372 
3373 /*
3374  * Check if the current event time stamp matches the deltas on
3375  * the buffer page.
3376  */
check_buffer(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info,unsigned long tail)3377 static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
3378 			 struct rb_event_info *info,
3379 			 unsigned long tail)
3380 {
3381 	struct ring_buffer_event *event;
3382 	struct buffer_data_page *bpage;
3383 	u64 ts, delta;
3384 	bool full = false;
3385 	int e;
3386 
3387 	bpage = info->tail_page->page;
3388 
3389 	if (tail == CHECK_FULL_PAGE) {
3390 		full = true;
3391 		tail = local_read(&bpage->commit);
3392 	} else if (info->add_timestamp &
3393 		   (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)) {
3394 		/* Ignore events with absolute time stamps */
3395 		return;
3396 	}
3397 
3398 	/*
3399 	 * Do not check the first event (skip possible extends too).
3400 	 * Also do not check if previous events have not been committed.
3401 	 */
3402 	if (tail <= 8 || tail > local_read(&bpage->commit))
3403 		return;
3404 
3405 	/*
3406 	 * If this interrupted another event,
3407 	 */
3408 	if (atomic_inc_return(this_cpu_ptr(&checking)) != 1)
3409 		goto out;
3410 
3411 	ts = bpage->time_stamp;
3412 
3413 	for (e = 0; e < tail; e += rb_event_length(event)) {
3414 
3415 		event = (struct ring_buffer_event *)(bpage->data + e);
3416 
3417 		switch (event->type_len) {
3418 
3419 		case RINGBUF_TYPE_TIME_EXTEND:
3420 			delta = rb_event_time_stamp(event);
3421 			ts += delta;
3422 			break;
3423 
3424 		case RINGBUF_TYPE_TIME_STAMP:
3425 			delta = rb_event_time_stamp(event);
3426 			ts = rb_fix_abs_ts(delta, ts);
3427 			break;
3428 
3429 		case RINGBUF_TYPE_PADDING:
3430 			if (event->time_delta == 1)
3431 				break;
3432 			fallthrough;
3433 		case RINGBUF_TYPE_DATA:
3434 			ts += event->time_delta;
3435 			break;
3436 
3437 		default:
3438 			RB_WARN_ON(cpu_buffer, 1);
3439 		}
3440 	}
3441 	if ((full && ts > info->ts) ||
3442 	    (!full && ts + info->delta != info->ts)) {
3443 		/* If another report is happening, ignore this one */
3444 		if (atomic_inc_return(&ts_dump) != 1) {
3445 			atomic_dec(&ts_dump);
3446 			goto out;
3447 		}
3448 		atomic_inc(&cpu_buffer->record_disabled);
3449 		/* There's some cases in boot up that this can happen */
3450 		WARN_ON_ONCE(system_state != SYSTEM_BOOTING);
3451 		pr_warn("[CPU: %d]TIME DOES NOT MATCH expected:%lld actual:%lld delta:%lld before:%lld after:%lld%s\n",
3452 			cpu_buffer->cpu,
3453 			ts + info->delta, info->ts, info->delta,
3454 			info->before, info->after,
3455 			full ? " (full)" : "");
3456 		dump_buffer_page(bpage, info, tail);
3457 		atomic_dec(&ts_dump);
3458 		/* Do not re-enable checking */
3459 		return;
3460 	}
3461 out:
3462 	atomic_dec(this_cpu_ptr(&checking));
3463 }
3464 #else
check_buffer(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info,unsigned long tail)3465 static inline void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
3466 			 struct rb_event_info *info,
3467 			 unsigned long tail)
3468 {
3469 }
3470 #endif /* CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS */
3471 
3472 static struct ring_buffer_event *
__rb_reserve_next(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)3473 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
3474 		  struct rb_event_info *info)
3475 {
3476 	struct ring_buffer_event *event;
3477 	struct buffer_page *tail_page;
3478 	unsigned long tail, write, w;
3479 	bool a_ok;
3480 	bool b_ok;
3481 
3482 	/* Don't let the compiler play games with cpu_buffer->tail_page */
3483 	tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
3484 
3485  /*A*/	w = local_read(&tail_page->write) & RB_WRITE_MASK;
3486 	barrier();
3487 	b_ok = rb_time_read(&cpu_buffer->before_stamp, &info->before);
3488 	a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3489 	barrier();
3490 	info->ts = rb_time_stamp(cpu_buffer->buffer);
3491 
3492 	if ((info->add_timestamp & RB_ADD_STAMP_ABSOLUTE)) {
3493 		info->delta = info->ts;
3494 	} else {
3495 		/*
3496 		 * If interrupting an event time update, we may need an
3497 		 * absolute timestamp.
3498 		 * Don't bother if this is the start of a new page (w == 0).
3499 		 */
3500 		if (unlikely(!a_ok || !b_ok || (info->before != info->after && w))) {
3501 			info->add_timestamp |= RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND;
3502 			info->length += RB_LEN_TIME_EXTEND;
3503 		} else {
3504 			info->delta = info->ts - info->after;
3505 			if (unlikely(test_time_stamp(info->delta))) {
3506 				info->add_timestamp |= RB_ADD_STAMP_EXTEND;
3507 				info->length += RB_LEN_TIME_EXTEND;
3508 			}
3509 		}
3510 	}
3511 
3512  /*B*/	rb_time_set(&cpu_buffer->before_stamp, info->ts);
3513 
3514  /*C*/	write = local_add_return(info->length, &tail_page->write);
3515 
3516 	/* set write to only the index of the write */
3517 	write &= RB_WRITE_MASK;
3518 
3519 	tail = write - info->length;
3520 
3521 	/* See if we shot pass the end of this buffer page */
3522 	if (unlikely(write > BUF_PAGE_SIZE)) {
3523 		/* before and after may now different, fix it up*/
3524 		b_ok = rb_time_read(&cpu_buffer->before_stamp, &info->before);
3525 		a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3526 		if (a_ok && b_ok && info->before != info->after)
3527 			(void)rb_time_cmpxchg(&cpu_buffer->before_stamp,
3528 					      info->before, info->after);
3529 		if (a_ok && b_ok)
3530 			check_buffer(cpu_buffer, info, CHECK_FULL_PAGE);
3531 		return rb_move_tail(cpu_buffer, tail, info);
3532 	}
3533 
3534 	if (likely(tail == w)) {
3535 		u64 save_before;
3536 		bool s_ok;
3537 
3538 		/* Nothing interrupted us between A and C */
3539  /*D*/		rb_time_set(&cpu_buffer->write_stamp, info->ts);
3540 		barrier();
3541  /*E*/		s_ok = rb_time_read(&cpu_buffer->before_stamp, &save_before);
3542 		RB_WARN_ON(cpu_buffer, !s_ok);
3543 		if (likely(!(info->add_timestamp &
3544 			     (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
3545 			/* This did not interrupt any time update */
3546 			info->delta = info->ts - info->after;
3547 		else
3548 			/* Just use full timestamp for interrupting event */
3549 			info->delta = info->ts;
3550 		barrier();
3551 		check_buffer(cpu_buffer, info, tail);
3552 		if (unlikely(info->ts != save_before)) {
3553 			/* SLOW PATH - Interrupted between C and E */
3554 
3555 			a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3556 			RB_WARN_ON(cpu_buffer, !a_ok);
3557 
3558 			/* Write stamp must only go forward */
3559 			if (save_before > info->after) {
3560 				/*
3561 				 * We do not care about the result, only that
3562 				 * it gets updated atomically.
3563 				 */
3564 				(void)rb_time_cmpxchg(&cpu_buffer->write_stamp,
3565 						      info->after, save_before);
3566 			}
3567 		}
3568 	} else {
3569 		u64 ts;
3570 		/* SLOW PATH - Interrupted between A and C */
3571 		a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3572 		/* Was interrupted before here, write_stamp must be valid */
3573 		RB_WARN_ON(cpu_buffer, !a_ok);
3574 		ts = rb_time_stamp(cpu_buffer->buffer);
3575 		barrier();
3576  /*E*/		if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) &&
3577 		    info->after < ts &&
3578 		    rb_time_cmpxchg(&cpu_buffer->write_stamp,
3579 				    info->after, ts)) {
3580 			/* Nothing came after this event between C and E */
3581 			info->delta = ts - info->after;
3582 		} else {
3583 			/*
3584 			 * Interrupted between C and E:
3585 			 * Lost the previous events time stamp. Just set the
3586 			 * delta to zero, and this will be the same time as
3587 			 * the event this event interrupted. And the events that
3588 			 * came after this will still be correct (as they would
3589 			 * have built their delta on the previous event.
3590 			 */
3591 			info->delta = 0;
3592 		}
3593 		info->ts = ts;
3594 		info->add_timestamp &= ~RB_ADD_STAMP_FORCE;
3595 	}
3596 
3597 	/*
3598 	 * If this is the first commit on the page, then it has the same
3599 	 * timestamp as the page itself.
3600 	 */
3601 	if (unlikely(!tail && !(info->add_timestamp &
3602 				(RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
3603 		info->delta = 0;
3604 
3605 	/* We reserved something on the buffer */
3606 
3607 	event = __rb_page_index(tail_page, tail);
3608 	rb_update_event(cpu_buffer, event, info);
3609 
3610 	local_inc(&tail_page->entries);
3611 
3612 	/*
3613 	 * If this is the first commit on the page, then update
3614 	 * its timestamp.
3615 	 */
3616 	if (unlikely(!tail))
3617 		tail_page->page->time_stamp = info->ts;
3618 
3619 	/* account for these added bytes */
3620 	local_add(info->length, &cpu_buffer->entries_bytes);
3621 
3622 	return event;
3623 }
3624 
3625 static __always_inline struct ring_buffer_event *
rb_reserve_next_event(struct trace_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer,unsigned long length)3626 rb_reserve_next_event(struct trace_buffer *buffer,
3627 		      struct ring_buffer_per_cpu *cpu_buffer,
3628 		      unsigned long length)
3629 {
3630 	struct ring_buffer_event *event;
3631 	struct rb_event_info info;
3632 	int nr_loops = 0;
3633 	int add_ts_default;
3634 
3635 	rb_start_commit(cpu_buffer);
3636 	/* The commit page can not change after this */
3637 
3638 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
3639 	/*
3640 	 * Due to the ability to swap a cpu buffer from a buffer
3641 	 * it is possible it was swapped before we committed.
3642 	 * (committing stops a swap). We check for it here and
3643 	 * if it happened, we have to fail the write.
3644 	 */
3645 	barrier();
3646 	if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
3647 		local_dec(&cpu_buffer->committing);
3648 		local_dec(&cpu_buffer->commits);
3649 		return NULL;
3650 	}
3651 #endif
3652 
3653 	info.length = rb_calculate_event_length(length);
3654 
3655 	if (ring_buffer_time_stamp_abs(cpu_buffer->buffer)) {
3656 		add_ts_default = RB_ADD_STAMP_ABSOLUTE;
3657 		info.length += RB_LEN_TIME_EXTEND;
3658 	} else {
3659 		add_ts_default = RB_ADD_STAMP_NONE;
3660 	}
3661 
3662  again:
3663 	info.add_timestamp = add_ts_default;
3664 	info.delta = 0;
3665 
3666 	/*
3667 	 * We allow for interrupts to reenter here and do a trace.
3668 	 * If one does, it will cause this original code to loop
3669 	 * back here. Even with heavy interrupts happening, this
3670 	 * should only happen a few times in a row. If this happens
3671 	 * 1000 times in a row, there must be either an interrupt
3672 	 * storm or we have something buggy.
3673 	 * Bail!
3674 	 */
3675 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
3676 		goto out_fail;
3677 
3678 	event = __rb_reserve_next(cpu_buffer, &info);
3679 
3680 	if (unlikely(PTR_ERR(event) == -EAGAIN)) {
3681 		if (info.add_timestamp & (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND))
3682 			info.length -= RB_LEN_TIME_EXTEND;
3683 		goto again;
3684 	}
3685 
3686 	if (likely(event))
3687 		return event;
3688  out_fail:
3689 	rb_end_commit(cpu_buffer);
3690 	return NULL;
3691 }
3692 
3693 /**
3694  * ring_buffer_lock_reserve - reserve a part of the buffer
3695  * @buffer: the ring buffer to reserve from
3696  * @length: the length of the data to reserve (excluding event header)
3697  *
3698  * Returns a reserved event on the ring buffer to copy directly to.
3699  * The user of this interface will need to get the body to write into
3700  * and can use the ring_buffer_event_data() interface.
3701  *
3702  * The length is the length of the data needed, not the event length
3703  * which also includes the event header.
3704  *
3705  * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
3706  * If NULL is returned, then nothing has been allocated or locked.
3707  */
3708 struct ring_buffer_event *
ring_buffer_lock_reserve(struct trace_buffer * buffer,unsigned long length)3709 ring_buffer_lock_reserve(struct trace_buffer *buffer, unsigned long length)
3710 {
3711 	struct ring_buffer_per_cpu *cpu_buffer;
3712 	struct ring_buffer_event *event;
3713 	int cpu;
3714 
3715 	/* If we are tracing schedule, we don't want to recurse */
3716 	preempt_disable_notrace();
3717 
3718 	if (unlikely(atomic_read(&buffer->record_disabled)))
3719 		goto out;
3720 
3721 	cpu = raw_smp_processor_id();
3722 
3723 	if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
3724 		goto out;
3725 
3726 	cpu_buffer = buffer->buffers[cpu];
3727 
3728 	if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
3729 		goto out;
3730 
3731 	if (unlikely(length > BUF_MAX_DATA_SIZE))
3732 		goto out;
3733 
3734 	if (unlikely(trace_recursive_lock(cpu_buffer)))
3735 		goto out;
3736 
3737 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
3738 	if (!event)
3739 		goto out_unlock;
3740 
3741 	return event;
3742 
3743  out_unlock:
3744 	trace_recursive_unlock(cpu_buffer);
3745  out:
3746 	preempt_enable_notrace();
3747 	return NULL;
3748 }
3749 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
3750 
3751 /*
3752  * Decrement the entries to the page that an event is on.
3753  * The event does not even need to exist, only the pointer
3754  * to the page it is on. This may only be called before the commit
3755  * takes place.
3756  */
3757 static inline void
rb_decrement_entry(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3758 rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
3759 		   struct ring_buffer_event *event)
3760 {
3761 	unsigned long addr = (unsigned long)event;
3762 	struct buffer_page *bpage = cpu_buffer->commit_page;
3763 	struct buffer_page *start;
3764 
3765 	addr &= PAGE_MASK;
3766 
3767 	/* Do the likely case first */
3768 	if (likely(bpage->page == (void *)addr)) {
3769 		local_dec(&bpage->entries);
3770 		return;
3771 	}
3772 
3773 	/*
3774 	 * Because the commit page may be on the reader page we
3775 	 * start with the next page and check the end loop there.
3776 	 */
3777 	rb_inc_page(&bpage);
3778 	start = bpage;
3779 	do {
3780 		if (bpage->page == (void *)addr) {
3781 			local_dec(&bpage->entries);
3782 			return;
3783 		}
3784 		rb_inc_page(&bpage);
3785 	} while (bpage != start);
3786 
3787 	/* commit not part of this buffer?? */
3788 	RB_WARN_ON(cpu_buffer, 1);
3789 }
3790 
3791 /**
3792  * ring_buffer_discard_commit - discard an event that has not been committed
3793  * @buffer: the ring buffer
3794  * @event: non committed event to discard
3795  *
3796  * Sometimes an event that is in the ring buffer needs to be ignored.
3797  * This function lets the user discard an event in the ring buffer
3798  * and then that event will not be read later.
3799  *
3800  * This function only works if it is called before the item has been
3801  * committed. It will try to free the event from the ring buffer
3802  * if another event has not been added behind it.
3803  *
3804  * If another event has been added behind it, it will set the event
3805  * up as discarded, and perform the commit.
3806  *
3807  * If this function is called, do not call ring_buffer_unlock_commit on
3808  * the event.
3809  */
ring_buffer_discard_commit(struct trace_buffer * buffer,struct ring_buffer_event * event)3810 void ring_buffer_discard_commit(struct trace_buffer *buffer,
3811 				struct ring_buffer_event *event)
3812 {
3813 	struct ring_buffer_per_cpu *cpu_buffer;
3814 	int cpu;
3815 
3816 	/* The event is discarded regardless */
3817 	rb_event_discard(event);
3818 
3819 	cpu = smp_processor_id();
3820 	cpu_buffer = buffer->buffers[cpu];
3821 
3822 	/*
3823 	 * This must only be called if the event has not been
3824 	 * committed yet. Thus we can assume that preemption
3825 	 * is still disabled.
3826 	 */
3827 	RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
3828 
3829 	rb_decrement_entry(cpu_buffer, event);
3830 	if (rb_try_to_discard(cpu_buffer, event))
3831 		goto out;
3832 
3833  out:
3834 	rb_end_commit(cpu_buffer);
3835 
3836 	trace_recursive_unlock(cpu_buffer);
3837 
3838 	preempt_enable_notrace();
3839 
3840 }
3841 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
3842 
3843 /**
3844  * ring_buffer_write - write data to the buffer without reserving
3845  * @buffer: The ring buffer to write to.
3846  * @length: The length of the data being written (excluding the event header)
3847  * @data: The data to write to the buffer.
3848  *
3849  * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
3850  * one function. If you already have the data to write to the buffer, it
3851  * may be easier to simply call this function.
3852  *
3853  * Note, like ring_buffer_lock_reserve, the length is the length of the data
3854  * and not the length of the event which would hold the header.
3855  */
ring_buffer_write(struct trace_buffer * buffer,unsigned long length,void * data)3856 int ring_buffer_write(struct trace_buffer *buffer,
3857 		      unsigned long length,
3858 		      void *data)
3859 {
3860 	struct ring_buffer_per_cpu *cpu_buffer;
3861 	struct ring_buffer_event *event;
3862 	void *body;
3863 	int ret = -EBUSY;
3864 	int cpu;
3865 
3866 	preempt_disable_notrace();
3867 
3868 	if (atomic_read(&buffer->record_disabled))
3869 		goto out;
3870 
3871 	cpu = raw_smp_processor_id();
3872 
3873 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
3874 		goto out;
3875 
3876 	cpu_buffer = buffer->buffers[cpu];
3877 
3878 	if (atomic_read(&cpu_buffer->record_disabled))
3879 		goto out;
3880 
3881 	if (length > BUF_MAX_DATA_SIZE)
3882 		goto out;
3883 
3884 	if (unlikely(trace_recursive_lock(cpu_buffer)))
3885 		goto out;
3886 
3887 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
3888 	if (!event)
3889 		goto out_unlock;
3890 
3891 	body = rb_event_data(event);
3892 
3893 	memcpy(body, data, length);
3894 
3895 	rb_commit(cpu_buffer, event);
3896 
3897 	rb_wakeups(buffer, cpu_buffer);
3898 
3899 	ret = 0;
3900 
3901  out_unlock:
3902 	trace_recursive_unlock(cpu_buffer);
3903 
3904  out:
3905 	preempt_enable_notrace();
3906 
3907 	return ret;
3908 }
3909 EXPORT_SYMBOL_GPL(ring_buffer_write);
3910 
rb_per_cpu_empty(struct ring_buffer_per_cpu * cpu_buffer)3911 static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
3912 {
3913 	struct buffer_page *reader = cpu_buffer->reader_page;
3914 	struct buffer_page *head = rb_set_head_page(cpu_buffer);
3915 	struct buffer_page *commit = cpu_buffer->commit_page;
3916 
3917 	/* In case of error, head will be NULL */
3918 	if (unlikely(!head))
3919 		return true;
3920 
3921 	/* Reader should exhaust content in reader page */
3922 	if (reader->read != rb_page_commit(reader))
3923 		return false;
3924 
3925 	/*
3926 	 * If writers are committing on the reader page, knowing all
3927 	 * committed content has been read, the ring buffer is empty.
3928 	 */
3929 	if (commit == reader)
3930 		return true;
3931 
3932 	/*
3933 	 * If writers are committing on a page other than reader page
3934 	 * and head page, there should always be content to read.
3935 	 */
3936 	if (commit != head)
3937 		return false;
3938 
3939 	/*
3940 	 * Writers are committing on the head page, we just need
3941 	 * to care about there're committed data, and the reader will
3942 	 * swap reader page with head page when it is to read data.
3943 	 */
3944 	return rb_page_commit(commit) == 0;
3945 }
3946 
3947 /**
3948  * ring_buffer_record_disable - stop all writes into the buffer
3949  * @buffer: The ring buffer to stop writes to.
3950  *
3951  * This prevents all writes to the buffer. Any attempt to write
3952  * to the buffer after this will fail and return NULL.
3953  *
3954  * The caller should call synchronize_rcu() after this.
3955  */
ring_buffer_record_disable(struct trace_buffer * buffer)3956 void ring_buffer_record_disable(struct trace_buffer *buffer)
3957 {
3958 	atomic_inc(&buffer->record_disabled);
3959 }
3960 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
3961 
3962 /**
3963  * ring_buffer_record_enable - enable writes to the buffer
3964  * @buffer: The ring buffer to enable writes
3965  *
3966  * Note, multiple disables will need the same number of enables
3967  * to truly enable the writing (much like preempt_disable).
3968  */
ring_buffer_record_enable(struct trace_buffer * buffer)3969 void ring_buffer_record_enable(struct trace_buffer *buffer)
3970 {
3971 	atomic_dec(&buffer->record_disabled);
3972 }
3973 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
3974 
3975 /**
3976  * ring_buffer_record_off - stop all writes into the buffer
3977  * @buffer: The ring buffer to stop writes to.
3978  *
3979  * This prevents all writes to the buffer. Any attempt to write
3980  * to the buffer after this will fail and return NULL.
3981  *
3982  * This is different than ring_buffer_record_disable() as
3983  * it works like an on/off switch, where as the disable() version
3984  * must be paired with a enable().
3985  */
ring_buffer_record_off(struct trace_buffer * buffer)3986 void ring_buffer_record_off(struct trace_buffer *buffer)
3987 {
3988 	unsigned int rd;
3989 	unsigned int new_rd;
3990 
3991 	do {
3992 		rd = atomic_read(&buffer->record_disabled);
3993 		new_rd = rd | RB_BUFFER_OFF;
3994 	} while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
3995 }
3996 EXPORT_SYMBOL_GPL(ring_buffer_record_off);
3997 
3998 /**
3999  * ring_buffer_record_on - restart writes into the buffer
4000  * @buffer: The ring buffer to start writes to.
4001  *
4002  * This enables all writes to the buffer that was disabled by
4003  * ring_buffer_record_off().
4004  *
4005  * This is different than ring_buffer_record_enable() as
4006  * it works like an on/off switch, where as the enable() version
4007  * must be paired with a disable().
4008  */
ring_buffer_record_on(struct trace_buffer * buffer)4009 void ring_buffer_record_on(struct trace_buffer *buffer)
4010 {
4011 	unsigned int rd;
4012 	unsigned int new_rd;
4013 
4014 	do {
4015 		rd = atomic_read(&buffer->record_disabled);
4016 		new_rd = rd & ~RB_BUFFER_OFF;
4017 	} while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
4018 }
4019 EXPORT_SYMBOL_GPL(ring_buffer_record_on);
4020 
4021 /**
4022  * ring_buffer_record_is_on - return true if the ring buffer can write
4023  * @buffer: The ring buffer to see if write is enabled
4024  *
4025  * Returns true if the ring buffer is in a state that it accepts writes.
4026  */
ring_buffer_record_is_on(struct trace_buffer * buffer)4027 bool ring_buffer_record_is_on(struct trace_buffer *buffer)
4028 {
4029 	return !atomic_read(&buffer->record_disabled);
4030 }
4031 
4032 /**
4033  * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
4034  * @buffer: The ring buffer to see if write is set enabled
4035  *
4036  * Returns true if the ring buffer is set writable by ring_buffer_record_on().
4037  * Note that this does NOT mean it is in a writable state.
4038  *
4039  * It may return true when the ring buffer has been disabled by
4040  * ring_buffer_record_disable(), as that is a temporary disabling of
4041  * the ring buffer.
4042  */
ring_buffer_record_is_set_on(struct trace_buffer * buffer)4043 bool ring_buffer_record_is_set_on(struct trace_buffer *buffer)
4044 {
4045 	return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF);
4046 }
4047 
4048 /**
4049  * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
4050  * @buffer: The ring buffer to stop writes to.
4051  * @cpu: The CPU buffer to stop
4052  *
4053  * This prevents all writes to the buffer. Any attempt to write
4054  * to the buffer after this will fail and return NULL.
4055  *
4056  * The caller should call synchronize_rcu() after this.
4057  */
ring_buffer_record_disable_cpu(struct trace_buffer * buffer,int cpu)4058 void ring_buffer_record_disable_cpu(struct trace_buffer *buffer, int cpu)
4059 {
4060 	struct ring_buffer_per_cpu *cpu_buffer;
4061 
4062 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4063 		return;
4064 
4065 	cpu_buffer = buffer->buffers[cpu];
4066 	atomic_inc(&cpu_buffer->record_disabled);
4067 }
4068 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
4069 
4070 /**
4071  * ring_buffer_record_enable_cpu - enable writes to the buffer
4072  * @buffer: The ring buffer to enable writes
4073  * @cpu: The CPU to enable.
4074  *
4075  * Note, multiple disables will need the same number of enables
4076  * to truly enable the writing (much like preempt_disable).
4077  */
ring_buffer_record_enable_cpu(struct trace_buffer * buffer,int cpu)4078 void ring_buffer_record_enable_cpu(struct trace_buffer *buffer, int cpu)
4079 {
4080 	struct ring_buffer_per_cpu *cpu_buffer;
4081 
4082 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4083 		return;
4084 
4085 	cpu_buffer = buffer->buffers[cpu];
4086 	atomic_dec(&cpu_buffer->record_disabled);
4087 }
4088 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
4089 
4090 /*
4091  * The total entries in the ring buffer is the running counter
4092  * of entries entered into the ring buffer, minus the sum of
4093  * the entries read from the ring buffer and the number of
4094  * entries that were overwritten.
4095  */
4096 static inline unsigned long
rb_num_of_entries(struct ring_buffer_per_cpu * cpu_buffer)4097 rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
4098 {
4099 	return local_read(&cpu_buffer->entries) -
4100 		(local_read(&cpu_buffer->overrun) + cpu_buffer->read);
4101 }
4102 
4103 /**
4104  * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
4105  * @buffer: The ring buffer
4106  * @cpu: The per CPU buffer to read from.
4107  */
ring_buffer_oldest_event_ts(struct trace_buffer * buffer,int cpu)4108 u64 ring_buffer_oldest_event_ts(struct trace_buffer *buffer, int cpu)
4109 {
4110 	unsigned long flags;
4111 	struct ring_buffer_per_cpu *cpu_buffer;
4112 	struct buffer_page *bpage;
4113 	u64 ret = 0;
4114 
4115 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4116 		return 0;
4117 
4118 	cpu_buffer = buffer->buffers[cpu];
4119 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4120 	/*
4121 	 * if the tail is on reader_page, oldest time stamp is on the reader
4122 	 * page
4123 	 */
4124 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
4125 		bpage = cpu_buffer->reader_page;
4126 	else
4127 		bpage = rb_set_head_page(cpu_buffer);
4128 	if (bpage)
4129 		ret = bpage->page->time_stamp;
4130 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4131 
4132 	return ret;
4133 }
4134 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
4135 
4136 /**
4137  * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
4138  * @buffer: The ring buffer
4139  * @cpu: The per CPU buffer to read from.
4140  */
ring_buffer_bytes_cpu(struct trace_buffer * buffer,int cpu)4141 unsigned long ring_buffer_bytes_cpu(struct trace_buffer *buffer, int cpu)
4142 {
4143 	struct ring_buffer_per_cpu *cpu_buffer;
4144 	unsigned long ret;
4145 
4146 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4147 		return 0;
4148 
4149 	cpu_buffer = buffer->buffers[cpu];
4150 	ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
4151 
4152 	return ret;
4153 }
4154 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
4155 
4156 /**
4157  * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
4158  * @buffer: The ring buffer
4159  * @cpu: The per CPU buffer to get the entries from.
4160  */
ring_buffer_entries_cpu(struct trace_buffer * buffer,int cpu)4161 unsigned long ring_buffer_entries_cpu(struct trace_buffer *buffer, int cpu)
4162 {
4163 	struct ring_buffer_per_cpu *cpu_buffer;
4164 
4165 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4166 		return 0;
4167 
4168 	cpu_buffer = buffer->buffers[cpu];
4169 
4170 	return rb_num_of_entries(cpu_buffer);
4171 }
4172 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
4173 
4174 /**
4175  * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
4176  * buffer wrapping around (only if RB_FL_OVERWRITE is on).
4177  * @buffer: The ring buffer
4178  * @cpu: The per CPU buffer to get the number of overruns from
4179  */
ring_buffer_overrun_cpu(struct trace_buffer * buffer,int cpu)4180 unsigned long ring_buffer_overrun_cpu(struct trace_buffer *buffer, int cpu)
4181 {
4182 	struct ring_buffer_per_cpu *cpu_buffer;
4183 	unsigned long ret;
4184 
4185 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4186 		return 0;
4187 
4188 	cpu_buffer = buffer->buffers[cpu];
4189 	ret = local_read(&cpu_buffer->overrun);
4190 
4191 	return ret;
4192 }
4193 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
4194 
4195 /**
4196  * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
4197  * commits failing due to the buffer wrapping around while there are uncommitted
4198  * events, such as during an interrupt storm.
4199  * @buffer: The ring buffer
4200  * @cpu: The per CPU buffer to get the number of overruns from
4201  */
4202 unsigned long
ring_buffer_commit_overrun_cpu(struct trace_buffer * buffer,int cpu)4203 ring_buffer_commit_overrun_cpu(struct trace_buffer *buffer, int cpu)
4204 {
4205 	struct ring_buffer_per_cpu *cpu_buffer;
4206 	unsigned long ret;
4207 
4208 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4209 		return 0;
4210 
4211 	cpu_buffer = buffer->buffers[cpu];
4212 	ret = local_read(&cpu_buffer->commit_overrun);
4213 
4214 	return ret;
4215 }
4216 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
4217 
4218 /**
4219  * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
4220  * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
4221  * @buffer: The ring buffer
4222  * @cpu: The per CPU buffer to get the number of overruns from
4223  */
4224 unsigned long
ring_buffer_dropped_events_cpu(struct trace_buffer * buffer,int cpu)4225 ring_buffer_dropped_events_cpu(struct trace_buffer *buffer, int cpu)
4226 {
4227 	struct ring_buffer_per_cpu *cpu_buffer;
4228 	unsigned long ret;
4229 
4230 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4231 		return 0;
4232 
4233 	cpu_buffer = buffer->buffers[cpu];
4234 	ret = local_read(&cpu_buffer->dropped_events);
4235 
4236 	return ret;
4237 }
4238 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
4239 
4240 /**
4241  * ring_buffer_read_events_cpu - get the number of events successfully read
4242  * @buffer: The ring buffer
4243  * @cpu: The per CPU buffer to get the number of events read
4244  */
4245 unsigned long
ring_buffer_read_events_cpu(struct trace_buffer * buffer,int cpu)4246 ring_buffer_read_events_cpu(struct trace_buffer *buffer, int cpu)
4247 {
4248 	struct ring_buffer_per_cpu *cpu_buffer;
4249 
4250 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4251 		return 0;
4252 
4253 	cpu_buffer = buffer->buffers[cpu];
4254 	return cpu_buffer->read;
4255 }
4256 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
4257 
4258 /**
4259  * ring_buffer_entries - get the number of entries in a buffer
4260  * @buffer: The ring buffer
4261  *
4262  * Returns the total number of entries in the ring buffer
4263  * (all CPU entries)
4264  */
ring_buffer_entries(struct trace_buffer * buffer)4265 unsigned long ring_buffer_entries(struct trace_buffer *buffer)
4266 {
4267 	struct ring_buffer_per_cpu *cpu_buffer;
4268 	unsigned long entries = 0;
4269 	int cpu;
4270 
4271 	/* if you care about this being correct, lock the buffer */
4272 	for_each_buffer_cpu(buffer, cpu) {
4273 		cpu_buffer = buffer->buffers[cpu];
4274 		entries += rb_num_of_entries(cpu_buffer);
4275 	}
4276 
4277 	return entries;
4278 }
4279 EXPORT_SYMBOL_GPL(ring_buffer_entries);
4280 
4281 /**
4282  * ring_buffer_overruns - get the number of overruns in buffer
4283  * @buffer: The ring buffer
4284  *
4285  * Returns the total number of overruns in the ring buffer
4286  * (all CPU entries)
4287  */
ring_buffer_overruns(struct trace_buffer * buffer)4288 unsigned long ring_buffer_overruns(struct trace_buffer *buffer)
4289 {
4290 	struct ring_buffer_per_cpu *cpu_buffer;
4291 	unsigned long overruns = 0;
4292 	int cpu;
4293 
4294 	/* if you care about this being correct, lock the buffer */
4295 	for_each_buffer_cpu(buffer, cpu) {
4296 		cpu_buffer = buffer->buffers[cpu];
4297 		overruns += local_read(&cpu_buffer->overrun);
4298 	}
4299 
4300 	return overruns;
4301 }
4302 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
4303 
rb_iter_reset(struct ring_buffer_iter * iter)4304 static void rb_iter_reset(struct ring_buffer_iter *iter)
4305 {
4306 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4307 
4308 	/* Iterator usage is expected to have record disabled */
4309 	iter->head_page = cpu_buffer->reader_page;
4310 	iter->head = cpu_buffer->reader_page->read;
4311 	iter->next_event = iter->head;
4312 
4313 	iter->cache_reader_page = iter->head_page;
4314 	iter->cache_read = cpu_buffer->read;
4315 
4316 	if (iter->head) {
4317 		iter->read_stamp = cpu_buffer->read_stamp;
4318 		iter->page_stamp = cpu_buffer->reader_page->page->time_stamp;
4319 	} else {
4320 		iter->read_stamp = iter->head_page->page->time_stamp;
4321 		iter->page_stamp = iter->read_stamp;
4322 	}
4323 }
4324 
4325 /**
4326  * ring_buffer_iter_reset - reset an iterator
4327  * @iter: The iterator to reset
4328  *
4329  * Resets the iterator, so that it will start from the beginning
4330  * again.
4331  */
ring_buffer_iter_reset(struct ring_buffer_iter * iter)4332 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
4333 {
4334 	struct ring_buffer_per_cpu *cpu_buffer;
4335 	unsigned long flags;
4336 
4337 	if (!iter)
4338 		return;
4339 
4340 	cpu_buffer = iter->cpu_buffer;
4341 
4342 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4343 	rb_iter_reset(iter);
4344 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4345 }
4346 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
4347 
4348 /**
4349  * ring_buffer_iter_empty - check if an iterator has no more to read
4350  * @iter: The iterator to check
4351  */
ring_buffer_iter_empty(struct ring_buffer_iter * iter)4352 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
4353 {
4354 	struct ring_buffer_per_cpu *cpu_buffer;
4355 	struct buffer_page *reader;
4356 	struct buffer_page *head_page;
4357 	struct buffer_page *commit_page;
4358 	struct buffer_page *curr_commit_page;
4359 	unsigned commit;
4360 	u64 curr_commit_ts;
4361 	u64 commit_ts;
4362 
4363 	cpu_buffer = iter->cpu_buffer;
4364 	reader = cpu_buffer->reader_page;
4365 	head_page = cpu_buffer->head_page;
4366 	commit_page = cpu_buffer->commit_page;
4367 	commit_ts = commit_page->page->time_stamp;
4368 
4369 	/*
4370 	 * When the writer goes across pages, it issues a cmpxchg which
4371 	 * is a mb(), which will synchronize with the rmb here.
4372 	 * (see rb_tail_page_update())
4373 	 */
4374 	smp_rmb();
4375 	commit = rb_page_commit(commit_page);
4376 	/* We want to make sure that the commit page doesn't change */
4377 	smp_rmb();
4378 
4379 	/* Make sure commit page didn't change */
4380 	curr_commit_page = READ_ONCE(cpu_buffer->commit_page);
4381 	curr_commit_ts = READ_ONCE(curr_commit_page->page->time_stamp);
4382 
4383 	/* If the commit page changed, then there's more data */
4384 	if (curr_commit_page != commit_page ||
4385 	    curr_commit_ts != commit_ts)
4386 		return 0;
4387 
4388 	/* Still racy, as it may return a false positive, but that's OK */
4389 	return ((iter->head_page == commit_page && iter->head >= commit) ||
4390 		(iter->head_page == reader && commit_page == head_page &&
4391 		 head_page->read == commit &&
4392 		 iter->head == rb_page_commit(cpu_buffer->reader_page)));
4393 }
4394 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
4395 
4396 static void
rb_update_read_stamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)4397 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
4398 		     struct ring_buffer_event *event)
4399 {
4400 	u64 delta;
4401 
4402 	switch (event->type_len) {
4403 	case RINGBUF_TYPE_PADDING:
4404 		return;
4405 
4406 	case RINGBUF_TYPE_TIME_EXTEND:
4407 		delta = rb_event_time_stamp(event);
4408 		cpu_buffer->read_stamp += delta;
4409 		return;
4410 
4411 	case RINGBUF_TYPE_TIME_STAMP:
4412 		delta = rb_event_time_stamp(event);
4413 		delta = rb_fix_abs_ts(delta, cpu_buffer->read_stamp);
4414 		cpu_buffer->read_stamp = delta;
4415 		return;
4416 
4417 	case RINGBUF_TYPE_DATA:
4418 		cpu_buffer->read_stamp += event->time_delta;
4419 		return;
4420 
4421 	default:
4422 		RB_WARN_ON(cpu_buffer, 1);
4423 	}
4424 	return;
4425 }
4426 
4427 static void
rb_update_iter_read_stamp(struct ring_buffer_iter * iter,struct ring_buffer_event * event)4428 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
4429 			  struct ring_buffer_event *event)
4430 {
4431 	u64 delta;
4432 
4433 	switch (event->type_len) {
4434 	case RINGBUF_TYPE_PADDING:
4435 		return;
4436 
4437 	case RINGBUF_TYPE_TIME_EXTEND:
4438 		delta = rb_event_time_stamp(event);
4439 		iter->read_stamp += delta;
4440 		return;
4441 
4442 	case RINGBUF_TYPE_TIME_STAMP:
4443 		delta = rb_event_time_stamp(event);
4444 		delta = rb_fix_abs_ts(delta, iter->read_stamp);
4445 		iter->read_stamp = delta;
4446 		return;
4447 
4448 	case RINGBUF_TYPE_DATA:
4449 		iter->read_stamp += event->time_delta;
4450 		return;
4451 
4452 	default:
4453 		RB_WARN_ON(iter->cpu_buffer, 1);
4454 	}
4455 	return;
4456 }
4457 
4458 static struct buffer_page *
rb_get_reader_page(struct ring_buffer_per_cpu * cpu_buffer)4459 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
4460 {
4461 	struct buffer_page *reader = NULL;
4462 	unsigned long overwrite;
4463 	unsigned long flags;
4464 	int nr_loops = 0;
4465 	int ret;
4466 
4467 	local_irq_save(flags);
4468 	arch_spin_lock(&cpu_buffer->lock);
4469 
4470  again:
4471 	/*
4472 	 * This should normally only loop twice. But because the
4473 	 * start of the reader inserts an empty page, it causes
4474 	 * a case where we will loop three times. There should be no
4475 	 * reason to loop four times (that I know of).
4476 	 */
4477 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
4478 		reader = NULL;
4479 		goto out;
4480 	}
4481 
4482 	reader = cpu_buffer->reader_page;
4483 
4484 	/* If there's more to read, return this page */
4485 	if (cpu_buffer->reader_page->read < rb_page_size(reader))
4486 		goto out;
4487 
4488 	/* Never should we have an index greater than the size */
4489 	if (RB_WARN_ON(cpu_buffer,
4490 		       cpu_buffer->reader_page->read > rb_page_size(reader)))
4491 		goto out;
4492 
4493 	/* check if we caught up to the tail */
4494 	reader = NULL;
4495 	if (cpu_buffer->commit_page == cpu_buffer->reader_page)
4496 		goto out;
4497 
4498 	/* Don't bother swapping if the ring buffer is empty */
4499 	if (rb_num_of_entries(cpu_buffer) == 0)
4500 		goto out;
4501 
4502 	/*
4503 	 * Reset the reader page to size zero.
4504 	 */
4505 	local_set(&cpu_buffer->reader_page->write, 0);
4506 	local_set(&cpu_buffer->reader_page->entries, 0);
4507 	local_set(&cpu_buffer->reader_page->page->commit, 0);
4508 	cpu_buffer->reader_page->real_end = 0;
4509 
4510  spin:
4511 	/*
4512 	 * Splice the empty reader page into the list around the head.
4513 	 */
4514 	reader = rb_set_head_page(cpu_buffer);
4515 	if (!reader)
4516 		goto out;
4517 	cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
4518 	cpu_buffer->reader_page->list.prev = reader->list.prev;
4519 
4520 	/*
4521 	 * cpu_buffer->pages just needs to point to the buffer, it
4522 	 *  has no specific buffer page to point to. Lets move it out
4523 	 *  of our way so we don't accidentally swap it.
4524 	 */
4525 	cpu_buffer->pages = reader->list.prev;
4526 
4527 	/* The reader page will be pointing to the new head */
4528 	rb_set_list_to_head(&cpu_buffer->reader_page->list);
4529 
4530 	/*
4531 	 * We want to make sure we read the overruns after we set up our
4532 	 * pointers to the next object. The writer side does a
4533 	 * cmpxchg to cross pages which acts as the mb on the writer
4534 	 * side. Note, the reader will constantly fail the swap
4535 	 * while the writer is updating the pointers, so this
4536 	 * guarantees that the overwrite recorded here is the one we
4537 	 * want to compare with the last_overrun.
4538 	 */
4539 	smp_mb();
4540 	overwrite = local_read(&(cpu_buffer->overrun));
4541 
4542 	/*
4543 	 * Here's the tricky part.
4544 	 *
4545 	 * We need to move the pointer past the header page.
4546 	 * But we can only do that if a writer is not currently
4547 	 * moving it. The page before the header page has the
4548 	 * flag bit '1' set if it is pointing to the page we want.
4549 	 * but if the writer is in the process of moving it
4550 	 * than it will be '2' or already moved '0'.
4551 	 */
4552 
4553 	ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
4554 
4555 	/*
4556 	 * If we did not convert it, then we must try again.
4557 	 */
4558 	if (!ret)
4559 		goto spin;
4560 
4561 	/*
4562 	 * Yay! We succeeded in replacing the page.
4563 	 *
4564 	 * Now make the new head point back to the reader page.
4565 	 */
4566 	rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
4567 	rb_inc_page(&cpu_buffer->head_page);
4568 
4569 	local_inc(&cpu_buffer->pages_read);
4570 
4571 	/* Finally update the reader page to the new head */
4572 	cpu_buffer->reader_page = reader;
4573 	cpu_buffer->reader_page->read = 0;
4574 
4575 	if (overwrite != cpu_buffer->last_overrun) {
4576 		cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
4577 		cpu_buffer->last_overrun = overwrite;
4578 	}
4579 
4580 	goto again;
4581 
4582  out:
4583 	/* Update the read_stamp on the first event */
4584 	if (reader && reader->read == 0)
4585 		cpu_buffer->read_stamp = reader->page->time_stamp;
4586 
4587 	arch_spin_unlock(&cpu_buffer->lock);
4588 	local_irq_restore(flags);
4589 
4590 	return reader;
4591 }
4592 
rb_advance_reader(struct ring_buffer_per_cpu * cpu_buffer)4593 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
4594 {
4595 	struct ring_buffer_event *event;
4596 	struct buffer_page *reader;
4597 	unsigned length;
4598 
4599 	reader = rb_get_reader_page(cpu_buffer);
4600 
4601 	/* This function should not be called when buffer is empty */
4602 	if (RB_WARN_ON(cpu_buffer, !reader))
4603 		return;
4604 
4605 	event = rb_reader_event(cpu_buffer);
4606 
4607 	if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
4608 		cpu_buffer->read++;
4609 
4610 	rb_update_read_stamp(cpu_buffer, event);
4611 
4612 	length = rb_event_length(event);
4613 	cpu_buffer->reader_page->read += length;
4614 }
4615 
rb_advance_iter(struct ring_buffer_iter * iter)4616 static void rb_advance_iter(struct ring_buffer_iter *iter)
4617 {
4618 	struct ring_buffer_per_cpu *cpu_buffer;
4619 
4620 	cpu_buffer = iter->cpu_buffer;
4621 
4622 	/* If head == next_event then we need to jump to the next event */
4623 	if (iter->head == iter->next_event) {
4624 		/* If the event gets overwritten again, there's nothing to do */
4625 		if (rb_iter_head_event(iter) == NULL)
4626 			return;
4627 	}
4628 
4629 	iter->head = iter->next_event;
4630 
4631 	/*
4632 	 * Check if we are at the end of the buffer.
4633 	 */
4634 	if (iter->next_event >= rb_page_size(iter->head_page)) {
4635 		/* discarded commits can make the page empty */
4636 		if (iter->head_page == cpu_buffer->commit_page)
4637 			return;
4638 		rb_inc_iter(iter);
4639 		return;
4640 	}
4641 
4642 	rb_update_iter_read_stamp(iter, iter->event);
4643 }
4644 
rb_lost_events(struct ring_buffer_per_cpu * cpu_buffer)4645 static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
4646 {
4647 	return cpu_buffer->lost_events;
4648 }
4649 
4650 static struct ring_buffer_event *
rb_buffer_peek(struct ring_buffer_per_cpu * cpu_buffer,u64 * ts,unsigned long * lost_events)4651 rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
4652 	       unsigned long *lost_events)
4653 {
4654 	struct ring_buffer_event *event;
4655 	struct buffer_page *reader;
4656 	int nr_loops = 0;
4657 
4658 	if (ts)
4659 		*ts = 0;
4660  again:
4661 	/*
4662 	 * We repeat when a time extend is encountered.
4663 	 * Since the time extend is always attached to a data event,
4664 	 * we should never loop more than once.
4665 	 * (We never hit the following condition more than twice).
4666 	 */
4667 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
4668 		return NULL;
4669 
4670 	reader = rb_get_reader_page(cpu_buffer);
4671 	if (!reader)
4672 		return NULL;
4673 
4674 	event = rb_reader_event(cpu_buffer);
4675 
4676 	switch (event->type_len) {
4677 	case RINGBUF_TYPE_PADDING:
4678 		if (rb_null_event(event))
4679 			RB_WARN_ON(cpu_buffer, 1);
4680 		/*
4681 		 * Because the writer could be discarding every
4682 		 * event it creates (which would probably be bad)
4683 		 * if we were to go back to "again" then we may never
4684 		 * catch up, and will trigger the warn on, or lock
4685 		 * the box. Return the padding, and we will release
4686 		 * the current locks, and try again.
4687 		 */
4688 		return event;
4689 
4690 	case RINGBUF_TYPE_TIME_EXTEND:
4691 		/* Internal data, OK to advance */
4692 		rb_advance_reader(cpu_buffer);
4693 		goto again;
4694 
4695 	case RINGBUF_TYPE_TIME_STAMP:
4696 		if (ts) {
4697 			*ts = rb_event_time_stamp(event);
4698 			*ts = rb_fix_abs_ts(*ts, reader->page->time_stamp);
4699 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4700 							 cpu_buffer->cpu, ts);
4701 		}
4702 		/* Internal data, OK to advance */
4703 		rb_advance_reader(cpu_buffer);
4704 		goto again;
4705 
4706 	case RINGBUF_TYPE_DATA:
4707 		if (ts && !(*ts)) {
4708 			*ts = cpu_buffer->read_stamp + event->time_delta;
4709 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4710 							 cpu_buffer->cpu, ts);
4711 		}
4712 		if (lost_events)
4713 			*lost_events = rb_lost_events(cpu_buffer);
4714 		return event;
4715 
4716 	default:
4717 		RB_WARN_ON(cpu_buffer, 1);
4718 	}
4719 
4720 	return NULL;
4721 }
4722 EXPORT_SYMBOL_GPL(ring_buffer_peek);
4723 
4724 static struct ring_buffer_event *
rb_iter_peek(struct ring_buffer_iter * iter,u64 * ts)4725 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
4726 {
4727 	struct trace_buffer *buffer;
4728 	struct ring_buffer_per_cpu *cpu_buffer;
4729 	struct ring_buffer_event *event;
4730 	int nr_loops = 0;
4731 
4732 	if (ts)
4733 		*ts = 0;
4734 
4735 	cpu_buffer = iter->cpu_buffer;
4736 	buffer = cpu_buffer->buffer;
4737 
4738 	/*
4739 	 * Check if someone performed a consuming read to
4740 	 * the buffer. A consuming read invalidates the iterator
4741 	 * and we need to reset the iterator in this case.
4742 	 */
4743 	if (unlikely(iter->cache_read != cpu_buffer->read ||
4744 		     iter->cache_reader_page != cpu_buffer->reader_page))
4745 		rb_iter_reset(iter);
4746 
4747  again:
4748 	if (ring_buffer_iter_empty(iter))
4749 		return NULL;
4750 
4751 	/*
4752 	 * As the writer can mess with what the iterator is trying
4753 	 * to read, just give up if we fail to get an event after
4754 	 * three tries. The iterator is not as reliable when reading
4755 	 * the ring buffer with an active write as the consumer is.
4756 	 * Do not warn if the three failures is reached.
4757 	 */
4758 	if (++nr_loops > 3)
4759 		return NULL;
4760 
4761 	if (rb_per_cpu_empty(cpu_buffer))
4762 		return NULL;
4763 
4764 	if (iter->head >= rb_page_size(iter->head_page)) {
4765 		rb_inc_iter(iter);
4766 		goto again;
4767 	}
4768 
4769 	event = rb_iter_head_event(iter);
4770 	if (!event)
4771 		goto again;
4772 
4773 	switch (event->type_len) {
4774 	case RINGBUF_TYPE_PADDING:
4775 		if (rb_null_event(event)) {
4776 			rb_inc_iter(iter);
4777 			goto again;
4778 		}
4779 		rb_advance_iter(iter);
4780 		return event;
4781 
4782 	case RINGBUF_TYPE_TIME_EXTEND:
4783 		/* Internal data, OK to advance */
4784 		rb_advance_iter(iter);
4785 		goto again;
4786 
4787 	case RINGBUF_TYPE_TIME_STAMP:
4788 		if (ts) {
4789 			*ts = rb_event_time_stamp(event);
4790 			*ts = rb_fix_abs_ts(*ts, iter->head_page->page->time_stamp);
4791 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4792 							 cpu_buffer->cpu, ts);
4793 		}
4794 		/* Internal data, OK to advance */
4795 		rb_advance_iter(iter);
4796 		goto again;
4797 
4798 	case RINGBUF_TYPE_DATA:
4799 		if (ts && !(*ts)) {
4800 			*ts = iter->read_stamp + event->time_delta;
4801 			ring_buffer_normalize_time_stamp(buffer,
4802 							 cpu_buffer->cpu, ts);
4803 		}
4804 		return event;
4805 
4806 	default:
4807 		RB_WARN_ON(cpu_buffer, 1);
4808 	}
4809 
4810 	return NULL;
4811 }
4812 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
4813 
rb_reader_lock(struct ring_buffer_per_cpu * cpu_buffer)4814 static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
4815 {
4816 	if (likely(!in_nmi())) {
4817 		raw_spin_lock(&cpu_buffer->reader_lock);
4818 		return true;
4819 	}
4820 
4821 	/*
4822 	 * If an NMI die dumps out the content of the ring buffer
4823 	 * trylock must be used to prevent a deadlock if the NMI
4824 	 * preempted a task that holds the ring buffer locks. If
4825 	 * we get the lock then all is fine, if not, then continue
4826 	 * to do the read, but this can corrupt the ring buffer,
4827 	 * so it must be permanently disabled from future writes.
4828 	 * Reading from NMI is a oneshot deal.
4829 	 */
4830 	if (raw_spin_trylock(&cpu_buffer->reader_lock))
4831 		return true;
4832 
4833 	/* Continue without locking, but disable the ring buffer */
4834 	atomic_inc(&cpu_buffer->record_disabled);
4835 	return false;
4836 }
4837 
4838 static inline void
rb_reader_unlock(struct ring_buffer_per_cpu * cpu_buffer,bool locked)4839 rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
4840 {
4841 	if (likely(locked))
4842 		raw_spin_unlock(&cpu_buffer->reader_lock);
4843 	return;
4844 }
4845 
4846 /**
4847  * ring_buffer_peek - peek at the next event to be read
4848  * @buffer: The ring buffer to read
4849  * @cpu: The cpu to peak at
4850  * @ts: The timestamp counter of this event.
4851  * @lost_events: a variable to store if events were lost (may be NULL)
4852  *
4853  * This will return the event that will be read next, but does
4854  * not consume the data.
4855  */
4856 struct ring_buffer_event *
ring_buffer_peek(struct trace_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4857 ring_buffer_peek(struct trace_buffer *buffer, int cpu, u64 *ts,
4858 		 unsigned long *lost_events)
4859 {
4860 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4861 	struct ring_buffer_event *event;
4862 	unsigned long flags;
4863 	bool dolock;
4864 
4865 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4866 		return NULL;
4867 
4868  again:
4869 	local_irq_save(flags);
4870 	dolock = rb_reader_lock(cpu_buffer);
4871 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4872 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4873 		rb_advance_reader(cpu_buffer);
4874 	rb_reader_unlock(cpu_buffer, dolock);
4875 	local_irq_restore(flags);
4876 
4877 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4878 		goto again;
4879 
4880 	return event;
4881 }
4882 
4883 /** ring_buffer_iter_dropped - report if there are dropped events
4884  * @iter: The ring buffer iterator
4885  *
4886  * Returns true if there was dropped events since the last peek.
4887  */
ring_buffer_iter_dropped(struct ring_buffer_iter * iter)4888 bool ring_buffer_iter_dropped(struct ring_buffer_iter *iter)
4889 {
4890 	bool ret = iter->missed_events != 0;
4891 
4892 	iter->missed_events = 0;
4893 	return ret;
4894 }
4895 EXPORT_SYMBOL_GPL(ring_buffer_iter_dropped);
4896 
4897 /**
4898  * ring_buffer_iter_peek - peek at the next event to be read
4899  * @iter: The ring buffer iterator
4900  * @ts: The timestamp counter of this event.
4901  *
4902  * This will return the event that will be read next, but does
4903  * not increment the iterator.
4904  */
4905 struct ring_buffer_event *
ring_buffer_iter_peek(struct ring_buffer_iter * iter,u64 * ts)4906 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
4907 {
4908 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4909 	struct ring_buffer_event *event;
4910 	unsigned long flags;
4911 
4912  again:
4913 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4914 	event = rb_iter_peek(iter, ts);
4915 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4916 
4917 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4918 		goto again;
4919 
4920 	return event;
4921 }
4922 
4923 /**
4924  * ring_buffer_consume - return an event and consume it
4925  * @buffer: The ring buffer to get the next event from
4926  * @cpu: the cpu to read the buffer from
4927  * @ts: a variable to store the timestamp (may be NULL)
4928  * @lost_events: a variable to store if events were lost (may be NULL)
4929  *
4930  * Returns the next event in the ring buffer, and that event is consumed.
4931  * Meaning, that sequential reads will keep returning a different event,
4932  * and eventually empty the ring buffer if the producer is slower.
4933  */
4934 struct ring_buffer_event *
ring_buffer_consume(struct trace_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4935 ring_buffer_consume(struct trace_buffer *buffer, int cpu, u64 *ts,
4936 		    unsigned long *lost_events)
4937 {
4938 	struct ring_buffer_per_cpu *cpu_buffer;
4939 	struct ring_buffer_event *event = NULL;
4940 	unsigned long flags;
4941 	bool dolock;
4942 
4943  again:
4944 	/* might be called in atomic */
4945 	preempt_disable();
4946 
4947 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4948 		goto out;
4949 
4950 	cpu_buffer = buffer->buffers[cpu];
4951 	local_irq_save(flags);
4952 	dolock = rb_reader_lock(cpu_buffer);
4953 
4954 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4955 	if (event) {
4956 		cpu_buffer->lost_events = 0;
4957 		rb_advance_reader(cpu_buffer);
4958 	}
4959 
4960 	rb_reader_unlock(cpu_buffer, dolock);
4961 	local_irq_restore(flags);
4962 
4963  out:
4964 	preempt_enable();
4965 
4966 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
4967 		goto again;
4968 
4969 	return event;
4970 }
4971 EXPORT_SYMBOL_GPL(ring_buffer_consume);
4972 
4973 /**
4974  * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
4975  * @buffer: The ring buffer to read from
4976  * @cpu: The cpu buffer to iterate over
4977  * @flags: gfp flags to use for memory allocation
4978  *
4979  * This performs the initial preparations necessary to iterate
4980  * through the buffer.  Memory is allocated, buffer recording
4981  * is disabled, and the iterator pointer is returned to the caller.
4982  *
4983  * Disabling buffer recording prevents the reading from being
4984  * corrupted. This is not a consuming read, so a producer is not
4985  * expected.
4986  *
4987  * After a sequence of ring_buffer_read_prepare calls, the user is
4988  * expected to make at least one call to ring_buffer_read_prepare_sync.
4989  * Afterwards, ring_buffer_read_start is invoked to get things going
4990  * for real.
4991  *
4992  * This overall must be paired with ring_buffer_read_finish.
4993  */
4994 struct ring_buffer_iter *
ring_buffer_read_prepare(struct trace_buffer * buffer,int cpu,gfp_t flags)4995 ring_buffer_read_prepare(struct trace_buffer *buffer, int cpu, gfp_t flags)
4996 {
4997 	struct ring_buffer_per_cpu *cpu_buffer;
4998 	struct ring_buffer_iter *iter;
4999 
5000 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5001 		return NULL;
5002 
5003 	iter = kzalloc(sizeof(*iter), flags);
5004 	if (!iter)
5005 		return NULL;
5006 
5007 	iter->event = kmalloc(BUF_MAX_DATA_SIZE, flags);
5008 	if (!iter->event) {
5009 		kfree(iter);
5010 		return NULL;
5011 	}
5012 
5013 	cpu_buffer = buffer->buffers[cpu];
5014 
5015 	iter->cpu_buffer = cpu_buffer;
5016 
5017 	atomic_inc(&cpu_buffer->resize_disabled);
5018 
5019 	return iter;
5020 }
5021 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
5022 
5023 /**
5024  * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
5025  *
5026  * All previously invoked ring_buffer_read_prepare calls to prepare
5027  * iterators will be synchronized.  Afterwards, read_buffer_read_start
5028  * calls on those iterators are allowed.
5029  */
5030 void
ring_buffer_read_prepare_sync(void)5031 ring_buffer_read_prepare_sync(void)
5032 {
5033 	synchronize_rcu();
5034 }
5035 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
5036 
5037 /**
5038  * ring_buffer_read_start - start a non consuming read of the buffer
5039  * @iter: The iterator returned by ring_buffer_read_prepare
5040  *
5041  * This finalizes the startup of an iteration through the buffer.
5042  * The iterator comes from a call to ring_buffer_read_prepare and
5043  * an intervening ring_buffer_read_prepare_sync must have been
5044  * performed.
5045  *
5046  * Must be paired with ring_buffer_read_finish.
5047  */
5048 void
ring_buffer_read_start(struct ring_buffer_iter * iter)5049 ring_buffer_read_start(struct ring_buffer_iter *iter)
5050 {
5051 	struct ring_buffer_per_cpu *cpu_buffer;
5052 	unsigned long flags;
5053 
5054 	if (!iter)
5055 		return;
5056 
5057 	cpu_buffer = iter->cpu_buffer;
5058 
5059 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5060 	arch_spin_lock(&cpu_buffer->lock);
5061 	rb_iter_reset(iter);
5062 	arch_spin_unlock(&cpu_buffer->lock);
5063 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5064 }
5065 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
5066 
5067 /**
5068  * ring_buffer_read_finish - finish reading the iterator of the buffer
5069  * @iter: The iterator retrieved by ring_buffer_start
5070  *
5071  * This re-enables the recording to the buffer, and frees the
5072  * iterator.
5073  */
5074 void
ring_buffer_read_finish(struct ring_buffer_iter * iter)5075 ring_buffer_read_finish(struct ring_buffer_iter *iter)
5076 {
5077 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5078 	unsigned long flags;
5079 
5080 	/*
5081 	 * Ring buffer is disabled from recording, here's a good place
5082 	 * to check the integrity of the ring buffer.
5083 	 * Must prevent readers from trying to read, as the check
5084 	 * clears the HEAD page and readers require it.
5085 	 */
5086 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5087 	rb_check_pages(cpu_buffer);
5088 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5089 
5090 	atomic_dec(&cpu_buffer->resize_disabled);
5091 	kfree(iter->event);
5092 	kfree(iter);
5093 }
5094 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
5095 
5096 /**
5097  * ring_buffer_iter_advance - advance the iterator to the next location
5098  * @iter: The ring buffer iterator
5099  *
5100  * Move the location of the iterator such that the next read will
5101  * be the next location of the iterator.
5102  */
ring_buffer_iter_advance(struct ring_buffer_iter * iter)5103 void ring_buffer_iter_advance(struct ring_buffer_iter *iter)
5104 {
5105 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5106 	unsigned long flags;
5107 
5108 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5109 
5110 	rb_advance_iter(iter);
5111 
5112 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5113 }
5114 EXPORT_SYMBOL_GPL(ring_buffer_iter_advance);
5115 
5116 /**
5117  * ring_buffer_size - return the size of the ring buffer (in bytes)
5118  * @buffer: The ring buffer.
5119  * @cpu: The CPU to get ring buffer size from.
5120  */
ring_buffer_size(struct trace_buffer * buffer,int cpu)5121 unsigned long ring_buffer_size(struct trace_buffer *buffer, int cpu)
5122 {
5123 	/*
5124 	 * Earlier, this method returned
5125 	 *	BUF_PAGE_SIZE * buffer->nr_pages
5126 	 * Since the nr_pages field is now removed, we have converted this to
5127 	 * return the per cpu buffer value.
5128 	 */
5129 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5130 		return 0;
5131 
5132 	return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages;
5133 }
5134 EXPORT_SYMBOL_GPL(ring_buffer_size);
5135 
5136 static void
rb_reset_cpu(struct ring_buffer_per_cpu * cpu_buffer)5137 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
5138 {
5139 	rb_head_page_deactivate(cpu_buffer);
5140 
5141 	cpu_buffer->head_page
5142 		= list_entry(cpu_buffer->pages, struct buffer_page, list);
5143 	local_set(&cpu_buffer->head_page->write, 0);
5144 	local_set(&cpu_buffer->head_page->entries, 0);
5145 	local_set(&cpu_buffer->head_page->page->commit, 0);
5146 
5147 	cpu_buffer->head_page->read = 0;
5148 
5149 	cpu_buffer->tail_page = cpu_buffer->head_page;
5150 	cpu_buffer->commit_page = cpu_buffer->head_page;
5151 
5152 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
5153 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
5154 	local_set(&cpu_buffer->reader_page->write, 0);
5155 	local_set(&cpu_buffer->reader_page->entries, 0);
5156 	local_set(&cpu_buffer->reader_page->page->commit, 0);
5157 	cpu_buffer->reader_page->read = 0;
5158 
5159 	local_set(&cpu_buffer->entries_bytes, 0);
5160 	local_set(&cpu_buffer->overrun, 0);
5161 	local_set(&cpu_buffer->commit_overrun, 0);
5162 	local_set(&cpu_buffer->dropped_events, 0);
5163 	local_set(&cpu_buffer->entries, 0);
5164 	local_set(&cpu_buffer->committing, 0);
5165 	local_set(&cpu_buffer->commits, 0);
5166 	local_set(&cpu_buffer->pages_touched, 0);
5167 	local_set(&cpu_buffer->pages_read, 0);
5168 	cpu_buffer->last_pages_touch = 0;
5169 	cpu_buffer->shortest_full = 0;
5170 	cpu_buffer->read = 0;
5171 	cpu_buffer->read_bytes = 0;
5172 
5173 	rb_time_set(&cpu_buffer->write_stamp, 0);
5174 	rb_time_set(&cpu_buffer->before_stamp, 0);
5175 
5176 	memset(cpu_buffer->event_stamp, 0, sizeof(cpu_buffer->event_stamp));
5177 
5178 	cpu_buffer->lost_events = 0;
5179 	cpu_buffer->last_overrun = 0;
5180 
5181 	rb_head_page_activate(cpu_buffer);
5182 }
5183 
5184 /* Must have disabled the cpu buffer then done a synchronize_rcu */
reset_disabled_cpu_buffer(struct ring_buffer_per_cpu * cpu_buffer)5185 static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
5186 {
5187 	unsigned long flags;
5188 
5189 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5190 
5191 	if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
5192 		goto out;
5193 
5194 	arch_spin_lock(&cpu_buffer->lock);
5195 
5196 	rb_reset_cpu(cpu_buffer);
5197 
5198 	arch_spin_unlock(&cpu_buffer->lock);
5199 
5200  out:
5201 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5202 }
5203 
5204 /**
5205  * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
5206  * @buffer: The ring buffer to reset a per cpu buffer of
5207  * @cpu: The CPU buffer to be reset
5208  */
ring_buffer_reset_cpu(struct trace_buffer * buffer,int cpu)5209 void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu)
5210 {
5211 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5212 
5213 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5214 		return;
5215 
5216 	/* prevent another thread from changing buffer sizes */
5217 	mutex_lock(&buffer->mutex);
5218 
5219 	atomic_inc(&cpu_buffer->resize_disabled);
5220 	atomic_inc(&cpu_buffer->record_disabled);
5221 
5222 	/* Make sure all commits have finished */
5223 	synchronize_rcu();
5224 
5225 	reset_disabled_cpu_buffer(cpu_buffer);
5226 
5227 	atomic_dec(&cpu_buffer->record_disabled);
5228 	atomic_dec(&cpu_buffer->resize_disabled);
5229 
5230 	mutex_unlock(&buffer->mutex);
5231 }
5232 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
5233 
5234 /**
5235  * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
5236  * @buffer: The ring buffer to reset a per cpu buffer of
5237  * @cpu: The CPU buffer to be reset
5238  */
ring_buffer_reset_online_cpus(struct trace_buffer * buffer)5239 void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
5240 {
5241 	struct ring_buffer_per_cpu *cpu_buffer;
5242 	int cpu;
5243 
5244 	/* prevent another thread from changing buffer sizes */
5245 	mutex_lock(&buffer->mutex);
5246 
5247 	for_each_online_buffer_cpu(buffer, cpu) {
5248 		cpu_buffer = buffer->buffers[cpu];
5249 
5250 		atomic_inc(&cpu_buffer->resize_disabled);
5251 		atomic_inc(&cpu_buffer->record_disabled);
5252 	}
5253 
5254 	/* Make sure all commits have finished */
5255 	synchronize_rcu();
5256 
5257 	for_each_online_buffer_cpu(buffer, cpu) {
5258 		cpu_buffer = buffer->buffers[cpu];
5259 
5260 		reset_disabled_cpu_buffer(cpu_buffer);
5261 
5262 		atomic_dec(&cpu_buffer->record_disabled);
5263 		atomic_dec(&cpu_buffer->resize_disabled);
5264 	}
5265 
5266 	mutex_unlock(&buffer->mutex);
5267 }
5268 
5269 /**
5270  * ring_buffer_reset - reset a ring buffer
5271  * @buffer: The ring buffer to reset all cpu buffers
5272  */
ring_buffer_reset(struct trace_buffer * buffer)5273 void ring_buffer_reset(struct trace_buffer *buffer)
5274 {
5275 	struct ring_buffer_per_cpu *cpu_buffer;
5276 	int cpu;
5277 
5278 	/* prevent another thread from changing buffer sizes */
5279 	mutex_lock(&buffer->mutex);
5280 
5281 	for_each_buffer_cpu(buffer, cpu) {
5282 		cpu_buffer = buffer->buffers[cpu];
5283 
5284 		atomic_inc(&cpu_buffer->resize_disabled);
5285 		atomic_inc(&cpu_buffer->record_disabled);
5286 	}
5287 
5288 	/* Make sure all commits have finished */
5289 	synchronize_rcu();
5290 
5291 	for_each_buffer_cpu(buffer, cpu) {
5292 		cpu_buffer = buffer->buffers[cpu];
5293 
5294 		reset_disabled_cpu_buffer(cpu_buffer);
5295 
5296 		atomic_dec(&cpu_buffer->record_disabled);
5297 		atomic_dec(&cpu_buffer->resize_disabled);
5298 	}
5299 
5300 	mutex_unlock(&buffer->mutex);
5301 }
5302 EXPORT_SYMBOL_GPL(ring_buffer_reset);
5303 
5304 /**
5305  * rind_buffer_empty - is the ring buffer empty?
5306  * @buffer: The ring buffer to test
5307  */
ring_buffer_empty(struct trace_buffer * buffer)5308 bool ring_buffer_empty(struct trace_buffer *buffer)
5309 {
5310 	struct ring_buffer_per_cpu *cpu_buffer;
5311 	unsigned long flags;
5312 	bool dolock;
5313 	int cpu;
5314 	int ret;
5315 
5316 	/* yes this is racy, but if you don't like the race, lock the buffer */
5317 	for_each_buffer_cpu(buffer, cpu) {
5318 		cpu_buffer = buffer->buffers[cpu];
5319 		local_irq_save(flags);
5320 		dolock = rb_reader_lock(cpu_buffer);
5321 		ret = rb_per_cpu_empty(cpu_buffer);
5322 		rb_reader_unlock(cpu_buffer, dolock);
5323 		local_irq_restore(flags);
5324 
5325 		if (!ret)
5326 			return false;
5327 	}
5328 
5329 	return true;
5330 }
5331 EXPORT_SYMBOL_GPL(ring_buffer_empty);
5332 
5333 /**
5334  * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
5335  * @buffer: The ring buffer
5336  * @cpu: The CPU buffer to test
5337  */
ring_buffer_empty_cpu(struct trace_buffer * buffer,int cpu)5338 bool ring_buffer_empty_cpu(struct trace_buffer *buffer, int cpu)
5339 {
5340 	struct ring_buffer_per_cpu *cpu_buffer;
5341 	unsigned long flags;
5342 	bool dolock;
5343 	int ret;
5344 
5345 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5346 		return true;
5347 
5348 	cpu_buffer = buffer->buffers[cpu];
5349 	local_irq_save(flags);
5350 	dolock = rb_reader_lock(cpu_buffer);
5351 	ret = rb_per_cpu_empty(cpu_buffer);
5352 	rb_reader_unlock(cpu_buffer, dolock);
5353 	local_irq_restore(flags);
5354 
5355 	return ret;
5356 }
5357 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
5358 
5359 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
5360 /**
5361  * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
5362  * @buffer_a: One buffer to swap with
5363  * @buffer_b: The other buffer to swap with
5364  * @cpu: the CPU of the buffers to swap
5365  *
5366  * This function is useful for tracers that want to take a "snapshot"
5367  * of a CPU buffer and has another back up buffer lying around.
5368  * it is expected that the tracer handles the cpu buffer not being
5369  * used at the moment.
5370  */
ring_buffer_swap_cpu(struct trace_buffer * buffer_a,struct trace_buffer * buffer_b,int cpu)5371 int ring_buffer_swap_cpu(struct trace_buffer *buffer_a,
5372 			 struct trace_buffer *buffer_b, int cpu)
5373 {
5374 	struct ring_buffer_per_cpu *cpu_buffer_a;
5375 	struct ring_buffer_per_cpu *cpu_buffer_b;
5376 	int ret = -EINVAL;
5377 
5378 	if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
5379 	    !cpumask_test_cpu(cpu, buffer_b->cpumask))
5380 		goto out;
5381 
5382 	cpu_buffer_a = buffer_a->buffers[cpu];
5383 	cpu_buffer_b = buffer_b->buffers[cpu];
5384 
5385 	/* At least make sure the two buffers are somewhat the same */
5386 	if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
5387 		goto out;
5388 
5389 	ret = -EAGAIN;
5390 
5391 	if (atomic_read(&buffer_a->record_disabled))
5392 		goto out;
5393 
5394 	if (atomic_read(&buffer_b->record_disabled))
5395 		goto out;
5396 
5397 	if (atomic_read(&cpu_buffer_a->record_disabled))
5398 		goto out;
5399 
5400 	if (atomic_read(&cpu_buffer_b->record_disabled))
5401 		goto out;
5402 
5403 	/*
5404 	 * We can't do a synchronize_rcu here because this
5405 	 * function can be called in atomic context.
5406 	 * Normally this will be called from the same CPU as cpu.
5407 	 * If not it's up to the caller to protect this.
5408 	 */
5409 	atomic_inc(&cpu_buffer_a->record_disabled);
5410 	atomic_inc(&cpu_buffer_b->record_disabled);
5411 
5412 	ret = -EBUSY;
5413 	if (local_read(&cpu_buffer_a->committing))
5414 		goto out_dec;
5415 	if (local_read(&cpu_buffer_b->committing))
5416 		goto out_dec;
5417 
5418 	buffer_a->buffers[cpu] = cpu_buffer_b;
5419 	buffer_b->buffers[cpu] = cpu_buffer_a;
5420 
5421 	cpu_buffer_b->buffer = buffer_a;
5422 	cpu_buffer_a->buffer = buffer_b;
5423 
5424 	ret = 0;
5425 
5426 out_dec:
5427 	atomic_dec(&cpu_buffer_a->record_disabled);
5428 	atomic_dec(&cpu_buffer_b->record_disabled);
5429 out:
5430 	return ret;
5431 }
5432 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
5433 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
5434 
5435 /**
5436  * ring_buffer_alloc_read_page - allocate a page to read from buffer
5437  * @buffer: the buffer to allocate for.
5438  * @cpu: the cpu buffer to allocate.
5439  *
5440  * This function is used in conjunction with ring_buffer_read_page.
5441  * When reading a full page from the ring buffer, these functions
5442  * can be used to speed up the process. The calling function should
5443  * allocate a few pages first with this function. Then when it
5444  * needs to get pages from the ring buffer, it passes the result
5445  * of this function into ring_buffer_read_page, which will swap
5446  * the page that was allocated, with the read page of the buffer.
5447  *
5448  * Returns:
5449  *  The page allocated, or ERR_PTR
5450  */
ring_buffer_alloc_read_page(struct trace_buffer * buffer,int cpu)5451 void *ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
5452 {
5453 	struct ring_buffer_per_cpu *cpu_buffer;
5454 	struct buffer_data_page *bpage = NULL;
5455 	unsigned long flags;
5456 	struct page *page;
5457 
5458 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5459 		return ERR_PTR(-ENODEV);
5460 
5461 	cpu_buffer = buffer->buffers[cpu];
5462 	local_irq_save(flags);
5463 	arch_spin_lock(&cpu_buffer->lock);
5464 
5465 	if (cpu_buffer->free_page) {
5466 		bpage = cpu_buffer->free_page;
5467 		cpu_buffer->free_page = NULL;
5468 	}
5469 
5470 	arch_spin_unlock(&cpu_buffer->lock);
5471 	local_irq_restore(flags);
5472 
5473 	if (bpage)
5474 		goto out;
5475 
5476 	page = alloc_pages_node(cpu_to_node(cpu),
5477 				GFP_KERNEL | __GFP_NORETRY, 0);
5478 	if (!page)
5479 		return ERR_PTR(-ENOMEM);
5480 
5481 	bpage = page_address(page);
5482 
5483  out:
5484 	rb_init_page(bpage);
5485 
5486 	return bpage;
5487 }
5488 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
5489 
5490 /**
5491  * ring_buffer_free_read_page - free an allocated read page
5492  * @buffer: the buffer the page was allocate for
5493  * @cpu: the cpu buffer the page came from
5494  * @data: the page to free
5495  *
5496  * Free a page allocated from ring_buffer_alloc_read_page.
5497  */
ring_buffer_free_read_page(struct trace_buffer * buffer,int cpu,void * data)5498 void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, void *data)
5499 {
5500 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5501 	struct buffer_data_page *bpage = data;
5502 	struct page *page = virt_to_page(bpage);
5503 	unsigned long flags;
5504 
5505 	/* If the page is still in use someplace else, we can't reuse it */
5506 	if (page_ref_count(page) > 1)
5507 		goto out;
5508 
5509 	local_irq_save(flags);
5510 	arch_spin_lock(&cpu_buffer->lock);
5511 
5512 	if (!cpu_buffer->free_page) {
5513 		cpu_buffer->free_page = bpage;
5514 		bpage = NULL;
5515 	}
5516 
5517 	arch_spin_unlock(&cpu_buffer->lock);
5518 	local_irq_restore(flags);
5519 
5520  out:
5521 	free_page((unsigned long)bpage);
5522 }
5523 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
5524 
5525 /**
5526  * ring_buffer_read_page - extract a page from the ring buffer
5527  * @buffer: buffer to extract from
5528  * @data_page: the page to use allocated from ring_buffer_alloc_read_page
5529  * @len: amount to extract
5530  * @cpu: the cpu of the buffer to extract
5531  * @full: should the extraction only happen when the page is full.
5532  *
5533  * This function will pull out a page from the ring buffer and consume it.
5534  * @data_page must be the address of the variable that was returned
5535  * from ring_buffer_alloc_read_page. This is because the page might be used
5536  * to swap with a page in the ring buffer.
5537  *
5538  * for example:
5539  *	rpage = ring_buffer_alloc_read_page(buffer, cpu);
5540  *	if (IS_ERR(rpage))
5541  *		return PTR_ERR(rpage);
5542  *	ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
5543  *	if (ret >= 0)
5544  *		process_page(rpage, ret);
5545  *
5546  * When @full is set, the function will not return true unless
5547  * the writer is off the reader page.
5548  *
5549  * Note: it is up to the calling functions to handle sleeps and wakeups.
5550  *  The ring buffer can be used anywhere in the kernel and can not
5551  *  blindly call wake_up. The layer that uses the ring buffer must be
5552  *  responsible for that.
5553  *
5554  * Returns:
5555  *  >=0 if data has been transferred, returns the offset of consumed data.
5556  *  <0 if no data has been transferred.
5557  */
ring_buffer_read_page(struct trace_buffer * buffer,void ** data_page,size_t len,int cpu,int full)5558 int ring_buffer_read_page(struct trace_buffer *buffer,
5559 			  void **data_page, size_t len, int cpu, int full)
5560 {
5561 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5562 	struct ring_buffer_event *event;
5563 	struct buffer_data_page *bpage;
5564 	struct buffer_page *reader;
5565 	unsigned long missed_events;
5566 	unsigned long flags;
5567 	unsigned int commit;
5568 	unsigned int read;
5569 	u64 save_timestamp;
5570 	int ret = -1;
5571 
5572 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5573 		goto out;
5574 
5575 	/*
5576 	 * If len is not big enough to hold the page header, then
5577 	 * we can not copy anything.
5578 	 */
5579 	if (len <= BUF_PAGE_HDR_SIZE)
5580 		goto out;
5581 
5582 	len -= BUF_PAGE_HDR_SIZE;
5583 
5584 	if (!data_page)
5585 		goto out;
5586 
5587 	bpage = *data_page;
5588 	if (!bpage)
5589 		goto out;
5590 
5591 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5592 
5593 	reader = rb_get_reader_page(cpu_buffer);
5594 	if (!reader)
5595 		goto out_unlock;
5596 
5597 	event = rb_reader_event(cpu_buffer);
5598 
5599 	read = reader->read;
5600 	commit = rb_page_commit(reader);
5601 
5602 	/* Check if any events were dropped */
5603 	missed_events = cpu_buffer->lost_events;
5604 
5605 	/*
5606 	 * If this page has been partially read or
5607 	 * if len is not big enough to read the rest of the page or
5608 	 * a writer is still on the page, then
5609 	 * we must copy the data from the page to the buffer.
5610 	 * Otherwise, we can simply swap the page with the one passed in.
5611 	 */
5612 	if (read || (len < (commit - read)) ||
5613 	    cpu_buffer->reader_page == cpu_buffer->commit_page) {
5614 		struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
5615 		unsigned int rpos = read;
5616 		unsigned int pos = 0;
5617 		unsigned int size;
5618 
5619 		if (full)
5620 			goto out_unlock;
5621 
5622 		if (len > (commit - read))
5623 			len = (commit - read);
5624 
5625 		/* Always keep the time extend and data together */
5626 		size = rb_event_ts_length(event);
5627 
5628 		if (len < size)
5629 			goto out_unlock;
5630 
5631 		/* save the current timestamp, since the user will need it */
5632 		save_timestamp = cpu_buffer->read_stamp;
5633 
5634 		/* Need to copy one event at a time */
5635 		do {
5636 			/* We need the size of one event, because
5637 			 * rb_advance_reader only advances by one event,
5638 			 * whereas rb_event_ts_length may include the size of
5639 			 * one or two events.
5640 			 * We have already ensured there's enough space if this
5641 			 * is a time extend. */
5642 			size = rb_event_length(event);
5643 			memcpy(bpage->data + pos, rpage->data + rpos, size);
5644 
5645 			len -= size;
5646 
5647 			rb_advance_reader(cpu_buffer);
5648 			rpos = reader->read;
5649 			pos += size;
5650 
5651 			if (rpos >= commit)
5652 				break;
5653 
5654 			event = rb_reader_event(cpu_buffer);
5655 			/* Always keep the time extend and data together */
5656 			size = rb_event_ts_length(event);
5657 		} while (len >= size);
5658 
5659 		/* update bpage */
5660 		local_set(&bpage->commit, pos);
5661 		bpage->time_stamp = save_timestamp;
5662 
5663 		/* we copied everything to the beginning */
5664 		read = 0;
5665 	} else {
5666 		/* update the entry counter */
5667 		cpu_buffer->read += rb_page_entries(reader);
5668 		cpu_buffer->read_bytes += BUF_PAGE_SIZE;
5669 
5670 		/* swap the pages */
5671 		rb_init_page(bpage);
5672 		bpage = reader->page;
5673 		reader->page = *data_page;
5674 		local_set(&reader->write, 0);
5675 		local_set(&reader->entries, 0);
5676 		reader->read = 0;
5677 		*data_page = bpage;
5678 
5679 		/*
5680 		 * Use the real_end for the data size,
5681 		 * This gives us a chance to store the lost events
5682 		 * on the page.
5683 		 */
5684 		if (reader->real_end)
5685 			local_set(&bpage->commit, reader->real_end);
5686 	}
5687 	ret = read;
5688 
5689 	cpu_buffer->lost_events = 0;
5690 
5691 	commit = local_read(&bpage->commit);
5692 	/*
5693 	 * Set a flag in the commit field if we lost events
5694 	 */
5695 	if (missed_events) {
5696 		/* If there is room at the end of the page to save the
5697 		 * missed events, then record it there.
5698 		 */
5699 		if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
5700 			memcpy(&bpage->data[commit], &missed_events,
5701 			       sizeof(missed_events));
5702 			local_add(RB_MISSED_STORED, &bpage->commit);
5703 			commit += sizeof(missed_events);
5704 		}
5705 		local_add(RB_MISSED_EVENTS, &bpage->commit);
5706 	}
5707 
5708 	/*
5709 	 * This page may be off to user land. Zero it out here.
5710 	 */
5711 	if (commit < BUF_PAGE_SIZE)
5712 		memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
5713 
5714  out_unlock:
5715 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5716 
5717  out:
5718 	return ret;
5719 }
5720 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
5721 
5722 /*
5723  * We only allocate new buffers, never free them if the CPU goes down.
5724  * If we were to free the buffer, then the user would lose any trace that was in
5725  * the buffer.
5726  */
trace_rb_cpu_prepare(unsigned int cpu,struct hlist_node * node)5727 int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
5728 {
5729 	struct trace_buffer *buffer;
5730 	long nr_pages_same;
5731 	int cpu_i;
5732 	unsigned long nr_pages;
5733 
5734 	buffer = container_of(node, struct trace_buffer, node);
5735 	if (cpumask_test_cpu(cpu, buffer->cpumask))
5736 		return 0;
5737 
5738 	nr_pages = 0;
5739 	nr_pages_same = 1;
5740 	/* check if all cpu sizes are same */
5741 	for_each_buffer_cpu(buffer, cpu_i) {
5742 		/* fill in the size from first enabled cpu */
5743 		if (nr_pages == 0)
5744 			nr_pages = buffer->buffers[cpu_i]->nr_pages;
5745 		if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
5746 			nr_pages_same = 0;
5747 			break;
5748 		}
5749 	}
5750 	/* allocate minimum pages, user can later expand it */
5751 	if (!nr_pages_same)
5752 		nr_pages = 2;
5753 	buffer->buffers[cpu] =
5754 		rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
5755 	if (!buffer->buffers[cpu]) {
5756 		WARN(1, "failed to allocate ring buffer on CPU %u\n",
5757 		     cpu);
5758 		return -ENOMEM;
5759 	}
5760 	smp_wmb();
5761 	cpumask_set_cpu(cpu, buffer->cpumask);
5762 	return 0;
5763 }
5764 
5765 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
5766 /*
5767  * This is a basic integrity check of the ring buffer.
5768  * Late in the boot cycle this test will run when configured in.
5769  * It will kick off a thread per CPU that will go into a loop
5770  * writing to the per cpu ring buffer various sizes of data.
5771  * Some of the data will be large items, some small.
5772  *
5773  * Another thread is created that goes into a spin, sending out
5774  * IPIs to the other CPUs to also write into the ring buffer.
5775  * this is to test the nesting ability of the buffer.
5776  *
5777  * Basic stats are recorded and reported. If something in the
5778  * ring buffer should happen that's not expected, a big warning
5779  * is displayed and all ring buffers are disabled.
5780  */
5781 static struct task_struct *rb_threads[NR_CPUS] __initdata;
5782 
5783 struct rb_test_data {
5784 	struct trace_buffer *buffer;
5785 	unsigned long		events;
5786 	unsigned long		bytes_written;
5787 	unsigned long		bytes_alloc;
5788 	unsigned long		bytes_dropped;
5789 	unsigned long		events_nested;
5790 	unsigned long		bytes_written_nested;
5791 	unsigned long		bytes_alloc_nested;
5792 	unsigned long		bytes_dropped_nested;
5793 	int			min_size_nested;
5794 	int			max_size_nested;
5795 	int			max_size;
5796 	int			min_size;
5797 	int			cpu;
5798 	int			cnt;
5799 };
5800 
5801 static struct rb_test_data rb_data[NR_CPUS] __initdata;
5802 
5803 /* 1 meg per cpu */
5804 #define RB_TEST_BUFFER_SIZE	1048576
5805 
5806 static char rb_string[] __initdata =
5807 	"abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
5808 	"?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
5809 	"!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
5810 
5811 static bool rb_test_started __initdata;
5812 
5813 struct rb_item {
5814 	int size;
5815 	char str[];
5816 };
5817 
rb_write_something(struct rb_test_data * data,bool nested)5818 static __init int rb_write_something(struct rb_test_data *data, bool nested)
5819 {
5820 	struct ring_buffer_event *event;
5821 	struct rb_item *item;
5822 	bool started;
5823 	int event_len;
5824 	int size;
5825 	int len;
5826 	int cnt;
5827 
5828 	/* Have nested writes different that what is written */
5829 	cnt = data->cnt + (nested ? 27 : 0);
5830 
5831 	/* Multiply cnt by ~e, to make some unique increment */
5832 	size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
5833 
5834 	len = size + sizeof(struct rb_item);
5835 
5836 	started = rb_test_started;
5837 	/* read rb_test_started before checking buffer enabled */
5838 	smp_rmb();
5839 
5840 	event = ring_buffer_lock_reserve(data->buffer, len);
5841 	if (!event) {
5842 		/* Ignore dropped events before test starts. */
5843 		if (started) {
5844 			if (nested)
5845 				data->bytes_dropped += len;
5846 			else
5847 				data->bytes_dropped_nested += len;
5848 		}
5849 		return len;
5850 	}
5851 
5852 	event_len = ring_buffer_event_length(event);
5853 
5854 	if (RB_WARN_ON(data->buffer, event_len < len))
5855 		goto out;
5856 
5857 	item = ring_buffer_event_data(event);
5858 	item->size = size;
5859 	memcpy(item->str, rb_string, size);
5860 
5861 	if (nested) {
5862 		data->bytes_alloc_nested += event_len;
5863 		data->bytes_written_nested += len;
5864 		data->events_nested++;
5865 		if (!data->min_size_nested || len < data->min_size_nested)
5866 			data->min_size_nested = len;
5867 		if (len > data->max_size_nested)
5868 			data->max_size_nested = len;
5869 	} else {
5870 		data->bytes_alloc += event_len;
5871 		data->bytes_written += len;
5872 		data->events++;
5873 		if (!data->min_size || len < data->min_size)
5874 			data->max_size = len;
5875 		if (len > data->max_size)
5876 			data->max_size = len;
5877 	}
5878 
5879  out:
5880 	ring_buffer_unlock_commit(data->buffer, event);
5881 
5882 	return 0;
5883 }
5884 
rb_test(void * arg)5885 static __init int rb_test(void *arg)
5886 {
5887 	struct rb_test_data *data = arg;
5888 
5889 	while (!kthread_should_stop()) {
5890 		rb_write_something(data, false);
5891 		data->cnt++;
5892 
5893 		set_current_state(TASK_INTERRUPTIBLE);
5894 		/* Now sleep between a min of 100-300us and a max of 1ms */
5895 		usleep_range(((data->cnt % 3) + 1) * 100, 1000);
5896 	}
5897 
5898 	return 0;
5899 }
5900 
rb_ipi(void * ignore)5901 static __init void rb_ipi(void *ignore)
5902 {
5903 	struct rb_test_data *data;
5904 	int cpu = smp_processor_id();
5905 
5906 	data = &rb_data[cpu];
5907 	rb_write_something(data, true);
5908 }
5909 
rb_hammer_test(void * arg)5910 static __init int rb_hammer_test(void *arg)
5911 {
5912 	while (!kthread_should_stop()) {
5913 
5914 		/* Send an IPI to all cpus to write data! */
5915 		smp_call_function(rb_ipi, NULL, 1);
5916 		/* No sleep, but for non preempt, let others run */
5917 		schedule();
5918 	}
5919 
5920 	return 0;
5921 }
5922 
test_ringbuffer(void)5923 static __init int test_ringbuffer(void)
5924 {
5925 	struct task_struct *rb_hammer;
5926 	struct trace_buffer *buffer;
5927 	int cpu;
5928 	int ret = 0;
5929 
5930 	if (security_locked_down(LOCKDOWN_TRACEFS)) {
5931 		pr_warn("Lockdown is enabled, skipping ring buffer tests\n");
5932 		return 0;
5933 	}
5934 
5935 	pr_info("Running ring buffer tests...\n");
5936 
5937 	buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
5938 	if (WARN_ON(!buffer))
5939 		return 0;
5940 
5941 	/* Disable buffer so that threads can't write to it yet */
5942 	ring_buffer_record_off(buffer);
5943 
5944 	for_each_online_cpu(cpu) {
5945 		rb_data[cpu].buffer = buffer;
5946 		rb_data[cpu].cpu = cpu;
5947 		rb_data[cpu].cnt = cpu;
5948 		rb_threads[cpu] = kthread_run_on_cpu(rb_test, &rb_data[cpu],
5949 						     cpu, "rbtester/%u");
5950 		if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
5951 			pr_cont("FAILED\n");
5952 			ret = PTR_ERR(rb_threads[cpu]);
5953 			goto out_free;
5954 		}
5955 	}
5956 
5957 	/* Now create the rb hammer! */
5958 	rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
5959 	if (WARN_ON(IS_ERR(rb_hammer))) {
5960 		pr_cont("FAILED\n");
5961 		ret = PTR_ERR(rb_hammer);
5962 		goto out_free;
5963 	}
5964 
5965 	ring_buffer_record_on(buffer);
5966 	/*
5967 	 * Show buffer is enabled before setting rb_test_started.
5968 	 * Yes there's a small race window where events could be
5969 	 * dropped and the thread wont catch it. But when a ring
5970 	 * buffer gets enabled, there will always be some kind of
5971 	 * delay before other CPUs see it. Thus, we don't care about
5972 	 * those dropped events. We care about events dropped after
5973 	 * the threads see that the buffer is active.
5974 	 */
5975 	smp_wmb();
5976 	rb_test_started = true;
5977 
5978 	set_current_state(TASK_INTERRUPTIBLE);
5979 	/* Just run for 10 seconds */;
5980 	schedule_timeout(10 * HZ);
5981 
5982 	kthread_stop(rb_hammer);
5983 
5984  out_free:
5985 	for_each_online_cpu(cpu) {
5986 		if (!rb_threads[cpu])
5987 			break;
5988 		kthread_stop(rb_threads[cpu]);
5989 	}
5990 	if (ret) {
5991 		ring_buffer_free(buffer);
5992 		return ret;
5993 	}
5994 
5995 	/* Report! */
5996 	pr_info("finished\n");
5997 	for_each_online_cpu(cpu) {
5998 		struct ring_buffer_event *event;
5999 		struct rb_test_data *data = &rb_data[cpu];
6000 		struct rb_item *item;
6001 		unsigned long total_events;
6002 		unsigned long total_dropped;
6003 		unsigned long total_written;
6004 		unsigned long total_alloc;
6005 		unsigned long total_read = 0;
6006 		unsigned long total_size = 0;
6007 		unsigned long total_len = 0;
6008 		unsigned long total_lost = 0;
6009 		unsigned long lost;
6010 		int big_event_size;
6011 		int small_event_size;
6012 
6013 		ret = -1;
6014 
6015 		total_events = data->events + data->events_nested;
6016 		total_written = data->bytes_written + data->bytes_written_nested;
6017 		total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
6018 		total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
6019 
6020 		big_event_size = data->max_size + data->max_size_nested;
6021 		small_event_size = data->min_size + data->min_size_nested;
6022 
6023 		pr_info("CPU %d:\n", cpu);
6024 		pr_info("              events:    %ld\n", total_events);
6025 		pr_info("       dropped bytes:    %ld\n", total_dropped);
6026 		pr_info("       alloced bytes:    %ld\n", total_alloc);
6027 		pr_info("       written bytes:    %ld\n", total_written);
6028 		pr_info("       biggest event:    %d\n", big_event_size);
6029 		pr_info("      smallest event:    %d\n", small_event_size);
6030 
6031 		if (RB_WARN_ON(buffer, total_dropped))
6032 			break;
6033 
6034 		ret = 0;
6035 
6036 		while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
6037 			total_lost += lost;
6038 			item = ring_buffer_event_data(event);
6039 			total_len += ring_buffer_event_length(event);
6040 			total_size += item->size + sizeof(struct rb_item);
6041 			if (memcmp(&item->str[0], rb_string, item->size) != 0) {
6042 				pr_info("FAILED!\n");
6043 				pr_info("buffer had: %.*s\n", item->size, item->str);
6044 				pr_info("expected:   %.*s\n", item->size, rb_string);
6045 				RB_WARN_ON(buffer, 1);
6046 				ret = -1;
6047 				break;
6048 			}
6049 			total_read++;
6050 		}
6051 		if (ret)
6052 			break;
6053 
6054 		ret = -1;
6055 
6056 		pr_info("         read events:   %ld\n", total_read);
6057 		pr_info("         lost events:   %ld\n", total_lost);
6058 		pr_info("        total events:   %ld\n", total_lost + total_read);
6059 		pr_info("  recorded len bytes:   %ld\n", total_len);
6060 		pr_info(" recorded size bytes:   %ld\n", total_size);
6061 		if (total_lost) {
6062 			pr_info(" With dropped events, record len and size may not match\n"
6063 				" alloced and written from above\n");
6064 		} else {
6065 			if (RB_WARN_ON(buffer, total_len != total_alloc ||
6066 				       total_size != total_written))
6067 				break;
6068 		}
6069 		if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
6070 			break;
6071 
6072 		ret = 0;
6073 	}
6074 	if (!ret)
6075 		pr_info("Ring buffer PASSED!\n");
6076 
6077 	ring_buffer_free(buffer);
6078 	return 0;
6079 }
6080 
6081 late_initcall(test_ringbuffer);
6082 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */
6083