xref: /linux/kernel/trace/ring_buffer.c (revision 8aa76aa415897f6c1ba47d9f131fa463499c4169)
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/sched/isolation.h>
8 #include <linux/trace_recursion.h>
9 #include <linux/trace_events.h>
10 #include <linux/ring_buffer.h>
11 #include <linux/trace_clock.h>
12 #include <linux/sched/clock.h>
13 #include <linux/cacheflush.h>
14 #include <linux/trace_seq.h>
15 #include <linux/spinlock.h>
16 #include <linux/irq_work.h>
17 #include <linux/security.h>
18 #include <linux/uaccess.h>
19 #include <linux/hardirq.h>
20 #include <linux/kthread.h>	/* for self test */
21 #include <linux/module.h>
22 #include <linux/percpu.h>
23 #include <linux/mutex.h>
24 #include <linux/delay.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/hash.h>
28 #include <linux/list.h>
29 #include <linux/cpu.h>
30 #include <linux/oom.h>
31 #include <linux/mm.h>
32 
33 #include <asm/local64.h>
34 #include <asm/local.h>
35 #include <asm/setup.h>
36 
37 #include "trace.h"
38 
39 /*
40  * The "absolute" timestamp in the buffer is only 59 bits.
41  * If a clock has the 5 MSBs set, it needs to be saved and
42  * reinserted.
43  */
44 #define TS_MSB		(0xf8ULL << 56)
45 #define ABS_TS_MASK	(~TS_MSB)
46 
47 static void update_pages_handler(struct work_struct *work);
48 
49 #define RING_BUFFER_META_MAGIC	0xBADFEED
50 
51 struct ring_buffer_meta {
52 	int		magic;
53 	int		struct_sizes;
54 	unsigned long	total_size;
55 	unsigned long	buffers_offset;
56 };
57 
58 struct ring_buffer_cpu_meta {
59 	unsigned long	first_buffer;
60 	unsigned long	head_buffer;
61 	unsigned long	commit_buffer;
62 	__u32		subbuf_size;
63 	__u32		nr_subbufs;
64 	int		buffers[];
65 };
66 
67 /*
68  * The ring buffer header is special. We must manually up keep it.
69  */
70 int ring_buffer_print_entry_header(struct trace_seq *s)
71 {
72 	trace_seq_puts(s, "# compressed entry header\n");
73 	trace_seq_puts(s, "\ttype_len    :    5 bits\n");
74 	trace_seq_puts(s, "\ttime_delta  :   27 bits\n");
75 	trace_seq_puts(s, "\tarray       :   32 bits\n");
76 	trace_seq_putc(s, '\n');
77 	trace_seq_printf(s, "\tpadding     : type == %d\n",
78 			 RINGBUF_TYPE_PADDING);
79 	trace_seq_printf(s, "\ttime_extend : type == %d\n",
80 			 RINGBUF_TYPE_TIME_EXTEND);
81 	trace_seq_printf(s, "\ttime_stamp : type == %d\n",
82 			 RINGBUF_TYPE_TIME_STAMP);
83 	trace_seq_printf(s, "\tdata max type_len  == %d\n",
84 			 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
85 
86 	return !trace_seq_has_overflowed(s);
87 }
88 
89 /*
90  * The ring buffer is made up of a list of pages. A separate list of pages is
91  * allocated for each CPU. A writer may only write to a buffer that is
92  * associated with the CPU it is currently executing on.  A reader may read
93  * from any per cpu buffer.
94  *
95  * The reader is special. For each per cpu buffer, the reader has its own
96  * reader page. When a reader has read the entire reader page, this reader
97  * page is swapped with another page in the ring buffer.
98  *
99  * Now, as long as the writer is off the reader page, the reader can do what
100  * ever it wants with that page. The writer will never write to that page
101  * again (as long as it is out of the ring buffer).
102  *
103  * Here's some silly ASCII art.
104  *
105  *   +------+
106  *   |reader|          RING BUFFER
107  *   |page  |
108  *   +------+        +---+   +---+   +---+
109  *                   |   |-->|   |-->|   |
110  *                   +---+   +---+   +---+
111  *                     ^               |
112  *                     |               |
113  *                     +---------------+
114  *
115  *
116  *   +------+
117  *   |reader|          RING BUFFER
118  *   |page  |------------------v
119  *   +------+        +---+   +---+   +---+
120  *                   |   |-->|   |-->|   |
121  *                   +---+   +---+   +---+
122  *                     ^               |
123  *                     |               |
124  *                     +---------------+
125  *
126  *
127  *   +------+
128  *   |reader|          RING BUFFER
129  *   |page  |------------------v
130  *   +------+        +---+   +---+   +---+
131  *      ^            |   |-->|   |-->|   |
132  *      |            +---+   +---+   +---+
133  *      |                              |
134  *      |                              |
135  *      +------------------------------+
136  *
137  *
138  *   +------+
139  *   |buffer|          RING BUFFER
140  *   |page  |------------------v
141  *   +------+        +---+   +---+   +---+
142  *      ^            |   |   |   |-->|   |
143  *      |   New      +---+   +---+   +---+
144  *      |  Reader------^               |
145  *      |   page                       |
146  *      +------------------------------+
147  *
148  *
149  * After we make this swap, the reader can hand this page off to the splice
150  * code and be done with it. It can even allocate a new page if it needs to
151  * and swap that into the ring buffer.
152  *
153  * We will be using cmpxchg soon to make all this lockless.
154  *
155  */
156 
157 /* Used for individual buffers (after the counter) */
158 #define RB_BUFFER_OFF		(1 << 20)
159 
160 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
161 
162 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
163 #define RB_ALIGNMENT		4U
164 #define RB_MAX_SMALL_DATA	(RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
165 #define RB_EVNT_MIN_SIZE	8U	/* two 32bit words */
166 
167 #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
168 # define RB_FORCE_8BYTE_ALIGNMENT	0
169 # define RB_ARCH_ALIGNMENT		RB_ALIGNMENT
170 #else
171 # define RB_FORCE_8BYTE_ALIGNMENT	1
172 # define RB_ARCH_ALIGNMENT		8U
173 #endif
174 
175 #define RB_ALIGN_DATA		__aligned(RB_ARCH_ALIGNMENT)
176 
177 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
178 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
179 
180 enum {
181 	RB_LEN_TIME_EXTEND = 8,
182 	RB_LEN_TIME_STAMP =  8,
183 };
184 
185 #define skip_time_extend(event) \
186 	((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
187 
188 #define extended_time(event) \
189 	(event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
190 
191 static inline bool rb_null_event(struct ring_buffer_event *event)
192 {
193 	return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
194 }
195 
196 static void rb_event_set_padding(struct ring_buffer_event *event)
197 {
198 	/* padding has a NULL time_delta */
199 	event->type_len = RINGBUF_TYPE_PADDING;
200 	event->time_delta = 0;
201 }
202 
203 static unsigned
204 rb_event_data_length(struct ring_buffer_event *event)
205 {
206 	unsigned length;
207 
208 	if (event->type_len)
209 		length = event->type_len * RB_ALIGNMENT;
210 	else
211 		length = event->array[0];
212 	return length + RB_EVNT_HDR_SIZE;
213 }
214 
215 /*
216  * Return the length of the given event. Will return
217  * the length of the time extend if the event is a
218  * time extend.
219  */
220 static inline unsigned
221 rb_event_length(struct ring_buffer_event *event)
222 {
223 	switch (event->type_len) {
224 	case RINGBUF_TYPE_PADDING:
225 		if (rb_null_event(event))
226 			/* undefined */
227 			return -1;
228 		return  event->array[0] + RB_EVNT_HDR_SIZE;
229 
230 	case RINGBUF_TYPE_TIME_EXTEND:
231 		return RB_LEN_TIME_EXTEND;
232 
233 	case RINGBUF_TYPE_TIME_STAMP:
234 		return RB_LEN_TIME_STAMP;
235 
236 	case RINGBUF_TYPE_DATA:
237 		return rb_event_data_length(event);
238 	default:
239 		WARN_ON_ONCE(1);
240 	}
241 	/* not hit */
242 	return 0;
243 }
244 
245 /*
246  * Return total length of time extend and data,
247  *   or just the event length for all other events.
248  */
249 static inline unsigned
250 rb_event_ts_length(struct ring_buffer_event *event)
251 {
252 	unsigned len = 0;
253 
254 	if (extended_time(event)) {
255 		/* time extends include the data event after it */
256 		len = RB_LEN_TIME_EXTEND;
257 		event = skip_time_extend(event);
258 	}
259 	return len + rb_event_length(event);
260 }
261 
262 /**
263  * ring_buffer_event_length - return the length of the event
264  * @event: the event to get the length of
265  *
266  * Returns the size of the data load of a data event.
267  * If the event is something other than a data event, it
268  * returns the size of the event itself. With the exception
269  * of a TIME EXTEND, where it still returns the size of the
270  * data load of the data event after it.
271  */
272 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
273 {
274 	unsigned length;
275 
276 	if (extended_time(event))
277 		event = skip_time_extend(event);
278 
279 	length = rb_event_length(event);
280 	if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
281 		return length;
282 	length -= RB_EVNT_HDR_SIZE;
283 	if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
284                 length -= sizeof(event->array[0]);
285 	return length;
286 }
287 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
288 
289 /* inline for ring buffer fast paths */
290 static __always_inline void *
291 rb_event_data(struct ring_buffer_event *event)
292 {
293 	if (extended_time(event))
294 		event = skip_time_extend(event);
295 	WARN_ON_ONCE(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
296 	/* If length is in len field, then array[0] has the data */
297 	if (event->type_len)
298 		return (void *)&event->array[0];
299 	/* Otherwise length is in array[0] and array[1] has the data */
300 	return (void *)&event->array[1];
301 }
302 
303 /**
304  * ring_buffer_event_data - return the data of the event
305  * @event: the event to get the data from
306  */
307 void *ring_buffer_event_data(struct ring_buffer_event *event)
308 {
309 	return rb_event_data(event);
310 }
311 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
312 
313 #define for_each_buffer_cpu(buffer, cpu)		\
314 	for_each_cpu(cpu, buffer->cpumask)
315 
316 #define for_each_online_buffer_cpu(buffer, cpu)		\
317 	for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask)
318 
319 #define TS_SHIFT	27
320 #define TS_MASK		((1ULL << TS_SHIFT) - 1)
321 #define TS_DELTA_TEST	(~TS_MASK)
322 
323 static u64 rb_event_time_stamp(struct ring_buffer_event *event)
324 {
325 	u64 ts;
326 
327 	ts = event->array[0];
328 	ts <<= TS_SHIFT;
329 	ts += event->time_delta;
330 
331 	return ts;
332 }
333 
334 /* Flag when events were overwritten */
335 #define RB_MISSED_EVENTS	(1 << 31)
336 /* Missed count stored at end */
337 #define RB_MISSED_STORED	(1 << 30)
338 
339 #define RB_MISSED_MASK		(3 << 30)
340 
341 struct buffer_data_page {
342 	u64		 time_stamp;	/* page time stamp */
343 	local_t		 commit;	/* write committed index */
344 	unsigned char	 data[] RB_ALIGN_DATA;	/* data of buffer page */
345 };
346 
347 struct buffer_data_read_page {
348 	unsigned		order;	/* order of the page */
349 	struct buffer_data_page	*data;	/* actual data, stored in this page */
350 };
351 
352 /*
353  * Note, the buffer_page list must be first. The buffer pages
354  * are allocated in cache lines, which means that each buffer
355  * page will be at the beginning of a cache line, and thus
356  * the least significant bits will be zero. We use this to
357  * add flags in the list struct pointers, to make the ring buffer
358  * lockless.
359  */
360 struct buffer_page {
361 	struct list_head list;		/* list of buffer pages */
362 	local_t		 write;		/* index for next write */
363 	unsigned	 read;		/* index for next read */
364 	local_t		 entries;	/* entries on this page */
365 	unsigned long	 real_end;	/* real end of data */
366 	unsigned	 order;		/* order of the page */
367 	u32		 id:30;		/* ID for external mapping */
368 	u32		 range:1;	/* Mapped via a range */
369 	struct buffer_data_page *page;	/* Actual data page */
370 };
371 
372 /*
373  * The buffer page counters, write and entries, must be reset
374  * atomically when crossing page boundaries. To synchronize this
375  * update, two counters are inserted into the number. One is
376  * the actual counter for the write position or count on the page.
377  *
378  * The other is a counter of updaters. Before an update happens
379  * the update partition of the counter is incremented. This will
380  * allow the updater to update the counter atomically.
381  *
382  * The counter is 20 bits, and the state data is 12.
383  */
384 #define RB_WRITE_MASK		0xfffff
385 #define RB_WRITE_INTCNT		(1 << 20)
386 
387 static void rb_init_page(struct buffer_data_page *bpage)
388 {
389 	local_set(&bpage->commit, 0);
390 }
391 
392 static __always_inline unsigned int rb_page_commit(struct buffer_page *bpage)
393 {
394 	return local_read(&bpage->page->commit);
395 }
396 
397 static void free_buffer_page(struct buffer_page *bpage)
398 {
399 	/* Range pages are not to be freed */
400 	if (!bpage->range)
401 		free_pages((unsigned long)bpage->page, bpage->order);
402 	kfree(bpage);
403 }
404 
405 /*
406  * For best performance, allocate cpu buffer data cache line sized
407  * and per CPU.
408  */
409 #define alloc_cpu_buffer(cpu) (struct ring_buffer_per_cpu *)		\
410 	kzalloc_node(ALIGN(sizeof(struct ring_buffer_per_cpu),		\
411 			   cache_line_size()), GFP_KERNEL, cpu_to_node(cpu));
412 
413 #define alloc_cpu_page(cpu) (struct buffer_page *)			\
414 	kzalloc_node(ALIGN(sizeof(struct buffer_page),			\
415 			   cache_line_size()), GFP_KERNEL, cpu_to_node(cpu));
416 
417 static struct buffer_data_page *alloc_cpu_data(int cpu, int order)
418 {
419 	struct buffer_data_page *dpage;
420 	struct page *page;
421 	gfp_t mflags;
422 
423 	/*
424 	 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
425 	 * gracefully without invoking oom-killer and the system is not
426 	 * destabilized.
427 	 */
428 	mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_COMP | __GFP_ZERO;
429 
430 	page = alloc_pages_node(cpu_to_node(cpu), mflags, order);
431 	if (!page)
432 		return NULL;
433 
434 	dpage = page_address(page);
435 	rb_init_page(dpage);
436 
437 	return dpage;
438 }
439 
440 /*
441  * We need to fit the time_stamp delta into 27 bits.
442  */
443 static inline bool test_time_stamp(u64 delta)
444 {
445 	return !!(delta & TS_DELTA_TEST);
446 }
447 
448 struct rb_irq_work {
449 	struct irq_work			work;
450 	wait_queue_head_t		waiters;
451 	wait_queue_head_t		full_waiters;
452 	atomic_t			seq;
453 	bool				waiters_pending;
454 	bool				full_waiters_pending;
455 	bool				wakeup_full;
456 };
457 
458 /*
459  * Structure to hold event state and handle nested events.
460  */
461 struct rb_event_info {
462 	u64			ts;
463 	u64			delta;
464 	u64			before;
465 	u64			after;
466 	unsigned long		length;
467 	struct buffer_page	*tail_page;
468 	int			add_timestamp;
469 };
470 
471 /*
472  * Used for the add_timestamp
473  *  NONE
474  *  EXTEND - wants a time extend
475  *  ABSOLUTE - the buffer requests all events to have absolute time stamps
476  *  FORCE - force a full time stamp.
477  */
478 enum {
479 	RB_ADD_STAMP_NONE		= 0,
480 	RB_ADD_STAMP_EXTEND		= BIT(1),
481 	RB_ADD_STAMP_ABSOLUTE		= BIT(2),
482 	RB_ADD_STAMP_FORCE		= BIT(3)
483 };
484 /*
485  * Used for which event context the event is in.
486  *  TRANSITION = 0
487  *  NMI     = 1
488  *  IRQ     = 2
489  *  SOFTIRQ = 3
490  *  NORMAL  = 4
491  *
492  * See trace_recursive_lock() comment below for more details.
493  */
494 enum {
495 	RB_CTX_TRANSITION,
496 	RB_CTX_NMI,
497 	RB_CTX_IRQ,
498 	RB_CTX_SOFTIRQ,
499 	RB_CTX_NORMAL,
500 	RB_CTX_MAX
501 };
502 
503 struct rb_time_struct {
504 	local64_t	time;
505 };
506 typedef struct rb_time_struct rb_time_t;
507 
508 #define MAX_NEST	5
509 
510 /*
511  * head_page == tail_page && head == tail then buffer is empty.
512  */
513 struct ring_buffer_per_cpu {
514 	int				cpu;
515 	atomic_t			record_disabled;
516 	atomic_t			resize_disabled;
517 	struct trace_buffer	*buffer;
518 	raw_spinlock_t			reader_lock;	/* serialize readers */
519 	arch_spinlock_t			lock;
520 	struct lock_class_key		lock_key;
521 	struct buffer_data_page		*free_page;
522 	unsigned long			nr_pages;
523 	unsigned int			current_context;
524 	struct list_head		*pages;
525 	/* pages generation counter, incremented when the list changes */
526 	unsigned long			cnt;
527 	struct buffer_page		*head_page;	/* read from head */
528 	struct buffer_page		*tail_page;	/* write to tail */
529 	struct buffer_page		*commit_page;	/* committed pages */
530 	struct buffer_page		*reader_page;
531 	unsigned long			lost_events;
532 	unsigned long			last_overrun;
533 	unsigned long			nest;
534 	local_t				entries_bytes;
535 	local_t				entries;
536 	local_t				overrun;
537 	local_t				commit_overrun;
538 	local_t				dropped_events;
539 	local_t				committing;
540 	local_t				commits;
541 	local_t				pages_touched;
542 	local_t				pages_lost;
543 	local_t				pages_read;
544 	long				last_pages_touch;
545 	size_t				shortest_full;
546 	unsigned long			read;
547 	unsigned long			read_bytes;
548 	rb_time_t			write_stamp;
549 	rb_time_t			before_stamp;
550 	u64				event_stamp[MAX_NEST];
551 	u64				read_stamp;
552 	/* pages removed since last reset */
553 	unsigned long			pages_removed;
554 
555 	unsigned int			mapped;
556 	unsigned int			user_mapped;	/* user space mapping */
557 	struct mutex			mapping_lock;
558 	unsigned long			*subbuf_ids;	/* ID to subbuf VA */
559 	struct trace_buffer_meta	*meta_page;
560 	struct ring_buffer_cpu_meta	*ring_meta;
561 
562 	/* ring buffer pages to update, > 0 to add, < 0 to remove */
563 	long				nr_pages_to_update;
564 	struct list_head		new_pages; /* new pages to add */
565 	struct work_struct		update_pages_work;
566 	struct completion		update_done;
567 
568 	struct rb_irq_work		irq_work;
569 };
570 
571 struct trace_buffer {
572 	unsigned			flags;
573 	int				cpus;
574 	atomic_t			record_disabled;
575 	atomic_t			resizing;
576 	cpumask_var_t			cpumask;
577 
578 	struct lock_class_key		*reader_lock_key;
579 
580 	struct mutex			mutex;
581 
582 	struct ring_buffer_per_cpu	**buffers;
583 
584 	struct hlist_node		node;
585 	u64				(*clock)(void);
586 
587 	struct rb_irq_work		irq_work;
588 	bool				time_stamp_abs;
589 
590 	unsigned long			range_addr_start;
591 	unsigned long			range_addr_end;
592 
593 	struct ring_buffer_meta		*meta;
594 
595 	unsigned int			subbuf_size;
596 	unsigned int			subbuf_order;
597 	unsigned int			max_data_size;
598 };
599 
600 struct ring_buffer_iter {
601 	struct ring_buffer_per_cpu	*cpu_buffer;
602 	unsigned long			head;
603 	unsigned long			next_event;
604 	struct buffer_page		*head_page;
605 	struct buffer_page		*cache_reader_page;
606 	unsigned long			cache_read;
607 	unsigned long			cache_pages_removed;
608 	u64				read_stamp;
609 	u64				page_stamp;
610 	struct ring_buffer_event	*event;
611 	size_t				event_size;
612 	int				missed_events;
613 };
614 
615 int ring_buffer_print_page_header(struct trace_buffer *buffer, struct trace_seq *s)
616 {
617 	struct buffer_data_page field;
618 
619 	trace_seq_printf(s, "\tfield: u64 timestamp;\t"
620 			 "offset:0;\tsize:%u;\tsigned:%u;\n",
621 			 (unsigned int)sizeof(field.time_stamp),
622 			 (unsigned int)is_signed_type(u64));
623 
624 	trace_seq_printf(s, "\tfield: local_t commit;\t"
625 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
626 			 (unsigned int)offsetof(typeof(field), commit),
627 			 (unsigned int)sizeof(field.commit),
628 			 (unsigned int)is_signed_type(long));
629 
630 	trace_seq_printf(s, "\tfield: int overwrite;\t"
631 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
632 			 (unsigned int)offsetof(typeof(field), commit),
633 			 1,
634 			 (unsigned int)is_signed_type(long));
635 
636 	trace_seq_printf(s, "\tfield: char data;\t"
637 			 "offset:%u;\tsize:%u;\tsigned:%u;\n",
638 			 (unsigned int)offsetof(typeof(field), data),
639 			 (unsigned int)buffer->subbuf_size,
640 			 (unsigned int)is_signed_type(char));
641 
642 	return !trace_seq_has_overflowed(s);
643 }
644 
645 static inline void rb_time_read(rb_time_t *t, u64 *ret)
646 {
647 	*ret = local64_read(&t->time);
648 }
649 static void rb_time_set(rb_time_t *t, u64 val)
650 {
651 	local64_set(&t->time, val);
652 }
653 
654 /*
655  * Enable this to make sure that the event passed to
656  * ring_buffer_event_time_stamp() is not committed and also
657  * is on the buffer that it passed in.
658  */
659 //#define RB_VERIFY_EVENT
660 #ifdef RB_VERIFY_EVENT
661 static struct list_head *rb_list_head(struct list_head *list);
662 static void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
663 			 void *event)
664 {
665 	struct buffer_page *page = cpu_buffer->commit_page;
666 	struct buffer_page *tail_page = READ_ONCE(cpu_buffer->tail_page);
667 	struct list_head *next;
668 	long commit, write;
669 	unsigned long addr = (unsigned long)event;
670 	bool done = false;
671 	int stop = 0;
672 
673 	/* Make sure the event exists and is not committed yet */
674 	do {
675 		if (page == tail_page || WARN_ON_ONCE(stop++ > 100))
676 			done = true;
677 		commit = local_read(&page->page->commit);
678 		write = local_read(&page->write);
679 		if (addr >= (unsigned long)&page->page->data[commit] &&
680 		    addr < (unsigned long)&page->page->data[write])
681 			return;
682 
683 		next = rb_list_head(page->list.next);
684 		page = list_entry(next, struct buffer_page, list);
685 	} while (!done);
686 	WARN_ON_ONCE(1);
687 }
688 #else
689 static inline void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
690 			 void *event)
691 {
692 }
693 #endif
694 
695 /*
696  * The absolute time stamp drops the 5 MSBs and some clocks may
697  * require them. The rb_fix_abs_ts() will take a previous full
698  * time stamp, and add the 5 MSB of that time stamp on to the
699  * saved absolute time stamp. Then they are compared in case of
700  * the unlikely event that the latest time stamp incremented
701  * the 5 MSB.
702  */
703 static inline u64 rb_fix_abs_ts(u64 abs, u64 save_ts)
704 {
705 	if (save_ts & TS_MSB) {
706 		abs |= save_ts & TS_MSB;
707 		/* Check for overflow */
708 		if (unlikely(abs < save_ts))
709 			abs += 1ULL << 59;
710 	}
711 	return abs;
712 }
713 
714 static inline u64 rb_time_stamp(struct trace_buffer *buffer);
715 
716 /**
717  * ring_buffer_event_time_stamp - return the event's current time stamp
718  * @buffer: The buffer that the event is on
719  * @event: the event to get the time stamp of
720  *
721  * Note, this must be called after @event is reserved, and before it is
722  * committed to the ring buffer. And must be called from the same
723  * context where the event was reserved (normal, softirq, irq, etc).
724  *
725  * Returns the time stamp associated with the current event.
726  * If the event has an extended time stamp, then that is used as
727  * the time stamp to return.
728  * In the highly unlikely case that the event was nested more than
729  * the max nesting, then the write_stamp of the buffer is returned,
730  * otherwise  current time is returned, but that really neither of
731  * the last two cases should ever happen.
732  */
733 u64 ring_buffer_event_time_stamp(struct trace_buffer *buffer,
734 				 struct ring_buffer_event *event)
735 {
736 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[smp_processor_id()];
737 	unsigned int nest;
738 	u64 ts;
739 
740 	/* If the event includes an absolute time, then just use that */
741 	if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
742 		ts = rb_event_time_stamp(event);
743 		return rb_fix_abs_ts(ts, cpu_buffer->tail_page->page->time_stamp);
744 	}
745 
746 	nest = local_read(&cpu_buffer->committing);
747 	verify_event(cpu_buffer, event);
748 	if (WARN_ON_ONCE(!nest))
749 		goto fail;
750 
751 	/* Read the current saved nesting level time stamp */
752 	if (likely(--nest < MAX_NEST))
753 		return cpu_buffer->event_stamp[nest];
754 
755 	/* Shouldn't happen, warn if it does */
756 	WARN_ONCE(1, "nest (%d) greater than max", nest);
757 
758  fail:
759 	rb_time_read(&cpu_buffer->write_stamp, &ts);
760 
761 	return ts;
762 }
763 
764 /**
765  * ring_buffer_nr_dirty_pages - get the number of used pages in the ring buffer
766  * @buffer: The ring_buffer to get the number of pages from
767  * @cpu: The cpu of the ring_buffer to get the number of pages from
768  *
769  * Returns the number of pages that have content in the ring buffer.
770  */
771 size_t ring_buffer_nr_dirty_pages(struct trace_buffer *buffer, int cpu)
772 {
773 	size_t read;
774 	size_t lost;
775 	size_t cnt;
776 
777 	read = local_read(&buffer->buffers[cpu]->pages_read);
778 	lost = local_read(&buffer->buffers[cpu]->pages_lost);
779 	cnt = local_read(&buffer->buffers[cpu]->pages_touched);
780 
781 	if (WARN_ON_ONCE(cnt < lost))
782 		return 0;
783 
784 	cnt -= lost;
785 
786 	/* The reader can read an empty page, but not more than that */
787 	if (cnt < read) {
788 		WARN_ON_ONCE(read > cnt + 1);
789 		return 0;
790 	}
791 
792 	return cnt - read;
793 }
794 
795 static __always_inline bool full_hit(struct trace_buffer *buffer, int cpu, int full)
796 {
797 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
798 	size_t nr_pages;
799 	size_t dirty;
800 
801 	nr_pages = cpu_buffer->nr_pages;
802 	if (!nr_pages || !full)
803 		return true;
804 
805 	/*
806 	 * Add one as dirty will never equal nr_pages, as the sub-buffer
807 	 * that the writer is on is not counted as dirty.
808 	 * This is needed if "buffer_percent" is set to 100.
809 	 */
810 	dirty = ring_buffer_nr_dirty_pages(buffer, cpu) + 1;
811 
812 	return (dirty * 100) >= (full * nr_pages);
813 }
814 
815 /*
816  * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
817  *
818  * Schedules a delayed work to wake up any task that is blocked on the
819  * ring buffer waiters queue.
820  */
821 static void rb_wake_up_waiters(struct irq_work *work)
822 {
823 	struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
824 
825 	/* For waiters waiting for the first wake up */
826 	(void)atomic_fetch_inc_release(&rbwork->seq);
827 
828 	wake_up_all(&rbwork->waiters);
829 	if (rbwork->full_waiters_pending || rbwork->wakeup_full) {
830 		/* Only cpu_buffer sets the above flags */
831 		struct ring_buffer_per_cpu *cpu_buffer =
832 			container_of(rbwork, struct ring_buffer_per_cpu, irq_work);
833 
834 		/* Called from interrupt context */
835 		raw_spin_lock(&cpu_buffer->reader_lock);
836 		rbwork->wakeup_full = false;
837 		rbwork->full_waiters_pending = false;
838 
839 		/* Waking up all waiters, they will reset the shortest full */
840 		cpu_buffer->shortest_full = 0;
841 		raw_spin_unlock(&cpu_buffer->reader_lock);
842 
843 		wake_up_all(&rbwork->full_waiters);
844 	}
845 }
846 
847 /**
848  * ring_buffer_wake_waiters - wake up any waiters on this ring buffer
849  * @buffer: The ring buffer to wake waiters on
850  * @cpu: The CPU buffer to wake waiters on
851  *
852  * In the case of a file that represents a ring buffer is closing,
853  * it is prudent to wake up any waiters that are on this.
854  */
855 void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu)
856 {
857 	struct ring_buffer_per_cpu *cpu_buffer;
858 	struct rb_irq_work *rbwork;
859 
860 	if (!buffer)
861 		return;
862 
863 	if (cpu == RING_BUFFER_ALL_CPUS) {
864 
865 		/* Wake up individual ones too. One level recursion */
866 		for_each_buffer_cpu(buffer, cpu)
867 			ring_buffer_wake_waiters(buffer, cpu);
868 
869 		rbwork = &buffer->irq_work;
870 	} else {
871 		if (WARN_ON_ONCE(!buffer->buffers))
872 			return;
873 		if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
874 			return;
875 
876 		cpu_buffer = buffer->buffers[cpu];
877 		/* The CPU buffer may not have been initialized yet */
878 		if (!cpu_buffer)
879 			return;
880 		rbwork = &cpu_buffer->irq_work;
881 	}
882 
883 	/* This can be called in any context */
884 	irq_work_queue(&rbwork->work);
885 }
886 
887 static bool rb_watermark_hit(struct trace_buffer *buffer, int cpu, int full)
888 {
889 	struct ring_buffer_per_cpu *cpu_buffer;
890 	bool ret = false;
891 
892 	/* Reads of all CPUs always waits for any data */
893 	if (cpu == RING_BUFFER_ALL_CPUS)
894 		return !ring_buffer_empty(buffer);
895 
896 	cpu_buffer = buffer->buffers[cpu];
897 
898 	if (!ring_buffer_empty_cpu(buffer, cpu)) {
899 		unsigned long flags;
900 		bool pagebusy;
901 
902 		if (!full)
903 			return true;
904 
905 		raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
906 		pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
907 		ret = !pagebusy && full_hit(buffer, cpu, full);
908 
909 		if (!ret && (!cpu_buffer->shortest_full ||
910 			     cpu_buffer->shortest_full > full)) {
911 		    cpu_buffer->shortest_full = full;
912 		}
913 		raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
914 	}
915 	return ret;
916 }
917 
918 static inline bool
919 rb_wait_cond(struct rb_irq_work *rbwork, struct trace_buffer *buffer,
920 	     int cpu, int full, ring_buffer_cond_fn cond, void *data)
921 {
922 	if (rb_watermark_hit(buffer, cpu, full))
923 		return true;
924 
925 	if (cond(data))
926 		return true;
927 
928 	/*
929 	 * The events can happen in critical sections where
930 	 * checking a work queue can cause deadlocks.
931 	 * After adding a task to the queue, this flag is set
932 	 * only to notify events to try to wake up the queue
933 	 * using irq_work.
934 	 *
935 	 * We don't clear it even if the buffer is no longer
936 	 * empty. The flag only causes the next event to run
937 	 * irq_work to do the work queue wake up. The worse
938 	 * that can happen if we race with !trace_empty() is that
939 	 * an event will cause an irq_work to try to wake up
940 	 * an empty queue.
941 	 *
942 	 * There's no reason to protect this flag either, as
943 	 * the work queue and irq_work logic will do the necessary
944 	 * synchronization for the wake ups. The only thing
945 	 * that is necessary is that the wake up happens after
946 	 * a task has been queued. It's OK for spurious wake ups.
947 	 */
948 	if (full)
949 		rbwork->full_waiters_pending = true;
950 	else
951 		rbwork->waiters_pending = true;
952 
953 	return false;
954 }
955 
956 struct rb_wait_data {
957 	struct rb_irq_work		*irq_work;
958 	int				seq;
959 };
960 
961 /*
962  * The default wait condition for ring_buffer_wait() is to just to exit the
963  * wait loop the first time it is woken up.
964  */
965 static bool rb_wait_once(void *data)
966 {
967 	struct rb_wait_data *rdata = data;
968 	struct rb_irq_work *rbwork = rdata->irq_work;
969 
970 	return atomic_read_acquire(&rbwork->seq) != rdata->seq;
971 }
972 
973 /**
974  * ring_buffer_wait - wait for input to the ring buffer
975  * @buffer: buffer to wait on
976  * @cpu: the cpu buffer to wait on
977  * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
978  * @cond: condition function to break out of wait (NULL to run once)
979  * @data: the data to pass to @cond.
980  *
981  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
982  * as data is added to any of the @buffer's cpu buffers. Otherwise
983  * it will wait for data to be added to a specific cpu buffer.
984  */
985 int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full,
986 		     ring_buffer_cond_fn cond, void *data)
987 {
988 	struct ring_buffer_per_cpu *cpu_buffer;
989 	struct wait_queue_head *waitq;
990 	struct rb_irq_work *rbwork;
991 	struct rb_wait_data rdata;
992 	int ret = 0;
993 
994 	/*
995 	 * Depending on what the caller is waiting for, either any
996 	 * data in any cpu buffer, or a specific buffer, put the
997 	 * caller on the appropriate wait queue.
998 	 */
999 	if (cpu == RING_BUFFER_ALL_CPUS) {
1000 		rbwork = &buffer->irq_work;
1001 		/* Full only makes sense on per cpu reads */
1002 		full = 0;
1003 	} else {
1004 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
1005 			return -ENODEV;
1006 		cpu_buffer = buffer->buffers[cpu];
1007 		rbwork = &cpu_buffer->irq_work;
1008 	}
1009 
1010 	if (full)
1011 		waitq = &rbwork->full_waiters;
1012 	else
1013 		waitq = &rbwork->waiters;
1014 
1015 	/* Set up to exit loop as soon as it is woken */
1016 	if (!cond) {
1017 		cond = rb_wait_once;
1018 		rdata.irq_work = rbwork;
1019 		rdata.seq = atomic_read_acquire(&rbwork->seq);
1020 		data = &rdata;
1021 	}
1022 
1023 	ret = wait_event_interruptible((*waitq),
1024 				rb_wait_cond(rbwork, buffer, cpu, full, cond, data));
1025 
1026 	return ret;
1027 }
1028 
1029 /**
1030  * ring_buffer_poll_wait - poll on buffer input
1031  * @buffer: buffer to wait on
1032  * @cpu: the cpu buffer to wait on
1033  * @filp: the file descriptor
1034  * @poll_table: The poll descriptor
1035  * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
1036  *
1037  * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
1038  * as data is added to any of the @buffer's cpu buffers. Otherwise
1039  * it will wait for data to be added to a specific cpu buffer.
1040  *
1041  * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
1042  * zero otherwise.
1043  */
1044 __poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu,
1045 			  struct file *filp, poll_table *poll_table, int full)
1046 {
1047 	struct ring_buffer_per_cpu *cpu_buffer;
1048 	struct rb_irq_work *rbwork;
1049 
1050 	if (cpu == RING_BUFFER_ALL_CPUS) {
1051 		rbwork = &buffer->irq_work;
1052 		full = 0;
1053 	} else {
1054 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
1055 			return EPOLLERR;
1056 
1057 		cpu_buffer = buffer->buffers[cpu];
1058 		rbwork = &cpu_buffer->irq_work;
1059 	}
1060 
1061 	if (full) {
1062 		poll_wait(filp, &rbwork->full_waiters, poll_table);
1063 
1064 		if (rb_watermark_hit(buffer, cpu, full))
1065 			return EPOLLIN | EPOLLRDNORM;
1066 		/*
1067 		 * Only allow full_waiters_pending update to be seen after
1068 		 * the shortest_full is set (in rb_watermark_hit). If the
1069 		 * writer sees the full_waiters_pending flag set, it will
1070 		 * compare the amount in the ring buffer to shortest_full.
1071 		 * If the amount in the ring buffer is greater than the
1072 		 * shortest_full percent, it will call the irq_work handler
1073 		 * to wake up this list. The irq_handler will reset shortest_full
1074 		 * back to zero. That's done under the reader_lock, but
1075 		 * the below smp_mb() makes sure that the update to
1076 		 * full_waiters_pending doesn't leak up into the above.
1077 		 */
1078 		smp_mb();
1079 		rbwork->full_waiters_pending = true;
1080 		return 0;
1081 	}
1082 
1083 	poll_wait(filp, &rbwork->waiters, poll_table);
1084 	rbwork->waiters_pending = true;
1085 
1086 	/*
1087 	 * There's a tight race between setting the waiters_pending and
1088 	 * checking if the ring buffer is empty.  Once the waiters_pending bit
1089 	 * is set, the next event will wake the task up, but we can get stuck
1090 	 * if there's only a single event in.
1091 	 *
1092 	 * FIXME: Ideally, we need a memory barrier on the writer side as well,
1093 	 * but adding a memory barrier to all events will cause too much of a
1094 	 * performance hit in the fast path.  We only need a memory barrier when
1095 	 * the buffer goes from empty to having content.  But as this race is
1096 	 * extremely small, and it's not a problem if another event comes in, we
1097 	 * will fix it later.
1098 	 */
1099 	smp_mb();
1100 
1101 	if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
1102 	    (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
1103 		return EPOLLIN | EPOLLRDNORM;
1104 	return 0;
1105 }
1106 
1107 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
1108 #define RB_WARN_ON(b, cond)						\
1109 	({								\
1110 		int _____ret = unlikely(cond);				\
1111 		if (_____ret) {						\
1112 			if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
1113 				struct ring_buffer_per_cpu *__b =	\
1114 					(void *)b;			\
1115 				atomic_inc(&__b->buffer->record_disabled); \
1116 			} else						\
1117 				atomic_inc(&b->record_disabled);	\
1118 			WARN_ON(1);					\
1119 		}							\
1120 		_____ret;						\
1121 	})
1122 
1123 /* Up this if you want to test the TIME_EXTENTS and normalization */
1124 #define DEBUG_SHIFT 0
1125 
1126 static inline u64 rb_time_stamp(struct trace_buffer *buffer)
1127 {
1128 	u64 ts;
1129 
1130 	/* Skip retpolines :-( */
1131 	if (IS_ENABLED(CONFIG_MITIGATION_RETPOLINE) && likely(buffer->clock == trace_clock_local))
1132 		ts = trace_clock_local();
1133 	else
1134 		ts = buffer->clock();
1135 
1136 	/* shift to debug/test normalization and TIME_EXTENTS */
1137 	return ts << DEBUG_SHIFT;
1138 }
1139 
1140 u64 ring_buffer_time_stamp(struct trace_buffer *buffer)
1141 {
1142 	u64 time;
1143 
1144 	preempt_disable_notrace();
1145 	time = rb_time_stamp(buffer);
1146 	preempt_enable_notrace();
1147 
1148 	return time;
1149 }
1150 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
1151 
1152 void ring_buffer_normalize_time_stamp(struct trace_buffer *buffer,
1153 				      int cpu, u64 *ts)
1154 {
1155 	/* Just stupid testing the normalize function and deltas */
1156 	*ts >>= DEBUG_SHIFT;
1157 }
1158 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
1159 
1160 /*
1161  * Making the ring buffer lockless makes things tricky.
1162  * Although writes only happen on the CPU that they are on,
1163  * and they only need to worry about interrupts. Reads can
1164  * happen on any CPU.
1165  *
1166  * The reader page is always off the ring buffer, but when the
1167  * reader finishes with a page, it needs to swap its page with
1168  * a new one from the buffer. The reader needs to take from
1169  * the head (writes go to the tail). But if a writer is in overwrite
1170  * mode and wraps, it must push the head page forward.
1171  *
1172  * Here lies the problem.
1173  *
1174  * The reader must be careful to replace only the head page, and
1175  * not another one. As described at the top of the file in the
1176  * ASCII art, the reader sets its old page to point to the next
1177  * page after head. It then sets the page after head to point to
1178  * the old reader page. But if the writer moves the head page
1179  * during this operation, the reader could end up with the tail.
1180  *
1181  * We use cmpxchg to help prevent this race. We also do something
1182  * special with the page before head. We set the LSB to 1.
1183  *
1184  * When the writer must push the page forward, it will clear the
1185  * bit that points to the head page, move the head, and then set
1186  * the bit that points to the new head page.
1187  *
1188  * We also don't want an interrupt coming in and moving the head
1189  * page on another writer. Thus we use the second LSB to catch
1190  * that too. Thus:
1191  *
1192  * head->list->prev->next        bit 1          bit 0
1193  *                              -------        -------
1194  * Normal page                     0              0
1195  * Points to head page             0              1
1196  * New head page                   1              0
1197  *
1198  * Note we can not trust the prev pointer of the head page, because:
1199  *
1200  * +----+       +-----+        +-----+
1201  * |    |------>|  T  |---X--->|  N  |
1202  * |    |<------|     |        |     |
1203  * +----+       +-----+        +-----+
1204  *   ^                           ^ |
1205  *   |          +-----+          | |
1206  *   +----------|  R  |----------+ |
1207  *              |     |<-----------+
1208  *              +-----+
1209  *
1210  * Key:  ---X-->  HEAD flag set in pointer
1211  *         T      Tail page
1212  *         R      Reader page
1213  *         N      Next page
1214  *
1215  * (see __rb_reserve_next() to see where this happens)
1216  *
1217  *  What the above shows is that the reader just swapped out
1218  *  the reader page with a page in the buffer, but before it
1219  *  could make the new header point back to the new page added
1220  *  it was preempted by a writer. The writer moved forward onto
1221  *  the new page added by the reader and is about to move forward
1222  *  again.
1223  *
1224  *  You can see, it is legitimate for the previous pointer of
1225  *  the head (or any page) not to point back to itself. But only
1226  *  temporarily.
1227  */
1228 
1229 #define RB_PAGE_NORMAL		0UL
1230 #define RB_PAGE_HEAD		1UL
1231 #define RB_PAGE_UPDATE		2UL
1232 
1233 
1234 #define RB_FLAG_MASK		3UL
1235 
1236 /* PAGE_MOVED is not part of the mask */
1237 #define RB_PAGE_MOVED		4UL
1238 
1239 /*
1240  * rb_list_head - remove any bit
1241  */
1242 static struct list_head *rb_list_head(struct list_head *list)
1243 {
1244 	unsigned long val = (unsigned long)list;
1245 
1246 	return (struct list_head *)(val & ~RB_FLAG_MASK);
1247 }
1248 
1249 /*
1250  * rb_is_head_page - test if the given page is the head page
1251  *
1252  * Because the reader may move the head_page pointer, we can
1253  * not trust what the head page is (it may be pointing to
1254  * the reader page). But if the next page is a header page,
1255  * its flags will be non zero.
1256  */
1257 static inline int
1258 rb_is_head_page(struct buffer_page *page, struct list_head *list)
1259 {
1260 	unsigned long val;
1261 
1262 	val = (unsigned long)list->next;
1263 
1264 	if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
1265 		return RB_PAGE_MOVED;
1266 
1267 	return val & RB_FLAG_MASK;
1268 }
1269 
1270 /*
1271  * rb_is_reader_page
1272  *
1273  * The unique thing about the reader page, is that, if the
1274  * writer is ever on it, the previous pointer never points
1275  * back to the reader page.
1276  */
1277 static bool rb_is_reader_page(struct buffer_page *page)
1278 {
1279 	struct list_head *list = page->list.prev;
1280 
1281 	return rb_list_head(list->next) != &page->list;
1282 }
1283 
1284 /*
1285  * rb_set_list_to_head - set a list_head to be pointing to head.
1286  */
1287 static void rb_set_list_to_head(struct list_head *list)
1288 {
1289 	unsigned long *ptr;
1290 
1291 	ptr = (unsigned long *)&list->next;
1292 	*ptr |= RB_PAGE_HEAD;
1293 	*ptr &= ~RB_PAGE_UPDATE;
1294 }
1295 
1296 /*
1297  * rb_head_page_activate - sets up head page
1298  */
1299 static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
1300 {
1301 	struct buffer_page *head;
1302 
1303 	head = cpu_buffer->head_page;
1304 	if (!head)
1305 		return;
1306 
1307 	/*
1308 	 * Set the previous list pointer to have the HEAD flag.
1309 	 */
1310 	rb_set_list_to_head(head->list.prev);
1311 
1312 	if (cpu_buffer->ring_meta) {
1313 		struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
1314 		meta->head_buffer = (unsigned long)head->page;
1315 	}
1316 }
1317 
1318 static void rb_list_head_clear(struct list_head *list)
1319 {
1320 	unsigned long *ptr = (unsigned long *)&list->next;
1321 
1322 	*ptr &= ~RB_FLAG_MASK;
1323 }
1324 
1325 /*
1326  * rb_head_page_deactivate - clears head page ptr (for free list)
1327  */
1328 static void
1329 rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
1330 {
1331 	struct list_head *hd;
1332 
1333 	/* Go through the whole list and clear any pointers found. */
1334 	rb_list_head_clear(cpu_buffer->pages);
1335 
1336 	list_for_each(hd, cpu_buffer->pages)
1337 		rb_list_head_clear(hd);
1338 }
1339 
1340 static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
1341 			    struct buffer_page *head,
1342 			    struct buffer_page *prev,
1343 			    int old_flag, int new_flag)
1344 {
1345 	struct list_head *list;
1346 	unsigned long val = (unsigned long)&head->list;
1347 	unsigned long ret;
1348 
1349 	list = &prev->list;
1350 
1351 	val &= ~RB_FLAG_MASK;
1352 
1353 	ret = cmpxchg((unsigned long *)&list->next,
1354 		      val | old_flag, val | new_flag);
1355 
1356 	/* check if the reader took the page */
1357 	if ((ret & ~RB_FLAG_MASK) != val)
1358 		return RB_PAGE_MOVED;
1359 
1360 	return ret & RB_FLAG_MASK;
1361 }
1362 
1363 static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
1364 				   struct buffer_page *head,
1365 				   struct buffer_page *prev,
1366 				   int old_flag)
1367 {
1368 	return rb_head_page_set(cpu_buffer, head, prev,
1369 				old_flag, RB_PAGE_UPDATE);
1370 }
1371 
1372 static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
1373 				 struct buffer_page *head,
1374 				 struct buffer_page *prev,
1375 				 int old_flag)
1376 {
1377 	return rb_head_page_set(cpu_buffer, head, prev,
1378 				old_flag, RB_PAGE_HEAD);
1379 }
1380 
1381 static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
1382 				   struct buffer_page *head,
1383 				   struct buffer_page *prev,
1384 				   int old_flag)
1385 {
1386 	return rb_head_page_set(cpu_buffer, head, prev,
1387 				old_flag, RB_PAGE_NORMAL);
1388 }
1389 
1390 static inline void rb_inc_page(struct buffer_page **bpage)
1391 {
1392 	struct list_head *p = rb_list_head((*bpage)->list.next);
1393 
1394 	*bpage = list_entry(p, struct buffer_page, list);
1395 }
1396 
1397 static inline void rb_dec_page(struct buffer_page **bpage)
1398 {
1399 	struct list_head *p = rb_list_head((*bpage)->list.prev);
1400 
1401 	*bpage = list_entry(p, struct buffer_page, list);
1402 }
1403 
1404 static struct buffer_page *
1405 rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1406 {
1407 	struct buffer_page *head;
1408 	struct buffer_page *page;
1409 	struct list_head *list;
1410 	int i;
1411 
1412 	if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1413 		return NULL;
1414 
1415 	/* sanity check */
1416 	list = cpu_buffer->pages;
1417 	if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1418 		return NULL;
1419 
1420 	page = head = cpu_buffer->head_page;
1421 	/*
1422 	 * It is possible that the writer moves the header behind
1423 	 * where we started, and we miss in one loop.
1424 	 * A second loop should grab the header, but we'll do
1425 	 * three loops just because I'm paranoid.
1426 	 */
1427 	for (i = 0; i < 3; i++) {
1428 		do {
1429 			if (rb_is_head_page(page, page->list.prev)) {
1430 				cpu_buffer->head_page = page;
1431 				return page;
1432 			}
1433 			rb_inc_page(&page);
1434 		} while (page != head);
1435 	}
1436 
1437 	RB_WARN_ON(cpu_buffer, 1);
1438 
1439 	return NULL;
1440 }
1441 
1442 static bool rb_head_page_replace(struct buffer_page *old,
1443 				struct buffer_page *new)
1444 {
1445 	unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1446 	unsigned long val;
1447 
1448 	val = *ptr & ~RB_FLAG_MASK;
1449 	val |= RB_PAGE_HEAD;
1450 
1451 	return try_cmpxchg(ptr, &val, (unsigned long)&new->list);
1452 }
1453 
1454 /*
1455  * rb_tail_page_update - move the tail page forward
1456  */
1457 static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1458 			       struct buffer_page *tail_page,
1459 			       struct buffer_page *next_page)
1460 {
1461 	unsigned long old_entries;
1462 	unsigned long old_write;
1463 
1464 	/*
1465 	 * The tail page now needs to be moved forward.
1466 	 *
1467 	 * We need to reset the tail page, but without messing
1468 	 * with possible erasing of data brought in by interrupts
1469 	 * that have moved the tail page and are currently on it.
1470 	 *
1471 	 * We add a counter to the write field to denote this.
1472 	 */
1473 	old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
1474 	old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
1475 
1476 	/*
1477 	 * Just make sure we have seen our old_write and synchronize
1478 	 * with any interrupts that come in.
1479 	 */
1480 	barrier();
1481 
1482 	/*
1483 	 * If the tail page is still the same as what we think
1484 	 * it is, then it is up to us to update the tail
1485 	 * pointer.
1486 	 */
1487 	if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1488 		/* Zero the write counter */
1489 		unsigned long val = old_write & ~RB_WRITE_MASK;
1490 		unsigned long eval = old_entries & ~RB_WRITE_MASK;
1491 
1492 		/*
1493 		 * This will only succeed if an interrupt did
1494 		 * not come in and change it. In which case, we
1495 		 * do not want to modify it.
1496 		 *
1497 		 * We add (void) to let the compiler know that we do not care
1498 		 * about the return value of these functions. We use the
1499 		 * cmpxchg to only update if an interrupt did not already
1500 		 * do it for us. If the cmpxchg fails, we don't care.
1501 		 */
1502 		(void)local_cmpxchg(&next_page->write, old_write, val);
1503 		(void)local_cmpxchg(&next_page->entries, old_entries, eval);
1504 
1505 		/*
1506 		 * No need to worry about races with clearing out the commit.
1507 		 * it only can increment when a commit takes place. But that
1508 		 * only happens in the outer most nested commit.
1509 		 */
1510 		local_set(&next_page->page->commit, 0);
1511 
1512 		/* Either we update tail_page or an interrupt does */
1513 		if (try_cmpxchg(&cpu_buffer->tail_page, &tail_page, next_page))
1514 			local_inc(&cpu_buffer->pages_touched);
1515 	}
1516 }
1517 
1518 static void rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1519 			  struct buffer_page *bpage)
1520 {
1521 	unsigned long val = (unsigned long)bpage;
1522 
1523 	RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK);
1524 }
1525 
1526 static bool rb_check_links(struct ring_buffer_per_cpu *cpu_buffer,
1527 			   struct list_head *list)
1528 {
1529 	if (RB_WARN_ON(cpu_buffer,
1530 		       rb_list_head(rb_list_head(list->next)->prev) != list))
1531 		return false;
1532 
1533 	if (RB_WARN_ON(cpu_buffer,
1534 		       rb_list_head(rb_list_head(list->prev)->next) != list))
1535 		return false;
1536 
1537 	return true;
1538 }
1539 
1540 /**
1541  * rb_check_pages - integrity check of buffer pages
1542  * @cpu_buffer: CPU buffer with pages to test
1543  *
1544  * As a safety measure we check to make sure the data pages have not
1545  * been corrupted.
1546  */
1547 static void rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1548 {
1549 	struct list_head *head, *tmp;
1550 	unsigned long buffer_cnt;
1551 	unsigned long flags;
1552 	int nr_loops = 0;
1553 
1554 	/*
1555 	 * Walk the linked list underpinning the ring buffer and validate all
1556 	 * its next and prev links.
1557 	 *
1558 	 * The check acquires the reader_lock to avoid concurrent processing
1559 	 * with code that could be modifying the list. However, the lock cannot
1560 	 * be held for the entire duration of the walk, as this would make the
1561 	 * time when interrupts are disabled non-deterministic, dependent on the
1562 	 * ring buffer size. Therefore, the code releases and re-acquires the
1563 	 * lock after checking each page. The ring_buffer_per_cpu.cnt variable
1564 	 * is then used to detect if the list was modified while the lock was
1565 	 * not held, in which case the check needs to be restarted.
1566 	 *
1567 	 * The code attempts to perform the check at most three times before
1568 	 * giving up. This is acceptable because this is only a self-validation
1569 	 * to detect problems early on. In practice, the list modification
1570 	 * operations are fairly spaced, and so this check typically succeeds at
1571 	 * most on the second try.
1572 	 */
1573 again:
1574 	if (++nr_loops > 3)
1575 		return;
1576 
1577 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1578 	head = rb_list_head(cpu_buffer->pages);
1579 	if (!rb_check_links(cpu_buffer, head))
1580 		goto out_locked;
1581 	buffer_cnt = cpu_buffer->cnt;
1582 	tmp = head;
1583 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1584 
1585 	while (true) {
1586 		raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1587 
1588 		if (buffer_cnt != cpu_buffer->cnt) {
1589 			/* The list was updated, try again. */
1590 			raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1591 			goto again;
1592 		}
1593 
1594 		tmp = rb_list_head(tmp->next);
1595 		if (tmp == head)
1596 			/* The iteration circled back, all is done. */
1597 			goto out_locked;
1598 
1599 		if (!rb_check_links(cpu_buffer, tmp))
1600 			goto out_locked;
1601 
1602 		raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1603 	}
1604 
1605 out_locked:
1606 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1607 }
1608 
1609 /*
1610  * Take an address, add the meta data size as well as the array of
1611  * array subbuffer indexes, then align it to a subbuffer size.
1612  *
1613  * This is used to help find the next per cpu subbuffer within a mapped range.
1614  */
1615 static unsigned long
1616 rb_range_align_subbuf(unsigned long addr, int subbuf_size, int nr_subbufs)
1617 {
1618 	addr += sizeof(struct ring_buffer_cpu_meta) +
1619 		sizeof(int) * nr_subbufs;
1620 	return ALIGN(addr, subbuf_size);
1621 }
1622 
1623 /*
1624  * Return the ring_buffer_meta for a given @cpu.
1625  */
1626 static void *rb_range_meta(struct trace_buffer *buffer, int nr_pages, int cpu)
1627 {
1628 	int subbuf_size = buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
1629 	struct ring_buffer_cpu_meta *meta;
1630 	struct ring_buffer_meta *bmeta;
1631 	unsigned long ptr;
1632 	int nr_subbufs;
1633 
1634 	bmeta = buffer->meta;
1635 	if (!bmeta)
1636 		return NULL;
1637 
1638 	ptr = (unsigned long)bmeta + bmeta->buffers_offset;
1639 	meta = (struct ring_buffer_cpu_meta *)ptr;
1640 
1641 	/* When nr_pages passed in is zero, the first meta has already been initialized */
1642 	if (!nr_pages) {
1643 		nr_subbufs = meta->nr_subbufs;
1644 	} else {
1645 		/* Include the reader page */
1646 		nr_subbufs = nr_pages + 1;
1647 	}
1648 
1649 	/*
1650 	 * The first chunk may not be subbuffer aligned, where as
1651 	 * the rest of the chunks are.
1652 	 */
1653 	if (cpu) {
1654 		ptr = rb_range_align_subbuf(ptr, subbuf_size, nr_subbufs);
1655 		ptr += subbuf_size * nr_subbufs;
1656 
1657 		/* We can use multiplication to find chunks greater than 1 */
1658 		if (cpu > 1) {
1659 			unsigned long size;
1660 			unsigned long p;
1661 
1662 			/* Save the beginning of this CPU chunk */
1663 			p = ptr;
1664 			ptr = rb_range_align_subbuf(ptr, subbuf_size, nr_subbufs);
1665 			ptr += subbuf_size * nr_subbufs;
1666 
1667 			/* Now all chunks after this are the same size */
1668 			size = ptr - p;
1669 			ptr += size * (cpu - 2);
1670 		}
1671 	}
1672 	return (void *)ptr;
1673 }
1674 
1675 /* Return the start of subbufs given the meta pointer */
1676 static void *rb_subbufs_from_meta(struct ring_buffer_cpu_meta *meta)
1677 {
1678 	int subbuf_size = meta->subbuf_size;
1679 	unsigned long ptr;
1680 
1681 	ptr = (unsigned long)meta;
1682 	ptr = rb_range_align_subbuf(ptr, subbuf_size, meta->nr_subbufs);
1683 
1684 	return (void *)ptr;
1685 }
1686 
1687 /*
1688  * Return a specific sub-buffer for a given @cpu defined by @idx.
1689  */
1690 static void *rb_range_buffer(struct ring_buffer_per_cpu *cpu_buffer, int idx)
1691 {
1692 	struct ring_buffer_cpu_meta *meta;
1693 	unsigned long ptr;
1694 	int subbuf_size;
1695 
1696 	meta = rb_range_meta(cpu_buffer->buffer, 0, cpu_buffer->cpu);
1697 	if (!meta)
1698 		return NULL;
1699 
1700 	if (WARN_ON_ONCE(idx >= meta->nr_subbufs))
1701 		return NULL;
1702 
1703 	subbuf_size = meta->subbuf_size;
1704 
1705 	/* Map this buffer to the order that's in meta->buffers[] */
1706 	idx = meta->buffers[idx];
1707 
1708 	ptr = (unsigned long)rb_subbufs_from_meta(meta);
1709 
1710 	ptr += subbuf_size * idx;
1711 	if (ptr + subbuf_size > cpu_buffer->buffer->range_addr_end)
1712 		return NULL;
1713 
1714 	return (void *)ptr;
1715 }
1716 
1717 /*
1718  * See if the existing memory contains a valid meta section.
1719  * if so, use that, otherwise initialize it.
1720  */
1721 static bool rb_meta_init(struct trace_buffer *buffer, int scratch_size)
1722 {
1723 	unsigned long ptr = buffer->range_addr_start;
1724 	struct ring_buffer_meta *bmeta;
1725 	unsigned long total_size;
1726 	int struct_sizes;
1727 
1728 	bmeta = (struct ring_buffer_meta *)ptr;
1729 	buffer->meta = bmeta;
1730 
1731 	total_size = buffer->range_addr_end - buffer->range_addr_start;
1732 
1733 	struct_sizes = sizeof(struct ring_buffer_cpu_meta);
1734 	struct_sizes |= sizeof(*bmeta) << 16;
1735 
1736 	/* The first buffer will start word size after the meta page */
1737 	ptr += sizeof(*bmeta);
1738 	ptr = ALIGN(ptr, sizeof(long));
1739 	ptr += scratch_size;
1740 
1741 	if (bmeta->magic != RING_BUFFER_META_MAGIC) {
1742 		pr_info("Ring buffer boot meta mismatch of magic\n");
1743 		goto init;
1744 	}
1745 
1746 	if (bmeta->struct_sizes != struct_sizes) {
1747 		pr_info("Ring buffer boot meta mismatch of struct size\n");
1748 		goto init;
1749 	}
1750 
1751 	if (bmeta->total_size != total_size) {
1752 		pr_info("Ring buffer boot meta mismatch of total size\n");
1753 		goto init;
1754 	}
1755 
1756 	if (bmeta->buffers_offset > bmeta->total_size) {
1757 		pr_info("Ring buffer boot meta mismatch of offset outside of total size\n");
1758 		goto init;
1759 	}
1760 
1761 	if (bmeta->buffers_offset != (void *)ptr - (void *)bmeta) {
1762 		pr_info("Ring buffer boot meta mismatch of first buffer offset\n");
1763 		goto init;
1764 	}
1765 
1766 	return true;
1767 
1768  init:
1769 	bmeta->magic = RING_BUFFER_META_MAGIC;
1770 	bmeta->struct_sizes = struct_sizes;
1771 	bmeta->total_size = total_size;
1772 	bmeta->buffers_offset = (void *)ptr - (void *)bmeta;
1773 
1774 	/* Zero out the scratch pad */
1775 	memset((void *)bmeta + sizeof(*bmeta), 0, bmeta->buffers_offset - sizeof(*bmeta));
1776 
1777 	return false;
1778 }
1779 
1780 /*
1781  * See if the existing memory contains valid ring buffer data.
1782  * As the previous kernel must be the same as this kernel, all
1783  * the calculations (size of buffers and number of buffers)
1784  * must be the same.
1785  */
1786 static bool rb_cpu_meta_valid(struct ring_buffer_cpu_meta *meta, int cpu,
1787 			      struct trace_buffer *buffer, int nr_pages,
1788 			      unsigned long *subbuf_mask)
1789 {
1790 	int subbuf_size = PAGE_SIZE;
1791 	struct buffer_data_page *subbuf;
1792 	unsigned long buffers_start;
1793 	unsigned long buffers_end;
1794 	int i;
1795 
1796 	if (!subbuf_mask)
1797 		return false;
1798 
1799 	buffers_start = meta->first_buffer;
1800 	buffers_end = meta->first_buffer + (subbuf_size * meta->nr_subbufs);
1801 
1802 	/* Is the head and commit buffers within the range of buffers? */
1803 	if (meta->head_buffer < buffers_start ||
1804 	    meta->head_buffer >= buffers_end) {
1805 		pr_info("Ring buffer boot meta [%d] head buffer out of range\n", cpu);
1806 		return false;
1807 	}
1808 
1809 	if (meta->commit_buffer < buffers_start ||
1810 	    meta->commit_buffer >= buffers_end) {
1811 		pr_info("Ring buffer boot meta [%d] commit buffer out of range\n", cpu);
1812 		return false;
1813 	}
1814 
1815 	subbuf = rb_subbufs_from_meta(meta);
1816 
1817 	bitmap_clear(subbuf_mask, 0, meta->nr_subbufs);
1818 
1819 	/* Is the meta buffers and the subbufs themselves have correct data? */
1820 	for (i = 0; i < meta->nr_subbufs; i++) {
1821 		if (meta->buffers[i] < 0 ||
1822 		    meta->buffers[i] >= meta->nr_subbufs) {
1823 			pr_info("Ring buffer boot meta [%d] array out of range\n", cpu);
1824 			return false;
1825 		}
1826 
1827 		if ((unsigned)local_read(&subbuf->commit) > subbuf_size) {
1828 			pr_info("Ring buffer boot meta [%d] buffer invalid commit\n", cpu);
1829 			return false;
1830 		}
1831 
1832 		if (test_bit(meta->buffers[i], subbuf_mask)) {
1833 			pr_info("Ring buffer boot meta [%d] array has duplicates\n", cpu);
1834 			return false;
1835 		}
1836 
1837 		set_bit(meta->buffers[i], subbuf_mask);
1838 		subbuf = (void *)subbuf + subbuf_size;
1839 	}
1840 
1841 	return true;
1842 }
1843 
1844 static int rb_meta_subbuf_idx(struct ring_buffer_cpu_meta *meta, void *subbuf);
1845 
1846 static int rb_read_data_buffer(struct buffer_data_page *dpage, int tail, int cpu,
1847 			       unsigned long long *timestamp, u64 *delta_ptr)
1848 {
1849 	struct ring_buffer_event *event;
1850 	u64 ts, delta;
1851 	int events = 0;
1852 	int e;
1853 
1854 	*delta_ptr = 0;
1855 	*timestamp = 0;
1856 
1857 	ts = dpage->time_stamp;
1858 
1859 	for (e = 0; e < tail; e += rb_event_length(event)) {
1860 
1861 		event = (struct ring_buffer_event *)(dpage->data + e);
1862 
1863 		switch (event->type_len) {
1864 
1865 		case RINGBUF_TYPE_TIME_EXTEND:
1866 			delta = rb_event_time_stamp(event);
1867 			ts += delta;
1868 			break;
1869 
1870 		case RINGBUF_TYPE_TIME_STAMP:
1871 			delta = rb_event_time_stamp(event);
1872 			delta = rb_fix_abs_ts(delta, ts);
1873 			if (delta < ts) {
1874 				*delta_ptr = delta;
1875 				*timestamp = ts;
1876 				return -1;
1877 			}
1878 			ts = delta;
1879 			break;
1880 
1881 		case RINGBUF_TYPE_PADDING:
1882 			if (event->time_delta == 1)
1883 				break;
1884 			fallthrough;
1885 		case RINGBUF_TYPE_DATA:
1886 			events++;
1887 			ts += event->time_delta;
1888 			break;
1889 
1890 		default:
1891 			return -1;
1892 		}
1893 	}
1894 	*timestamp = ts;
1895 	return events;
1896 }
1897 
1898 static int rb_validate_buffer(struct buffer_data_page *dpage, int cpu)
1899 {
1900 	unsigned long long ts;
1901 	u64 delta;
1902 	int tail;
1903 
1904 	tail = local_read(&dpage->commit);
1905 	return rb_read_data_buffer(dpage, tail, cpu, &ts, &delta);
1906 }
1907 
1908 /* If the meta data has been validated, now validate the events */
1909 static void rb_meta_validate_events(struct ring_buffer_per_cpu *cpu_buffer)
1910 {
1911 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
1912 	struct buffer_page *head_page, *orig_head;
1913 	unsigned long entry_bytes = 0;
1914 	unsigned long entries = 0;
1915 	int ret;
1916 	u64 ts;
1917 	int i;
1918 
1919 	if (!meta || !meta->head_buffer)
1920 		return;
1921 
1922 	/* Do the reader page first */
1923 	ret = rb_validate_buffer(cpu_buffer->reader_page->page, cpu_buffer->cpu);
1924 	if (ret < 0) {
1925 		pr_info("Ring buffer reader page is invalid\n");
1926 		goto invalid;
1927 	}
1928 	entries += ret;
1929 	entry_bytes += local_read(&cpu_buffer->reader_page->page->commit);
1930 	local_set(&cpu_buffer->reader_page->entries, ret);
1931 
1932 	orig_head = head_page = cpu_buffer->head_page;
1933 	ts = head_page->page->time_stamp;
1934 
1935 	/*
1936 	 * Try to rewind the head so that we can read the pages which already
1937 	 * read in the previous boot.
1938 	 */
1939 	if (head_page == cpu_buffer->tail_page)
1940 		goto skip_rewind;
1941 
1942 	rb_dec_page(&head_page);
1943 	for (i = 0; i < meta->nr_subbufs + 1; i++, rb_dec_page(&head_page)) {
1944 
1945 		/* Rewind until tail (writer) page. */
1946 		if (head_page == cpu_buffer->tail_page)
1947 			break;
1948 
1949 		/* Ensure the page has older data than head. */
1950 		if (ts < head_page->page->time_stamp)
1951 			break;
1952 
1953 		ts = head_page->page->time_stamp;
1954 		/* Ensure the page has correct timestamp and some data. */
1955 		if (!ts || rb_page_commit(head_page) == 0)
1956 			break;
1957 
1958 		/* Stop rewind if the page is invalid. */
1959 		ret = rb_validate_buffer(head_page->page, cpu_buffer->cpu);
1960 		if (ret < 0)
1961 			break;
1962 
1963 		/* Recover the number of entries and update stats. */
1964 		local_set(&head_page->entries, ret);
1965 		if (ret)
1966 			local_inc(&cpu_buffer->pages_touched);
1967 		entries += ret;
1968 		entry_bytes += rb_page_commit(head_page);
1969 	}
1970 	if (i)
1971 		pr_info("Ring buffer [%d] rewound %d pages\n", cpu_buffer->cpu, i);
1972 
1973 	/* The last rewound page must be skipped. */
1974 	if (head_page != orig_head)
1975 		rb_inc_page(&head_page);
1976 
1977 	/*
1978 	 * If the ring buffer was rewound, then inject the reader page
1979 	 * into the location just before the original head page.
1980 	 */
1981 	if (head_page != orig_head) {
1982 		struct buffer_page *bpage = orig_head;
1983 
1984 		rb_dec_page(&bpage);
1985 		/*
1986 		 * Insert the reader_page before the original head page.
1987 		 * Since the list encode RB_PAGE flags, general list
1988 		 * operations should be avoided.
1989 		 */
1990 		cpu_buffer->reader_page->list.next = &orig_head->list;
1991 		cpu_buffer->reader_page->list.prev = orig_head->list.prev;
1992 		orig_head->list.prev = &cpu_buffer->reader_page->list;
1993 		bpage->list.next = &cpu_buffer->reader_page->list;
1994 
1995 		/* Make the head_page the reader page */
1996 		cpu_buffer->reader_page = head_page;
1997 		bpage = head_page;
1998 		rb_inc_page(&head_page);
1999 		head_page->list.prev = bpage->list.prev;
2000 		rb_dec_page(&bpage);
2001 		bpage->list.next = &head_page->list;
2002 		rb_set_list_to_head(&bpage->list);
2003 		cpu_buffer->pages = &head_page->list;
2004 
2005 		cpu_buffer->head_page = head_page;
2006 		meta->head_buffer = (unsigned long)head_page->page;
2007 
2008 		/* Reset all the indexes */
2009 		bpage = cpu_buffer->reader_page;
2010 		meta->buffers[0] = rb_meta_subbuf_idx(meta, bpage->page);
2011 		bpage->id = 0;
2012 
2013 		for (i = 1, bpage = head_page; i < meta->nr_subbufs;
2014 		     i++, rb_inc_page(&bpage)) {
2015 			meta->buffers[i] = rb_meta_subbuf_idx(meta, bpage->page);
2016 			bpage->id = i;
2017 		}
2018 
2019 		/* We'll restart verifying from orig_head */
2020 		head_page = orig_head;
2021 	}
2022 
2023  skip_rewind:
2024 	/* If the commit_buffer is the reader page, update the commit page */
2025 	if (meta->commit_buffer == (unsigned long)cpu_buffer->reader_page->page) {
2026 		cpu_buffer->commit_page = cpu_buffer->reader_page;
2027 		/* Nothing more to do, the only page is the reader page */
2028 		goto done;
2029 	}
2030 
2031 	/* Iterate until finding the commit page */
2032 	for (i = 0; i < meta->nr_subbufs + 1; i++, rb_inc_page(&head_page)) {
2033 
2034 		/* Reader page has already been done */
2035 		if (head_page == cpu_buffer->reader_page)
2036 			continue;
2037 
2038 		ret = rb_validate_buffer(head_page->page, cpu_buffer->cpu);
2039 		if (ret < 0) {
2040 			pr_info("Ring buffer meta [%d] invalid buffer page\n",
2041 				cpu_buffer->cpu);
2042 			goto invalid;
2043 		}
2044 
2045 		/* If the buffer has content, update pages_touched */
2046 		if (ret)
2047 			local_inc(&cpu_buffer->pages_touched);
2048 
2049 		entries += ret;
2050 		entry_bytes += local_read(&head_page->page->commit);
2051 		local_set(&cpu_buffer->head_page->entries, ret);
2052 
2053 		if (head_page == cpu_buffer->commit_page)
2054 			break;
2055 	}
2056 
2057 	if (head_page != cpu_buffer->commit_page) {
2058 		pr_info("Ring buffer meta [%d] commit page not found\n",
2059 			cpu_buffer->cpu);
2060 		goto invalid;
2061 	}
2062  done:
2063 	local_set(&cpu_buffer->entries, entries);
2064 	local_set(&cpu_buffer->entries_bytes, entry_bytes);
2065 
2066 	pr_info("Ring buffer meta [%d] is from previous boot!\n", cpu_buffer->cpu);
2067 	return;
2068 
2069  invalid:
2070 	/* The content of the buffers are invalid, reset the meta data */
2071 	meta->head_buffer = 0;
2072 	meta->commit_buffer = 0;
2073 
2074 	/* Reset the reader page */
2075 	local_set(&cpu_buffer->reader_page->entries, 0);
2076 	local_set(&cpu_buffer->reader_page->page->commit, 0);
2077 
2078 	/* Reset all the subbuffers */
2079 	for (i = 0; i < meta->nr_subbufs - 1; i++, rb_inc_page(&head_page)) {
2080 		local_set(&head_page->entries, 0);
2081 		local_set(&head_page->page->commit, 0);
2082 	}
2083 }
2084 
2085 static void rb_range_meta_init(struct trace_buffer *buffer, int nr_pages, int scratch_size)
2086 {
2087 	struct ring_buffer_cpu_meta *meta;
2088 	unsigned long *subbuf_mask;
2089 	unsigned long delta;
2090 	void *subbuf;
2091 	bool valid = false;
2092 	int cpu;
2093 	int i;
2094 
2095 	/* Create a mask to test the subbuf array */
2096 	subbuf_mask = bitmap_alloc(nr_pages + 1, GFP_KERNEL);
2097 	/* If subbuf_mask fails to allocate, then rb_meta_valid() will return false */
2098 
2099 	if (rb_meta_init(buffer, scratch_size))
2100 		valid = true;
2101 
2102 	for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
2103 		void *next_meta;
2104 
2105 		meta = rb_range_meta(buffer, nr_pages, cpu);
2106 
2107 		if (valid && rb_cpu_meta_valid(meta, cpu, buffer, nr_pages, subbuf_mask)) {
2108 			/* Make the mappings match the current address */
2109 			subbuf = rb_subbufs_from_meta(meta);
2110 			delta = (unsigned long)subbuf - meta->first_buffer;
2111 			meta->first_buffer += delta;
2112 			meta->head_buffer += delta;
2113 			meta->commit_buffer += delta;
2114 			continue;
2115 		}
2116 
2117 		if (cpu < nr_cpu_ids - 1)
2118 			next_meta = rb_range_meta(buffer, nr_pages, cpu + 1);
2119 		else
2120 			next_meta = (void *)buffer->range_addr_end;
2121 
2122 		memset(meta, 0, next_meta - (void *)meta);
2123 
2124 		meta->nr_subbufs = nr_pages + 1;
2125 		meta->subbuf_size = PAGE_SIZE;
2126 
2127 		subbuf = rb_subbufs_from_meta(meta);
2128 
2129 		meta->first_buffer = (unsigned long)subbuf;
2130 
2131 		/*
2132 		 * The buffers[] array holds the order of the sub-buffers
2133 		 * that are after the meta data. The sub-buffers may
2134 		 * be swapped out when read and inserted into a different
2135 		 * location of the ring buffer. Although their addresses
2136 		 * remain the same, the buffers[] array contains the
2137 		 * index into the sub-buffers holding their actual order.
2138 		 */
2139 		for (i = 0; i < meta->nr_subbufs; i++) {
2140 			meta->buffers[i] = i;
2141 			rb_init_page(subbuf);
2142 			subbuf += meta->subbuf_size;
2143 		}
2144 	}
2145 	bitmap_free(subbuf_mask);
2146 }
2147 
2148 static void *rbm_start(struct seq_file *m, loff_t *pos)
2149 {
2150 	struct ring_buffer_per_cpu *cpu_buffer = m->private;
2151 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2152 	unsigned long val;
2153 
2154 	if (!meta)
2155 		return NULL;
2156 
2157 	if (*pos > meta->nr_subbufs)
2158 		return NULL;
2159 
2160 	val = *pos;
2161 	val++;
2162 
2163 	return (void *)val;
2164 }
2165 
2166 static void *rbm_next(struct seq_file *m, void *v, loff_t *pos)
2167 {
2168 	(*pos)++;
2169 
2170 	return rbm_start(m, pos);
2171 }
2172 
2173 static int rbm_show(struct seq_file *m, void *v)
2174 {
2175 	struct ring_buffer_per_cpu *cpu_buffer = m->private;
2176 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2177 	unsigned long val = (unsigned long)v;
2178 
2179 	if (val == 1) {
2180 		seq_printf(m, "head_buffer:   %d\n",
2181 			   rb_meta_subbuf_idx(meta, (void *)meta->head_buffer));
2182 		seq_printf(m, "commit_buffer: %d\n",
2183 			   rb_meta_subbuf_idx(meta, (void *)meta->commit_buffer));
2184 		seq_printf(m, "subbuf_size:   %d\n", meta->subbuf_size);
2185 		seq_printf(m, "nr_subbufs:    %d\n", meta->nr_subbufs);
2186 		return 0;
2187 	}
2188 
2189 	val -= 2;
2190 	seq_printf(m, "buffer[%ld]:    %d\n", val, meta->buffers[val]);
2191 
2192 	return 0;
2193 }
2194 
2195 static void rbm_stop(struct seq_file *m, void *p)
2196 {
2197 }
2198 
2199 static const struct seq_operations rb_meta_seq_ops = {
2200 	.start		= rbm_start,
2201 	.next		= rbm_next,
2202 	.show		= rbm_show,
2203 	.stop		= rbm_stop,
2204 };
2205 
2206 int ring_buffer_meta_seq_init(struct file *file, struct trace_buffer *buffer, int cpu)
2207 {
2208 	struct seq_file *m;
2209 	int ret;
2210 
2211 	ret = seq_open(file, &rb_meta_seq_ops);
2212 	if (ret)
2213 		return ret;
2214 
2215 	m = file->private_data;
2216 	m->private = buffer->buffers[cpu];
2217 
2218 	return 0;
2219 }
2220 
2221 /* Map the buffer_pages to the previous head and commit pages */
2222 static void rb_meta_buffer_update(struct ring_buffer_per_cpu *cpu_buffer,
2223 				  struct buffer_page *bpage)
2224 {
2225 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2226 
2227 	if (meta->head_buffer == (unsigned long)bpage->page)
2228 		cpu_buffer->head_page = bpage;
2229 
2230 	if (meta->commit_buffer == (unsigned long)bpage->page) {
2231 		cpu_buffer->commit_page = bpage;
2232 		cpu_buffer->tail_page = bpage;
2233 	}
2234 }
2235 
2236 static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
2237 		long nr_pages, struct list_head *pages)
2238 {
2239 	struct trace_buffer *buffer = cpu_buffer->buffer;
2240 	struct ring_buffer_cpu_meta *meta = NULL;
2241 	struct buffer_page *bpage, *tmp;
2242 	bool user_thread = current->mm != NULL;
2243 	long i;
2244 
2245 	/*
2246 	 * Check if the available memory is there first.
2247 	 * Note, si_mem_available() only gives us a rough estimate of available
2248 	 * memory. It may not be accurate. But we don't care, we just want
2249 	 * to prevent doing any allocation when it is obvious that it is
2250 	 * not going to succeed.
2251 	 */
2252 	i = si_mem_available();
2253 	if (i < nr_pages)
2254 		return -ENOMEM;
2255 
2256 	/*
2257 	 * If a user thread allocates too much, and si_mem_available()
2258 	 * reports there's enough memory, even though there is not.
2259 	 * Make sure the OOM killer kills this thread. This can happen
2260 	 * even with RETRY_MAYFAIL because another task may be doing
2261 	 * an allocation after this task has taken all memory.
2262 	 * This is the task the OOM killer needs to take out during this
2263 	 * loop, even if it was triggered by an allocation somewhere else.
2264 	 */
2265 	if (user_thread)
2266 		set_current_oom_origin();
2267 
2268 	if (buffer->range_addr_start)
2269 		meta = rb_range_meta(buffer, nr_pages, cpu_buffer->cpu);
2270 
2271 	for (i = 0; i < nr_pages; i++) {
2272 
2273 		bpage = alloc_cpu_page(cpu_buffer->cpu);
2274 		if (!bpage)
2275 			goto free_pages;
2276 
2277 		rb_check_bpage(cpu_buffer, bpage);
2278 
2279 		/*
2280 		 * Append the pages as for mapped buffers we want to keep
2281 		 * the order
2282 		 */
2283 		list_add_tail(&bpage->list, pages);
2284 
2285 		if (meta) {
2286 			/* A range was given. Use that for the buffer page */
2287 			bpage->page = rb_range_buffer(cpu_buffer, i + 1);
2288 			if (!bpage->page)
2289 				goto free_pages;
2290 			/* If this is valid from a previous boot */
2291 			if (meta->head_buffer)
2292 				rb_meta_buffer_update(cpu_buffer, bpage);
2293 			bpage->range = 1;
2294 			bpage->id = i + 1;
2295 		} else {
2296 			int order = cpu_buffer->buffer->subbuf_order;
2297 			bpage->page = alloc_cpu_data(cpu_buffer->cpu, order);
2298 			if (!bpage->page)
2299 				goto free_pages;
2300 		}
2301 		bpage->order = cpu_buffer->buffer->subbuf_order;
2302 
2303 		if (user_thread && fatal_signal_pending(current))
2304 			goto free_pages;
2305 	}
2306 	if (user_thread)
2307 		clear_current_oom_origin();
2308 
2309 	return 0;
2310 
2311 free_pages:
2312 	list_for_each_entry_safe(bpage, tmp, pages, list) {
2313 		list_del_init(&bpage->list);
2314 		free_buffer_page(bpage);
2315 	}
2316 	if (user_thread)
2317 		clear_current_oom_origin();
2318 
2319 	return -ENOMEM;
2320 }
2321 
2322 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
2323 			     unsigned long nr_pages)
2324 {
2325 	LIST_HEAD(pages);
2326 
2327 	WARN_ON(!nr_pages);
2328 
2329 	if (__rb_allocate_pages(cpu_buffer, nr_pages, &pages))
2330 		return -ENOMEM;
2331 
2332 	/*
2333 	 * The ring buffer page list is a circular list that does not
2334 	 * start and end with a list head. All page list items point to
2335 	 * other pages.
2336 	 */
2337 	cpu_buffer->pages = pages.next;
2338 	list_del(&pages);
2339 
2340 	cpu_buffer->nr_pages = nr_pages;
2341 
2342 	rb_check_pages(cpu_buffer);
2343 
2344 	return 0;
2345 }
2346 
2347 static struct ring_buffer_per_cpu *
2348 rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu)
2349 {
2350 	struct ring_buffer_per_cpu *cpu_buffer __free(kfree) =
2351 		alloc_cpu_buffer(cpu);
2352 	struct ring_buffer_cpu_meta *meta;
2353 	struct buffer_page *bpage;
2354 	int ret;
2355 
2356 	if (!cpu_buffer)
2357 		return NULL;
2358 
2359 	cpu_buffer->cpu = cpu;
2360 	cpu_buffer->buffer = buffer;
2361 	raw_spin_lock_init(&cpu_buffer->reader_lock);
2362 	lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
2363 	cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
2364 	INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
2365 	init_completion(&cpu_buffer->update_done);
2366 	init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
2367 	init_waitqueue_head(&cpu_buffer->irq_work.waiters);
2368 	init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
2369 	mutex_init(&cpu_buffer->mapping_lock);
2370 
2371 	bpage = alloc_cpu_page(cpu);
2372 	if (!bpage)
2373 		return NULL;
2374 
2375 	rb_check_bpage(cpu_buffer, bpage);
2376 
2377 	cpu_buffer->reader_page = bpage;
2378 
2379 	if (buffer->range_addr_start) {
2380 		/*
2381 		 * Range mapped buffers have the same restrictions as memory
2382 		 * mapped ones do.
2383 		 */
2384 		cpu_buffer->mapped = 1;
2385 		cpu_buffer->ring_meta = rb_range_meta(buffer, nr_pages, cpu);
2386 		bpage->page = rb_range_buffer(cpu_buffer, 0);
2387 		if (!bpage->page)
2388 			goto fail_free_reader;
2389 		if (cpu_buffer->ring_meta->head_buffer)
2390 			rb_meta_buffer_update(cpu_buffer, bpage);
2391 		bpage->range = 1;
2392 	} else {
2393 		int order = cpu_buffer->buffer->subbuf_order;
2394 		bpage->page = alloc_cpu_data(cpu, order);
2395 		if (!bpage->page)
2396 			goto fail_free_reader;
2397 	}
2398 
2399 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
2400 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
2401 
2402 	ret = rb_allocate_pages(cpu_buffer, nr_pages);
2403 	if (ret < 0)
2404 		goto fail_free_reader;
2405 
2406 	rb_meta_validate_events(cpu_buffer);
2407 
2408 	/* If the boot meta was valid then this has already been updated */
2409 	meta = cpu_buffer->ring_meta;
2410 	if (!meta || !meta->head_buffer ||
2411 	    !cpu_buffer->head_page || !cpu_buffer->commit_page || !cpu_buffer->tail_page) {
2412 		if (meta && meta->head_buffer &&
2413 		    (cpu_buffer->head_page || cpu_buffer->commit_page || cpu_buffer->tail_page)) {
2414 			pr_warn("Ring buffer meta buffers not all mapped\n");
2415 			if (!cpu_buffer->head_page)
2416 				pr_warn("   Missing head_page\n");
2417 			if (!cpu_buffer->commit_page)
2418 				pr_warn("   Missing commit_page\n");
2419 			if (!cpu_buffer->tail_page)
2420 				pr_warn("   Missing tail_page\n");
2421 		}
2422 
2423 		cpu_buffer->head_page
2424 			= list_entry(cpu_buffer->pages, struct buffer_page, list);
2425 		cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
2426 
2427 		rb_head_page_activate(cpu_buffer);
2428 
2429 		if (cpu_buffer->ring_meta)
2430 			meta->commit_buffer = meta->head_buffer;
2431 	} else {
2432 		/* The valid meta buffer still needs to activate the head page */
2433 		rb_head_page_activate(cpu_buffer);
2434 	}
2435 
2436 	return_ptr(cpu_buffer);
2437 
2438  fail_free_reader:
2439 	free_buffer_page(cpu_buffer->reader_page);
2440 
2441 	return NULL;
2442 }
2443 
2444 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
2445 {
2446 	struct list_head *head = cpu_buffer->pages;
2447 	struct buffer_page *bpage, *tmp;
2448 
2449 	irq_work_sync(&cpu_buffer->irq_work.work);
2450 
2451 	free_buffer_page(cpu_buffer->reader_page);
2452 
2453 	if (head) {
2454 		rb_head_page_deactivate(cpu_buffer);
2455 
2456 		list_for_each_entry_safe(bpage, tmp, head, list) {
2457 			list_del_init(&bpage->list);
2458 			free_buffer_page(bpage);
2459 		}
2460 		bpage = list_entry(head, struct buffer_page, list);
2461 		free_buffer_page(bpage);
2462 	}
2463 
2464 	free_page((unsigned long)cpu_buffer->free_page);
2465 
2466 	kfree(cpu_buffer);
2467 }
2468 
2469 static struct trace_buffer *alloc_buffer(unsigned long size, unsigned flags,
2470 					 int order, unsigned long start,
2471 					 unsigned long end,
2472 					 unsigned long scratch_size,
2473 					 struct lock_class_key *key)
2474 {
2475 	struct trace_buffer *buffer __free(kfree) = NULL;
2476 	long nr_pages;
2477 	int subbuf_size;
2478 	int bsize;
2479 	int cpu;
2480 	int ret;
2481 
2482 	/* keep it in its own cache line */
2483 	buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
2484 			 GFP_KERNEL);
2485 	if (!buffer)
2486 		return NULL;
2487 
2488 	if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
2489 		return NULL;
2490 
2491 	buffer->subbuf_order = order;
2492 	subbuf_size = (PAGE_SIZE << order);
2493 	buffer->subbuf_size = subbuf_size - BUF_PAGE_HDR_SIZE;
2494 
2495 	/* Max payload is buffer page size - header (8bytes) */
2496 	buffer->max_data_size = buffer->subbuf_size - (sizeof(u32) * 2);
2497 
2498 	buffer->flags = flags;
2499 	buffer->clock = trace_clock_local;
2500 	buffer->reader_lock_key = key;
2501 
2502 	init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters);
2503 	init_waitqueue_head(&buffer->irq_work.waiters);
2504 
2505 	buffer->cpus = nr_cpu_ids;
2506 
2507 	bsize = sizeof(void *) * nr_cpu_ids;
2508 	buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
2509 				  GFP_KERNEL);
2510 	if (!buffer->buffers)
2511 		goto fail_free_cpumask;
2512 
2513 	/* If start/end are specified, then that overrides size */
2514 	if (start && end) {
2515 		unsigned long buffers_start;
2516 		unsigned long ptr;
2517 		int n;
2518 
2519 		/* Make sure that start is word aligned */
2520 		start = ALIGN(start, sizeof(long));
2521 
2522 		/* scratch_size needs to be aligned too */
2523 		scratch_size = ALIGN(scratch_size, sizeof(long));
2524 
2525 		/* Subtract the buffer meta data and word aligned */
2526 		buffers_start = start + sizeof(struct ring_buffer_cpu_meta);
2527 		buffers_start = ALIGN(buffers_start, sizeof(long));
2528 		buffers_start += scratch_size;
2529 
2530 		/* Calculate the size for the per CPU data */
2531 		size = end - buffers_start;
2532 		size = size / nr_cpu_ids;
2533 
2534 		/*
2535 		 * The number of sub-buffers (nr_pages) is determined by the
2536 		 * total size allocated minus the meta data size.
2537 		 * Then that is divided by the number of per CPU buffers
2538 		 * needed, plus account for the integer array index that
2539 		 * will be appended to the meta data.
2540 		 */
2541 		nr_pages = (size - sizeof(struct ring_buffer_cpu_meta)) /
2542 			(subbuf_size + sizeof(int));
2543 		/* Need at least two pages plus the reader page */
2544 		if (nr_pages < 3)
2545 			goto fail_free_buffers;
2546 
2547  again:
2548 		/* Make sure that the size fits aligned */
2549 		for (n = 0, ptr = buffers_start; n < nr_cpu_ids; n++) {
2550 			ptr += sizeof(struct ring_buffer_cpu_meta) +
2551 				sizeof(int) * nr_pages;
2552 			ptr = ALIGN(ptr, subbuf_size);
2553 			ptr += subbuf_size * nr_pages;
2554 		}
2555 		if (ptr > end) {
2556 			if (nr_pages <= 3)
2557 				goto fail_free_buffers;
2558 			nr_pages--;
2559 			goto again;
2560 		}
2561 
2562 		/* nr_pages should not count the reader page */
2563 		nr_pages--;
2564 		buffer->range_addr_start = start;
2565 		buffer->range_addr_end = end;
2566 
2567 		rb_range_meta_init(buffer, nr_pages, scratch_size);
2568 	} else {
2569 
2570 		/* need at least two pages */
2571 		nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size);
2572 		if (nr_pages < 2)
2573 			nr_pages = 2;
2574 	}
2575 
2576 	cpu = raw_smp_processor_id();
2577 	cpumask_set_cpu(cpu, buffer->cpumask);
2578 	buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
2579 	if (!buffer->buffers[cpu])
2580 		goto fail_free_buffers;
2581 
2582 	ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
2583 	if (ret < 0)
2584 		goto fail_free_buffers;
2585 
2586 	mutex_init(&buffer->mutex);
2587 
2588 	return_ptr(buffer);
2589 
2590  fail_free_buffers:
2591 	for_each_buffer_cpu(buffer, cpu) {
2592 		if (buffer->buffers[cpu])
2593 			rb_free_cpu_buffer(buffer->buffers[cpu]);
2594 	}
2595 	kfree(buffer->buffers);
2596 
2597  fail_free_cpumask:
2598 	free_cpumask_var(buffer->cpumask);
2599 
2600 	return NULL;
2601 }
2602 
2603 /**
2604  * __ring_buffer_alloc - allocate a new ring_buffer
2605  * @size: the size in bytes per cpu that is needed.
2606  * @flags: attributes to set for the ring buffer.
2607  * @key: ring buffer reader_lock_key.
2608  *
2609  * Currently the only flag that is available is the RB_FL_OVERWRITE
2610  * flag. This flag means that the buffer will overwrite old data
2611  * when the buffer wraps. If this flag is not set, the buffer will
2612  * drop data when the tail hits the head.
2613  */
2614 struct trace_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
2615 					struct lock_class_key *key)
2616 {
2617 	/* Default buffer page size - one system page */
2618 	return alloc_buffer(size, flags, 0, 0, 0, 0, key);
2619 
2620 }
2621 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
2622 
2623 /**
2624  * __ring_buffer_alloc_range - allocate a new ring_buffer from existing memory
2625  * @size: the size in bytes per cpu that is needed.
2626  * @flags: attributes to set for the ring buffer.
2627  * @order: sub-buffer order
2628  * @start: start of allocated range
2629  * @range_size: size of allocated range
2630  * @scratch_size: size of scratch area (for preallocated memory buffers)
2631  * @key: ring buffer reader_lock_key.
2632  *
2633  * Currently the only flag that is available is the RB_FL_OVERWRITE
2634  * flag. This flag means that the buffer will overwrite old data
2635  * when the buffer wraps. If this flag is not set, the buffer will
2636  * drop data when the tail hits the head.
2637  */
2638 struct trace_buffer *__ring_buffer_alloc_range(unsigned long size, unsigned flags,
2639 					       int order, unsigned long start,
2640 					       unsigned long range_size,
2641 					       unsigned long scratch_size,
2642 					       struct lock_class_key *key)
2643 {
2644 	return alloc_buffer(size, flags, order, start, start + range_size,
2645 			    scratch_size, key);
2646 }
2647 
2648 void *ring_buffer_meta_scratch(struct trace_buffer *buffer, unsigned int *size)
2649 {
2650 	struct ring_buffer_meta *meta;
2651 	void *ptr;
2652 
2653 	if (!buffer || !buffer->meta)
2654 		return NULL;
2655 
2656 	meta = buffer->meta;
2657 
2658 	ptr = (void *)ALIGN((unsigned long)meta + sizeof(*meta), sizeof(long));
2659 
2660 	if (size)
2661 		*size = (void *)meta + meta->buffers_offset - ptr;
2662 
2663 	return ptr;
2664 }
2665 
2666 /**
2667  * ring_buffer_free - free a ring buffer.
2668  * @buffer: the buffer to free.
2669  */
2670 void
2671 ring_buffer_free(struct trace_buffer *buffer)
2672 {
2673 	int cpu;
2674 
2675 	cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
2676 
2677 	irq_work_sync(&buffer->irq_work.work);
2678 
2679 	for_each_buffer_cpu(buffer, cpu)
2680 		rb_free_cpu_buffer(buffer->buffers[cpu]);
2681 
2682 	kfree(buffer->buffers);
2683 	free_cpumask_var(buffer->cpumask);
2684 
2685 	kfree(buffer);
2686 }
2687 EXPORT_SYMBOL_GPL(ring_buffer_free);
2688 
2689 void ring_buffer_set_clock(struct trace_buffer *buffer,
2690 			   u64 (*clock)(void))
2691 {
2692 	buffer->clock = clock;
2693 }
2694 
2695 void ring_buffer_set_time_stamp_abs(struct trace_buffer *buffer, bool abs)
2696 {
2697 	buffer->time_stamp_abs = abs;
2698 }
2699 
2700 bool ring_buffer_time_stamp_abs(struct trace_buffer *buffer)
2701 {
2702 	return buffer->time_stamp_abs;
2703 }
2704 
2705 static inline unsigned long rb_page_entries(struct buffer_page *bpage)
2706 {
2707 	return local_read(&bpage->entries) & RB_WRITE_MASK;
2708 }
2709 
2710 static inline unsigned long rb_page_write(struct buffer_page *bpage)
2711 {
2712 	return local_read(&bpage->write) & RB_WRITE_MASK;
2713 }
2714 
2715 static bool
2716 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
2717 {
2718 	struct list_head *tail_page, *to_remove, *next_page;
2719 	struct buffer_page *to_remove_page, *tmp_iter_page;
2720 	struct buffer_page *last_page, *first_page;
2721 	unsigned long nr_removed;
2722 	unsigned long head_bit;
2723 	int page_entries;
2724 
2725 	head_bit = 0;
2726 
2727 	raw_spin_lock_irq(&cpu_buffer->reader_lock);
2728 	atomic_inc(&cpu_buffer->record_disabled);
2729 	/*
2730 	 * We don't race with the readers since we have acquired the reader
2731 	 * lock. We also don't race with writers after disabling recording.
2732 	 * This makes it easy to figure out the first and the last page to be
2733 	 * removed from the list. We unlink all the pages in between including
2734 	 * the first and last pages. This is done in a busy loop so that we
2735 	 * lose the least number of traces.
2736 	 * The pages are freed after we restart recording and unlock readers.
2737 	 */
2738 	tail_page = &cpu_buffer->tail_page->list;
2739 
2740 	/*
2741 	 * tail page might be on reader page, we remove the next page
2742 	 * from the ring buffer
2743 	 */
2744 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
2745 		tail_page = rb_list_head(tail_page->next);
2746 	to_remove = tail_page;
2747 
2748 	/* start of pages to remove */
2749 	first_page = list_entry(rb_list_head(to_remove->next),
2750 				struct buffer_page, list);
2751 
2752 	for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
2753 		to_remove = rb_list_head(to_remove)->next;
2754 		head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
2755 	}
2756 	/* Read iterators need to reset themselves when some pages removed */
2757 	cpu_buffer->pages_removed += nr_removed;
2758 
2759 	next_page = rb_list_head(to_remove)->next;
2760 
2761 	/*
2762 	 * Now we remove all pages between tail_page and next_page.
2763 	 * Make sure that we have head_bit value preserved for the
2764 	 * next page
2765 	 */
2766 	tail_page->next = (struct list_head *)((unsigned long)next_page |
2767 						head_bit);
2768 	next_page = rb_list_head(next_page);
2769 	next_page->prev = tail_page;
2770 
2771 	/* make sure pages points to a valid page in the ring buffer */
2772 	cpu_buffer->pages = next_page;
2773 	cpu_buffer->cnt++;
2774 
2775 	/* update head page */
2776 	if (head_bit)
2777 		cpu_buffer->head_page = list_entry(next_page,
2778 						struct buffer_page, list);
2779 
2780 	/* pages are removed, resume tracing and then free the pages */
2781 	atomic_dec(&cpu_buffer->record_disabled);
2782 	raw_spin_unlock_irq(&cpu_buffer->reader_lock);
2783 
2784 	RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
2785 
2786 	/* last buffer page to remove */
2787 	last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
2788 				list);
2789 	tmp_iter_page = first_page;
2790 
2791 	do {
2792 		cond_resched();
2793 
2794 		to_remove_page = tmp_iter_page;
2795 		rb_inc_page(&tmp_iter_page);
2796 
2797 		/* update the counters */
2798 		page_entries = rb_page_entries(to_remove_page);
2799 		if (page_entries) {
2800 			/*
2801 			 * If something was added to this page, it was full
2802 			 * since it is not the tail page. So we deduct the
2803 			 * bytes consumed in ring buffer from here.
2804 			 * Increment overrun to account for the lost events.
2805 			 */
2806 			local_add(page_entries, &cpu_buffer->overrun);
2807 			local_sub(rb_page_commit(to_remove_page), &cpu_buffer->entries_bytes);
2808 			local_inc(&cpu_buffer->pages_lost);
2809 		}
2810 
2811 		/*
2812 		 * We have already removed references to this list item, just
2813 		 * free up the buffer_page and its page
2814 		 */
2815 		free_buffer_page(to_remove_page);
2816 		nr_removed--;
2817 
2818 	} while (to_remove_page != last_page);
2819 
2820 	RB_WARN_ON(cpu_buffer, nr_removed);
2821 
2822 	return nr_removed == 0;
2823 }
2824 
2825 static bool
2826 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
2827 {
2828 	struct list_head *pages = &cpu_buffer->new_pages;
2829 	unsigned long flags;
2830 	bool success;
2831 	int retries;
2832 
2833 	/* Can be called at early boot up, where interrupts must not been enabled */
2834 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2835 	/*
2836 	 * We are holding the reader lock, so the reader page won't be swapped
2837 	 * in the ring buffer. Now we are racing with the writer trying to
2838 	 * move head page and the tail page.
2839 	 * We are going to adapt the reader page update process where:
2840 	 * 1. We first splice the start and end of list of new pages between
2841 	 *    the head page and its previous page.
2842 	 * 2. We cmpxchg the prev_page->next to point from head page to the
2843 	 *    start of new pages list.
2844 	 * 3. Finally, we update the head->prev to the end of new list.
2845 	 *
2846 	 * We will try this process 10 times, to make sure that we don't keep
2847 	 * spinning.
2848 	 */
2849 	retries = 10;
2850 	success = false;
2851 	while (retries--) {
2852 		struct list_head *head_page, *prev_page;
2853 		struct list_head *last_page, *first_page;
2854 		struct list_head *head_page_with_bit;
2855 		struct buffer_page *hpage = rb_set_head_page(cpu_buffer);
2856 
2857 		if (!hpage)
2858 			break;
2859 		head_page = &hpage->list;
2860 		prev_page = head_page->prev;
2861 
2862 		first_page = pages->next;
2863 		last_page  = pages->prev;
2864 
2865 		head_page_with_bit = (struct list_head *)
2866 				     ((unsigned long)head_page | RB_PAGE_HEAD);
2867 
2868 		last_page->next = head_page_with_bit;
2869 		first_page->prev = prev_page;
2870 
2871 		/* caution: head_page_with_bit gets updated on cmpxchg failure */
2872 		if (try_cmpxchg(&prev_page->next,
2873 				&head_page_with_bit, first_page)) {
2874 			/*
2875 			 * yay, we replaced the page pointer to our new list,
2876 			 * now, we just have to update to head page's prev
2877 			 * pointer to point to end of list
2878 			 */
2879 			head_page->prev = last_page;
2880 			cpu_buffer->cnt++;
2881 			success = true;
2882 			break;
2883 		}
2884 	}
2885 
2886 	if (success)
2887 		INIT_LIST_HEAD(pages);
2888 	/*
2889 	 * If we weren't successful in adding in new pages, warn and stop
2890 	 * tracing
2891 	 */
2892 	RB_WARN_ON(cpu_buffer, !success);
2893 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2894 
2895 	/* free pages if they weren't inserted */
2896 	if (!success) {
2897 		struct buffer_page *bpage, *tmp;
2898 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2899 					 list) {
2900 			list_del_init(&bpage->list);
2901 			free_buffer_page(bpage);
2902 		}
2903 	}
2904 	return success;
2905 }
2906 
2907 static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
2908 {
2909 	bool success;
2910 
2911 	if (cpu_buffer->nr_pages_to_update > 0)
2912 		success = rb_insert_pages(cpu_buffer);
2913 	else
2914 		success = rb_remove_pages(cpu_buffer,
2915 					-cpu_buffer->nr_pages_to_update);
2916 
2917 	if (success)
2918 		cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
2919 }
2920 
2921 static void update_pages_handler(struct work_struct *work)
2922 {
2923 	struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
2924 			struct ring_buffer_per_cpu, update_pages_work);
2925 	rb_update_pages(cpu_buffer);
2926 	complete(&cpu_buffer->update_done);
2927 }
2928 
2929 /**
2930  * ring_buffer_resize - resize the ring buffer
2931  * @buffer: the buffer to resize.
2932  * @size: the new size.
2933  * @cpu_id: the cpu buffer to resize
2934  *
2935  * Minimum size is 2 * buffer->subbuf_size.
2936  *
2937  * Returns 0 on success and < 0 on failure.
2938  */
2939 int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size,
2940 			int cpu_id)
2941 {
2942 	struct ring_buffer_per_cpu *cpu_buffer;
2943 	unsigned long nr_pages;
2944 	int cpu, err;
2945 
2946 	/*
2947 	 * Always succeed at resizing a non-existent buffer:
2948 	 */
2949 	if (!buffer)
2950 		return 0;
2951 
2952 	/* Make sure the requested buffer exists */
2953 	if (cpu_id != RING_BUFFER_ALL_CPUS &&
2954 	    !cpumask_test_cpu(cpu_id, buffer->cpumask))
2955 		return 0;
2956 
2957 	nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size);
2958 
2959 	/* we need a minimum of two pages */
2960 	if (nr_pages < 2)
2961 		nr_pages = 2;
2962 
2963 	/*
2964 	 * Keep CPUs from coming online while resizing to synchronize
2965 	 * with new per CPU buffers being created.
2966 	 */
2967 	guard(cpus_read_lock)();
2968 
2969 	/* prevent another thread from changing buffer sizes */
2970 	mutex_lock(&buffer->mutex);
2971 	atomic_inc(&buffer->resizing);
2972 
2973 	if (cpu_id == RING_BUFFER_ALL_CPUS) {
2974 		/*
2975 		 * Don't succeed if resizing is disabled, as a reader might be
2976 		 * manipulating the ring buffer and is expecting a sane state while
2977 		 * this is true.
2978 		 */
2979 		for_each_buffer_cpu(buffer, cpu) {
2980 			cpu_buffer = buffer->buffers[cpu];
2981 			if (atomic_read(&cpu_buffer->resize_disabled)) {
2982 				err = -EBUSY;
2983 				goto out_err_unlock;
2984 			}
2985 		}
2986 
2987 		/* calculate the pages to update */
2988 		for_each_buffer_cpu(buffer, cpu) {
2989 			cpu_buffer = buffer->buffers[cpu];
2990 
2991 			cpu_buffer->nr_pages_to_update = nr_pages -
2992 							cpu_buffer->nr_pages;
2993 			/*
2994 			 * nothing more to do for removing pages or no update
2995 			 */
2996 			if (cpu_buffer->nr_pages_to_update <= 0)
2997 				continue;
2998 			/*
2999 			 * to add pages, make sure all new pages can be
3000 			 * allocated without receiving ENOMEM
3001 			 */
3002 			INIT_LIST_HEAD(&cpu_buffer->new_pages);
3003 			if (__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
3004 						&cpu_buffer->new_pages)) {
3005 				/* not enough memory for new pages */
3006 				err = -ENOMEM;
3007 				goto out_err;
3008 			}
3009 
3010 			cond_resched();
3011 		}
3012 
3013 		/*
3014 		 * Fire off all the required work handlers
3015 		 * We can't schedule on offline CPUs, but it's not necessary
3016 		 * since we can change their buffer sizes without any race.
3017 		 */
3018 		for_each_buffer_cpu(buffer, cpu) {
3019 			cpu_buffer = buffer->buffers[cpu];
3020 			if (!cpu_buffer->nr_pages_to_update)
3021 				continue;
3022 
3023 			/* Can't run something on an offline CPU. */
3024 			if (!cpu_online(cpu)) {
3025 				rb_update_pages(cpu_buffer);
3026 				cpu_buffer->nr_pages_to_update = 0;
3027 			} else {
3028 				/* Run directly if possible. */
3029 				migrate_disable();
3030 				if (cpu != smp_processor_id()) {
3031 					migrate_enable();
3032 					schedule_work_on(cpu,
3033 							 &cpu_buffer->update_pages_work);
3034 				} else {
3035 					update_pages_handler(&cpu_buffer->update_pages_work);
3036 					migrate_enable();
3037 				}
3038 			}
3039 		}
3040 
3041 		/* wait for all the updates to complete */
3042 		for_each_buffer_cpu(buffer, cpu) {
3043 			cpu_buffer = buffer->buffers[cpu];
3044 			if (!cpu_buffer->nr_pages_to_update)
3045 				continue;
3046 
3047 			if (cpu_online(cpu))
3048 				wait_for_completion(&cpu_buffer->update_done);
3049 			cpu_buffer->nr_pages_to_update = 0;
3050 		}
3051 
3052 	} else {
3053 		cpu_buffer = buffer->buffers[cpu_id];
3054 
3055 		if (nr_pages == cpu_buffer->nr_pages)
3056 			goto out;
3057 
3058 		/*
3059 		 * Don't succeed if resizing is disabled, as a reader might be
3060 		 * manipulating the ring buffer and is expecting a sane state while
3061 		 * this is true.
3062 		 */
3063 		if (atomic_read(&cpu_buffer->resize_disabled)) {
3064 			err = -EBUSY;
3065 			goto out_err_unlock;
3066 		}
3067 
3068 		cpu_buffer->nr_pages_to_update = nr_pages -
3069 						cpu_buffer->nr_pages;
3070 
3071 		INIT_LIST_HEAD(&cpu_buffer->new_pages);
3072 		if (cpu_buffer->nr_pages_to_update > 0 &&
3073 			__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
3074 					    &cpu_buffer->new_pages)) {
3075 			err = -ENOMEM;
3076 			goto out_err;
3077 		}
3078 
3079 		/* Can't run something on an offline CPU. */
3080 		if (!cpu_online(cpu_id))
3081 			rb_update_pages(cpu_buffer);
3082 		else {
3083 			/* Run directly if possible. */
3084 			migrate_disable();
3085 			if (cpu_id == smp_processor_id()) {
3086 				rb_update_pages(cpu_buffer);
3087 				migrate_enable();
3088 			} else {
3089 				migrate_enable();
3090 				schedule_work_on(cpu_id,
3091 						 &cpu_buffer->update_pages_work);
3092 				wait_for_completion(&cpu_buffer->update_done);
3093 			}
3094 		}
3095 
3096 		cpu_buffer->nr_pages_to_update = 0;
3097 	}
3098 
3099  out:
3100 	/*
3101 	 * The ring buffer resize can happen with the ring buffer
3102 	 * enabled, so that the update disturbs the tracing as little
3103 	 * as possible. But if the buffer is disabled, we do not need
3104 	 * to worry about that, and we can take the time to verify
3105 	 * that the buffer is not corrupt.
3106 	 */
3107 	if (atomic_read(&buffer->record_disabled)) {
3108 		atomic_inc(&buffer->record_disabled);
3109 		/*
3110 		 * Even though the buffer was disabled, we must make sure
3111 		 * that it is truly disabled before calling rb_check_pages.
3112 		 * There could have been a race between checking
3113 		 * record_disable and incrementing it.
3114 		 */
3115 		synchronize_rcu();
3116 		for_each_buffer_cpu(buffer, cpu) {
3117 			cpu_buffer = buffer->buffers[cpu];
3118 			rb_check_pages(cpu_buffer);
3119 		}
3120 		atomic_dec(&buffer->record_disabled);
3121 	}
3122 
3123 	atomic_dec(&buffer->resizing);
3124 	mutex_unlock(&buffer->mutex);
3125 	return 0;
3126 
3127  out_err:
3128 	for_each_buffer_cpu(buffer, cpu) {
3129 		struct buffer_page *bpage, *tmp;
3130 
3131 		cpu_buffer = buffer->buffers[cpu];
3132 		cpu_buffer->nr_pages_to_update = 0;
3133 
3134 		if (list_empty(&cpu_buffer->new_pages))
3135 			continue;
3136 
3137 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
3138 					list) {
3139 			list_del_init(&bpage->list);
3140 			free_buffer_page(bpage);
3141 
3142 			cond_resched();
3143 		}
3144 	}
3145  out_err_unlock:
3146 	atomic_dec(&buffer->resizing);
3147 	mutex_unlock(&buffer->mutex);
3148 	return err;
3149 }
3150 EXPORT_SYMBOL_GPL(ring_buffer_resize);
3151 
3152 void ring_buffer_change_overwrite(struct trace_buffer *buffer, int val)
3153 {
3154 	mutex_lock(&buffer->mutex);
3155 	if (val)
3156 		buffer->flags |= RB_FL_OVERWRITE;
3157 	else
3158 		buffer->flags &= ~RB_FL_OVERWRITE;
3159 	mutex_unlock(&buffer->mutex);
3160 }
3161 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
3162 
3163 static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
3164 {
3165 	return bpage->page->data + index;
3166 }
3167 
3168 static __always_inline struct ring_buffer_event *
3169 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
3170 {
3171 	return __rb_page_index(cpu_buffer->reader_page,
3172 			       cpu_buffer->reader_page->read);
3173 }
3174 
3175 static struct ring_buffer_event *
3176 rb_iter_head_event(struct ring_buffer_iter *iter)
3177 {
3178 	struct ring_buffer_event *event;
3179 	struct buffer_page *iter_head_page = iter->head_page;
3180 	unsigned long commit;
3181 	unsigned length;
3182 
3183 	if (iter->head != iter->next_event)
3184 		return iter->event;
3185 
3186 	/*
3187 	 * When the writer goes across pages, it issues a cmpxchg which
3188 	 * is a mb(), which will synchronize with the rmb here.
3189 	 * (see rb_tail_page_update() and __rb_reserve_next())
3190 	 */
3191 	commit = rb_page_commit(iter_head_page);
3192 	smp_rmb();
3193 
3194 	/* An event needs to be at least 8 bytes in size */
3195 	if (iter->head > commit - 8)
3196 		goto reset;
3197 
3198 	event = __rb_page_index(iter_head_page, iter->head);
3199 	length = rb_event_length(event);
3200 
3201 	/*
3202 	 * READ_ONCE() doesn't work on functions and we don't want the
3203 	 * compiler doing any crazy optimizations with length.
3204 	 */
3205 	barrier();
3206 
3207 	if ((iter->head + length) > commit || length > iter->event_size)
3208 		/* Writer corrupted the read? */
3209 		goto reset;
3210 
3211 	memcpy(iter->event, event, length);
3212 	/*
3213 	 * If the page stamp is still the same after this rmb() then the
3214 	 * event was safely copied without the writer entering the page.
3215 	 */
3216 	smp_rmb();
3217 
3218 	/* Make sure the page didn't change since we read this */
3219 	if (iter->page_stamp != iter_head_page->page->time_stamp ||
3220 	    commit > rb_page_commit(iter_head_page))
3221 		goto reset;
3222 
3223 	iter->next_event = iter->head + length;
3224 	return iter->event;
3225  reset:
3226 	/* Reset to the beginning */
3227 	iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
3228 	iter->head = 0;
3229 	iter->next_event = 0;
3230 	iter->missed_events = 1;
3231 	return NULL;
3232 }
3233 
3234 /* Size is determined by what has been committed */
3235 static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
3236 {
3237 	return rb_page_commit(bpage) & ~RB_MISSED_MASK;
3238 }
3239 
3240 static __always_inline unsigned
3241 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
3242 {
3243 	return rb_page_commit(cpu_buffer->commit_page);
3244 }
3245 
3246 static __always_inline unsigned
3247 rb_event_index(struct ring_buffer_per_cpu *cpu_buffer, struct ring_buffer_event *event)
3248 {
3249 	unsigned long addr = (unsigned long)event;
3250 
3251 	addr &= (PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1;
3252 
3253 	return addr - BUF_PAGE_HDR_SIZE;
3254 }
3255 
3256 static void rb_inc_iter(struct ring_buffer_iter *iter)
3257 {
3258 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3259 
3260 	/*
3261 	 * The iterator could be on the reader page (it starts there).
3262 	 * But the head could have moved, since the reader was
3263 	 * found. Check for this case and assign the iterator
3264 	 * to the head page instead of next.
3265 	 */
3266 	if (iter->head_page == cpu_buffer->reader_page)
3267 		iter->head_page = rb_set_head_page(cpu_buffer);
3268 	else
3269 		rb_inc_page(&iter->head_page);
3270 
3271 	iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
3272 	iter->head = 0;
3273 	iter->next_event = 0;
3274 }
3275 
3276 /* Return the index into the sub-buffers for a given sub-buffer */
3277 static int rb_meta_subbuf_idx(struct ring_buffer_cpu_meta *meta, void *subbuf)
3278 {
3279 	void *subbuf_array;
3280 
3281 	subbuf_array = (void *)meta + sizeof(int) * meta->nr_subbufs;
3282 	subbuf_array = (void *)ALIGN((unsigned long)subbuf_array, meta->subbuf_size);
3283 	return (subbuf - subbuf_array) / meta->subbuf_size;
3284 }
3285 
3286 static void rb_update_meta_head(struct ring_buffer_per_cpu *cpu_buffer,
3287 				struct buffer_page *next_page)
3288 {
3289 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3290 	unsigned long old_head = (unsigned long)next_page->page;
3291 	unsigned long new_head;
3292 
3293 	rb_inc_page(&next_page);
3294 	new_head = (unsigned long)next_page->page;
3295 
3296 	/*
3297 	 * Only move it forward once, if something else came in and
3298 	 * moved it forward, then we don't want to touch it.
3299 	 */
3300 	(void)cmpxchg(&meta->head_buffer, old_head, new_head);
3301 }
3302 
3303 static void rb_update_meta_reader(struct ring_buffer_per_cpu *cpu_buffer,
3304 				  struct buffer_page *reader)
3305 {
3306 	struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3307 	void *old_reader = cpu_buffer->reader_page->page;
3308 	void *new_reader = reader->page;
3309 	int id;
3310 
3311 	id = reader->id;
3312 	cpu_buffer->reader_page->id = id;
3313 	reader->id = 0;
3314 
3315 	meta->buffers[0] = rb_meta_subbuf_idx(meta, new_reader);
3316 	meta->buffers[id] = rb_meta_subbuf_idx(meta, old_reader);
3317 
3318 	/* The head pointer is the one after the reader */
3319 	rb_update_meta_head(cpu_buffer, reader);
3320 }
3321 
3322 /*
3323  * rb_handle_head_page - writer hit the head page
3324  *
3325  * Returns: +1 to retry page
3326  *           0 to continue
3327  *          -1 on error
3328  */
3329 static int
3330 rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
3331 		    struct buffer_page *tail_page,
3332 		    struct buffer_page *next_page)
3333 {
3334 	struct buffer_page *new_head;
3335 	int entries;
3336 	int type;
3337 	int ret;
3338 
3339 	entries = rb_page_entries(next_page);
3340 
3341 	/*
3342 	 * The hard part is here. We need to move the head
3343 	 * forward, and protect against both readers on
3344 	 * other CPUs and writers coming in via interrupts.
3345 	 */
3346 	type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
3347 				       RB_PAGE_HEAD);
3348 
3349 	/*
3350 	 * type can be one of four:
3351 	 *  NORMAL - an interrupt already moved it for us
3352 	 *  HEAD   - we are the first to get here.
3353 	 *  UPDATE - we are the interrupt interrupting
3354 	 *           a current move.
3355 	 *  MOVED  - a reader on another CPU moved the next
3356 	 *           pointer to its reader page. Give up
3357 	 *           and try again.
3358 	 */
3359 
3360 	switch (type) {
3361 	case RB_PAGE_HEAD:
3362 		/*
3363 		 * We changed the head to UPDATE, thus
3364 		 * it is our responsibility to update
3365 		 * the counters.
3366 		 */
3367 		local_add(entries, &cpu_buffer->overrun);
3368 		local_sub(rb_page_commit(next_page), &cpu_buffer->entries_bytes);
3369 		local_inc(&cpu_buffer->pages_lost);
3370 
3371 		if (cpu_buffer->ring_meta)
3372 			rb_update_meta_head(cpu_buffer, next_page);
3373 		/*
3374 		 * The entries will be zeroed out when we move the
3375 		 * tail page.
3376 		 */
3377 
3378 		/* still more to do */
3379 		break;
3380 
3381 	case RB_PAGE_UPDATE:
3382 		/*
3383 		 * This is an interrupt that interrupt the
3384 		 * previous update. Still more to do.
3385 		 */
3386 		break;
3387 	case RB_PAGE_NORMAL:
3388 		/*
3389 		 * An interrupt came in before the update
3390 		 * and processed this for us.
3391 		 * Nothing left to do.
3392 		 */
3393 		return 1;
3394 	case RB_PAGE_MOVED:
3395 		/*
3396 		 * The reader is on another CPU and just did
3397 		 * a swap with our next_page.
3398 		 * Try again.
3399 		 */
3400 		return 1;
3401 	default:
3402 		RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
3403 		return -1;
3404 	}
3405 
3406 	/*
3407 	 * Now that we are here, the old head pointer is
3408 	 * set to UPDATE. This will keep the reader from
3409 	 * swapping the head page with the reader page.
3410 	 * The reader (on another CPU) will spin till
3411 	 * we are finished.
3412 	 *
3413 	 * We just need to protect against interrupts
3414 	 * doing the job. We will set the next pointer
3415 	 * to HEAD. After that, we set the old pointer
3416 	 * to NORMAL, but only if it was HEAD before.
3417 	 * otherwise we are an interrupt, and only
3418 	 * want the outer most commit to reset it.
3419 	 */
3420 	new_head = next_page;
3421 	rb_inc_page(&new_head);
3422 
3423 	ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
3424 				    RB_PAGE_NORMAL);
3425 
3426 	/*
3427 	 * Valid returns are:
3428 	 *  HEAD   - an interrupt came in and already set it.
3429 	 *  NORMAL - One of two things:
3430 	 *            1) We really set it.
3431 	 *            2) A bunch of interrupts came in and moved
3432 	 *               the page forward again.
3433 	 */
3434 	switch (ret) {
3435 	case RB_PAGE_HEAD:
3436 	case RB_PAGE_NORMAL:
3437 		/* OK */
3438 		break;
3439 	default:
3440 		RB_WARN_ON(cpu_buffer, 1);
3441 		return -1;
3442 	}
3443 
3444 	/*
3445 	 * It is possible that an interrupt came in,
3446 	 * set the head up, then more interrupts came in
3447 	 * and moved it again. When we get back here,
3448 	 * the page would have been set to NORMAL but we
3449 	 * just set it back to HEAD.
3450 	 *
3451 	 * How do you detect this? Well, if that happened
3452 	 * the tail page would have moved.
3453 	 */
3454 	if (ret == RB_PAGE_NORMAL) {
3455 		struct buffer_page *buffer_tail_page;
3456 
3457 		buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
3458 		/*
3459 		 * If the tail had moved passed next, then we need
3460 		 * to reset the pointer.
3461 		 */
3462 		if (buffer_tail_page != tail_page &&
3463 		    buffer_tail_page != next_page)
3464 			rb_head_page_set_normal(cpu_buffer, new_head,
3465 						next_page,
3466 						RB_PAGE_HEAD);
3467 	}
3468 
3469 	/*
3470 	 * If this was the outer most commit (the one that
3471 	 * changed the original pointer from HEAD to UPDATE),
3472 	 * then it is up to us to reset it to NORMAL.
3473 	 */
3474 	if (type == RB_PAGE_HEAD) {
3475 		ret = rb_head_page_set_normal(cpu_buffer, next_page,
3476 					      tail_page,
3477 					      RB_PAGE_UPDATE);
3478 		if (RB_WARN_ON(cpu_buffer,
3479 			       ret != RB_PAGE_UPDATE))
3480 			return -1;
3481 	}
3482 
3483 	return 0;
3484 }
3485 
3486 static inline void
3487 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
3488 	      unsigned long tail, struct rb_event_info *info)
3489 {
3490 	unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size);
3491 	struct buffer_page *tail_page = info->tail_page;
3492 	struct ring_buffer_event *event;
3493 	unsigned long length = info->length;
3494 
3495 	/*
3496 	 * Only the event that crossed the page boundary
3497 	 * must fill the old tail_page with padding.
3498 	 */
3499 	if (tail >= bsize) {
3500 		/*
3501 		 * If the page was filled, then we still need
3502 		 * to update the real_end. Reset it to zero
3503 		 * and the reader will ignore it.
3504 		 */
3505 		if (tail == bsize)
3506 			tail_page->real_end = 0;
3507 
3508 		local_sub(length, &tail_page->write);
3509 		return;
3510 	}
3511 
3512 	event = __rb_page_index(tail_page, tail);
3513 
3514 	/*
3515 	 * Save the original length to the meta data.
3516 	 * This will be used by the reader to add lost event
3517 	 * counter.
3518 	 */
3519 	tail_page->real_end = tail;
3520 
3521 	/*
3522 	 * If this event is bigger than the minimum size, then
3523 	 * we need to be careful that we don't subtract the
3524 	 * write counter enough to allow another writer to slip
3525 	 * in on this page.
3526 	 * We put in a discarded commit instead, to make sure
3527 	 * that this space is not used again, and this space will
3528 	 * not be accounted into 'entries_bytes'.
3529 	 *
3530 	 * If we are less than the minimum size, we don't need to
3531 	 * worry about it.
3532 	 */
3533 	if (tail > (bsize - RB_EVNT_MIN_SIZE)) {
3534 		/* No room for any events */
3535 
3536 		/* Mark the rest of the page with padding */
3537 		rb_event_set_padding(event);
3538 
3539 		/* Make sure the padding is visible before the write update */
3540 		smp_wmb();
3541 
3542 		/* Set the write back to the previous setting */
3543 		local_sub(length, &tail_page->write);
3544 		return;
3545 	}
3546 
3547 	/* Put in a discarded event */
3548 	event->array[0] = (bsize - tail) - RB_EVNT_HDR_SIZE;
3549 	event->type_len = RINGBUF_TYPE_PADDING;
3550 	/* time delta must be non zero */
3551 	event->time_delta = 1;
3552 
3553 	/* account for padding bytes */
3554 	local_add(bsize - tail, &cpu_buffer->entries_bytes);
3555 
3556 	/* Make sure the padding is visible before the tail_page->write update */
3557 	smp_wmb();
3558 
3559 	/* Set write to end of buffer */
3560 	length = (tail + length) - bsize;
3561 	local_sub(length, &tail_page->write);
3562 }
3563 
3564 static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
3565 
3566 /*
3567  * This is the slow path, force gcc not to inline it.
3568  */
3569 static noinline struct ring_buffer_event *
3570 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
3571 	     unsigned long tail, struct rb_event_info *info)
3572 {
3573 	struct buffer_page *tail_page = info->tail_page;
3574 	struct buffer_page *commit_page = cpu_buffer->commit_page;
3575 	struct trace_buffer *buffer = cpu_buffer->buffer;
3576 	struct buffer_page *next_page;
3577 	int ret;
3578 
3579 	next_page = tail_page;
3580 
3581 	rb_inc_page(&next_page);
3582 
3583 	/*
3584 	 * If for some reason, we had an interrupt storm that made
3585 	 * it all the way around the buffer, bail, and warn
3586 	 * about it.
3587 	 */
3588 	if (unlikely(next_page == commit_page)) {
3589 		local_inc(&cpu_buffer->commit_overrun);
3590 		goto out_reset;
3591 	}
3592 
3593 	/*
3594 	 * This is where the fun begins!
3595 	 *
3596 	 * We are fighting against races between a reader that
3597 	 * could be on another CPU trying to swap its reader
3598 	 * page with the buffer head.
3599 	 *
3600 	 * We are also fighting against interrupts coming in and
3601 	 * moving the head or tail on us as well.
3602 	 *
3603 	 * If the next page is the head page then we have filled
3604 	 * the buffer, unless the commit page is still on the
3605 	 * reader page.
3606 	 */
3607 	if (rb_is_head_page(next_page, &tail_page->list)) {
3608 
3609 		/*
3610 		 * If the commit is not on the reader page, then
3611 		 * move the header page.
3612 		 */
3613 		if (!rb_is_reader_page(cpu_buffer->commit_page)) {
3614 			/*
3615 			 * If we are not in overwrite mode,
3616 			 * this is easy, just stop here.
3617 			 */
3618 			if (!(buffer->flags & RB_FL_OVERWRITE)) {
3619 				local_inc(&cpu_buffer->dropped_events);
3620 				goto out_reset;
3621 			}
3622 
3623 			ret = rb_handle_head_page(cpu_buffer,
3624 						  tail_page,
3625 						  next_page);
3626 			if (ret < 0)
3627 				goto out_reset;
3628 			if (ret)
3629 				goto out_again;
3630 		} else {
3631 			/*
3632 			 * We need to be careful here too. The
3633 			 * commit page could still be on the reader
3634 			 * page. We could have a small buffer, and
3635 			 * have filled up the buffer with events
3636 			 * from interrupts and such, and wrapped.
3637 			 *
3638 			 * Note, if the tail page is also on the
3639 			 * reader_page, we let it move out.
3640 			 */
3641 			if (unlikely((cpu_buffer->commit_page !=
3642 				      cpu_buffer->tail_page) &&
3643 				     (cpu_buffer->commit_page ==
3644 				      cpu_buffer->reader_page))) {
3645 				local_inc(&cpu_buffer->commit_overrun);
3646 				goto out_reset;
3647 			}
3648 		}
3649 	}
3650 
3651 	rb_tail_page_update(cpu_buffer, tail_page, next_page);
3652 
3653  out_again:
3654 
3655 	rb_reset_tail(cpu_buffer, tail, info);
3656 
3657 	/* Commit what we have for now. */
3658 	rb_end_commit(cpu_buffer);
3659 	/* rb_end_commit() decs committing */
3660 	local_inc(&cpu_buffer->committing);
3661 
3662 	/* fail and let the caller try again */
3663 	return ERR_PTR(-EAGAIN);
3664 
3665  out_reset:
3666 	/* reset write */
3667 	rb_reset_tail(cpu_buffer, tail, info);
3668 
3669 	return NULL;
3670 }
3671 
3672 /* Slow path */
3673 static struct ring_buffer_event *
3674 rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
3675 		  struct ring_buffer_event *event, u64 delta, bool abs)
3676 {
3677 	if (abs)
3678 		event->type_len = RINGBUF_TYPE_TIME_STAMP;
3679 	else
3680 		event->type_len = RINGBUF_TYPE_TIME_EXTEND;
3681 
3682 	/* Not the first event on the page, or not delta? */
3683 	if (abs || rb_event_index(cpu_buffer, event)) {
3684 		event->time_delta = delta & TS_MASK;
3685 		event->array[0] = delta >> TS_SHIFT;
3686 	} else {
3687 		/* nope, just zero it */
3688 		event->time_delta = 0;
3689 		event->array[0] = 0;
3690 	}
3691 
3692 	return skip_time_extend(event);
3693 }
3694 
3695 #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
3696 static inline bool sched_clock_stable(void)
3697 {
3698 	return true;
3699 }
3700 #endif
3701 
3702 static void
3703 rb_check_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
3704 		   struct rb_event_info *info)
3705 {
3706 	u64 write_stamp;
3707 
3708 	WARN_ONCE(1, "Delta way too big! %llu ts=%llu before=%llu after=%llu write stamp=%llu\n%s",
3709 		  (unsigned long long)info->delta,
3710 		  (unsigned long long)info->ts,
3711 		  (unsigned long long)info->before,
3712 		  (unsigned long long)info->after,
3713 		  (unsigned long long)({rb_time_read(&cpu_buffer->write_stamp, &write_stamp); write_stamp;}),
3714 		  sched_clock_stable() ? "" :
3715 		  "If you just came from a suspend/resume,\n"
3716 		  "please switch to the trace global clock:\n"
3717 		  "  echo global > /sys/kernel/tracing/trace_clock\n"
3718 		  "or add trace_clock=global to the kernel command line\n");
3719 }
3720 
3721 static void rb_add_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
3722 				      struct ring_buffer_event **event,
3723 				      struct rb_event_info *info,
3724 				      u64 *delta,
3725 				      unsigned int *length)
3726 {
3727 	bool abs = info->add_timestamp &
3728 		(RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE);
3729 
3730 	if (unlikely(info->delta > (1ULL << 59))) {
3731 		/*
3732 		 * Some timers can use more than 59 bits, and when a timestamp
3733 		 * is added to the buffer, it will lose those bits.
3734 		 */
3735 		if (abs && (info->ts & TS_MSB)) {
3736 			info->delta &= ABS_TS_MASK;
3737 
3738 		/* did the clock go backwards */
3739 		} else if (info->before == info->after && info->before > info->ts) {
3740 			/* not interrupted */
3741 			static int once;
3742 
3743 			/*
3744 			 * This is possible with a recalibrating of the TSC.
3745 			 * Do not produce a call stack, but just report it.
3746 			 */
3747 			if (!once) {
3748 				once++;
3749 				pr_warn("Ring buffer clock went backwards: %llu -> %llu\n",
3750 					info->before, info->ts);
3751 			}
3752 		} else
3753 			rb_check_timestamp(cpu_buffer, info);
3754 		if (!abs)
3755 			info->delta = 0;
3756 	}
3757 	*event = rb_add_time_stamp(cpu_buffer, *event, info->delta, abs);
3758 	*length -= RB_LEN_TIME_EXTEND;
3759 	*delta = 0;
3760 }
3761 
3762 /**
3763  * rb_update_event - update event type and data
3764  * @cpu_buffer: The per cpu buffer of the @event
3765  * @event: the event to update
3766  * @info: The info to update the @event with (contains length and delta)
3767  *
3768  * Update the type and data fields of the @event. The length
3769  * is the actual size that is written to the ring buffer,
3770  * and with this, we can determine what to place into the
3771  * data field.
3772  */
3773 static void
3774 rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
3775 		struct ring_buffer_event *event,
3776 		struct rb_event_info *info)
3777 {
3778 	unsigned length = info->length;
3779 	u64 delta = info->delta;
3780 	unsigned int nest = local_read(&cpu_buffer->committing) - 1;
3781 
3782 	if (!WARN_ON_ONCE(nest >= MAX_NEST))
3783 		cpu_buffer->event_stamp[nest] = info->ts;
3784 
3785 	/*
3786 	 * If we need to add a timestamp, then we
3787 	 * add it to the start of the reserved space.
3788 	 */
3789 	if (unlikely(info->add_timestamp))
3790 		rb_add_timestamp(cpu_buffer, &event, info, &delta, &length);
3791 
3792 	event->time_delta = delta;
3793 	length -= RB_EVNT_HDR_SIZE;
3794 	if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
3795 		event->type_len = 0;
3796 		event->array[0] = length;
3797 	} else
3798 		event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
3799 }
3800 
3801 static unsigned rb_calculate_event_length(unsigned length)
3802 {
3803 	struct ring_buffer_event event; /* Used only for sizeof array */
3804 
3805 	/* zero length can cause confusions */
3806 	if (!length)
3807 		length++;
3808 
3809 	if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
3810 		length += sizeof(event.array[0]);
3811 
3812 	length += RB_EVNT_HDR_SIZE;
3813 	length = ALIGN(length, RB_ARCH_ALIGNMENT);
3814 
3815 	/*
3816 	 * In case the time delta is larger than the 27 bits for it
3817 	 * in the header, we need to add a timestamp. If another
3818 	 * event comes in when trying to discard this one to increase
3819 	 * the length, then the timestamp will be added in the allocated
3820 	 * space of this event. If length is bigger than the size needed
3821 	 * for the TIME_EXTEND, then padding has to be used. The events
3822 	 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
3823 	 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
3824 	 * As length is a multiple of 4, we only need to worry if it
3825 	 * is 12 (RB_LEN_TIME_EXTEND + 4).
3826 	 */
3827 	if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
3828 		length += RB_ALIGNMENT;
3829 
3830 	return length;
3831 }
3832 
3833 static inline bool
3834 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
3835 		  struct ring_buffer_event *event)
3836 {
3837 	unsigned long new_index, old_index;
3838 	struct buffer_page *bpage;
3839 	unsigned long addr;
3840 
3841 	new_index = rb_event_index(cpu_buffer, event);
3842 	old_index = new_index + rb_event_ts_length(event);
3843 	addr = (unsigned long)event;
3844 	addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1);
3845 
3846 	bpage = READ_ONCE(cpu_buffer->tail_page);
3847 
3848 	/*
3849 	 * Make sure the tail_page is still the same and
3850 	 * the next write location is the end of this event
3851 	 */
3852 	if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
3853 		unsigned long write_mask =
3854 			local_read(&bpage->write) & ~RB_WRITE_MASK;
3855 		unsigned long event_length = rb_event_length(event);
3856 
3857 		/*
3858 		 * For the before_stamp to be different than the write_stamp
3859 		 * to make sure that the next event adds an absolute
3860 		 * value and does not rely on the saved write stamp, which
3861 		 * is now going to be bogus.
3862 		 *
3863 		 * By setting the before_stamp to zero, the next event
3864 		 * is not going to use the write_stamp and will instead
3865 		 * create an absolute timestamp. This means there's no
3866 		 * reason to update the wirte_stamp!
3867 		 */
3868 		rb_time_set(&cpu_buffer->before_stamp, 0);
3869 
3870 		/*
3871 		 * If an event were to come in now, it would see that the
3872 		 * write_stamp and the before_stamp are different, and assume
3873 		 * that this event just added itself before updating
3874 		 * the write stamp. The interrupting event will fix the
3875 		 * write stamp for us, and use an absolute timestamp.
3876 		 */
3877 
3878 		/*
3879 		 * This is on the tail page. It is possible that
3880 		 * a write could come in and move the tail page
3881 		 * and write to the next page. That is fine
3882 		 * because we just shorten what is on this page.
3883 		 */
3884 		old_index += write_mask;
3885 		new_index += write_mask;
3886 
3887 		/* caution: old_index gets updated on cmpxchg failure */
3888 		if (local_try_cmpxchg(&bpage->write, &old_index, new_index)) {
3889 			/* update counters */
3890 			local_sub(event_length, &cpu_buffer->entries_bytes);
3891 			return true;
3892 		}
3893 	}
3894 
3895 	/* could not discard */
3896 	return false;
3897 }
3898 
3899 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
3900 {
3901 	local_inc(&cpu_buffer->committing);
3902 	local_inc(&cpu_buffer->commits);
3903 }
3904 
3905 static __always_inline void
3906 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
3907 {
3908 	unsigned long max_count;
3909 
3910 	/*
3911 	 * We only race with interrupts and NMIs on this CPU.
3912 	 * If we own the commit event, then we can commit
3913 	 * all others that interrupted us, since the interruptions
3914 	 * are in stack format (they finish before they come
3915 	 * back to us). This allows us to do a simple loop to
3916 	 * assign the commit to the tail.
3917 	 */
3918  again:
3919 	max_count = cpu_buffer->nr_pages * 100;
3920 
3921 	while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
3922 		if (RB_WARN_ON(cpu_buffer, !(--max_count)))
3923 			return;
3924 		if (RB_WARN_ON(cpu_buffer,
3925 			       rb_is_reader_page(cpu_buffer->tail_page)))
3926 			return;
3927 		/*
3928 		 * No need for a memory barrier here, as the update
3929 		 * of the tail_page did it for this page.
3930 		 */
3931 		local_set(&cpu_buffer->commit_page->page->commit,
3932 			  rb_page_write(cpu_buffer->commit_page));
3933 		rb_inc_page(&cpu_buffer->commit_page);
3934 		if (cpu_buffer->ring_meta) {
3935 			struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3936 			meta->commit_buffer = (unsigned long)cpu_buffer->commit_page->page;
3937 		}
3938 		/* add barrier to keep gcc from optimizing too much */
3939 		barrier();
3940 	}
3941 	while (rb_commit_index(cpu_buffer) !=
3942 	       rb_page_write(cpu_buffer->commit_page)) {
3943 
3944 		/* Make sure the readers see the content of what is committed. */
3945 		smp_wmb();
3946 		local_set(&cpu_buffer->commit_page->page->commit,
3947 			  rb_page_write(cpu_buffer->commit_page));
3948 		RB_WARN_ON(cpu_buffer,
3949 			   local_read(&cpu_buffer->commit_page->page->commit) &
3950 			   ~RB_WRITE_MASK);
3951 		barrier();
3952 	}
3953 
3954 	/* again, keep gcc from optimizing */
3955 	barrier();
3956 
3957 	/*
3958 	 * If an interrupt came in just after the first while loop
3959 	 * and pushed the tail page forward, we will be left with
3960 	 * a dangling commit that will never go forward.
3961 	 */
3962 	if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
3963 		goto again;
3964 }
3965 
3966 static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
3967 {
3968 	unsigned long commits;
3969 
3970 	if (RB_WARN_ON(cpu_buffer,
3971 		       !local_read(&cpu_buffer->committing)))
3972 		return;
3973 
3974  again:
3975 	commits = local_read(&cpu_buffer->commits);
3976 	/* synchronize with interrupts */
3977 	barrier();
3978 	if (local_read(&cpu_buffer->committing) == 1)
3979 		rb_set_commit_to_write(cpu_buffer);
3980 
3981 	local_dec(&cpu_buffer->committing);
3982 
3983 	/* synchronize with interrupts */
3984 	barrier();
3985 
3986 	/*
3987 	 * Need to account for interrupts coming in between the
3988 	 * updating of the commit page and the clearing of the
3989 	 * committing counter.
3990 	 */
3991 	if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
3992 	    !local_read(&cpu_buffer->committing)) {
3993 		local_inc(&cpu_buffer->committing);
3994 		goto again;
3995 	}
3996 }
3997 
3998 static inline void rb_event_discard(struct ring_buffer_event *event)
3999 {
4000 	if (extended_time(event))
4001 		event = skip_time_extend(event);
4002 
4003 	/* array[0] holds the actual length for the discarded event */
4004 	event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
4005 	event->type_len = RINGBUF_TYPE_PADDING;
4006 	/* time delta must be non zero */
4007 	if (!event->time_delta)
4008 		event->time_delta = 1;
4009 }
4010 
4011 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer)
4012 {
4013 	local_inc(&cpu_buffer->entries);
4014 	rb_end_commit(cpu_buffer);
4015 }
4016 
4017 static bool
4018 rb_irq_work_queue(struct rb_irq_work *irq_work)
4019 {
4020 	int cpu;
4021 
4022 	/* irq_work_queue_on() is not NMI-safe */
4023 	if (unlikely(in_nmi()))
4024 		return irq_work_queue(&irq_work->work);
4025 
4026 	/*
4027 	 * If CPU isolation is not active, cpu is always the current
4028 	 * CPU, and the following is equivallent to irq_work_queue().
4029 	 */
4030 	cpu = housekeeping_any_cpu(HK_TYPE_KERNEL_NOISE);
4031 	return irq_work_queue_on(&irq_work->work, cpu);
4032 }
4033 
4034 static __always_inline void
4035 rb_wakeups(struct trace_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
4036 {
4037 	if (buffer->irq_work.waiters_pending) {
4038 		buffer->irq_work.waiters_pending = false;
4039 		/* irq_work_queue() supplies it's own memory barriers */
4040 		rb_irq_work_queue(&buffer->irq_work);
4041 	}
4042 
4043 	if (cpu_buffer->irq_work.waiters_pending) {
4044 		cpu_buffer->irq_work.waiters_pending = false;
4045 		/* irq_work_queue() supplies it's own memory barriers */
4046 		rb_irq_work_queue(&cpu_buffer->irq_work);
4047 	}
4048 
4049 	if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
4050 		return;
4051 
4052 	if (cpu_buffer->reader_page == cpu_buffer->commit_page)
4053 		return;
4054 
4055 	if (!cpu_buffer->irq_work.full_waiters_pending)
4056 		return;
4057 
4058 	cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
4059 
4060 	if (!full_hit(buffer, cpu_buffer->cpu, cpu_buffer->shortest_full))
4061 		return;
4062 
4063 	cpu_buffer->irq_work.wakeup_full = true;
4064 	cpu_buffer->irq_work.full_waiters_pending = false;
4065 	/* irq_work_queue() supplies it's own memory barriers */
4066 	rb_irq_work_queue(&cpu_buffer->irq_work);
4067 }
4068 
4069 #ifdef CONFIG_RING_BUFFER_RECORD_RECURSION
4070 # define do_ring_buffer_record_recursion()	\
4071 	do_ftrace_record_recursion(_THIS_IP_, _RET_IP_)
4072 #else
4073 # define do_ring_buffer_record_recursion() do { } while (0)
4074 #endif
4075 
4076 /*
4077  * The lock and unlock are done within a preempt disable section.
4078  * The current_context per_cpu variable can only be modified
4079  * by the current task between lock and unlock. But it can
4080  * be modified more than once via an interrupt. To pass this
4081  * information from the lock to the unlock without having to
4082  * access the 'in_interrupt()' functions again (which do show
4083  * a bit of overhead in something as critical as function tracing,
4084  * we use a bitmask trick.
4085  *
4086  *  bit 1 =  NMI context
4087  *  bit 2 =  IRQ context
4088  *  bit 3 =  SoftIRQ context
4089  *  bit 4 =  normal context.
4090  *
4091  * This works because this is the order of contexts that can
4092  * preempt other contexts. A SoftIRQ never preempts an IRQ
4093  * context.
4094  *
4095  * When the context is determined, the corresponding bit is
4096  * checked and set (if it was set, then a recursion of that context
4097  * happened).
4098  *
4099  * On unlock, we need to clear this bit. To do so, just subtract
4100  * 1 from the current_context and AND it to itself.
4101  *
4102  * (binary)
4103  *  101 - 1 = 100
4104  *  101 & 100 = 100 (clearing bit zero)
4105  *
4106  *  1010 - 1 = 1001
4107  *  1010 & 1001 = 1000 (clearing bit 1)
4108  *
4109  * The least significant bit can be cleared this way, and it
4110  * just so happens that it is the same bit corresponding to
4111  * the current context.
4112  *
4113  * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
4114  * is set when a recursion is detected at the current context, and if
4115  * the TRANSITION bit is already set, it will fail the recursion.
4116  * This is needed because there's a lag between the changing of
4117  * interrupt context and updating the preempt count. In this case,
4118  * a false positive will be found. To handle this, one extra recursion
4119  * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
4120  * bit is already set, then it is considered a recursion and the function
4121  * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
4122  *
4123  * On the trace_recursive_unlock(), the TRANSITION bit will be the first
4124  * to be cleared. Even if it wasn't the context that set it. That is,
4125  * if an interrupt comes in while NORMAL bit is set and the ring buffer
4126  * is called before preempt_count() is updated, since the check will
4127  * be on the NORMAL bit, the TRANSITION bit will then be set. If an
4128  * NMI then comes in, it will set the NMI bit, but when the NMI code
4129  * does the trace_recursive_unlock() it will clear the TRANSITION bit
4130  * and leave the NMI bit set. But this is fine, because the interrupt
4131  * code that set the TRANSITION bit will then clear the NMI bit when it
4132  * calls trace_recursive_unlock(). If another NMI comes in, it will
4133  * set the TRANSITION bit and continue.
4134  *
4135  * Note: The TRANSITION bit only handles a single transition between context.
4136  */
4137 
4138 static __always_inline bool
4139 trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
4140 {
4141 	unsigned int val = cpu_buffer->current_context;
4142 	int bit = interrupt_context_level();
4143 
4144 	bit = RB_CTX_NORMAL - bit;
4145 
4146 	if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) {
4147 		/*
4148 		 * It is possible that this was called by transitioning
4149 		 * between interrupt context, and preempt_count() has not
4150 		 * been updated yet. In this case, use the TRANSITION bit.
4151 		 */
4152 		bit = RB_CTX_TRANSITION;
4153 		if (val & (1 << (bit + cpu_buffer->nest))) {
4154 			do_ring_buffer_record_recursion();
4155 			return true;
4156 		}
4157 	}
4158 
4159 	val |= (1 << (bit + cpu_buffer->nest));
4160 	cpu_buffer->current_context = val;
4161 
4162 	return false;
4163 }
4164 
4165 static __always_inline void
4166 trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
4167 {
4168 	cpu_buffer->current_context &=
4169 		cpu_buffer->current_context - (1 << cpu_buffer->nest);
4170 }
4171 
4172 /* The recursive locking above uses 5 bits */
4173 #define NESTED_BITS 5
4174 
4175 /**
4176  * ring_buffer_nest_start - Allow to trace while nested
4177  * @buffer: The ring buffer to modify
4178  *
4179  * The ring buffer has a safety mechanism to prevent recursion.
4180  * But there may be a case where a trace needs to be done while
4181  * tracing something else. In this case, calling this function
4182  * will allow this function to nest within a currently active
4183  * ring_buffer_lock_reserve().
4184  *
4185  * Call this function before calling another ring_buffer_lock_reserve() and
4186  * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
4187  */
4188 void ring_buffer_nest_start(struct trace_buffer *buffer)
4189 {
4190 	struct ring_buffer_per_cpu *cpu_buffer;
4191 	int cpu;
4192 
4193 	/* Enabled by ring_buffer_nest_end() */
4194 	preempt_disable_notrace();
4195 	cpu = raw_smp_processor_id();
4196 	cpu_buffer = buffer->buffers[cpu];
4197 	/* This is the shift value for the above recursive locking */
4198 	cpu_buffer->nest += NESTED_BITS;
4199 }
4200 
4201 /**
4202  * ring_buffer_nest_end - Allow to trace while nested
4203  * @buffer: The ring buffer to modify
4204  *
4205  * Must be called after ring_buffer_nest_start() and after the
4206  * ring_buffer_unlock_commit().
4207  */
4208 void ring_buffer_nest_end(struct trace_buffer *buffer)
4209 {
4210 	struct ring_buffer_per_cpu *cpu_buffer;
4211 	int cpu;
4212 
4213 	/* disabled by ring_buffer_nest_start() */
4214 	cpu = raw_smp_processor_id();
4215 	cpu_buffer = buffer->buffers[cpu];
4216 	/* This is the shift value for the above recursive locking */
4217 	cpu_buffer->nest -= NESTED_BITS;
4218 	preempt_enable_notrace();
4219 }
4220 
4221 /**
4222  * ring_buffer_unlock_commit - commit a reserved
4223  * @buffer: The buffer to commit to
4224  *
4225  * This commits the data to the ring buffer, and releases any locks held.
4226  *
4227  * Must be paired with ring_buffer_lock_reserve.
4228  */
4229 int ring_buffer_unlock_commit(struct trace_buffer *buffer)
4230 {
4231 	struct ring_buffer_per_cpu *cpu_buffer;
4232 	int cpu = raw_smp_processor_id();
4233 
4234 	cpu_buffer = buffer->buffers[cpu];
4235 
4236 	rb_commit(cpu_buffer);
4237 
4238 	rb_wakeups(buffer, cpu_buffer);
4239 
4240 	trace_recursive_unlock(cpu_buffer);
4241 
4242 	preempt_enable_notrace();
4243 
4244 	return 0;
4245 }
4246 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
4247 
4248 /* Special value to validate all deltas on a page. */
4249 #define CHECK_FULL_PAGE		1L
4250 
4251 #ifdef CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS
4252 
4253 static const char *show_irq_str(int bits)
4254 {
4255 	static const char * type[] = {
4256 		".",	// 0
4257 		"s",	// 1
4258 		"h",	// 2
4259 		"Hs",	// 3
4260 		"n",	// 4
4261 		"Ns",	// 5
4262 		"Nh",	// 6
4263 		"NHs",	// 7
4264 	};
4265 
4266 	return type[bits];
4267 }
4268 
4269 /* Assume this is a trace event */
4270 static const char *show_flags(struct ring_buffer_event *event)
4271 {
4272 	struct trace_entry *entry;
4273 	int bits = 0;
4274 
4275 	if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry))
4276 		return "X";
4277 
4278 	entry = ring_buffer_event_data(event);
4279 
4280 	if (entry->flags & TRACE_FLAG_SOFTIRQ)
4281 		bits |= 1;
4282 
4283 	if (entry->flags & TRACE_FLAG_HARDIRQ)
4284 		bits |= 2;
4285 
4286 	if (entry->flags & TRACE_FLAG_NMI)
4287 		bits |= 4;
4288 
4289 	return show_irq_str(bits);
4290 }
4291 
4292 static const char *show_irq(struct ring_buffer_event *event)
4293 {
4294 	struct trace_entry *entry;
4295 
4296 	if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry))
4297 		return "";
4298 
4299 	entry = ring_buffer_event_data(event);
4300 	if (entry->flags & TRACE_FLAG_IRQS_OFF)
4301 		return "d";
4302 	return "";
4303 }
4304 
4305 static const char *show_interrupt_level(void)
4306 {
4307 	unsigned long pc = preempt_count();
4308 	unsigned char level = 0;
4309 
4310 	if (pc & SOFTIRQ_OFFSET)
4311 		level |= 1;
4312 
4313 	if (pc & HARDIRQ_MASK)
4314 		level |= 2;
4315 
4316 	if (pc & NMI_MASK)
4317 		level |= 4;
4318 
4319 	return show_irq_str(level);
4320 }
4321 
4322 static void dump_buffer_page(struct buffer_data_page *bpage,
4323 			     struct rb_event_info *info,
4324 			     unsigned long tail)
4325 {
4326 	struct ring_buffer_event *event;
4327 	u64 ts, delta;
4328 	int e;
4329 
4330 	ts = bpage->time_stamp;
4331 	pr_warn("  [%lld] PAGE TIME STAMP\n", ts);
4332 
4333 	for (e = 0; e < tail; e += rb_event_length(event)) {
4334 
4335 		event = (struct ring_buffer_event *)(bpage->data + e);
4336 
4337 		switch (event->type_len) {
4338 
4339 		case RINGBUF_TYPE_TIME_EXTEND:
4340 			delta = rb_event_time_stamp(event);
4341 			ts += delta;
4342 			pr_warn(" 0x%x: [%lld] delta:%lld TIME EXTEND\n",
4343 				e, ts, delta);
4344 			break;
4345 
4346 		case RINGBUF_TYPE_TIME_STAMP:
4347 			delta = rb_event_time_stamp(event);
4348 			ts = rb_fix_abs_ts(delta, ts);
4349 			pr_warn(" 0x%x:  [%lld] absolute:%lld TIME STAMP\n",
4350 				e, ts, delta);
4351 			break;
4352 
4353 		case RINGBUF_TYPE_PADDING:
4354 			ts += event->time_delta;
4355 			pr_warn(" 0x%x:  [%lld] delta:%d PADDING\n",
4356 				e, ts, event->time_delta);
4357 			break;
4358 
4359 		case RINGBUF_TYPE_DATA:
4360 			ts += event->time_delta;
4361 			pr_warn(" 0x%x:  [%lld] delta:%d %s%s\n",
4362 				e, ts, event->time_delta,
4363 				show_flags(event), show_irq(event));
4364 			break;
4365 
4366 		default:
4367 			break;
4368 		}
4369 	}
4370 	pr_warn("expected end:0x%lx last event actually ended at:0x%x\n", tail, e);
4371 }
4372 
4373 static DEFINE_PER_CPU(atomic_t, checking);
4374 static atomic_t ts_dump;
4375 
4376 #define buffer_warn_return(fmt, ...)					\
4377 	do {								\
4378 		/* If another report is happening, ignore this one */	\
4379 		if (atomic_inc_return(&ts_dump) != 1) {			\
4380 			atomic_dec(&ts_dump);				\
4381 			goto out;					\
4382 		}							\
4383 		atomic_inc(&cpu_buffer->record_disabled);		\
4384 		pr_warn(fmt, ##__VA_ARGS__);				\
4385 		dump_buffer_page(bpage, info, tail);			\
4386 		atomic_dec(&ts_dump);					\
4387 		/* There's some cases in boot up that this can happen */ \
4388 		if (WARN_ON_ONCE(system_state != SYSTEM_BOOTING))	\
4389 			/* Do not re-enable checking */			\
4390 			return;						\
4391 	} while (0)
4392 
4393 /*
4394  * Check if the current event time stamp matches the deltas on
4395  * the buffer page.
4396  */
4397 static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
4398 			 struct rb_event_info *info,
4399 			 unsigned long tail)
4400 {
4401 	struct buffer_data_page *bpage;
4402 	u64 ts, delta;
4403 	bool full = false;
4404 	int ret;
4405 
4406 	bpage = info->tail_page->page;
4407 
4408 	if (tail == CHECK_FULL_PAGE) {
4409 		full = true;
4410 		tail = local_read(&bpage->commit);
4411 	} else if (info->add_timestamp &
4412 		   (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)) {
4413 		/* Ignore events with absolute time stamps */
4414 		return;
4415 	}
4416 
4417 	/*
4418 	 * Do not check the first event (skip possible extends too).
4419 	 * Also do not check if previous events have not been committed.
4420 	 */
4421 	if (tail <= 8 || tail > local_read(&bpage->commit))
4422 		return;
4423 
4424 	/*
4425 	 * If this interrupted another event,
4426 	 */
4427 	if (atomic_inc_return(this_cpu_ptr(&checking)) != 1)
4428 		goto out;
4429 
4430 	ret = rb_read_data_buffer(bpage, tail, cpu_buffer->cpu, &ts, &delta);
4431 	if (ret < 0) {
4432 		if (delta < ts) {
4433 			buffer_warn_return("[CPU: %d]ABSOLUTE TIME WENT BACKWARDS: last ts: %lld absolute ts: %lld\n",
4434 					   cpu_buffer->cpu, ts, delta);
4435 			goto out;
4436 		}
4437 	}
4438 	if ((full && ts > info->ts) ||
4439 	    (!full && ts + info->delta != info->ts)) {
4440 		buffer_warn_return("[CPU: %d]TIME DOES NOT MATCH expected:%lld actual:%lld delta:%lld before:%lld after:%lld%s context:%s\n",
4441 				   cpu_buffer->cpu,
4442 				   ts + info->delta, info->ts, info->delta,
4443 				   info->before, info->after,
4444 				   full ? " (full)" : "", show_interrupt_level());
4445 	}
4446 out:
4447 	atomic_dec(this_cpu_ptr(&checking));
4448 }
4449 #else
4450 static inline void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
4451 			 struct rb_event_info *info,
4452 			 unsigned long tail)
4453 {
4454 }
4455 #endif /* CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS */
4456 
4457 static struct ring_buffer_event *
4458 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
4459 		  struct rb_event_info *info)
4460 {
4461 	struct ring_buffer_event *event;
4462 	struct buffer_page *tail_page;
4463 	unsigned long tail, write, w;
4464 
4465 	/* Don't let the compiler play games with cpu_buffer->tail_page */
4466 	tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
4467 
4468  /*A*/	w = local_read(&tail_page->write) & RB_WRITE_MASK;
4469 	barrier();
4470 	rb_time_read(&cpu_buffer->before_stamp, &info->before);
4471 	rb_time_read(&cpu_buffer->write_stamp, &info->after);
4472 	barrier();
4473 	info->ts = rb_time_stamp(cpu_buffer->buffer);
4474 
4475 	if ((info->add_timestamp & RB_ADD_STAMP_ABSOLUTE)) {
4476 		info->delta = info->ts;
4477 	} else {
4478 		/*
4479 		 * If interrupting an event time update, we may need an
4480 		 * absolute timestamp.
4481 		 * Don't bother if this is the start of a new page (w == 0).
4482 		 */
4483 		if (!w) {
4484 			/* Use the sub-buffer timestamp */
4485 			info->delta = 0;
4486 		} else if (unlikely(info->before != info->after)) {
4487 			info->add_timestamp |= RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND;
4488 			info->length += RB_LEN_TIME_EXTEND;
4489 		} else {
4490 			info->delta = info->ts - info->after;
4491 			if (unlikely(test_time_stamp(info->delta))) {
4492 				info->add_timestamp |= RB_ADD_STAMP_EXTEND;
4493 				info->length += RB_LEN_TIME_EXTEND;
4494 			}
4495 		}
4496 	}
4497 
4498  /*B*/	rb_time_set(&cpu_buffer->before_stamp, info->ts);
4499 
4500  /*C*/	write = local_add_return(info->length, &tail_page->write);
4501 
4502 	/* set write to only the index of the write */
4503 	write &= RB_WRITE_MASK;
4504 
4505 	tail = write - info->length;
4506 
4507 	/* See if we shot pass the end of this buffer page */
4508 	if (unlikely(write > cpu_buffer->buffer->subbuf_size)) {
4509 		check_buffer(cpu_buffer, info, CHECK_FULL_PAGE);
4510 		return rb_move_tail(cpu_buffer, tail, info);
4511 	}
4512 
4513 	if (likely(tail == w)) {
4514 		/* Nothing interrupted us between A and C */
4515  /*D*/		rb_time_set(&cpu_buffer->write_stamp, info->ts);
4516 		/*
4517 		 * If something came in between C and D, the write stamp
4518 		 * may now not be in sync. But that's fine as the before_stamp
4519 		 * will be different and then next event will just be forced
4520 		 * to use an absolute timestamp.
4521 		 */
4522 		if (likely(!(info->add_timestamp &
4523 			     (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
4524 			/* This did not interrupt any time update */
4525 			info->delta = info->ts - info->after;
4526 		else
4527 			/* Just use full timestamp for interrupting event */
4528 			info->delta = info->ts;
4529 		check_buffer(cpu_buffer, info, tail);
4530 	} else {
4531 		u64 ts;
4532 		/* SLOW PATH - Interrupted between A and C */
4533 
4534 		/* Save the old before_stamp */
4535 		rb_time_read(&cpu_buffer->before_stamp, &info->before);
4536 
4537 		/*
4538 		 * Read a new timestamp and update the before_stamp to make
4539 		 * the next event after this one force using an absolute
4540 		 * timestamp. This is in case an interrupt were to come in
4541 		 * between E and F.
4542 		 */
4543 		ts = rb_time_stamp(cpu_buffer->buffer);
4544 		rb_time_set(&cpu_buffer->before_stamp, ts);
4545 
4546 		barrier();
4547  /*E*/		rb_time_read(&cpu_buffer->write_stamp, &info->after);
4548 		barrier();
4549  /*F*/		if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) &&
4550 		    info->after == info->before && info->after < ts) {
4551 			/*
4552 			 * Nothing came after this event between C and F, it is
4553 			 * safe to use info->after for the delta as it
4554 			 * matched info->before and is still valid.
4555 			 */
4556 			info->delta = ts - info->after;
4557 		} else {
4558 			/*
4559 			 * Interrupted between C and F:
4560 			 * Lost the previous events time stamp. Just set the
4561 			 * delta to zero, and this will be the same time as
4562 			 * the event this event interrupted. And the events that
4563 			 * came after this will still be correct (as they would
4564 			 * have built their delta on the previous event.
4565 			 */
4566 			info->delta = 0;
4567 		}
4568 		info->ts = ts;
4569 		info->add_timestamp &= ~RB_ADD_STAMP_FORCE;
4570 	}
4571 
4572 	/*
4573 	 * If this is the first commit on the page, then it has the same
4574 	 * timestamp as the page itself.
4575 	 */
4576 	if (unlikely(!tail && !(info->add_timestamp &
4577 				(RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
4578 		info->delta = 0;
4579 
4580 	/* We reserved something on the buffer */
4581 
4582 	event = __rb_page_index(tail_page, tail);
4583 	rb_update_event(cpu_buffer, event, info);
4584 
4585 	local_inc(&tail_page->entries);
4586 
4587 	/*
4588 	 * If this is the first commit on the page, then update
4589 	 * its timestamp.
4590 	 */
4591 	if (unlikely(!tail))
4592 		tail_page->page->time_stamp = info->ts;
4593 
4594 	/* account for these added bytes */
4595 	local_add(info->length, &cpu_buffer->entries_bytes);
4596 
4597 	return event;
4598 }
4599 
4600 static __always_inline struct ring_buffer_event *
4601 rb_reserve_next_event(struct trace_buffer *buffer,
4602 		      struct ring_buffer_per_cpu *cpu_buffer,
4603 		      unsigned long length)
4604 {
4605 	struct ring_buffer_event *event;
4606 	struct rb_event_info info;
4607 	int nr_loops = 0;
4608 	int add_ts_default;
4609 
4610 	/*
4611 	 * ring buffer does cmpxchg as well as atomic64 operations
4612 	 * (which some archs use locking for atomic64), make sure this
4613 	 * is safe in NMI context
4614 	 */
4615 	if ((!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) ||
4616 	     IS_ENABLED(CONFIG_GENERIC_ATOMIC64)) &&
4617 	    (unlikely(in_nmi()))) {
4618 		return NULL;
4619 	}
4620 
4621 	rb_start_commit(cpu_buffer);
4622 	/* The commit page can not change after this */
4623 
4624 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4625 	/*
4626 	 * Due to the ability to swap a cpu buffer from a buffer
4627 	 * it is possible it was swapped before we committed.
4628 	 * (committing stops a swap). We check for it here and
4629 	 * if it happened, we have to fail the write.
4630 	 */
4631 	barrier();
4632 	if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
4633 		local_dec(&cpu_buffer->committing);
4634 		local_dec(&cpu_buffer->commits);
4635 		return NULL;
4636 	}
4637 #endif
4638 
4639 	info.length = rb_calculate_event_length(length);
4640 
4641 	if (ring_buffer_time_stamp_abs(cpu_buffer->buffer)) {
4642 		add_ts_default = RB_ADD_STAMP_ABSOLUTE;
4643 		info.length += RB_LEN_TIME_EXTEND;
4644 		if (info.length > cpu_buffer->buffer->max_data_size)
4645 			goto out_fail;
4646 	} else {
4647 		add_ts_default = RB_ADD_STAMP_NONE;
4648 	}
4649 
4650  again:
4651 	info.add_timestamp = add_ts_default;
4652 	info.delta = 0;
4653 
4654 	/*
4655 	 * We allow for interrupts to reenter here and do a trace.
4656 	 * If one does, it will cause this original code to loop
4657 	 * back here. Even with heavy interrupts happening, this
4658 	 * should only happen a few times in a row. If this happens
4659 	 * 1000 times in a row, there must be either an interrupt
4660 	 * storm or we have something buggy.
4661 	 * Bail!
4662 	 */
4663 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
4664 		goto out_fail;
4665 
4666 	event = __rb_reserve_next(cpu_buffer, &info);
4667 
4668 	if (unlikely(PTR_ERR(event) == -EAGAIN)) {
4669 		if (info.add_timestamp & (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND))
4670 			info.length -= RB_LEN_TIME_EXTEND;
4671 		goto again;
4672 	}
4673 
4674 	if (likely(event))
4675 		return event;
4676  out_fail:
4677 	rb_end_commit(cpu_buffer);
4678 	return NULL;
4679 }
4680 
4681 /**
4682  * ring_buffer_lock_reserve - reserve a part of the buffer
4683  * @buffer: the ring buffer to reserve from
4684  * @length: the length of the data to reserve (excluding event header)
4685  *
4686  * Returns a reserved event on the ring buffer to copy directly to.
4687  * The user of this interface will need to get the body to write into
4688  * and can use the ring_buffer_event_data() interface.
4689  *
4690  * The length is the length of the data needed, not the event length
4691  * which also includes the event header.
4692  *
4693  * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
4694  * If NULL is returned, then nothing has been allocated or locked.
4695  */
4696 struct ring_buffer_event *
4697 ring_buffer_lock_reserve(struct trace_buffer *buffer, unsigned long length)
4698 {
4699 	struct ring_buffer_per_cpu *cpu_buffer;
4700 	struct ring_buffer_event *event;
4701 	int cpu;
4702 
4703 	/* If we are tracing schedule, we don't want to recurse */
4704 	preempt_disable_notrace();
4705 
4706 	if (unlikely(atomic_read(&buffer->record_disabled)))
4707 		goto out;
4708 
4709 	cpu = raw_smp_processor_id();
4710 
4711 	if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
4712 		goto out;
4713 
4714 	cpu_buffer = buffer->buffers[cpu];
4715 
4716 	if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
4717 		goto out;
4718 
4719 	if (unlikely(length > buffer->max_data_size))
4720 		goto out;
4721 
4722 	if (unlikely(trace_recursive_lock(cpu_buffer)))
4723 		goto out;
4724 
4725 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
4726 	if (!event)
4727 		goto out_unlock;
4728 
4729 	return event;
4730 
4731  out_unlock:
4732 	trace_recursive_unlock(cpu_buffer);
4733  out:
4734 	preempt_enable_notrace();
4735 	return NULL;
4736 }
4737 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
4738 
4739 /*
4740  * Decrement the entries to the page that an event is on.
4741  * The event does not even need to exist, only the pointer
4742  * to the page it is on. This may only be called before the commit
4743  * takes place.
4744  */
4745 static inline void
4746 rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
4747 		   struct ring_buffer_event *event)
4748 {
4749 	unsigned long addr = (unsigned long)event;
4750 	struct buffer_page *bpage = cpu_buffer->commit_page;
4751 	struct buffer_page *start;
4752 
4753 	addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1);
4754 
4755 	/* Do the likely case first */
4756 	if (likely(bpage->page == (void *)addr)) {
4757 		local_dec(&bpage->entries);
4758 		return;
4759 	}
4760 
4761 	/*
4762 	 * Because the commit page may be on the reader page we
4763 	 * start with the next page and check the end loop there.
4764 	 */
4765 	rb_inc_page(&bpage);
4766 	start = bpage;
4767 	do {
4768 		if (bpage->page == (void *)addr) {
4769 			local_dec(&bpage->entries);
4770 			return;
4771 		}
4772 		rb_inc_page(&bpage);
4773 	} while (bpage != start);
4774 
4775 	/* commit not part of this buffer?? */
4776 	RB_WARN_ON(cpu_buffer, 1);
4777 }
4778 
4779 /**
4780  * ring_buffer_discard_commit - discard an event that has not been committed
4781  * @buffer: the ring buffer
4782  * @event: non committed event to discard
4783  *
4784  * Sometimes an event that is in the ring buffer needs to be ignored.
4785  * This function lets the user discard an event in the ring buffer
4786  * and then that event will not be read later.
4787  *
4788  * This function only works if it is called before the item has been
4789  * committed. It will try to free the event from the ring buffer
4790  * if another event has not been added behind it.
4791  *
4792  * If another event has been added behind it, it will set the event
4793  * up as discarded, and perform the commit.
4794  *
4795  * If this function is called, do not call ring_buffer_unlock_commit on
4796  * the event.
4797  */
4798 void ring_buffer_discard_commit(struct trace_buffer *buffer,
4799 				struct ring_buffer_event *event)
4800 {
4801 	struct ring_buffer_per_cpu *cpu_buffer;
4802 	int cpu;
4803 
4804 	/* The event is discarded regardless */
4805 	rb_event_discard(event);
4806 
4807 	cpu = smp_processor_id();
4808 	cpu_buffer = buffer->buffers[cpu];
4809 
4810 	/*
4811 	 * This must only be called if the event has not been
4812 	 * committed yet. Thus we can assume that preemption
4813 	 * is still disabled.
4814 	 */
4815 	RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
4816 
4817 	rb_decrement_entry(cpu_buffer, event);
4818 	rb_try_to_discard(cpu_buffer, event);
4819 	rb_end_commit(cpu_buffer);
4820 
4821 	trace_recursive_unlock(cpu_buffer);
4822 
4823 	preempt_enable_notrace();
4824 
4825 }
4826 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
4827 
4828 /**
4829  * ring_buffer_write - write data to the buffer without reserving
4830  * @buffer: The ring buffer to write to.
4831  * @length: The length of the data being written (excluding the event header)
4832  * @data: The data to write to the buffer.
4833  *
4834  * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
4835  * one function. If you already have the data to write to the buffer, it
4836  * may be easier to simply call this function.
4837  *
4838  * Note, like ring_buffer_lock_reserve, the length is the length of the data
4839  * and not the length of the event which would hold the header.
4840  */
4841 int ring_buffer_write(struct trace_buffer *buffer,
4842 		      unsigned long length,
4843 		      void *data)
4844 {
4845 	struct ring_buffer_per_cpu *cpu_buffer;
4846 	struct ring_buffer_event *event;
4847 	void *body;
4848 	int ret = -EBUSY;
4849 	int cpu;
4850 
4851 	guard(preempt_notrace)();
4852 
4853 	if (atomic_read(&buffer->record_disabled))
4854 		return -EBUSY;
4855 
4856 	cpu = raw_smp_processor_id();
4857 
4858 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
4859 		return -EBUSY;
4860 
4861 	cpu_buffer = buffer->buffers[cpu];
4862 
4863 	if (atomic_read(&cpu_buffer->record_disabled))
4864 		return -EBUSY;
4865 
4866 	if (length > buffer->max_data_size)
4867 		return -EBUSY;
4868 
4869 	if (unlikely(trace_recursive_lock(cpu_buffer)))
4870 		return -EBUSY;
4871 
4872 	event = rb_reserve_next_event(buffer, cpu_buffer, length);
4873 	if (!event)
4874 		goto out_unlock;
4875 
4876 	body = rb_event_data(event);
4877 
4878 	memcpy(body, data, length);
4879 
4880 	rb_commit(cpu_buffer);
4881 
4882 	rb_wakeups(buffer, cpu_buffer);
4883 
4884 	ret = 0;
4885 
4886  out_unlock:
4887 	trace_recursive_unlock(cpu_buffer);
4888 	return ret;
4889 }
4890 EXPORT_SYMBOL_GPL(ring_buffer_write);
4891 
4892 /*
4893  * The total entries in the ring buffer is the running counter
4894  * of entries entered into the ring buffer, minus the sum of
4895  * the entries read from the ring buffer and the number of
4896  * entries that were overwritten.
4897  */
4898 static inline unsigned long
4899 rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
4900 {
4901 	return local_read(&cpu_buffer->entries) -
4902 		(local_read(&cpu_buffer->overrun) + cpu_buffer->read);
4903 }
4904 
4905 static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
4906 {
4907 	return !rb_num_of_entries(cpu_buffer);
4908 }
4909 
4910 /**
4911  * ring_buffer_record_disable - stop all writes into the buffer
4912  * @buffer: The ring buffer to stop writes to.
4913  *
4914  * This prevents all writes to the buffer. Any attempt to write
4915  * to the buffer after this will fail and return NULL.
4916  *
4917  * The caller should call synchronize_rcu() after this.
4918  */
4919 void ring_buffer_record_disable(struct trace_buffer *buffer)
4920 {
4921 	atomic_inc(&buffer->record_disabled);
4922 }
4923 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
4924 
4925 /**
4926  * ring_buffer_record_enable - enable writes to the buffer
4927  * @buffer: The ring buffer to enable writes
4928  *
4929  * Note, multiple disables will need the same number of enables
4930  * to truly enable the writing (much like preempt_disable).
4931  */
4932 void ring_buffer_record_enable(struct trace_buffer *buffer)
4933 {
4934 	atomic_dec(&buffer->record_disabled);
4935 }
4936 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
4937 
4938 /**
4939  * ring_buffer_record_off - stop all writes into the buffer
4940  * @buffer: The ring buffer to stop writes to.
4941  *
4942  * This prevents all writes to the buffer. Any attempt to write
4943  * to the buffer after this will fail and return NULL.
4944  *
4945  * This is different than ring_buffer_record_disable() as
4946  * it works like an on/off switch, where as the disable() version
4947  * must be paired with a enable().
4948  */
4949 void ring_buffer_record_off(struct trace_buffer *buffer)
4950 {
4951 	unsigned int rd;
4952 	unsigned int new_rd;
4953 
4954 	rd = atomic_read(&buffer->record_disabled);
4955 	do {
4956 		new_rd = rd | RB_BUFFER_OFF;
4957 	} while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd));
4958 }
4959 EXPORT_SYMBOL_GPL(ring_buffer_record_off);
4960 
4961 /**
4962  * ring_buffer_record_on - restart writes into the buffer
4963  * @buffer: The ring buffer to start writes to.
4964  *
4965  * This enables all writes to the buffer that was disabled by
4966  * ring_buffer_record_off().
4967  *
4968  * This is different than ring_buffer_record_enable() as
4969  * it works like an on/off switch, where as the enable() version
4970  * must be paired with a disable().
4971  */
4972 void ring_buffer_record_on(struct trace_buffer *buffer)
4973 {
4974 	unsigned int rd;
4975 	unsigned int new_rd;
4976 
4977 	rd = atomic_read(&buffer->record_disabled);
4978 	do {
4979 		new_rd = rd & ~RB_BUFFER_OFF;
4980 	} while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd));
4981 }
4982 EXPORT_SYMBOL_GPL(ring_buffer_record_on);
4983 
4984 /**
4985  * ring_buffer_record_is_on - return true if the ring buffer can write
4986  * @buffer: The ring buffer to see if write is enabled
4987  *
4988  * Returns true if the ring buffer is in a state that it accepts writes.
4989  */
4990 bool ring_buffer_record_is_on(struct trace_buffer *buffer)
4991 {
4992 	return !atomic_read(&buffer->record_disabled);
4993 }
4994 
4995 /**
4996  * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
4997  * @buffer: The ring buffer to see if write is set enabled
4998  *
4999  * Returns true if the ring buffer is set writable by ring_buffer_record_on().
5000  * Note that this does NOT mean it is in a writable state.
5001  *
5002  * It may return true when the ring buffer has been disabled by
5003  * ring_buffer_record_disable(), as that is a temporary disabling of
5004  * the ring buffer.
5005  */
5006 bool ring_buffer_record_is_set_on(struct trace_buffer *buffer)
5007 {
5008 	return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF);
5009 }
5010 
5011 /**
5012  * ring_buffer_record_is_on_cpu - return true if the ring buffer can write
5013  * @buffer: The ring buffer to see if write is enabled
5014  * @cpu: The CPU to test if the ring buffer can write too
5015  *
5016  * Returns true if the ring buffer is in a state that it accepts writes
5017  *   for a particular CPU.
5018  */
5019 bool ring_buffer_record_is_on_cpu(struct trace_buffer *buffer, int cpu)
5020 {
5021 	struct ring_buffer_per_cpu *cpu_buffer;
5022 
5023 	cpu_buffer = buffer->buffers[cpu];
5024 
5025 	return ring_buffer_record_is_set_on(buffer) &&
5026 		!atomic_read(&cpu_buffer->record_disabled);
5027 }
5028 
5029 /**
5030  * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
5031  * @buffer: The ring buffer to stop writes to.
5032  * @cpu: The CPU buffer to stop
5033  *
5034  * This prevents all writes to the buffer. Any attempt to write
5035  * to the buffer after this will fail and return NULL.
5036  *
5037  * The caller should call synchronize_rcu() after this.
5038  */
5039 void ring_buffer_record_disable_cpu(struct trace_buffer *buffer, int cpu)
5040 {
5041 	struct ring_buffer_per_cpu *cpu_buffer;
5042 
5043 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5044 		return;
5045 
5046 	cpu_buffer = buffer->buffers[cpu];
5047 	atomic_inc(&cpu_buffer->record_disabled);
5048 }
5049 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
5050 
5051 /**
5052  * ring_buffer_record_enable_cpu - enable writes to the buffer
5053  * @buffer: The ring buffer to enable writes
5054  * @cpu: The CPU to enable.
5055  *
5056  * Note, multiple disables will need the same number of enables
5057  * to truly enable the writing (much like preempt_disable).
5058  */
5059 void ring_buffer_record_enable_cpu(struct trace_buffer *buffer, int cpu)
5060 {
5061 	struct ring_buffer_per_cpu *cpu_buffer;
5062 
5063 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5064 		return;
5065 
5066 	cpu_buffer = buffer->buffers[cpu];
5067 	atomic_dec(&cpu_buffer->record_disabled);
5068 }
5069 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
5070 
5071 /**
5072  * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
5073  * @buffer: The ring buffer
5074  * @cpu: The per CPU buffer to read from.
5075  */
5076 u64 ring_buffer_oldest_event_ts(struct trace_buffer *buffer, int cpu)
5077 {
5078 	unsigned long flags;
5079 	struct ring_buffer_per_cpu *cpu_buffer;
5080 	struct buffer_page *bpage;
5081 	u64 ret = 0;
5082 
5083 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5084 		return 0;
5085 
5086 	cpu_buffer = buffer->buffers[cpu];
5087 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5088 	/*
5089 	 * if the tail is on reader_page, oldest time stamp is on the reader
5090 	 * page
5091 	 */
5092 	if (cpu_buffer->tail_page == cpu_buffer->reader_page)
5093 		bpage = cpu_buffer->reader_page;
5094 	else
5095 		bpage = rb_set_head_page(cpu_buffer);
5096 	if (bpage)
5097 		ret = bpage->page->time_stamp;
5098 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5099 
5100 	return ret;
5101 }
5102 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
5103 
5104 /**
5105  * ring_buffer_bytes_cpu - get the number of bytes unconsumed in a cpu buffer
5106  * @buffer: The ring buffer
5107  * @cpu: The per CPU buffer to read from.
5108  */
5109 unsigned long ring_buffer_bytes_cpu(struct trace_buffer *buffer, int cpu)
5110 {
5111 	struct ring_buffer_per_cpu *cpu_buffer;
5112 	unsigned long ret;
5113 
5114 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5115 		return 0;
5116 
5117 	cpu_buffer = buffer->buffers[cpu];
5118 	ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
5119 
5120 	return ret;
5121 }
5122 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
5123 
5124 /**
5125  * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
5126  * @buffer: The ring buffer
5127  * @cpu: The per CPU buffer to get the entries from.
5128  */
5129 unsigned long ring_buffer_entries_cpu(struct trace_buffer *buffer, int cpu)
5130 {
5131 	struct ring_buffer_per_cpu *cpu_buffer;
5132 
5133 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5134 		return 0;
5135 
5136 	cpu_buffer = buffer->buffers[cpu];
5137 
5138 	return rb_num_of_entries(cpu_buffer);
5139 }
5140 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
5141 
5142 /**
5143  * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
5144  * buffer wrapping around (only if RB_FL_OVERWRITE is on).
5145  * @buffer: The ring buffer
5146  * @cpu: The per CPU buffer to get the number of overruns from
5147  */
5148 unsigned long ring_buffer_overrun_cpu(struct trace_buffer *buffer, int cpu)
5149 {
5150 	struct ring_buffer_per_cpu *cpu_buffer;
5151 	unsigned long ret;
5152 
5153 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5154 		return 0;
5155 
5156 	cpu_buffer = buffer->buffers[cpu];
5157 	ret = local_read(&cpu_buffer->overrun);
5158 
5159 	return ret;
5160 }
5161 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
5162 
5163 /**
5164  * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
5165  * commits failing due to the buffer wrapping around while there are uncommitted
5166  * events, such as during an interrupt storm.
5167  * @buffer: The ring buffer
5168  * @cpu: The per CPU buffer to get the number of overruns from
5169  */
5170 unsigned long
5171 ring_buffer_commit_overrun_cpu(struct trace_buffer *buffer, int cpu)
5172 {
5173 	struct ring_buffer_per_cpu *cpu_buffer;
5174 	unsigned long ret;
5175 
5176 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5177 		return 0;
5178 
5179 	cpu_buffer = buffer->buffers[cpu];
5180 	ret = local_read(&cpu_buffer->commit_overrun);
5181 
5182 	return ret;
5183 }
5184 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
5185 
5186 /**
5187  * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
5188  * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
5189  * @buffer: The ring buffer
5190  * @cpu: The per CPU buffer to get the number of overruns from
5191  */
5192 unsigned long
5193 ring_buffer_dropped_events_cpu(struct trace_buffer *buffer, int cpu)
5194 {
5195 	struct ring_buffer_per_cpu *cpu_buffer;
5196 	unsigned long ret;
5197 
5198 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5199 		return 0;
5200 
5201 	cpu_buffer = buffer->buffers[cpu];
5202 	ret = local_read(&cpu_buffer->dropped_events);
5203 
5204 	return ret;
5205 }
5206 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
5207 
5208 /**
5209  * ring_buffer_read_events_cpu - get the number of events successfully read
5210  * @buffer: The ring buffer
5211  * @cpu: The per CPU buffer to get the number of events read
5212  */
5213 unsigned long
5214 ring_buffer_read_events_cpu(struct trace_buffer *buffer, int cpu)
5215 {
5216 	struct ring_buffer_per_cpu *cpu_buffer;
5217 
5218 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5219 		return 0;
5220 
5221 	cpu_buffer = buffer->buffers[cpu];
5222 	return cpu_buffer->read;
5223 }
5224 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
5225 
5226 /**
5227  * ring_buffer_entries - get the number of entries in a buffer
5228  * @buffer: The ring buffer
5229  *
5230  * Returns the total number of entries in the ring buffer
5231  * (all CPU entries)
5232  */
5233 unsigned long ring_buffer_entries(struct trace_buffer *buffer)
5234 {
5235 	struct ring_buffer_per_cpu *cpu_buffer;
5236 	unsigned long entries = 0;
5237 	int cpu;
5238 
5239 	/* if you care about this being correct, lock the buffer */
5240 	for_each_buffer_cpu(buffer, cpu) {
5241 		cpu_buffer = buffer->buffers[cpu];
5242 		entries += rb_num_of_entries(cpu_buffer);
5243 	}
5244 
5245 	return entries;
5246 }
5247 EXPORT_SYMBOL_GPL(ring_buffer_entries);
5248 
5249 /**
5250  * ring_buffer_overruns - get the number of overruns in buffer
5251  * @buffer: The ring buffer
5252  *
5253  * Returns the total number of overruns in the ring buffer
5254  * (all CPU entries)
5255  */
5256 unsigned long ring_buffer_overruns(struct trace_buffer *buffer)
5257 {
5258 	struct ring_buffer_per_cpu *cpu_buffer;
5259 	unsigned long overruns = 0;
5260 	int cpu;
5261 
5262 	/* if you care about this being correct, lock the buffer */
5263 	for_each_buffer_cpu(buffer, cpu) {
5264 		cpu_buffer = buffer->buffers[cpu];
5265 		overruns += local_read(&cpu_buffer->overrun);
5266 	}
5267 
5268 	return overruns;
5269 }
5270 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
5271 
5272 static void rb_iter_reset(struct ring_buffer_iter *iter)
5273 {
5274 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5275 
5276 	/* Iterator usage is expected to have record disabled */
5277 	iter->head_page = cpu_buffer->reader_page;
5278 	iter->head = cpu_buffer->reader_page->read;
5279 	iter->next_event = iter->head;
5280 
5281 	iter->cache_reader_page = iter->head_page;
5282 	iter->cache_read = cpu_buffer->read;
5283 	iter->cache_pages_removed = cpu_buffer->pages_removed;
5284 
5285 	if (iter->head) {
5286 		iter->read_stamp = cpu_buffer->read_stamp;
5287 		iter->page_stamp = cpu_buffer->reader_page->page->time_stamp;
5288 	} else {
5289 		iter->read_stamp = iter->head_page->page->time_stamp;
5290 		iter->page_stamp = iter->read_stamp;
5291 	}
5292 }
5293 
5294 /**
5295  * ring_buffer_iter_reset - reset an iterator
5296  * @iter: The iterator to reset
5297  *
5298  * Resets the iterator, so that it will start from the beginning
5299  * again.
5300  */
5301 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
5302 {
5303 	struct ring_buffer_per_cpu *cpu_buffer;
5304 	unsigned long flags;
5305 
5306 	if (!iter)
5307 		return;
5308 
5309 	cpu_buffer = iter->cpu_buffer;
5310 
5311 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5312 	rb_iter_reset(iter);
5313 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5314 }
5315 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
5316 
5317 /**
5318  * ring_buffer_iter_empty - check if an iterator has no more to read
5319  * @iter: The iterator to check
5320  */
5321 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
5322 {
5323 	struct ring_buffer_per_cpu *cpu_buffer;
5324 	struct buffer_page *reader;
5325 	struct buffer_page *head_page;
5326 	struct buffer_page *commit_page;
5327 	struct buffer_page *curr_commit_page;
5328 	unsigned commit;
5329 	u64 curr_commit_ts;
5330 	u64 commit_ts;
5331 
5332 	cpu_buffer = iter->cpu_buffer;
5333 	reader = cpu_buffer->reader_page;
5334 	head_page = cpu_buffer->head_page;
5335 	commit_page = READ_ONCE(cpu_buffer->commit_page);
5336 	commit_ts = commit_page->page->time_stamp;
5337 
5338 	/*
5339 	 * When the writer goes across pages, it issues a cmpxchg which
5340 	 * is a mb(), which will synchronize with the rmb here.
5341 	 * (see rb_tail_page_update())
5342 	 */
5343 	smp_rmb();
5344 	commit = rb_page_commit(commit_page);
5345 	/* We want to make sure that the commit page doesn't change */
5346 	smp_rmb();
5347 
5348 	/* Make sure commit page didn't change */
5349 	curr_commit_page = READ_ONCE(cpu_buffer->commit_page);
5350 	curr_commit_ts = READ_ONCE(curr_commit_page->page->time_stamp);
5351 
5352 	/* If the commit page changed, then there's more data */
5353 	if (curr_commit_page != commit_page ||
5354 	    curr_commit_ts != commit_ts)
5355 		return 0;
5356 
5357 	/* Still racy, as it may return a false positive, but that's OK */
5358 	return ((iter->head_page == commit_page && iter->head >= commit) ||
5359 		(iter->head_page == reader && commit_page == head_page &&
5360 		 head_page->read == commit &&
5361 		 iter->head == rb_page_size(cpu_buffer->reader_page)));
5362 }
5363 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
5364 
5365 static void
5366 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
5367 		     struct ring_buffer_event *event)
5368 {
5369 	u64 delta;
5370 
5371 	switch (event->type_len) {
5372 	case RINGBUF_TYPE_PADDING:
5373 		return;
5374 
5375 	case RINGBUF_TYPE_TIME_EXTEND:
5376 		delta = rb_event_time_stamp(event);
5377 		cpu_buffer->read_stamp += delta;
5378 		return;
5379 
5380 	case RINGBUF_TYPE_TIME_STAMP:
5381 		delta = rb_event_time_stamp(event);
5382 		delta = rb_fix_abs_ts(delta, cpu_buffer->read_stamp);
5383 		cpu_buffer->read_stamp = delta;
5384 		return;
5385 
5386 	case RINGBUF_TYPE_DATA:
5387 		cpu_buffer->read_stamp += event->time_delta;
5388 		return;
5389 
5390 	default:
5391 		RB_WARN_ON(cpu_buffer, 1);
5392 	}
5393 }
5394 
5395 static void
5396 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
5397 			  struct ring_buffer_event *event)
5398 {
5399 	u64 delta;
5400 
5401 	switch (event->type_len) {
5402 	case RINGBUF_TYPE_PADDING:
5403 		return;
5404 
5405 	case RINGBUF_TYPE_TIME_EXTEND:
5406 		delta = rb_event_time_stamp(event);
5407 		iter->read_stamp += delta;
5408 		return;
5409 
5410 	case RINGBUF_TYPE_TIME_STAMP:
5411 		delta = rb_event_time_stamp(event);
5412 		delta = rb_fix_abs_ts(delta, iter->read_stamp);
5413 		iter->read_stamp = delta;
5414 		return;
5415 
5416 	case RINGBUF_TYPE_DATA:
5417 		iter->read_stamp += event->time_delta;
5418 		return;
5419 
5420 	default:
5421 		RB_WARN_ON(iter->cpu_buffer, 1);
5422 	}
5423 }
5424 
5425 static struct buffer_page *
5426 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
5427 {
5428 	struct buffer_page *reader = NULL;
5429 	unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size);
5430 	unsigned long overwrite;
5431 	unsigned long flags;
5432 	int nr_loops = 0;
5433 	bool ret;
5434 
5435 	local_irq_save(flags);
5436 	arch_spin_lock(&cpu_buffer->lock);
5437 
5438  again:
5439 	/*
5440 	 * This should normally only loop twice. But because the
5441 	 * start of the reader inserts an empty page, it causes
5442 	 * a case where we will loop three times. There should be no
5443 	 * reason to loop four times (that I know of).
5444 	 */
5445 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
5446 		reader = NULL;
5447 		goto out;
5448 	}
5449 
5450 	reader = cpu_buffer->reader_page;
5451 
5452 	/* If there's more to read, return this page */
5453 	if (cpu_buffer->reader_page->read < rb_page_size(reader))
5454 		goto out;
5455 
5456 	/* Never should we have an index greater than the size */
5457 	if (RB_WARN_ON(cpu_buffer,
5458 		       cpu_buffer->reader_page->read > rb_page_size(reader)))
5459 		goto out;
5460 
5461 	/* check if we caught up to the tail */
5462 	reader = NULL;
5463 	if (cpu_buffer->commit_page == cpu_buffer->reader_page)
5464 		goto out;
5465 
5466 	/* Don't bother swapping if the ring buffer is empty */
5467 	if (rb_num_of_entries(cpu_buffer) == 0)
5468 		goto out;
5469 
5470 	/*
5471 	 * Reset the reader page to size zero.
5472 	 */
5473 	local_set(&cpu_buffer->reader_page->write, 0);
5474 	local_set(&cpu_buffer->reader_page->entries, 0);
5475 	cpu_buffer->reader_page->real_end = 0;
5476 
5477  spin:
5478 	/*
5479 	 * Splice the empty reader page into the list around the head.
5480 	 */
5481 	reader = rb_set_head_page(cpu_buffer);
5482 	if (!reader)
5483 		goto out;
5484 	cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
5485 	cpu_buffer->reader_page->list.prev = reader->list.prev;
5486 
5487 	/*
5488 	 * cpu_buffer->pages just needs to point to the buffer, it
5489 	 *  has no specific buffer page to point to. Lets move it out
5490 	 *  of our way so we don't accidentally swap it.
5491 	 */
5492 	cpu_buffer->pages = reader->list.prev;
5493 
5494 	/* The reader page will be pointing to the new head */
5495 	rb_set_list_to_head(&cpu_buffer->reader_page->list);
5496 
5497 	/*
5498 	 * We want to make sure we read the overruns after we set up our
5499 	 * pointers to the next object. The writer side does a
5500 	 * cmpxchg to cross pages which acts as the mb on the writer
5501 	 * side. Note, the reader will constantly fail the swap
5502 	 * while the writer is updating the pointers, so this
5503 	 * guarantees that the overwrite recorded here is the one we
5504 	 * want to compare with the last_overrun.
5505 	 */
5506 	smp_mb();
5507 	overwrite = local_read(&(cpu_buffer->overrun));
5508 
5509 	/*
5510 	 * Here's the tricky part.
5511 	 *
5512 	 * We need to move the pointer past the header page.
5513 	 * But we can only do that if a writer is not currently
5514 	 * moving it. The page before the header page has the
5515 	 * flag bit '1' set if it is pointing to the page we want.
5516 	 * but if the writer is in the process of moving it
5517 	 * then it will be '2' or already moved '0'.
5518 	 */
5519 
5520 	ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
5521 
5522 	/*
5523 	 * If we did not convert it, then we must try again.
5524 	 */
5525 	if (!ret)
5526 		goto spin;
5527 
5528 	if (cpu_buffer->ring_meta)
5529 		rb_update_meta_reader(cpu_buffer, reader);
5530 
5531 	/*
5532 	 * Yay! We succeeded in replacing the page.
5533 	 *
5534 	 * Now make the new head point back to the reader page.
5535 	 */
5536 	rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
5537 	rb_inc_page(&cpu_buffer->head_page);
5538 
5539 	cpu_buffer->cnt++;
5540 	local_inc(&cpu_buffer->pages_read);
5541 
5542 	/* Finally update the reader page to the new head */
5543 	cpu_buffer->reader_page = reader;
5544 	cpu_buffer->reader_page->read = 0;
5545 
5546 	if (overwrite != cpu_buffer->last_overrun) {
5547 		cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
5548 		cpu_buffer->last_overrun = overwrite;
5549 	}
5550 
5551 	goto again;
5552 
5553  out:
5554 	/* Update the read_stamp on the first event */
5555 	if (reader && reader->read == 0)
5556 		cpu_buffer->read_stamp = reader->page->time_stamp;
5557 
5558 	arch_spin_unlock(&cpu_buffer->lock);
5559 	local_irq_restore(flags);
5560 
5561 	/*
5562 	 * The writer has preempt disable, wait for it. But not forever
5563 	 * Although, 1 second is pretty much "forever"
5564 	 */
5565 #define USECS_WAIT	1000000
5566         for (nr_loops = 0; nr_loops < USECS_WAIT; nr_loops++) {
5567 		/* If the write is past the end of page, a writer is still updating it */
5568 		if (likely(!reader || rb_page_write(reader) <= bsize))
5569 			break;
5570 
5571 		udelay(1);
5572 
5573 		/* Get the latest version of the reader write value */
5574 		smp_rmb();
5575 	}
5576 
5577 	/* The writer is not moving forward? Something is wrong */
5578 	if (RB_WARN_ON(cpu_buffer, nr_loops == USECS_WAIT))
5579 		reader = NULL;
5580 
5581 	/*
5582 	 * Make sure we see any padding after the write update
5583 	 * (see rb_reset_tail()).
5584 	 *
5585 	 * In addition, a writer may be writing on the reader page
5586 	 * if the page has not been fully filled, so the read barrier
5587 	 * is also needed to make sure we see the content of what is
5588 	 * committed by the writer (see rb_set_commit_to_write()).
5589 	 */
5590 	smp_rmb();
5591 
5592 
5593 	return reader;
5594 }
5595 
5596 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
5597 {
5598 	struct ring_buffer_event *event;
5599 	struct buffer_page *reader;
5600 	unsigned length;
5601 
5602 	reader = rb_get_reader_page(cpu_buffer);
5603 
5604 	/* This function should not be called when buffer is empty */
5605 	if (RB_WARN_ON(cpu_buffer, !reader))
5606 		return;
5607 
5608 	event = rb_reader_event(cpu_buffer);
5609 
5610 	if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
5611 		cpu_buffer->read++;
5612 
5613 	rb_update_read_stamp(cpu_buffer, event);
5614 
5615 	length = rb_event_length(event);
5616 	cpu_buffer->reader_page->read += length;
5617 	cpu_buffer->read_bytes += length;
5618 }
5619 
5620 static void rb_advance_iter(struct ring_buffer_iter *iter)
5621 {
5622 	struct ring_buffer_per_cpu *cpu_buffer;
5623 
5624 	cpu_buffer = iter->cpu_buffer;
5625 
5626 	/* If head == next_event then we need to jump to the next event */
5627 	if (iter->head == iter->next_event) {
5628 		/* If the event gets overwritten again, there's nothing to do */
5629 		if (rb_iter_head_event(iter) == NULL)
5630 			return;
5631 	}
5632 
5633 	iter->head = iter->next_event;
5634 
5635 	/*
5636 	 * Check if we are at the end of the buffer.
5637 	 */
5638 	if (iter->next_event >= rb_page_size(iter->head_page)) {
5639 		/* discarded commits can make the page empty */
5640 		if (iter->head_page == cpu_buffer->commit_page)
5641 			return;
5642 		rb_inc_iter(iter);
5643 		return;
5644 	}
5645 
5646 	rb_update_iter_read_stamp(iter, iter->event);
5647 }
5648 
5649 static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
5650 {
5651 	return cpu_buffer->lost_events;
5652 }
5653 
5654 static struct ring_buffer_event *
5655 rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
5656 	       unsigned long *lost_events)
5657 {
5658 	struct ring_buffer_event *event;
5659 	struct buffer_page *reader;
5660 	int nr_loops = 0;
5661 
5662 	if (ts)
5663 		*ts = 0;
5664  again:
5665 	/*
5666 	 * We repeat when a time extend is encountered.
5667 	 * Since the time extend is always attached to a data event,
5668 	 * we should never loop more than once.
5669 	 * (We never hit the following condition more than twice).
5670 	 */
5671 	if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
5672 		return NULL;
5673 
5674 	reader = rb_get_reader_page(cpu_buffer);
5675 	if (!reader)
5676 		return NULL;
5677 
5678 	event = rb_reader_event(cpu_buffer);
5679 
5680 	switch (event->type_len) {
5681 	case RINGBUF_TYPE_PADDING:
5682 		if (rb_null_event(event))
5683 			RB_WARN_ON(cpu_buffer, 1);
5684 		/*
5685 		 * Because the writer could be discarding every
5686 		 * event it creates (which would probably be bad)
5687 		 * if we were to go back to "again" then we may never
5688 		 * catch up, and will trigger the warn on, or lock
5689 		 * the box. Return the padding, and we will release
5690 		 * the current locks, and try again.
5691 		 */
5692 		return event;
5693 
5694 	case RINGBUF_TYPE_TIME_EXTEND:
5695 		/* Internal data, OK to advance */
5696 		rb_advance_reader(cpu_buffer);
5697 		goto again;
5698 
5699 	case RINGBUF_TYPE_TIME_STAMP:
5700 		if (ts) {
5701 			*ts = rb_event_time_stamp(event);
5702 			*ts = rb_fix_abs_ts(*ts, reader->page->time_stamp);
5703 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5704 							 cpu_buffer->cpu, ts);
5705 		}
5706 		/* Internal data, OK to advance */
5707 		rb_advance_reader(cpu_buffer);
5708 		goto again;
5709 
5710 	case RINGBUF_TYPE_DATA:
5711 		if (ts && !(*ts)) {
5712 			*ts = cpu_buffer->read_stamp + event->time_delta;
5713 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5714 							 cpu_buffer->cpu, ts);
5715 		}
5716 		if (lost_events)
5717 			*lost_events = rb_lost_events(cpu_buffer);
5718 		return event;
5719 
5720 	default:
5721 		RB_WARN_ON(cpu_buffer, 1);
5722 	}
5723 
5724 	return NULL;
5725 }
5726 EXPORT_SYMBOL_GPL(ring_buffer_peek);
5727 
5728 static struct ring_buffer_event *
5729 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
5730 {
5731 	struct trace_buffer *buffer;
5732 	struct ring_buffer_per_cpu *cpu_buffer;
5733 	struct ring_buffer_event *event;
5734 	int nr_loops = 0;
5735 
5736 	if (ts)
5737 		*ts = 0;
5738 
5739 	cpu_buffer = iter->cpu_buffer;
5740 	buffer = cpu_buffer->buffer;
5741 
5742 	/*
5743 	 * Check if someone performed a consuming read to the buffer
5744 	 * or removed some pages from the buffer. In these cases,
5745 	 * iterator was invalidated and we need to reset it.
5746 	 */
5747 	if (unlikely(iter->cache_read != cpu_buffer->read ||
5748 		     iter->cache_reader_page != cpu_buffer->reader_page ||
5749 		     iter->cache_pages_removed != cpu_buffer->pages_removed))
5750 		rb_iter_reset(iter);
5751 
5752  again:
5753 	if (ring_buffer_iter_empty(iter))
5754 		return NULL;
5755 
5756 	/*
5757 	 * As the writer can mess with what the iterator is trying
5758 	 * to read, just give up if we fail to get an event after
5759 	 * three tries. The iterator is not as reliable when reading
5760 	 * the ring buffer with an active write as the consumer is.
5761 	 * Do not warn if the three failures is reached.
5762 	 */
5763 	if (++nr_loops > 3)
5764 		return NULL;
5765 
5766 	if (rb_per_cpu_empty(cpu_buffer))
5767 		return NULL;
5768 
5769 	if (iter->head >= rb_page_size(iter->head_page)) {
5770 		rb_inc_iter(iter);
5771 		goto again;
5772 	}
5773 
5774 	event = rb_iter_head_event(iter);
5775 	if (!event)
5776 		goto again;
5777 
5778 	switch (event->type_len) {
5779 	case RINGBUF_TYPE_PADDING:
5780 		if (rb_null_event(event)) {
5781 			rb_inc_iter(iter);
5782 			goto again;
5783 		}
5784 		rb_advance_iter(iter);
5785 		return event;
5786 
5787 	case RINGBUF_TYPE_TIME_EXTEND:
5788 		/* Internal data, OK to advance */
5789 		rb_advance_iter(iter);
5790 		goto again;
5791 
5792 	case RINGBUF_TYPE_TIME_STAMP:
5793 		if (ts) {
5794 			*ts = rb_event_time_stamp(event);
5795 			*ts = rb_fix_abs_ts(*ts, iter->head_page->page->time_stamp);
5796 			ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5797 							 cpu_buffer->cpu, ts);
5798 		}
5799 		/* Internal data, OK to advance */
5800 		rb_advance_iter(iter);
5801 		goto again;
5802 
5803 	case RINGBUF_TYPE_DATA:
5804 		if (ts && !(*ts)) {
5805 			*ts = iter->read_stamp + event->time_delta;
5806 			ring_buffer_normalize_time_stamp(buffer,
5807 							 cpu_buffer->cpu, ts);
5808 		}
5809 		return event;
5810 
5811 	default:
5812 		RB_WARN_ON(cpu_buffer, 1);
5813 	}
5814 
5815 	return NULL;
5816 }
5817 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
5818 
5819 static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
5820 {
5821 	if (likely(!in_nmi())) {
5822 		raw_spin_lock(&cpu_buffer->reader_lock);
5823 		return true;
5824 	}
5825 
5826 	/*
5827 	 * If an NMI die dumps out the content of the ring buffer
5828 	 * trylock must be used to prevent a deadlock if the NMI
5829 	 * preempted a task that holds the ring buffer locks. If
5830 	 * we get the lock then all is fine, if not, then continue
5831 	 * to do the read, but this can corrupt the ring buffer,
5832 	 * so it must be permanently disabled from future writes.
5833 	 * Reading from NMI is a oneshot deal.
5834 	 */
5835 	if (raw_spin_trylock(&cpu_buffer->reader_lock))
5836 		return true;
5837 
5838 	/* Continue without locking, but disable the ring buffer */
5839 	atomic_inc(&cpu_buffer->record_disabled);
5840 	return false;
5841 }
5842 
5843 static inline void
5844 rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
5845 {
5846 	if (likely(locked))
5847 		raw_spin_unlock(&cpu_buffer->reader_lock);
5848 }
5849 
5850 /**
5851  * ring_buffer_peek - peek at the next event to be read
5852  * @buffer: The ring buffer to read
5853  * @cpu: The cpu to peak at
5854  * @ts: The timestamp counter of this event.
5855  * @lost_events: a variable to store if events were lost (may be NULL)
5856  *
5857  * This will return the event that will be read next, but does
5858  * not consume the data.
5859  */
5860 struct ring_buffer_event *
5861 ring_buffer_peek(struct trace_buffer *buffer, int cpu, u64 *ts,
5862 		 unsigned long *lost_events)
5863 {
5864 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5865 	struct ring_buffer_event *event;
5866 	unsigned long flags;
5867 	bool dolock;
5868 
5869 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5870 		return NULL;
5871 
5872  again:
5873 	local_irq_save(flags);
5874 	dolock = rb_reader_lock(cpu_buffer);
5875 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
5876 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
5877 		rb_advance_reader(cpu_buffer);
5878 	rb_reader_unlock(cpu_buffer, dolock);
5879 	local_irq_restore(flags);
5880 
5881 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
5882 		goto again;
5883 
5884 	return event;
5885 }
5886 
5887 /** ring_buffer_iter_dropped - report if there are dropped events
5888  * @iter: The ring buffer iterator
5889  *
5890  * Returns true if there was dropped events since the last peek.
5891  */
5892 bool ring_buffer_iter_dropped(struct ring_buffer_iter *iter)
5893 {
5894 	bool ret = iter->missed_events != 0;
5895 
5896 	iter->missed_events = 0;
5897 	return ret;
5898 }
5899 EXPORT_SYMBOL_GPL(ring_buffer_iter_dropped);
5900 
5901 /**
5902  * ring_buffer_iter_peek - peek at the next event to be read
5903  * @iter: The ring buffer iterator
5904  * @ts: The timestamp counter of this event.
5905  *
5906  * This will return the event that will be read next, but does
5907  * not increment the iterator.
5908  */
5909 struct ring_buffer_event *
5910 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
5911 {
5912 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5913 	struct ring_buffer_event *event;
5914 	unsigned long flags;
5915 
5916  again:
5917 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5918 	event = rb_iter_peek(iter, ts);
5919 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5920 
5921 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
5922 		goto again;
5923 
5924 	return event;
5925 }
5926 
5927 /**
5928  * ring_buffer_consume - return an event and consume it
5929  * @buffer: The ring buffer to get the next event from
5930  * @cpu: the cpu to read the buffer from
5931  * @ts: a variable to store the timestamp (may be NULL)
5932  * @lost_events: a variable to store if events were lost (may be NULL)
5933  *
5934  * Returns the next event in the ring buffer, and that event is consumed.
5935  * Meaning, that sequential reads will keep returning a different event,
5936  * and eventually empty the ring buffer if the producer is slower.
5937  */
5938 struct ring_buffer_event *
5939 ring_buffer_consume(struct trace_buffer *buffer, int cpu, u64 *ts,
5940 		    unsigned long *lost_events)
5941 {
5942 	struct ring_buffer_per_cpu *cpu_buffer;
5943 	struct ring_buffer_event *event = NULL;
5944 	unsigned long flags;
5945 	bool dolock;
5946 
5947  again:
5948 	/* might be called in atomic */
5949 	preempt_disable();
5950 
5951 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5952 		goto out;
5953 
5954 	cpu_buffer = buffer->buffers[cpu];
5955 	local_irq_save(flags);
5956 	dolock = rb_reader_lock(cpu_buffer);
5957 
5958 	event = rb_buffer_peek(cpu_buffer, ts, lost_events);
5959 	if (event) {
5960 		cpu_buffer->lost_events = 0;
5961 		rb_advance_reader(cpu_buffer);
5962 	}
5963 
5964 	rb_reader_unlock(cpu_buffer, dolock);
5965 	local_irq_restore(flags);
5966 
5967  out:
5968 	preempt_enable();
5969 
5970 	if (event && event->type_len == RINGBUF_TYPE_PADDING)
5971 		goto again;
5972 
5973 	return event;
5974 }
5975 EXPORT_SYMBOL_GPL(ring_buffer_consume);
5976 
5977 /**
5978  * ring_buffer_read_start - start a non consuming read of the buffer
5979  * @buffer: The ring buffer to read from
5980  * @cpu: The cpu buffer to iterate over
5981  * @flags: gfp flags to use for memory allocation
5982  *
5983  * This creates an iterator to allow non-consuming iteration through
5984  * the buffer. If the buffer is disabled for writing, it will produce
5985  * the same information each time, but if the buffer is still writing
5986  * then the first hit of a write will cause the iteration to stop.
5987  *
5988  * Must be paired with ring_buffer_read_finish.
5989  */
5990 struct ring_buffer_iter *
5991 ring_buffer_read_start(struct trace_buffer *buffer, int cpu, gfp_t flags)
5992 {
5993 	struct ring_buffer_per_cpu *cpu_buffer;
5994 	struct ring_buffer_iter *iter;
5995 
5996 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
5997 		return NULL;
5998 
5999 	iter = kzalloc(sizeof(*iter), flags);
6000 	if (!iter)
6001 		return NULL;
6002 
6003 	/* Holds the entire event: data and meta data */
6004 	iter->event_size = buffer->subbuf_size;
6005 	iter->event = kmalloc(iter->event_size, flags);
6006 	if (!iter->event) {
6007 		kfree(iter);
6008 		return NULL;
6009 	}
6010 
6011 	cpu_buffer = buffer->buffers[cpu];
6012 
6013 	iter->cpu_buffer = cpu_buffer;
6014 
6015 	atomic_inc(&cpu_buffer->resize_disabled);
6016 
6017 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
6018 	arch_spin_lock(&cpu_buffer->lock);
6019 	rb_iter_reset(iter);
6020 	arch_spin_unlock(&cpu_buffer->lock);
6021 
6022 	return iter;
6023 }
6024 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
6025 
6026 /**
6027  * ring_buffer_read_finish - finish reading the iterator of the buffer
6028  * @iter: The iterator retrieved by ring_buffer_start
6029  *
6030  * This re-enables resizing of the buffer, and frees the iterator.
6031  */
6032 void
6033 ring_buffer_read_finish(struct ring_buffer_iter *iter)
6034 {
6035 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
6036 
6037 	/* Use this opportunity to check the integrity of the ring buffer. */
6038 	rb_check_pages(cpu_buffer);
6039 
6040 	atomic_dec(&cpu_buffer->resize_disabled);
6041 	kfree(iter->event);
6042 	kfree(iter);
6043 }
6044 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
6045 
6046 /**
6047  * ring_buffer_iter_advance - advance the iterator to the next location
6048  * @iter: The ring buffer iterator
6049  *
6050  * Move the location of the iterator such that the next read will
6051  * be the next location of the iterator.
6052  */
6053 void ring_buffer_iter_advance(struct ring_buffer_iter *iter)
6054 {
6055 	struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
6056 	unsigned long flags;
6057 
6058 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
6059 
6060 	rb_advance_iter(iter);
6061 
6062 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
6063 }
6064 EXPORT_SYMBOL_GPL(ring_buffer_iter_advance);
6065 
6066 /**
6067  * ring_buffer_size - return the size of the ring buffer (in bytes)
6068  * @buffer: The ring buffer.
6069  * @cpu: The CPU to get ring buffer size from.
6070  */
6071 unsigned long ring_buffer_size(struct trace_buffer *buffer, int cpu)
6072 {
6073 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
6074 		return 0;
6075 
6076 	return buffer->subbuf_size * buffer->buffers[cpu]->nr_pages;
6077 }
6078 EXPORT_SYMBOL_GPL(ring_buffer_size);
6079 
6080 /**
6081  * ring_buffer_max_event_size - return the max data size of an event
6082  * @buffer: The ring buffer.
6083  *
6084  * Returns the maximum size an event can be.
6085  */
6086 unsigned long ring_buffer_max_event_size(struct trace_buffer *buffer)
6087 {
6088 	/* If abs timestamp is requested, events have a timestamp too */
6089 	if (ring_buffer_time_stamp_abs(buffer))
6090 		return buffer->max_data_size - RB_LEN_TIME_EXTEND;
6091 	return buffer->max_data_size;
6092 }
6093 EXPORT_SYMBOL_GPL(ring_buffer_max_event_size);
6094 
6095 static void rb_clear_buffer_page(struct buffer_page *page)
6096 {
6097 	local_set(&page->write, 0);
6098 	local_set(&page->entries, 0);
6099 	rb_init_page(page->page);
6100 	page->read = 0;
6101 }
6102 
6103 /*
6104  * When the buffer is memory mapped to user space, each sub buffer
6105  * has a unique id that is used by the meta data to tell the user
6106  * where the current reader page is.
6107  *
6108  * For a normal allocated ring buffer, the id is saved in the buffer page
6109  * id field, and updated via this function.
6110  *
6111  * But for a fixed memory mapped buffer, the id is already assigned for
6112  * fixed memory ordering in the memory layout and can not be used. Instead
6113  * the index of where the page lies in the memory layout is used.
6114  *
6115  * For the normal pages, set the buffer page id with the passed in @id
6116  * value and return that.
6117  *
6118  * For fixed memory mapped pages, get the page index in the memory layout
6119  * and return that as the id.
6120  */
6121 static int rb_page_id(struct ring_buffer_per_cpu *cpu_buffer,
6122 		      struct buffer_page *bpage, int id)
6123 {
6124 	/*
6125 	 * For boot buffers, the id is the index,
6126 	 * otherwise, set the buffer page with this id
6127 	 */
6128 	if (cpu_buffer->ring_meta)
6129 		id = rb_meta_subbuf_idx(cpu_buffer->ring_meta, bpage->page);
6130 	else
6131 		bpage->id = id;
6132 
6133 	return id;
6134 }
6135 
6136 static void rb_update_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
6137 {
6138 	struct trace_buffer_meta *meta = cpu_buffer->meta_page;
6139 
6140 	if (!meta)
6141 		return;
6142 
6143 	meta->reader.read = cpu_buffer->reader_page->read;
6144 	meta->reader.id = rb_page_id(cpu_buffer, cpu_buffer->reader_page,
6145 				     cpu_buffer->reader_page->id);
6146 
6147 	meta->reader.lost_events = cpu_buffer->lost_events;
6148 
6149 	meta->entries = local_read(&cpu_buffer->entries);
6150 	meta->overrun = local_read(&cpu_buffer->overrun);
6151 	meta->read = cpu_buffer->read;
6152 
6153 	/* Some archs do not have data cache coherency between kernel and user-space */
6154 	flush_kernel_vmap_range(cpu_buffer->meta_page, PAGE_SIZE);
6155 }
6156 
6157 static void
6158 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
6159 {
6160 	struct buffer_page *page;
6161 
6162 	rb_head_page_deactivate(cpu_buffer);
6163 
6164 	cpu_buffer->head_page
6165 		= list_entry(cpu_buffer->pages, struct buffer_page, list);
6166 	rb_clear_buffer_page(cpu_buffer->head_page);
6167 	list_for_each_entry(page, cpu_buffer->pages, list) {
6168 		rb_clear_buffer_page(page);
6169 	}
6170 
6171 	cpu_buffer->tail_page = cpu_buffer->head_page;
6172 	cpu_buffer->commit_page = cpu_buffer->head_page;
6173 
6174 	INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
6175 	INIT_LIST_HEAD(&cpu_buffer->new_pages);
6176 	rb_clear_buffer_page(cpu_buffer->reader_page);
6177 
6178 	local_set(&cpu_buffer->entries_bytes, 0);
6179 	local_set(&cpu_buffer->overrun, 0);
6180 	local_set(&cpu_buffer->commit_overrun, 0);
6181 	local_set(&cpu_buffer->dropped_events, 0);
6182 	local_set(&cpu_buffer->entries, 0);
6183 	local_set(&cpu_buffer->committing, 0);
6184 	local_set(&cpu_buffer->commits, 0);
6185 	local_set(&cpu_buffer->pages_touched, 0);
6186 	local_set(&cpu_buffer->pages_lost, 0);
6187 	local_set(&cpu_buffer->pages_read, 0);
6188 	cpu_buffer->last_pages_touch = 0;
6189 	cpu_buffer->shortest_full = 0;
6190 	cpu_buffer->read = 0;
6191 	cpu_buffer->read_bytes = 0;
6192 
6193 	rb_time_set(&cpu_buffer->write_stamp, 0);
6194 	rb_time_set(&cpu_buffer->before_stamp, 0);
6195 
6196 	memset(cpu_buffer->event_stamp, 0, sizeof(cpu_buffer->event_stamp));
6197 
6198 	cpu_buffer->lost_events = 0;
6199 	cpu_buffer->last_overrun = 0;
6200 
6201 	rb_head_page_activate(cpu_buffer);
6202 	cpu_buffer->pages_removed = 0;
6203 
6204 	if (cpu_buffer->mapped) {
6205 		rb_update_meta_page(cpu_buffer);
6206 		if (cpu_buffer->ring_meta) {
6207 			struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
6208 			meta->commit_buffer = meta->head_buffer;
6209 		}
6210 	}
6211 }
6212 
6213 /* Must have disabled the cpu buffer then done a synchronize_rcu */
6214 static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
6215 {
6216 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
6217 
6218 	if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
6219 		return;
6220 
6221 	arch_spin_lock(&cpu_buffer->lock);
6222 
6223 	rb_reset_cpu(cpu_buffer);
6224 
6225 	arch_spin_unlock(&cpu_buffer->lock);
6226 }
6227 
6228 /**
6229  * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
6230  * @buffer: The ring buffer to reset a per cpu buffer of
6231  * @cpu: The CPU buffer to be reset
6232  */
6233 void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu)
6234 {
6235 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
6236 
6237 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
6238 		return;
6239 
6240 	/* prevent another thread from changing buffer sizes */
6241 	mutex_lock(&buffer->mutex);
6242 
6243 	atomic_inc(&cpu_buffer->resize_disabled);
6244 	atomic_inc(&cpu_buffer->record_disabled);
6245 
6246 	/* Make sure all commits have finished */
6247 	synchronize_rcu();
6248 
6249 	reset_disabled_cpu_buffer(cpu_buffer);
6250 
6251 	atomic_dec(&cpu_buffer->record_disabled);
6252 	atomic_dec(&cpu_buffer->resize_disabled);
6253 
6254 	mutex_unlock(&buffer->mutex);
6255 }
6256 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
6257 
6258 /* Flag to ensure proper resetting of atomic variables */
6259 #define RESET_BIT	(1 << 30)
6260 
6261 /**
6262  * ring_buffer_reset_online_cpus - reset a ring buffer per CPU buffer
6263  * @buffer: The ring buffer to reset a per cpu buffer of
6264  */
6265 void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
6266 {
6267 	struct ring_buffer_per_cpu *cpu_buffer;
6268 	int cpu;
6269 
6270 	/* prevent another thread from changing buffer sizes */
6271 	mutex_lock(&buffer->mutex);
6272 
6273 	for_each_online_buffer_cpu(buffer, cpu) {
6274 		cpu_buffer = buffer->buffers[cpu];
6275 
6276 		atomic_add(RESET_BIT, &cpu_buffer->resize_disabled);
6277 		atomic_inc(&cpu_buffer->record_disabled);
6278 	}
6279 
6280 	/* Make sure all commits have finished */
6281 	synchronize_rcu();
6282 
6283 	for_each_buffer_cpu(buffer, cpu) {
6284 		cpu_buffer = buffer->buffers[cpu];
6285 
6286 		/*
6287 		 * If a CPU came online during the synchronize_rcu(), then
6288 		 * ignore it.
6289 		 */
6290 		if (!(atomic_read(&cpu_buffer->resize_disabled) & RESET_BIT))
6291 			continue;
6292 
6293 		reset_disabled_cpu_buffer(cpu_buffer);
6294 
6295 		atomic_dec(&cpu_buffer->record_disabled);
6296 		atomic_sub(RESET_BIT, &cpu_buffer->resize_disabled);
6297 	}
6298 
6299 	mutex_unlock(&buffer->mutex);
6300 }
6301 
6302 /**
6303  * ring_buffer_reset - reset a ring buffer
6304  * @buffer: The ring buffer to reset all cpu buffers
6305  */
6306 void ring_buffer_reset(struct trace_buffer *buffer)
6307 {
6308 	struct ring_buffer_per_cpu *cpu_buffer;
6309 	int cpu;
6310 
6311 	/* prevent another thread from changing buffer sizes */
6312 	mutex_lock(&buffer->mutex);
6313 
6314 	for_each_buffer_cpu(buffer, cpu) {
6315 		cpu_buffer = buffer->buffers[cpu];
6316 
6317 		atomic_inc(&cpu_buffer->resize_disabled);
6318 		atomic_inc(&cpu_buffer->record_disabled);
6319 	}
6320 
6321 	/* Make sure all commits have finished */
6322 	synchronize_rcu();
6323 
6324 	for_each_buffer_cpu(buffer, cpu) {
6325 		cpu_buffer = buffer->buffers[cpu];
6326 
6327 		reset_disabled_cpu_buffer(cpu_buffer);
6328 
6329 		atomic_dec(&cpu_buffer->record_disabled);
6330 		atomic_dec(&cpu_buffer->resize_disabled);
6331 	}
6332 
6333 	mutex_unlock(&buffer->mutex);
6334 }
6335 EXPORT_SYMBOL_GPL(ring_buffer_reset);
6336 
6337 /**
6338  * ring_buffer_empty - is the ring buffer empty?
6339  * @buffer: The ring buffer to test
6340  */
6341 bool ring_buffer_empty(struct trace_buffer *buffer)
6342 {
6343 	struct ring_buffer_per_cpu *cpu_buffer;
6344 	unsigned long flags;
6345 	bool dolock;
6346 	bool ret;
6347 	int cpu;
6348 
6349 	/* yes this is racy, but if you don't like the race, lock the buffer */
6350 	for_each_buffer_cpu(buffer, cpu) {
6351 		cpu_buffer = buffer->buffers[cpu];
6352 		local_irq_save(flags);
6353 		dolock = rb_reader_lock(cpu_buffer);
6354 		ret = rb_per_cpu_empty(cpu_buffer);
6355 		rb_reader_unlock(cpu_buffer, dolock);
6356 		local_irq_restore(flags);
6357 
6358 		if (!ret)
6359 			return false;
6360 	}
6361 
6362 	return true;
6363 }
6364 EXPORT_SYMBOL_GPL(ring_buffer_empty);
6365 
6366 /**
6367  * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
6368  * @buffer: The ring buffer
6369  * @cpu: The CPU buffer to test
6370  */
6371 bool ring_buffer_empty_cpu(struct trace_buffer *buffer, int cpu)
6372 {
6373 	struct ring_buffer_per_cpu *cpu_buffer;
6374 	unsigned long flags;
6375 	bool dolock;
6376 	bool ret;
6377 
6378 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
6379 		return true;
6380 
6381 	cpu_buffer = buffer->buffers[cpu];
6382 	local_irq_save(flags);
6383 	dolock = rb_reader_lock(cpu_buffer);
6384 	ret = rb_per_cpu_empty(cpu_buffer);
6385 	rb_reader_unlock(cpu_buffer, dolock);
6386 	local_irq_restore(flags);
6387 
6388 	return ret;
6389 }
6390 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
6391 
6392 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
6393 /**
6394  * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
6395  * @buffer_a: One buffer to swap with
6396  * @buffer_b: The other buffer to swap with
6397  * @cpu: the CPU of the buffers to swap
6398  *
6399  * This function is useful for tracers that want to take a "snapshot"
6400  * of a CPU buffer and has another back up buffer lying around.
6401  * it is expected that the tracer handles the cpu buffer not being
6402  * used at the moment.
6403  */
6404 int ring_buffer_swap_cpu(struct trace_buffer *buffer_a,
6405 			 struct trace_buffer *buffer_b, int cpu)
6406 {
6407 	struct ring_buffer_per_cpu *cpu_buffer_a;
6408 	struct ring_buffer_per_cpu *cpu_buffer_b;
6409 	int ret = -EINVAL;
6410 
6411 	if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
6412 	    !cpumask_test_cpu(cpu, buffer_b->cpumask))
6413 		return -EINVAL;
6414 
6415 	cpu_buffer_a = buffer_a->buffers[cpu];
6416 	cpu_buffer_b = buffer_b->buffers[cpu];
6417 
6418 	/* It's up to the callers to not try to swap mapped buffers */
6419 	if (WARN_ON_ONCE(cpu_buffer_a->mapped || cpu_buffer_b->mapped))
6420 		return -EBUSY;
6421 
6422 	/* At least make sure the two buffers are somewhat the same */
6423 	if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
6424 		return -EINVAL;
6425 
6426 	if (buffer_a->subbuf_order != buffer_b->subbuf_order)
6427 		return -EINVAL;
6428 
6429 	if (atomic_read(&buffer_a->record_disabled))
6430 		return -EAGAIN;
6431 
6432 	if (atomic_read(&buffer_b->record_disabled))
6433 		return -EAGAIN;
6434 
6435 	if (atomic_read(&cpu_buffer_a->record_disabled))
6436 		return -EAGAIN;
6437 
6438 	if (atomic_read(&cpu_buffer_b->record_disabled))
6439 		return -EAGAIN;
6440 
6441 	/*
6442 	 * We can't do a synchronize_rcu here because this
6443 	 * function can be called in atomic context.
6444 	 * Normally this will be called from the same CPU as cpu.
6445 	 * If not it's up to the caller to protect this.
6446 	 */
6447 	atomic_inc(&cpu_buffer_a->record_disabled);
6448 	atomic_inc(&cpu_buffer_b->record_disabled);
6449 
6450 	ret = -EBUSY;
6451 	if (local_read(&cpu_buffer_a->committing))
6452 		goto out_dec;
6453 	if (local_read(&cpu_buffer_b->committing))
6454 		goto out_dec;
6455 
6456 	/*
6457 	 * When resize is in progress, we cannot swap it because
6458 	 * it will mess the state of the cpu buffer.
6459 	 */
6460 	if (atomic_read(&buffer_a->resizing))
6461 		goto out_dec;
6462 	if (atomic_read(&buffer_b->resizing))
6463 		goto out_dec;
6464 
6465 	buffer_a->buffers[cpu] = cpu_buffer_b;
6466 	buffer_b->buffers[cpu] = cpu_buffer_a;
6467 
6468 	cpu_buffer_b->buffer = buffer_a;
6469 	cpu_buffer_a->buffer = buffer_b;
6470 
6471 	ret = 0;
6472 
6473 out_dec:
6474 	atomic_dec(&cpu_buffer_a->record_disabled);
6475 	atomic_dec(&cpu_buffer_b->record_disabled);
6476 	return ret;
6477 }
6478 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
6479 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
6480 
6481 /**
6482  * ring_buffer_alloc_read_page - allocate a page to read from buffer
6483  * @buffer: the buffer to allocate for.
6484  * @cpu: the cpu buffer to allocate.
6485  *
6486  * This function is used in conjunction with ring_buffer_read_page.
6487  * When reading a full page from the ring buffer, these functions
6488  * can be used to speed up the process. The calling function should
6489  * allocate a few pages first with this function. Then when it
6490  * needs to get pages from the ring buffer, it passes the result
6491  * of this function into ring_buffer_read_page, which will swap
6492  * the page that was allocated, with the read page of the buffer.
6493  *
6494  * Returns:
6495  *  The page allocated, or ERR_PTR
6496  */
6497 struct buffer_data_read_page *
6498 ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
6499 {
6500 	struct ring_buffer_per_cpu *cpu_buffer;
6501 	struct buffer_data_read_page *bpage = NULL;
6502 	unsigned long flags;
6503 
6504 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
6505 		return ERR_PTR(-ENODEV);
6506 
6507 	bpage = kzalloc(sizeof(*bpage), GFP_KERNEL);
6508 	if (!bpage)
6509 		return ERR_PTR(-ENOMEM);
6510 
6511 	bpage->order = buffer->subbuf_order;
6512 	cpu_buffer = buffer->buffers[cpu];
6513 	local_irq_save(flags);
6514 	arch_spin_lock(&cpu_buffer->lock);
6515 
6516 	if (cpu_buffer->free_page) {
6517 		bpage->data = cpu_buffer->free_page;
6518 		cpu_buffer->free_page = NULL;
6519 	}
6520 
6521 	arch_spin_unlock(&cpu_buffer->lock);
6522 	local_irq_restore(flags);
6523 
6524 	if (bpage->data) {
6525 		rb_init_page(bpage->data);
6526 	} else {
6527 		bpage->data = alloc_cpu_data(cpu, cpu_buffer->buffer->subbuf_order);
6528 		if (!bpage->data) {
6529 			kfree(bpage);
6530 			return ERR_PTR(-ENOMEM);
6531 		}
6532 	}
6533 
6534 	return bpage;
6535 }
6536 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
6537 
6538 /**
6539  * ring_buffer_free_read_page - free an allocated read page
6540  * @buffer: the buffer the page was allocate for
6541  * @cpu: the cpu buffer the page came from
6542  * @data_page: the page to free
6543  *
6544  * Free a page allocated from ring_buffer_alloc_read_page.
6545  */
6546 void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
6547 				struct buffer_data_read_page *data_page)
6548 {
6549 	struct ring_buffer_per_cpu *cpu_buffer;
6550 	struct buffer_data_page *bpage = data_page->data;
6551 	struct page *page = virt_to_page(bpage);
6552 	unsigned long flags;
6553 
6554 	if (!buffer || !buffer->buffers || !buffer->buffers[cpu])
6555 		return;
6556 
6557 	cpu_buffer = buffer->buffers[cpu];
6558 
6559 	/*
6560 	 * If the page is still in use someplace else, or order of the page
6561 	 * is different from the subbuffer order of the buffer -
6562 	 * we can't reuse it
6563 	 */
6564 	if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order)
6565 		goto out;
6566 
6567 	local_irq_save(flags);
6568 	arch_spin_lock(&cpu_buffer->lock);
6569 
6570 	if (!cpu_buffer->free_page) {
6571 		cpu_buffer->free_page = bpage;
6572 		bpage = NULL;
6573 	}
6574 
6575 	arch_spin_unlock(&cpu_buffer->lock);
6576 	local_irq_restore(flags);
6577 
6578  out:
6579 	free_pages((unsigned long)bpage, data_page->order);
6580 	kfree(data_page);
6581 }
6582 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
6583 
6584 /**
6585  * ring_buffer_read_page - extract a page from the ring buffer
6586  * @buffer: buffer to extract from
6587  * @data_page: the page to use allocated from ring_buffer_alloc_read_page
6588  * @len: amount to extract
6589  * @cpu: the cpu of the buffer to extract
6590  * @full: should the extraction only happen when the page is full.
6591  *
6592  * This function will pull out a page from the ring buffer and consume it.
6593  * @data_page must be the address of the variable that was returned
6594  * from ring_buffer_alloc_read_page. This is because the page might be used
6595  * to swap with a page in the ring buffer.
6596  *
6597  * for example:
6598  *	rpage = ring_buffer_alloc_read_page(buffer, cpu);
6599  *	if (IS_ERR(rpage))
6600  *		return PTR_ERR(rpage);
6601  *	ret = ring_buffer_read_page(buffer, rpage, len, cpu, 0);
6602  *	if (ret >= 0)
6603  *		process_page(ring_buffer_read_page_data(rpage), ret);
6604  *	ring_buffer_free_read_page(buffer, cpu, rpage);
6605  *
6606  * When @full is set, the function will not return true unless
6607  * the writer is off the reader page.
6608  *
6609  * Note: it is up to the calling functions to handle sleeps and wakeups.
6610  *  The ring buffer can be used anywhere in the kernel and can not
6611  *  blindly call wake_up. The layer that uses the ring buffer must be
6612  *  responsible for that.
6613  *
6614  * Returns:
6615  *  >=0 if data has been transferred, returns the offset of consumed data.
6616  *  <0 if no data has been transferred.
6617  */
6618 int ring_buffer_read_page(struct trace_buffer *buffer,
6619 			  struct buffer_data_read_page *data_page,
6620 			  size_t len, int cpu, int full)
6621 {
6622 	struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
6623 	struct ring_buffer_event *event;
6624 	struct buffer_data_page *bpage;
6625 	struct buffer_page *reader;
6626 	unsigned long missed_events;
6627 	unsigned int commit;
6628 	unsigned int read;
6629 	u64 save_timestamp;
6630 
6631 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
6632 		return -1;
6633 
6634 	/*
6635 	 * If len is not big enough to hold the page header, then
6636 	 * we can not copy anything.
6637 	 */
6638 	if (len <= BUF_PAGE_HDR_SIZE)
6639 		return -1;
6640 
6641 	len -= BUF_PAGE_HDR_SIZE;
6642 
6643 	if (!data_page || !data_page->data)
6644 		return -1;
6645 
6646 	if (data_page->order != buffer->subbuf_order)
6647 		return -1;
6648 
6649 	bpage = data_page->data;
6650 	if (!bpage)
6651 		return -1;
6652 
6653 	guard(raw_spinlock_irqsave)(&cpu_buffer->reader_lock);
6654 
6655 	reader = rb_get_reader_page(cpu_buffer);
6656 	if (!reader)
6657 		return -1;
6658 
6659 	event = rb_reader_event(cpu_buffer);
6660 
6661 	read = reader->read;
6662 	commit = rb_page_size(reader);
6663 
6664 	/* Check if any events were dropped */
6665 	missed_events = cpu_buffer->lost_events;
6666 
6667 	/*
6668 	 * If this page has been partially read or
6669 	 * if len is not big enough to read the rest of the page or
6670 	 * a writer is still on the page, then
6671 	 * we must copy the data from the page to the buffer.
6672 	 * Otherwise, we can simply swap the page with the one passed in.
6673 	 */
6674 	if (read || (len < (commit - read)) ||
6675 	    cpu_buffer->reader_page == cpu_buffer->commit_page ||
6676 	    cpu_buffer->mapped) {
6677 		struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
6678 		unsigned int rpos = read;
6679 		unsigned int pos = 0;
6680 		unsigned int size;
6681 
6682 		/*
6683 		 * If a full page is expected, this can still be returned
6684 		 * if there's been a previous partial read and the
6685 		 * rest of the page can be read and the commit page is off
6686 		 * the reader page.
6687 		 */
6688 		if (full &&
6689 		    (!read || (len < (commit - read)) ||
6690 		     cpu_buffer->reader_page == cpu_buffer->commit_page))
6691 			return -1;
6692 
6693 		if (len > (commit - read))
6694 			len = (commit - read);
6695 
6696 		/* Always keep the time extend and data together */
6697 		size = rb_event_ts_length(event);
6698 
6699 		if (len < size)
6700 			return -1;
6701 
6702 		/* save the current timestamp, since the user will need it */
6703 		save_timestamp = cpu_buffer->read_stamp;
6704 
6705 		/* Need to copy one event at a time */
6706 		do {
6707 			/* We need the size of one event, because
6708 			 * rb_advance_reader only advances by one event,
6709 			 * whereas rb_event_ts_length may include the size of
6710 			 * one or two events.
6711 			 * We have already ensured there's enough space if this
6712 			 * is a time extend. */
6713 			size = rb_event_length(event);
6714 			memcpy(bpage->data + pos, rpage->data + rpos, size);
6715 
6716 			len -= size;
6717 
6718 			rb_advance_reader(cpu_buffer);
6719 			rpos = reader->read;
6720 			pos += size;
6721 
6722 			if (rpos >= commit)
6723 				break;
6724 
6725 			event = rb_reader_event(cpu_buffer);
6726 			/* Always keep the time extend and data together */
6727 			size = rb_event_ts_length(event);
6728 		} while (len >= size);
6729 
6730 		/* update bpage */
6731 		local_set(&bpage->commit, pos);
6732 		bpage->time_stamp = save_timestamp;
6733 
6734 		/* we copied everything to the beginning */
6735 		read = 0;
6736 	} else {
6737 		/* update the entry counter */
6738 		cpu_buffer->read += rb_page_entries(reader);
6739 		cpu_buffer->read_bytes += rb_page_size(reader);
6740 
6741 		/* swap the pages */
6742 		rb_init_page(bpage);
6743 		bpage = reader->page;
6744 		reader->page = data_page->data;
6745 		local_set(&reader->write, 0);
6746 		local_set(&reader->entries, 0);
6747 		reader->read = 0;
6748 		data_page->data = bpage;
6749 
6750 		/*
6751 		 * Use the real_end for the data size,
6752 		 * This gives us a chance to store the lost events
6753 		 * on the page.
6754 		 */
6755 		if (reader->real_end)
6756 			local_set(&bpage->commit, reader->real_end);
6757 	}
6758 
6759 	cpu_buffer->lost_events = 0;
6760 
6761 	commit = local_read(&bpage->commit);
6762 	/*
6763 	 * Set a flag in the commit field if we lost events
6764 	 */
6765 	if (missed_events) {
6766 		/* If there is room at the end of the page to save the
6767 		 * missed events, then record it there.
6768 		 */
6769 		if (buffer->subbuf_size - commit >= sizeof(missed_events)) {
6770 			memcpy(&bpage->data[commit], &missed_events,
6771 			       sizeof(missed_events));
6772 			local_add(RB_MISSED_STORED, &bpage->commit);
6773 			commit += sizeof(missed_events);
6774 		}
6775 		local_add(RB_MISSED_EVENTS, &bpage->commit);
6776 	}
6777 
6778 	/*
6779 	 * This page may be off to user land. Zero it out here.
6780 	 */
6781 	if (commit < buffer->subbuf_size)
6782 		memset(&bpage->data[commit], 0, buffer->subbuf_size - commit);
6783 
6784 	return read;
6785 }
6786 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
6787 
6788 /**
6789  * ring_buffer_read_page_data - get pointer to the data in the page.
6790  * @page:  the page to get the data from
6791  *
6792  * Returns pointer to the actual data in this page.
6793  */
6794 void *ring_buffer_read_page_data(struct buffer_data_read_page *page)
6795 {
6796 	return page->data;
6797 }
6798 EXPORT_SYMBOL_GPL(ring_buffer_read_page_data);
6799 
6800 /**
6801  * ring_buffer_subbuf_size_get - get size of the sub buffer.
6802  * @buffer: the buffer to get the sub buffer size from
6803  *
6804  * Returns size of the sub buffer, in bytes.
6805  */
6806 int ring_buffer_subbuf_size_get(struct trace_buffer *buffer)
6807 {
6808 	return buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
6809 }
6810 EXPORT_SYMBOL_GPL(ring_buffer_subbuf_size_get);
6811 
6812 /**
6813  * ring_buffer_subbuf_order_get - get order of system sub pages in one buffer page.
6814  * @buffer: The ring_buffer to get the system sub page order from
6815  *
6816  * By default, one ring buffer sub page equals to one system page. This parameter
6817  * is configurable, per ring buffer. The size of the ring buffer sub page can be
6818  * extended, but must be an order of system page size.
6819  *
6820  * Returns the order of buffer sub page size, in system pages:
6821  * 0 means the sub buffer size is 1 system page and so forth.
6822  * In case of an error < 0 is returned.
6823  */
6824 int ring_buffer_subbuf_order_get(struct trace_buffer *buffer)
6825 {
6826 	if (!buffer)
6827 		return -EINVAL;
6828 
6829 	return buffer->subbuf_order;
6830 }
6831 EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_get);
6832 
6833 /**
6834  * ring_buffer_subbuf_order_set - set the size of ring buffer sub page.
6835  * @buffer: The ring_buffer to set the new page size.
6836  * @order: Order of the system pages in one sub buffer page
6837  *
6838  * By default, one ring buffer pages equals to one system page. This API can be
6839  * used to set new size of the ring buffer page. The size must be order of
6840  * system page size, that's why the input parameter @order is the order of
6841  * system pages that are allocated for one ring buffer page:
6842  *  0 - 1 system page
6843  *  1 - 2 system pages
6844  *  3 - 4 system pages
6845  *  ...
6846  *
6847  * Returns 0 on success or < 0 in case of an error.
6848  */
6849 int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
6850 {
6851 	struct ring_buffer_per_cpu *cpu_buffer;
6852 	struct buffer_page *bpage, *tmp;
6853 	int old_order, old_size;
6854 	int nr_pages;
6855 	int psize;
6856 	int err;
6857 	int cpu;
6858 
6859 	if (!buffer || order < 0)
6860 		return -EINVAL;
6861 
6862 	if (buffer->subbuf_order == order)
6863 		return 0;
6864 
6865 	psize = (1 << order) * PAGE_SIZE;
6866 	if (psize <= BUF_PAGE_HDR_SIZE)
6867 		return -EINVAL;
6868 
6869 	/* Size of a subbuf cannot be greater than the write counter */
6870 	if (psize > RB_WRITE_MASK + 1)
6871 		return -EINVAL;
6872 
6873 	old_order = buffer->subbuf_order;
6874 	old_size = buffer->subbuf_size;
6875 
6876 	/* prevent another thread from changing buffer sizes */
6877 	guard(mutex)(&buffer->mutex);
6878 	atomic_inc(&buffer->record_disabled);
6879 
6880 	/* Make sure all commits have finished */
6881 	synchronize_rcu();
6882 
6883 	buffer->subbuf_order = order;
6884 	buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;
6885 
6886 	/* Make sure all new buffers are allocated, before deleting the old ones */
6887 	for_each_buffer_cpu(buffer, cpu) {
6888 
6889 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
6890 			continue;
6891 
6892 		cpu_buffer = buffer->buffers[cpu];
6893 
6894 		if (cpu_buffer->mapped) {
6895 			err = -EBUSY;
6896 			goto error;
6897 		}
6898 
6899 		/* Update the number of pages to match the new size */
6900 		nr_pages = old_size * buffer->buffers[cpu]->nr_pages;
6901 		nr_pages = DIV_ROUND_UP(nr_pages, buffer->subbuf_size);
6902 
6903 		/* we need a minimum of two pages */
6904 		if (nr_pages < 2)
6905 			nr_pages = 2;
6906 
6907 		cpu_buffer->nr_pages_to_update = nr_pages;
6908 
6909 		/* Include the reader page */
6910 		nr_pages++;
6911 
6912 		/* Allocate the new size buffer */
6913 		INIT_LIST_HEAD(&cpu_buffer->new_pages);
6914 		if (__rb_allocate_pages(cpu_buffer, nr_pages,
6915 					&cpu_buffer->new_pages)) {
6916 			/* not enough memory for new pages */
6917 			err = -ENOMEM;
6918 			goto error;
6919 		}
6920 	}
6921 
6922 	for_each_buffer_cpu(buffer, cpu) {
6923 		struct buffer_data_page *old_free_data_page;
6924 		struct list_head old_pages;
6925 		unsigned long flags;
6926 
6927 		if (!cpumask_test_cpu(cpu, buffer->cpumask))
6928 			continue;
6929 
6930 		cpu_buffer = buffer->buffers[cpu];
6931 
6932 		raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
6933 
6934 		/* Clear the head bit to make the link list normal to read */
6935 		rb_head_page_deactivate(cpu_buffer);
6936 
6937 		/*
6938 		 * Collect buffers from the cpu_buffer pages list and the
6939 		 * reader_page on old_pages, so they can be freed later when not
6940 		 * under a spinlock. The pages list is a linked list with no
6941 		 * head, adding old_pages turns it into a regular list with
6942 		 * old_pages being the head.
6943 		 */
6944 		list_add(&old_pages, cpu_buffer->pages);
6945 		list_add(&cpu_buffer->reader_page->list, &old_pages);
6946 
6947 		/* One page was allocated for the reader page */
6948 		cpu_buffer->reader_page = list_entry(cpu_buffer->new_pages.next,
6949 						     struct buffer_page, list);
6950 		list_del_init(&cpu_buffer->reader_page->list);
6951 
6952 		/* Install the new pages, remove the head from the list */
6953 		cpu_buffer->pages = cpu_buffer->new_pages.next;
6954 		list_del_init(&cpu_buffer->new_pages);
6955 		cpu_buffer->cnt++;
6956 
6957 		cpu_buffer->head_page
6958 			= list_entry(cpu_buffer->pages, struct buffer_page, list);
6959 		cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
6960 
6961 		cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
6962 		cpu_buffer->nr_pages_to_update = 0;
6963 
6964 		old_free_data_page = cpu_buffer->free_page;
6965 		cpu_buffer->free_page = NULL;
6966 
6967 		rb_head_page_activate(cpu_buffer);
6968 
6969 		raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
6970 
6971 		/* Free old sub buffers */
6972 		list_for_each_entry_safe(bpage, tmp, &old_pages, list) {
6973 			list_del_init(&bpage->list);
6974 			free_buffer_page(bpage);
6975 		}
6976 		free_pages((unsigned long)old_free_data_page, old_order);
6977 
6978 		rb_check_pages(cpu_buffer);
6979 	}
6980 
6981 	atomic_dec(&buffer->record_disabled);
6982 
6983 	return 0;
6984 
6985 error:
6986 	buffer->subbuf_order = old_order;
6987 	buffer->subbuf_size = old_size;
6988 
6989 	atomic_dec(&buffer->record_disabled);
6990 
6991 	for_each_buffer_cpu(buffer, cpu) {
6992 		cpu_buffer = buffer->buffers[cpu];
6993 
6994 		if (!cpu_buffer->nr_pages_to_update)
6995 			continue;
6996 
6997 		list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, list) {
6998 			list_del_init(&bpage->list);
6999 			free_buffer_page(bpage);
7000 		}
7001 	}
7002 
7003 	return err;
7004 }
7005 EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_set);
7006 
7007 static int rb_alloc_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
7008 {
7009 	struct page *page;
7010 
7011 	if (cpu_buffer->meta_page)
7012 		return 0;
7013 
7014 	page = alloc_page(GFP_USER | __GFP_ZERO);
7015 	if (!page)
7016 		return -ENOMEM;
7017 
7018 	cpu_buffer->meta_page = page_to_virt(page);
7019 
7020 	return 0;
7021 }
7022 
7023 static void rb_free_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
7024 {
7025 	unsigned long addr = (unsigned long)cpu_buffer->meta_page;
7026 
7027 	free_page(addr);
7028 	cpu_buffer->meta_page = NULL;
7029 }
7030 
7031 static void rb_setup_ids_meta_page(struct ring_buffer_per_cpu *cpu_buffer,
7032 				   unsigned long *subbuf_ids)
7033 {
7034 	struct trace_buffer_meta *meta = cpu_buffer->meta_page;
7035 	unsigned int nr_subbufs = cpu_buffer->nr_pages + 1;
7036 	struct buffer_page *first_subbuf, *subbuf;
7037 	int cnt = 0;
7038 	int id = 0;
7039 
7040 	id = rb_page_id(cpu_buffer, cpu_buffer->reader_page, id);
7041 	subbuf_ids[id++] = (unsigned long)cpu_buffer->reader_page->page;
7042 	cnt++;
7043 
7044 	first_subbuf = subbuf = rb_set_head_page(cpu_buffer);
7045 	do {
7046 		id = rb_page_id(cpu_buffer, subbuf, id);
7047 
7048 		if (WARN_ON(id >= nr_subbufs))
7049 			break;
7050 
7051 		subbuf_ids[id] = (unsigned long)subbuf->page;
7052 
7053 		rb_inc_page(&subbuf);
7054 		id++;
7055 		cnt++;
7056 	} while (subbuf != first_subbuf);
7057 
7058 	WARN_ON(cnt != nr_subbufs);
7059 
7060 	/* install subbuf ID to kern VA translation */
7061 	cpu_buffer->subbuf_ids = subbuf_ids;
7062 
7063 	meta->meta_struct_len = sizeof(*meta);
7064 	meta->nr_subbufs = nr_subbufs;
7065 	meta->subbuf_size = cpu_buffer->buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
7066 	meta->meta_page_size = meta->subbuf_size;
7067 
7068 	rb_update_meta_page(cpu_buffer);
7069 }
7070 
7071 static struct ring_buffer_per_cpu *
7072 rb_get_mapped_buffer(struct trace_buffer *buffer, int cpu)
7073 {
7074 	struct ring_buffer_per_cpu *cpu_buffer;
7075 
7076 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
7077 		return ERR_PTR(-EINVAL);
7078 
7079 	cpu_buffer = buffer->buffers[cpu];
7080 
7081 	mutex_lock(&cpu_buffer->mapping_lock);
7082 
7083 	if (!cpu_buffer->user_mapped) {
7084 		mutex_unlock(&cpu_buffer->mapping_lock);
7085 		return ERR_PTR(-ENODEV);
7086 	}
7087 
7088 	return cpu_buffer;
7089 }
7090 
7091 static void rb_put_mapped_buffer(struct ring_buffer_per_cpu *cpu_buffer)
7092 {
7093 	mutex_unlock(&cpu_buffer->mapping_lock);
7094 }
7095 
7096 /*
7097  * Fast-path for rb_buffer_(un)map(). Called whenever the meta-page doesn't need
7098  * to be set-up or torn-down.
7099  */
7100 static int __rb_inc_dec_mapped(struct ring_buffer_per_cpu *cpu_buffer,
7101 			       bool inc)
7102 {
7103 	unsigned long flags;
7104 
7105 	lockdep_assert_held(&cpu_buffer->mapping_lock);
7106 
7107 	/* mapped is always greater or equal to user_mapped */
7108 	if (WARN_ON(cpu_buffer->mapped < cpu_buffer->user_mapped))
7109 		return -EINVAL;
7110 
7111 	if (inc && cpu_buffer->mapped == UINT_MAX)
7112 		return -EBUSY;
7113 
7114 	if (WARN_ON(!inc && cpu_buffer->user_mapped == 0))
7115 		return -EINVAL;
7116 
7117 	mutex_lock(&cpu_buffer->buffer->mutex);
7118 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7119 
7120 	if (inc) {
7121 		cpu_buffer->user_mapped++;
7122 		cpu_buffer->mapped++;
7123 	} else {
7124 		cpu_buffer->user_mapped--;
7125 		cpu_buffer->mapped--;
7126 	}
7127 
7128 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7129 	mutex_unlock(&cpu_buffer->buffer->mutex);
7130 
7131 	return 0;
7132 }
7133 
7134 /*
7135  *   +--------------+  pgoff == 0
7136  *   |   meta page  |
7137  *   +--------------+  pgoff == 1
7138  *   | subbuffer 0  |
7139  *   |              |
7140  *   +--------------+  pgoff == (1 + (1 << subbuf_order))
7141  *   | subbuffer 1  |
7142  *   |              |
7143  *         ...
7144  */
7145 #ifdef CONFIG_MMU
7146 static int __rb_map_vma(struct ring_buffer_per_cpu *cpu_buffer,
7147 			struct vm_area_struct *vma)
7148 {
7149 	unsigned long nr_subbufs, nr_pages, nr_vma_pages, pgoff = vma->vm_pgoff;
7150 	unsigned int subbuf_pages, subbuf_order;
7151 	struct page **pages __free(kfree) = NULL;
7152 	int p = 0, s = 0;
7153 	int err;
7154 
7155 	/* Refuse MP_PRIVATE or writable mappings */
7156 	if (vma->vm_flags & VM_WRITE || vma->vm_flags & VM_EXEC ||
7157 	    !(vma->vm_flags & VM_MAYSHARE))
7158 		return -EPERM;
7159 
7160 	subbuf_order = cpu_buffer->buffer->subbuf_order;
7161 	subbuf_pages = 1 << subbuf_order;
7162 
7163 	if (subbuf_order && pgoff % subbuf_pages)
7164 		return -EINVAL;
7165 
7166 	/*
7167 	 * Make sure the mapping cannot become writable later. Also tell the VM
7168 	 * to not touch these pages (VM_DONTCOPY | VM_DONTEXPAND).
7169 	 */
7170 	vm_flags_mod(vma, VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP,
7171 		     VM_MAYWRITE);
7172 
7173 	lockdep_assert_held(&cpu_buffer->mapping_lock);
7174 
7175 	nr_subbufs = cpu_buffer->nr_pages + 1; /* + reader-subbuf */
7176 	nr_pages = ((nr_subbufs + 1) << subbuf_order); /* + meta-page */
7177 	if (nr_pages <= pgoff)
7178 		return -EINVAL;
7179 
7180 	nr_pages -= pgoff;
7181 
7182 	nr_vma_pages = vma_pages(vma);
7183 	if (!nr_vma_pages || nr_vma_pages > nr_pages)
7184 		return -EINVAL;
7185 
7186 	nr_pages = nr_vma_pages;
7187 
7188 	pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
7189 	if (!pages)
7190 		return -ENOMEM;
7191 
7192 	if (!pgoff) {
7193 		unsigned long meta_page_padding;
7194 
7195 		pages[p++] = virt_to_page(cpu_buffer->meta_page);
7196 
7197 		/*
7198 		 * Pad with the zero-page to align the meta-page with the
7199 		 * sub-buffers.
7200 		 */
7201 		meta_page_padding = subbuf_pages - 1;
7202 		while (meta_page_padding-- && p < nr_pages) {
7203 			unsigned long __maybe_unused zero_addr =
7204 				vma->vm_start + (PAGE_SIZE * p);
7205 
7206 			pages[p++] = ZERO_PAGE(zero_addr);
7207 		}
7208 	} else {
7209 		/* Skip the meta-page */
7210 		pgoff -= subbuf_pages;
7211 
7212 		s += pgoff / subbuf_pages;
7213 	}
7214 
7215 	while (p < nr_pages) {
7216 		struct page *page;
7217 		int off = 0;
7218 
7219 		if (WARN_ON_ONCE(s >= nr_subbufs))
7220 			return -EINVAL;
7221 
7222 		page = virt_to_page((void *)cpu_buffer->subbuf_ids[s]);
7223 
7224 		for (; off < (1 << (subbuf_order)); off++, page++) {
7225 			if (p >= nr_pages)
7226 				break;
7227 
7228 			pages[p++] = page;
7229 		}
7230 		s++;
7231 	}
7232 
7233 	err = vm_insert_pages(vma, vma->vm_start, pages, &nr_pages);
7234 
7235 	return err;
7236 }
7237 #else
7238 static int __rb_map_vma(struct ring_buffer_per_cpu *cpu_buffer,
7239 			struct vm_area_struct *vma)
7240 {
7241 	return -EOPNOTSUPP;
7242 }
7243 #endif
7244 
7245 int ring_buffer_map(struct trace_buffer *buffer, int cpu,
7246 		    struct vm_area_struct *vma)
7247 {
7248 	struct ring_buffer_per_cpu *cpu_buffer;
7249 	unsigned long flags, *subbuf_ids;
7250 	int err;
7251 
7252 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
7253 		return -EINVAL;
7254 
7255 	cpu_buffer = buffer->buffers[cpu];
7256 
7257 	guard(mutex)(&cpu_buffer->mapping_lock);
7258 
7259 	if (cpu_buffer->user_mapped) {
7260 		err = __rb_map_vma(cpu_buffer, vma);
7261 		if (!err)
7262 			err = __rb_inc_dec_mapped(cpu_buffer, true);
7263 		return err;
7264 	}
7265 
7266 	/* prevent another thread from changing buffer/sub-buffer sizes */
7267 	guard(mutex)(&buffer->mutex);
7268 
7269 	err = rb_alloc_meta_page(cpu_buffer);
7270 	if (err)
7271 		return err;
7272 
7273 	/* subbuf_ids include the reader while nr_pages does not */
7274 	subbuf_ids = kcalloc(cpu_buffer->nr_pages + 1, sizeof(*subbuf_ids), GFP_KERNEL);
7275 	if (!subbuf_ids) {
7276 		rb_free_meta_page(cpu_buffer);
7277 		return -ENOMEM;
7278 	}
7279 
7280 	atomic_inc(&cpu_buffer->resize_disabled);
7281 
7282 	/*
7283 	 * Lock all readers to block any subbuf swap until the subbuf IDs are
7284 	 * assigned.
7285 	 */
7286 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7287 	rb_setup_ids_meta_page(cpu_buffer, subbuf_ids);
7288 
7289 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7290 
7291 	err = __rb_map_vma(cpu_buffer, vma);
7292 	if (!err) {
7293 		raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7294 		/* This is the first time it is mapped by user */
7295 		cpu_buffer->mapped++;
7296 		cpu_buffer->user_mapped = 1;
7297 		raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7298 	} else {
7299 		kfree(cpu_buffer->subbuf_ids);
7300 		cpu_buffer->subbuf_ids = NULL;
7301 		rb_free_meta_page(cpu_buffer);
7302 		atomic_dec(&cpu_buffer->resize_disabled);
7303 	}
7304 
7305 	return err;
7306 }
7307 
7308 int ring_buffer_unmap(struct trace_buffer *buffer, int cpu)
7309 {
7310 	struct ring_buffer_per_cpu *cpu_buffer;
7311 	unsigned long flags;
7312 
7313 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
7314 		return -EINVAL;
7315 
7316 	cpu_buffer = buffer->buffers[cpu];
7317 
7318 	guard(mutex)(&cpu_buffer->mapping_lock);
7319 
7320 	if (!cpu_buffer->user_mapped) {
7321 		return -ENODEV;
7322 	} else if (cpu_buffer->user_mapped > 1) {
7323 		__rb_inc_dec_mapped(cpu_buffer, false);
7324 		return 0;
7325 	}
7326 
7327 	guard(mutex)(&buffer->mutex);
7328 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7329 
7330 	/* This is the last user space mapping */
7331 	if (!WARN_ON_ONCE(cpu_buffer->mapped < cpu_buffer->user_mapped))
7332 		cpu_buffer->mapped--;
7333 	cpu_buffer->user_mapped = 0;
7334 
7335 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7336 
7337 	kfree(cpu_buffer->subbuf_ids);
7338 	cpu_buffer->subbuf_ids = NULL;
7339 	rb_free_meta_page(cpu_buffer);
7340 	atomic_dec(&cpu_buffer->resize_disabled);
7341 
7342 	return 0;
7343 }
7344 
7345 int ring_buffer_map_get_reader(struct trace_buffer *buffer, int cpu)
7346 {
7347 	struct ring_buffer_per_cpu *cpu_buffer;
7348 	struct buffer_page *reader;
7349 	unsigned long missed_events;
7350 	unsigned long reader_size;
7351 	unsigned long flags;
7352 
7353 	cpu_buffer = rb_get_mapped_buffer(buffer, cpu);
7354 	if (IS_ERR(cpu_buffer))
7355 		return (int)PTR_ERR(cpu_buffer);
7356 
7357 	raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7358 
7359 consume:
7360 	if (rb_per_cpu_empty(cpu_buffer))
7361 		goto out;
7362 
7363 	reader_size = rb_page_size(cpu_buffer->reader_page);
7364 
7365 	/*
7366 	 * There are data to be read on the current reader page, we can
7367 	 * return to the caller. But before that, we assume the latter will read
7368 	 * everything. Let's update the kernel reader accordingly.
7369 	 */
7370 	if (cpu_buffer->reader_page->read < reader_size) {
7371 		while (cpu_buffer->reader_page->read < reader_size)
7372 			rb_advance_reader(cpu_buffer);
7373 		goto out;
7374 	}
7375 
7376 	/* Did the reader catch up with the writer? */
7377 	if (cpu_buffer->reader_page == cpu_buffer->commit_page)
7378 		goto out;
7379 
7380 	reader = rb_get_reader_page(cpu_buffer);
7381 	if (WARN_ON(!reader))
7382 		goto out;
7383 
7384 	/* Check if any events were dropped */
7385 	missed_events = cpu_buffer->lost_events;
7386 
7387 	if (missed_events) {
7388 		if (cpu_buffer->reader_page != cpu_buffer->commit_page) {
7389 			struct buffer_data_page *bpage = reader->page;
7390 			unsigned int commit;
7391 			/*
7392 			 * Use the real_end for the data size,
7393 			 * This gives us a chance to store the lost events
7394 			 * on the page.
7395 			 */
7396 			if (reader->real_end)
7397 				local_set(&bpage->commit, reader->real_end);
7398 			/*
7399 			 * If there is room at the end of the page to save the
7400 			 * missed events, then record it there.
7401 			 */
7402 			commit = rb_page_size(reader);
7403 			if (buffer->subbuf_size - commit >= sizeof(missed_events)) {
7404 				memcpy(&bpage->data[commit], &missed_events,
7405 				       sizeof(missed_events));
7406 				local_add(RB_MISSED_STORED, &bpage->commit);
7407 			}
7408 			local_add(RB_MISSED_EVENTS, &bpage->commit);
7409 		} else if (!WARN_ONCE(cpu_buffer->reader_page == cpu_buffer->tail_page,
7410 				      "Reader on commit with %ld missed events",
7411 				      missed_events)) {
7412 			/*
7413 			 * There shouldn't be any missed events if the tail_page
7414 			 * is on the reader page. But if the tail page is not on the
7415 			 * reader page and the commit_page is, that would mean that
7416 			 * there's a commit_overrun (an interrupt preempted an
7417 			 * addition of an event and then filled the buffer
7418 			 * with new events). In this case it's not an
7419 			 * error, but it should still be reported.
7420 			 *
7421 			 * TODO: Add missed events to the page for user space to know.
7422 			 */
7423 			pr_info("Ring buffer [%d] commit overrun lost %ld events at timestamp:%lld\n",
7424 				cpu, missed_events, cpu_buffer->reader_page->page->time_stamp);
7425 		}
7426 	}
7427 
7428 	cpu_buffer->lost_events = 0;
7429 
7430 	goto consume;
7431 
7432 out:
7433 	/* Some archs do not have data cache coherency between kernel and user-space */
7434 	flush_kernel_vmap_range(cpu_buffer->reader_page->page,
7435 				buffer->subbuf_size + BUF_PAGE_HDR_SIZE);
7436 
7437 	rb_update_meta_page(cpu_buffer);
7438 
7439 	raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7440 	rb_put_mapped_buffer(cpu_buffer);
7441 
7442 	return 0;
7443 }
7444 
7445 /*
7446  * We only allocate new buffers, never free them if the CPU goes down.
7447  * If we were to free the buffer, then the user would lose any trace that was in
7448  * the buffer.
7449  */
7450 int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
7451 {
7452 	struct trace_buffer *buffer;
7453 	long nr_pages_same;
7454 	int cpu_i;
7455 	unsigned long nr_pages;
7456 
7457 	buffer = container_of(node, struct trace_buffer, node);
7458 	if (cpumask_test_cpu(cpu, buffer->cpumask))
7459 		return 0;
7460 
7461 	nr_pages = 0;
7462 	nr_pages_same = 1;
7463 	/* check if all cpu sizes are same */
7464 	for_each_buffer_cpu(buffer, cpu_i) {
7465 		/* fill in the size from first enabled cpu */
7466 		if (nr_pages == 0)
7467 			nr_pages = buffer->buffers[cpu_i]->nr_pages;
7468 		if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
7469 			nr_pages_same = 0;
7470 			break;
7471 		}
7472 	}
7473 	/* allocate minimum pages, user can later expand it */
7474 	if (!nr_pages_same)
7475 		nr_pages = 2;
7476 	buffer->buffers[cpu] =
7477 		rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
7478 	if (!buffer->buffers[cpu]) {
7479 		WARN(1, "failed to allocate ring buffer on CPU %u\n",
7480 		     cpu);
7481 		return -ENOMEM;
7482 	}
7483 	smp_wmb();
7484 	cpumask_set_cpu(cpu, buffer->cpumask);
7485 	return 0;
7486 }
7487 
7488 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
7489 /*
7490  * This is a basic integrity check of the ring buffer.
7491  * Late in the boot cycle this test will run when configured in.
7492  * It will kick off a thread per CPU that will go into a loop
7493  * writing to the per cpu ring buffer various sizes of data.
7494  * Some of the data will be large items, some small.
7495  *
7496  * Another thread is created that goes into a spin, sending out
7497  * IPIs to the other CPUs to also write into the ring buffer.
7498  * this is to test the nesting ability of the buffer.
7499  *
7500  * Basic stats are recorded and reported. If something in the
7501  * ring buffer should happen that's not expected, a big warning
7502  * is displayed and all ring buffers are disabled.
7503  */
7504 static struct task_struct *rb_threads[NR_CPUS] __initdata;
7505 
7506 struct rb_test_data {
7507 	struct trace_buffer *buffer;
7508 	unsigned long		events;
7509 	unsigned long		bytes_written;
7510 	unsigned long		bytes_alloc;
7511 	unsigned long		bytes_dropped;
7512 	unsigned long		events_nested;
7513 	unsigned long		bytes_written_nested;
7514 	unsigned long		bytes_alloc_nested;
7515 	unsigned long		bytes_dropped_nested;
7516 	int			min_size_nested;
7517 	int			max_size_nested;
7518 	int			max_size;
7519 	int			min_size;
7520 	int			cpu;
7521 	int			cnt;
7522 };
7523 
7524 static struct rb_test_data rb_data[NR_CPUS] __initdata;
7525 
7526 /* 1 meg per cpu */
7527 #define RB_TEST_BUFFER_SIZE	1048576
7528 
7529 static char rb_string[] __initdata =
7530 	"abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
7531 	"?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
7532 	"!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
7533 
7534 static bool rb_test_started __initdata;
7535 
7536 struct rb_item {
7537 	int size;
7538 	char str[];
7539 };
7540 
7541 static __init int rb_write_something(struct rb_test_data *data, bool nested)
7542 {
7543 	struct ring_buffer_event *event;
7544 	struct rb_item *item;
7545 	bool started;
7546 	int event_len;
7547 	int size;
7548 	int len;
7549 	int cnt;
7550 
7551 	/* Have nested writes different that what is written */
7552 	cnt = data->cnt + (nested ? 27 : 0);
7553 
7554 	/* Multiply cnt by ~e, to make some unique increment */
7555 	size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
7556 
7557 	len = size + sizeof(struct rb_item);
7558 
7559 	started = rb_test_started;
7560 	/* read rb_test_started before checking buffer enabled */
7561 	smp_rmb();
7562 
7563 	event = ring_buffer_lock_reserve(data->buffer, len);
7564 	if (!event) {
7565 		/* Ignore dropped events before test starts. */
7566 		if (started) {
7567 			if (nested)
7568 				data->bytes_dropped_nested += len;
7569 			else
7570 				data->bytes_dropped += len;
7571 		}
7572 		return len;
7573 	}
7574 
7575 	event_len = ring_buffer_event_length(event);
7576 
7577 	if (RB_WARN_ON(data->buffer, event_len < len))
7578 		goto out;
7579 
7580 	item = ring_buffer_event_data(event);
7581 	item->size = size;
7582 	memcpy(item->str, rb_string, size);
7583 
7584 	if (nested) {
7585 		data->bytes_alloc_nested += event_len;
7586 		data->bytes_written_nested += len;
7587 		data->events_nested++;
7588 		if (!data->min_size_nested || len < data->min_size_nested)
7589 			data->min_size_nested = len;
7590 		if (len > data->max_size_nested)
7591 			data->max_size_nested = len;
7592 	} else {
7593 		data->bytes_alloc += event_len;
7594 		data->bytes_written += len;
7595 		data->events++;
7596 		if (!data->min_size || len < data->min_size)
7597 			data->max_size = len;
7598 		if (len > data->max_size)
7599 			data->max_size = len;
7600 	}
7601 
7602  out:
7603 	ring_buffer_unlock_commit(data->buffer);
7604 
7605 	return 0;
7606 }
7607 
7608 static __init int rb_test(void *arg)
7609 {
7610 	struct rb_test_data *data = arg;
7611 
7612 	while (!kthread_should_stop()) {
7613 		rb_write_something(data, false);
7614 		data->cnt++;
7615 
7616 		set_current_state(TASK_INTERRUPTIBLE);
7617 		/* Now sleep between a min of 100-300us and a max of 1ms */
7618 		usleep_range(((data->cnt % 3) + 1) * 100, 1000);
7619 	}
7620 
7621 	return 0;
7622 }
7623 
7624 static __init void rb_ipi(void *ignore)
7625 {
7626 	struct rb_test_data *data;
7627 	int cpu = smp_processor_id();
7628 
7629 	data = &rb_data[cpu];
7630 	rb_write_something(data, true);
7631 }
7632 
7633 static __init int rb_hammer_test(void *arg)
7634 {
7635 	while (!kthread_should_stop()) {
7636 
7637 		/* Send an IPI to all cpus to write data! */
7638 		smp_call_function(rb_ipi, NULL, 1);
7639 		/* No sleep, but for non preempt, let others run */
7640 		schedule();
7641 	}
7642 
7643 	return 0;
7644 }
7645 
7646 static __init int test_ringbuffer(void)
7647 {
7648 	struct task_struct *rb_hammer;
7649 	struct trace_buffer *buffer;
7650 	int cpu;
7651 	int ret = 0;
7652 
7653 	if (security_locked_down(LOCKDOWN_TRACEFS)) {
7654 		pr_warn("Lockdown is enabled, skipping ring buffer tests\n");
7655 		return 0;
7656 	}
7657 
7658 	pr_info("Running ring buffer tests...\n");
7659 
7660 	buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
7661 	if (WARN_ON(!buffer))
7662 		return 0;
7663 
7664 	/* Disable buffer so that threads can't write to it yet */
7665 	ring_buffer_record_off(buffer);
7666 
7667 	for_each_online_cpu(cpu) {
7668 		rb_data[cpu].buffer = buffer;
7669 		rb_data[cpu].cpu = cpu;
7670 		rb_data[cpu].cnt = cpu;
7671 		rb_threads[cpu] = kthread_run_on_cpu(rb_test, &rb_data[cpu],
7672 						     cpu, "rbtester/%u");
7673 		if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
7674 			pr_cont("FAILED\n");
7675 			ret = PTR_ERR(rb_threads[cpu]);
7676 			goto out_free;
7677 		}
7678 	}
7679 
7680 	/* Now create the rb hammer! */
7681 	rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
7682 	if (WARN_ON(IS_ERR(rb_hammer))) {
7683 		pr_cont("FAILED\n");
7684 		ret = PTR_ERR(rb_hammer);
7685 		goto out_free;
7686 	}
7687 
7688 	ring_buffer_record_on(buffer);
7689 	/*
7690 	 * Show buffer is enabled before setting rb_test_started.
7691 	 * Yes there's a small race window where events could be
7692 	 * dropped and the thread won't catch it. But when a ring
7693 	 * buffer gets enabled, there will always be some kind of
7694 	 * delay before other CPUs see it. Thus, we don't care about
7695 	 * those dropped events. We care about events dropped after
7696 	 * the threads see that the buffer is active.
7697 	 */
7698 	smp_wmb();
7699 	rb_test_started = true;
7700 
7701 	set_current_state(TASK_INTERRUPTIBLE);
7702 	/* Just run for 10 seconds */
7703 	schedule_timeout(10 * HZ);
7704 
7705 	kthread_stop(rb_hammer);
7706 
7707  out_free:
7708 	for_each_online_cpu(cpu) {
7709 		if (!rb_threads[cpu])
7710 			break;
7711 		kthread_stop(rb_threads[cpu]);
7712 	}
7713 	if (ret) {
7714 		ring_buffer_free(buffer);
7715 		return ret;
7716 	}
7717 
7718 	/* Report! */
7719 	pr_info("finished\n");
7720 	for_each_online_cpu(cpu) {
7721 		struct ring_buffer_event *event;
7722 		struct rb_test_data *data = &rb_data[cpu];
7723 		struct rb_item *item;
7724 		unsigned long total_events;
7725 		unsigned long total_dropped;
7726 		unsigned long total_written;
7727 		unsigned long total_alloc;
7728 		unsigned long total_read = 0;
7729 		unsigned long total_size = 0;
7730 		unsigned long total_len = 0;
7731 		unsigned long total_lost = 0;
7732 		unsigned long lost;
7733 		int big_event_size;
7734 		int small_event_size;
7735 
7736 		ret = -1;
7737 
7738 		total_events = data->events + data->events_nested;
7739 		total_written = data->bytes_written + data->bytes_written_nested;
7740 		total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
7741 		total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
7742 
7743 		big_event_size = data->max_size + data->max_size_nested;
7744 		small_event_size = data->min_size + data->min_size_nested;
7745 
7746 		pr_info("CPU %d:\n", cpu);
7747 		pr_info("              events:    %ld\n", total_events);
7748 		pr_info("       dropped bytes:    %ld\n", total_dropped);
7749 		pr_info("       alloced bytes:    %ld\n", total_alloc);
7750 		pr_info("       written bytes:    %ld\n", total_written);
7751 		pr_info("       biggest event:    %d\n", big_event_size);
7752 		pr_info("      smallest event:    %d\n", small_event_size);
7753 
7754 		if (RB_WARN_ON(buffer, total_dropped))
7755 			break;
7756 
7757 		ret = 0;
7758 
7759 		while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
7760 			total_lost += lost;
7761 			item = ring_buffer_event_data(event);
7762 			total_len += ring_buffer_event_length(event);
7763 			total_size += item->size + sizeof(struct rb_item);
7764 			if (memcmp(&item->str[0], rb_string, item->size) != 0) {
7765 				pr_info("FAILED!\n");
7766 				pr_info("buffer had: %.*s\n", item->size, item->str);
7767 				pr_info("expected:   %.*s\n", item->size, rb_string);
7768 				RB_WARN_ON(buffer, 1);
7769 				ret = -1;
7770 				break;
7771 			}
7772 			total_read++;
7773 		}
7774 		if (ret)
7775 			break;
7776 
7777 		ret = -1;
7778 
7779 		pr_info("         read events:   %ld\n", total_read);
7780 		pr_info("         lost events:   %ld\n", total_lost);
7781 		pr_info("        total events:   %ld\n", total_lost + total_read);
7782 		pr_info("  recorded len bytes:   %ld\n", total_len);
7783 		pr_info(" recorded size bytes:   %ld\n", total_size);
7784 		if (total_lost) {
7785 			pr_info(" With dropped events, record len and size may not match\n"
7786 				" alloced and written from above\n");
7787 		} else {
7788 			if (RB_WARN_ON(buffer, total_len != total_alloc ||
7789 				       total_size != total_written))
7790 				break;
7791 		}
7792 		if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
7793 			break;
7794 
7795 		ret = 0;
7796 	}
7797 	if (!ret)
7798 		pr_info("Ring buffer PASSED!\n");
7799 
7800 	ring_buffer_free(buffer);
7801 	return 0;
7802 }
7803 
7804 late_initcall(test_ringbuffer);
7805 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */
7806