xref: /linux/drivers/md/dm-vdo/funnel-queue.h (revision b4db9f840283caca0d904436f187ef56a9126eaa)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright 2023 Red Hat
4  */
5 
6 #ifndef VDO_FUNNEL_QUEUE_H
7 #define VDO_FUNNEL_QUEUE_H
8 
9 #include <linux/atomic.h>
10 #include <linux/cache.h>
11 
12 /*
13  * A funnel queue is a simple (almost) lock-free queue that accepts entries from multiple threads
14  * (multi-producer) and delivers them to a single thread (single-consumer). "Funnel" is an attempt
15  * to evoke the image of requests from more than one producer being "funneled down" to a single
16  * consumer.
17  *
18  * This is an unsynchronized but thread-safe data structure when used as intended. There is no
19  * mechanism to ensure that only one thread is consuming from the queue. If more than one thread
20  * attempts to consume from the queue, the resulting behavior is undefined. Clients must not
21  * directly access or manipulate the internals of the queue, which are only exposed for the purpose
22  * of allowing the very simple enqueue operation to be inlined.
23  *
24  * The implementation requires that a funnel_queue_entry structure (a link pointer) is embedded in
25  * the queue entries, and pointers to those structures are used exclusively by the queue. No macros
26  * are defined to template the queue, so the offset of the funnel_queue_entry in the records placed
27  * in the queue must all be the same so the client can derive their structure pointer from the
28  * entry pointer returned by vdo_funnel_queue_poll().
29  *
30  * Callers are wholly responsible for allocating and freeing the entries. Entries may be freed as
31  * soon as they are returned since this queue is not susceptible to the "ABA problem" present in
32  * many lock-free data structures. The queue is dynamically allocated to ensure cache-line
33  * alignment, but no other dynamic allocation is used.
34  *
35  * The algorithm is not actually 100% lock-free. There is a single point in vdo_funnel_queue_put()
36  * at which a preempted producer will prevent the consumers from seeing items added to the queue by
37  * later producers, and only if the queue is short enough or the consumer fast enough for it to
38  * reach what was the end of the queue at the time of the preemption.
39  *
40  * The consumer function, vdo_funnel_queue_poll(), will return NULL when the queue is empty. To
41  * wait for data to consume, spin (if safe) or combine the queue with a struct event_count to
42  * signal the presence of new entries.
43  */
44 
45 /* This queue link structure must be embedded in client entries. */
46 struct funnel_queue_entry {
47 	/* The next (newer) entry in the queue. */
48 	struct funnel_queue_entry *next;
49 };
50 
51 /*
52  * The dynamically allocated queue structure, which is allocated on a cache line boundary so the
53  * producer and consumer fields in the structure will land on separate cache lines. This should be
54  * consider opaque but it is exposed here so vdo_funnel_queue_put() can be inlined.
55  */
56 struct __aligned(L1_CACHE_BYTES) funnel_queue {
57 	/*
58 	 * The producers' end of the queue, an atomically exchanged pointer that will never be
59 	 * NULL.
60 	 */
61 	struct funnel_queue_entry *newest;
62 
63 	/* The consumer's end of the queue, which is owned by the consumer and never NULL. */
64 	struct funnel_queue_entry *oldest __aligned(L1_CACHE_BYTES);
65 
66 	/* A dummy entry used to provide the non-NULL invariants above. */
67 	struct funnel_queue_entry stub;
68 };
69 
70 int __must_check vdo_make_funnel_queue(struct funnel_queue **queue_ptr);
71 
72 void vdo_free_funnel_queue(struct funnel_queue *queue);
73 
74 /*
75  * Put an entry on the end of the queue.
76  *
77  * The entry pointer must be to the struct funnel_queue_entry embedded in the caller's data
78  * structure. The caller must be able to derive the address of the start of their data structure
79  * from the pointer that passed in here, so every entry in the queue must have the struct
80  * funnel_queue_entry at the same offset within the client's structure.
81  */
82 static inline void vdo_funnel_queue_put(struct funnel_queue *queue,
83 					struct funnel_queue_entry *entry)
84 {
85 	struct funnel_queue_entry *previous;
86 
87 	/*
88 	 * Barrier requirements: All stores relating to the entry ("next" pointer, containing data
89 	 * structure fields) must happen before the previous->next store making it visible to the
90 	 * consumer. Also, the entry's "next" field initialization to NULL must happen before any
91 	 * other producer threads can see the entry (the xchg) and try to update the "next" field.
92 	 *
93 	 * xchg implements a full barrier.
94 	 */
95 	WRITE_ONCE(entry->next, NULL);
96 	previous = xchg(&queue->newest, entry);
97 	/*
98 	 * Preemptions between these two statements hide the rest of the queue from the consumer,
99 	 * preventing consumption until the following assignment runs.
100 	 */
101 	WRITE_ONCE(previous->next, entry);
102 }
103 
104 struct funnel_queue_entry *__must_check vdo_funnel_queue_poll(struct funnel_queue *queue);
105 
106 bool __must_check vdo_is_funnel_queue_empty(struct funnel_queue *queue);
107 
108 bool __must_check vdo_is_funnel_queue_idle(struct funnel_queue *queue);
109 
110 #endif /* VDO_FUNNEL_QUEUE_H */
111