xref: /illumos-gate/usr/src/uts/common/fs/zfs/aggsum.c (revision 66582b606a8194f7f3ba5b3a3a6dca5b0d346361)
1 /*
2  * CDDL HEADER START
3  *
4  * This file and its contents are supplied under the terms of the
5  * Common Development and Distribution License ("CDDL"), version 1.0.
6  * You may only use this file in accordance with the terms of version
7  * 1.0 of the CDDL.
8  *
9  * A full copy of the text of the CDDL should have accompanied this
10  * source.  A copy of the CDDL is also available via the Internet at
11  * http://www.illumos.org/license/CDDL.
12  *
13  * CDDL HEADER END
14  */
15 /*
16  * Copyright (c) 2017, 2018 by Delphix. All rights reserved.
17  */
18 
19 #include <sys/zfs_context.h>
20 #include <sys/aggsum.h>
21 
22 /*
23  * Aggregate-sum counters are a form of fanned-out counter, used when atomic
24  * instructions on a single field cause enough CPU cache line contention to
25  * slow system performance. Due to their increased overhead and the expense
26  * involved with precisely reading from them, they should only be used in cases
27  * where the write rate (increment/decrement) is much higher than the read rate
28  * (get value).
29  *
30  * Aggregate sum counters are comprised of two basic parts, the core and the
31  * buckets. The core counter contains a lock for the entire counter, as well
32  * as the current upper and lower bounds on the value of the counter. The
33  * aggsum_bucket structure contains a per-bucket lock to protect the contents of
34  * the bucket, the current amount that this bucket has changed from the global
35  * counter (called the delta), and the amount of increment and decrement we have
36  * "borrowed" from the core counter.
37  *
38  * The basic operation of an aggsum is simple. Threads that wish to modify the
39  * counter will modify one bucket's counter (determined by their current CPU, to
40  * help minimize lock and cache contention). If the bucket already has
41  * sufficient capacity borrowed from the core structure to handle their request,
42  * they simply modify the delta and return.  If the bucket does not, we clear
43  * the bucket's current state (to prevent the borrowed amounts from getting too
44  * large), and borrow more from the core counter. Borrowing is done by adding to
45  * the upper bound (or subtracting from the lower bound) of the core counter,
46  * and setting the borrow value for the bucket to the amount added (or
47  * subtracted).  Clearing the bucket is the opposite; we add the current delta
48  * to both the lower and upper bounds of the core counter, subtract the borrowed
49  * incremental from the upper bound, and add the borrowed decrement from the
50  * lower bound.  Note that only borrowing and clearing require access to the
51  * core counter; since all other operations access CPU-local resources,
52  * performance can be much higher than a traditional counter.
53  *
54  * Threads that wish to read from the counter have a slightly more challenging
55  * task. It is fast to determine the upper and lower bounds of the aggum; this
56  * does not require grabbing any locks. This suffices for cases where an
57  * approximation of the aggsum's value is acceptable. However, if one needs to
58  * know whether some specific value is above or below the current value in the
59  * aggsum, they invoke aggsum_compare(). This function operates by repeatedly
60  * comparing the target value to the upper and lower bounds of the aggsum, and
61  * then clearing a bucket. This proceeds until the target is outside of the
62  * upper and lower bounds and we return a response, or the last bucket has been
63  * cleared and we know that the target is equal to the aggsum's value. Finally,
64  * the most expensive operation is determining the precise value of the aggsum.
65  * To do this, we clear every bucket and then return the upper bound (which must
66  * be equal to the lower bound). What makes aggsum_compare() and aggsum_value()
67  * expensive is clearing buckets. This involves grabbing the global lock
68  * (serializing against themselves and borrow operations), grabbing a bucket's
69  * lock (preventing threads on those CPUs from modifying their delta), and
70  * zeroing out the borrowed value (forcing that thread to borrow on its next
71  * request, which will also be expensive).  This is what makes aggsums well
72  * suited for write-many read-rarely operations.
73  */
74 
75 /*
76  * We will borrow aggsum_borrow_multiplier times the current request, so we will
77  * have to get the as_lock approximately every aggsum_borrow_multiplier calls to
78  * aggsum_delta().
79  */
80 static uint_t aggsum_borrow_multiplier = 10;
81 
82 void
83 aggsum_init(aggsum_t *as, uint64_t value)
84 {
85 	bzero(as, sizeof (*as));
86 	as->as_lower_bound = as->as_upper_bound = value;
87 	mutex_init(&as->as_lock, NULL, MUTEX_DEFAULT, NULL);
88 	as->as_numbuckets = boot_ncpus;
89 	as->as_buckets = kmem_zalloc(boot_ncpus * sizeof (aggsum_bucket_t),
90 	    KM_SLEEP);
91 	for (int i = 0; i < as->as_numbuckets; i++) {
92 		mutex_init(&as->as_buckets[i].asc_lock,
93 		    NULL, MUTEX_DEFAULT, NULL);
94 	}
95 }
96 
97 void
98 aggsum_fini(aggsum_t *as)
99 {
100 	for (int i = 0; i < as->as_numbuckets; i++)
101 		mutex_destroy(&as->as_buckets[i].asc_lock);
102 	kmem_free(as->as_buckets, as->as_numbuckets * sizeof (aggsum_bucket_t));
103 	mutex_destroy(&as->as_lock);
104 }
105 
106 int64_t
107 aggsum_lower_bound(aggsum_t *as)
108 {
109 	return (as->as_lower_bound);
110 }
111 
112 int64_t
113 aggsum_upper_bound(aggsum_t *as)
114 {
115 	return (as->as_upper_bound);
116 }
117 
118 static void
119 aggsum_flush_bucket(aggsum_t *as, struct aggsum_bucket *asb)
120 {
121 	ASSERT(MUTEX_HELD(&as->as_lock));
122 	ASSERT(MUTEX_HELD(&asb->asc_lock));
123 
124 	/*
125 	 * We use atomic instructions for this because we read the upper and
126 	 * lower bounds without the lock, so we need stores to be atomic.
127 	 */
128 	atomic_add_64((volatile uint64_t *)&as->as_lower_bound, asb->asc_delta);
129 	atomic_add_64((volatile uint64_t *)&as->as_upper_bound, asb->asc_delta);
130 	asb->asc_delta = 0;
131 	atomic_add_64((volatile uint64_t *)&as->as_upper_bound,
132 	    -asb->asc_borrowed);
133 	atomic_add_64((volatile uint64_t *)&as->as_lower_bound,
134 	    asb->asc_borrowed);
135 	asb->asc_borrowed = 0;
136 }
137 
138 uint64_t
139 aggsum_value(aggsum_t *as)
140 {
141 	int64_t rv;
142 
143 	mutex_enter(&as->as_lock);
144 	if (as->as_lower_bound == as->as_upper_bound) {
145 		rv = as->as_lower_bound;
146 		for (int i = 0; i < as->as_numbuckets; i++) {
147 			ASSERT0(as->as_buckets[i].asc_delta);
148 			ASSERT0(as->as_buckets[i].asc_borrowed);
149 		}
150 		mutex_exit(&as->as_lock);
151 		return (rv);
152 	}
153 	for (int i = 0; i < as->as_numbuckets; i++) {
154 		struct aggsum_bucket *asb = &as->as_buckets[i];
155 		mutex_enter(&asb->asc_lock);
156 		aggsum_flush_bucket(as, asb);
157 		mutex_exit(&asb->asc_lock);
158 	}
159 	VERIFY3U(as->as_lower_bound, ==, as->as_upper_bound);
160 	rv = as->as_lower_bound;
161 	mutex_exit(&as->as_lock);
162 
163 	return (rv);
164 }
165 
166 static void
167 aggsum_borrow(aggsum_t *as, int64_t delta, struct aggsum_bucket *asb)
168 {
169 	int64_t abs_delta = (delta < 0 ? -delta : delta);
170 	mutex_enter(&as->as_lock);
171 	mutex_enter(&asb->asc_lock);
172 
173 	aggsum_flush_bucket(as, asb);
174 
175 	atomic_add_64((volatile uint64_t *)&as->as_upper_bound, abs_delta);
176 	atomic_add_64((volatile uint64_t *)&as->as_lower_bound, -abs_delta);
177 	asb->asc_borrowed = abs_delta;
178 
179 	mutex_exit(&asb->asc_lock);
180 	mutex_exit(&as->as_lock);
181 }
182 
183 void
184 aggsum_add(aggsum_t *as, int64_t delta)
185 {
186 	struct aggsum_bucket *asb =
187 	    &as->as_buckets[CPU_SEQID % as->as_numbuckets];
188 
189 	for (;;) {
190 		mutex_enter(&asb->asc_lock);
191 		if (asb->asc_delta + delta <= (int64_t)asb->asc_borrowed &&
192 		    asb->asc_delta + delta >= -(int64_t)asb->asc_borrowed) {
193 			asb->asc_delta += delta;
194 			mutex_exit(&asb->asc_lock);
195 			return;
196 		}
197 		mutex_exit(&asb->asc_lock);
198 		aggsum_borrow(as, delta * aggsum_borrow_multiplier, asb);
199 	}
200 }
201 
202 /*
203  * Compare the aggsum value to target efficiently. Returns -1 if the value
204  * represented by the aggsum is less than target, 1 if it's greater, and 0 if
205  * they are equal.
206  */
207 int
208 aggsum_compare(aggsum_t *as, uint64_t target)
209 {
210 	if (as->as_upper_bound < target)
211 		return (-1);
212 	if (as->as_lower_bound > target)
213 		return (1);
214 	mutex_enter(&as->as_lock);
215 	for (int i = 0; i < as->as_numbuckets; i++) {
216 		struct aggsum_bucket *asb = &as->as_buckets[i];
217 		mutex_enter(&asb->asc_lock);
218 		aggsum_flush_bucket(as, asb);
219 		mutex_exit(&asb->asc_lock);
220 		if (as->as_upper_bound < target) {
221 			mutex_exit(&as->as_lock);
222 			return (-1);
223 		}
224 		if (as->as_lower_bound > target) {
225 			mutex_exit(&as->as_lock);
226 			return (1);
227 		}
228 	}
229 	VERIFY3U(as->as_lower_bound, ==, as->as_upper_bound);
230 	ASSERT3U(as->as_lower_bound, ==, target);
231 	mutex_exit(&as->as_lock);
232 	return (0);
233 }
234