1 .\" Copyright (c) 2000-2001 John H. Baldwin <jhb@FreeBSD.org>
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms, with or without
5 .\" modification, are permitted provided that the following conditions
7 .\" 1. Redistributions of source code must retain the above copyright
8 .\" notice, this list of conditions and the following disclaimer.
9 .\" 2. Redistributions in binary form must reproduce the above copyright
10 .\" notice, this list of conditions and the following disclaimer in the
11 .\" documentation and/or other materials provided with the distribution.
13 .\" THIS SOFTWARE IS PROVIDED BY THE DEVELOPERS ``AS IS'' AND ANY EXPRESS OR
14 .\" IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
15 .\" OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
16 .\" IN NO EVENT SHALL THE DEVELOPERS BE LIABLE FOR ANY DIRECT, INDIRECT,
17 .\" INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
18 .\" NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
19 .\" DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
20 .\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
21 .\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
22 .\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 .Dd September 27, 2005
35 .Nm atomic_readandclear ,
44 .Fn atomic_add_[acq_|rel_]<type> "volatile <type> *p" "<type> v"
46 .Fn atomic_clear_[acq_|rel_]<type> "volatile <type> *p" "<type> v"
48 .Fo atomic_cmpset_[acq_|rel_]<type>
49 .Fa "volatile <type> *dst"
54 .Fn atomic_fetchadd_<type> "volatile <type> *p" "<type> v"
56 .Fn atomic_load_acq_<type> "volatile <type> *p"
58 .Fn atomic_readandclear_<type> "volatile <type> *p"
60 .Fn atomic_set_[acq_|rel_]<type> "volatile <type> *p" "<type> v"
62 .Fn atomic_subtract_[acq_|rel_]<type> "volatile <type> *p" "<type> v"
64 .Fn atomic_store_rel_<type> "volatile <type> *p" "<type> v"
67 Each of the atomic operations is guaranteed to be atomic in the presence of
69 They can be used to implement reference counts or as building blocks for more
70 advanced synchronization primitives such as mutexes.
72 Each atomic operation operates on a specific
74 The type to use is indicated in the function name.
75 The available types that can be used are:
77 .Bl -tag -offset indent -width short -compact
83 unsigned integer the size of a pointer
85 unsigned 32-bit integer
87 unsigned 64-bit integer
90 For example, the function to atomically add two integers is called
93 Certain architectures also provide operations for types smaller than
96 .Bl -tag -offset indent -width short -compact
100 unsigned short integer
102 unsigned 8-bit integer
104 unsigned 16-bit integer
107 These must not be used in MI code because the instructions to implement them
108 efficiently may not be available.
110 Memory barriers are used to guarantee the order of data accesses in
112 First, they specify hints to the compiler to not re-order or optimize the
114 Second, on architectures that do not guarantee ordered data accesses,
115 special instructions or special variants of instructions are used to indicate
116 to the processor that data accesses need to occur in a certain order.
117 As a result, most of the atomic operations have three variants in order to
118 include optional memory barriers.
119 The first form just performs the operation without any explicit barriers.
120 The second form uses a read memory barrier, and the third variant uses a write
123 The second variant of each operation includes a read memory barrier.
124 This barrier ensures that the effects of this operation are completed before the
125 effects of any later data accesses.
126 As a result, the operation is said to have acquire semantics as it acquires a
127 pseudo-lock requiring further operations to wait until it has completed.
128 To denote this, the suffix
130 is inserted into the function name immediately prior to the
131 .Dq Li _ Ns Aq Fa type
133 For example, to subtract two integers ensuring that any later writes will
134 happen after the subtraction is performed, use
135 .Fn atomic_subtract_acq_int .
137 The third variant of each operation includes a write memory barrier.
138 This ensures that all effects of all previous data accesses are completed
139 before this operation takes place.
140 As a result, the operation is said to have release semantics as it releases
141 any pending data accesses to be completed before its operation is performed.
142 To denote this, the suffix
144 is inserted into the function name immediately prior to the
145 .Dq Li _ Ns Aq Fa type
147 For example, to add two long integers ensuring that all previous
148 writes will happen first, use
149 .Fn atomic_add_rel_long .
151 A practical example of using memory barriers is to ensure that data accesses
152 that are protected by a lock are all performed while the lock is held.
153 To achieve this, one would use a read barrier when acquiring the lock to
154 guarantee that the lock is held before any protected operations are performed.
155 Finally, one would use a write barrier when releasing the lock to ensure that
156 all of the protected operations are completed before the lock is released.
157 .Ss Multiple Processors
158 The current set of atomic operations do not necessarily guarantee atomicity
159 across multiple processors.
160 To guarantee atomicity across processors, not only does the individual
161 operation need to be atomic on the processor performing the operation, but
162 the result of the operation needs to be pushed out to stable storage and the
163 caches of all other processors on the system need to invalidate any cache
164 lines that include the affected memory region.
167 architecture, the cache coherency model requires that the hardware perform
168 this task, thus the atomic operations are atomic across multiple processors.
171 architecture, coherency is only guaranteed for pages that are configured to
172 using a caching policy of either uncached or write back.
174 This section describes the semantics of each operation using a C like notation.
176 .It Fn atomic_add p v
177 .Bd -literal -compact
180 .It Fn atomic_clear p v
181 .Bd -literal -compact
184 .It Fn atomic_cmpset dst old new
185 .Bd -literal -compact
196 functions are not implemented for the types
203 .It Fn atomic_fetchadd p v
204 .Bd -literal -compact
213 functions are only implemented for the types
217 and do not have any variants with memory barriers at this time.
219 .It Fn atomic_load addr
220 .Bd -literal -compact
227 functions always have acquire semantics.
229 .It Fn atomic_readandclear addr
230 .Bd -literal -compact
238 .Fn atomic_readandclear
239 functions are not implemented for the types
247 not have any variants with memory barriers at this time.
249 .It Fn atomic_set p v
250 .Bd -literal -compact
253 .It Fn atomic_subtract p v
254 .Bd -literal -compact
257 .It Fn atomic_store p v
258 .Bd -literal -compact
265 functions always have release semantics.
269 is currently not implemented for any of the atomic operations on the
279 returns the result of the compare operation.
281 .Fn atomic_fetchadd ,
284 .Fn atomic_readandclear
286 return the value at the specified address.
288 This example uses the
289 .Fn atomic_cmpset_acq_ptr
292 functions to obtain a sleep mutex and handle recursion.
301 /* Try to obtain mtx_lock once. */
302 #define _obtain_lock(mp, tid) \\
303 atomic_cmpset_acq_ptr(&(mp)->mtx_lock, MTX_UNOWNED, (tid))
305 /* Get a sleep lock, deal with recursion inline. */
306 #define _get_sleep_lock(mp, tid, opts, file, line) do { \\
307 uintptr_t _tid = (uintptr_t)(tid); \\
309 if (!_obtain_lock(mp, tid)) { \\
310 if (((mp)->mtx_lock & MTX_FLAGMASK) != _tid) \\
311 _mtx_lock_sleep((mp), _tid, (opts), (file), (line));\\
313 atomic_set_ptr(&(mp)->mtx_lock, MTX_RECURSE); \\
314 (mp)->mtx_recurse++; \\
326 operations were first introduced in
328 This first set only supported the types
337 .Fn atomic_readandclear ,
340 operations were added in
349 and all of the acquire and release variants
355 operations were added in