1.\" Copyright (c) 2000-2001 John H. Baldwin <jhb@FreeBSD.org> 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 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. 12.\" 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. 23.\" 24.\" $FreeBSD$ 25.\" 26.Dd September 27, 2005 27.Dt ATOMIC 9 28.Os 29.Sh NAME 30.Nm atomic_add , 31.Nm atomic_clear , 32.Nm atomic_cmpset , 33.Nm atomic_fetchadd , 34.Nm atomic_load , 35.Nm atomic_readandclear , 36.Nm atomic_set , 37.Nm atomic_subtract , 38.Nm atomic_store 39.Nd atomic operations 40.Sh SYNOPSIS 41.In sys/types.h 42.In machine/atomic.h 43.Ft void 44.Fn atomic_add_[acq_|rel_]<type> "volatile <type> *p" "<type> v" 45.Ft void 46.Fn atomic_clear_[acq_|rel_]<type> "volatile <type> *p" "<type> v" 47.Ft int 48.Fo atomic_cmpset_[acq_|rel_]<type> 49.Fa "volatile <type> *dst" 50.Fa "<type> old" 51.Fa "<type> new" 52.Fc 53.Ft <type> 54.Fn atomic_fetchadd_<type> "volatile <type> *p" "<type> v" 55.Ft <type> 56.Fn atomic_load_acq_<type> "volatile <type> *p" 57.Ft <type> 58.Fn atomic_readandclear_<type> "volatile <type> *p" 59.Ft void 60.Fn atomic_set_[acq_|rel_]<type> "volatile <type> *p" "<type> v" 61.Ft void 62.Fn atomic_subtract_[acq_|rel_]<type> "volatile <type> *p" "<type> v" 63.Ft void 64.Fn atomic_store_rel_<type> "volatile <type> *p" "<type> v" 65.Sh DESCRIPTION 66Each of the atomic operations is guaranteed to be atomic in the presence of 67interrupts. 68They can be used to implement reference counts or as building blocks for more 69advanced synchronization primitives such as mutexes. 70.Ss Types 71Each atomic operation operates on a specific 72.Fa type . 73The type to use is indicated in the function name. 74The available types that can be used are: 75.Pp 76.Bl -tag -offset indent -width short -compact 77.It Li int 78unsigned integer 79.It Li long 80unsigned long integer 81.It Li ptr 82unsigned integer the size of a pointer 83.It Li 32 84unsigned 32-bit integer 85.It Li 64 86unsigned 64-bit integer 87.El 88.Pp 89For example, the function to atomically add two integers is called 90.Fn atomic_add_int . 91.Pp 92Certain architectures also provide operations for types smaller than 93.Dq Li int . 94.Pp 95.Bl -tag -offset indent -width short -compact 96.It Li char 97unsigned character 98.It Li short 99unsigned short integer 100.It Li 8 101unsigned 8-bit integer 102.It Li 16 103unsigned 16-bit integer 104.El 105.Pp 106These must not be used in MI code because the instructions to implement them 107efficiently may not be available. 108.Ss Memory Barriers 109Memory barriers are used to guarantee the order of data accesses in 110two ways. 111First, they specify hints to the compiler to not re-order or optimize the 112operations. 113Second, on architectures that do not guarantee ordered data accesses, 114special instructions or special variants of instructions are used to indicate 115to the processor that data accesses need to occur in a certain order. 116As a result, most of the atomic operations have three variants in order to 117include optional memory barriers. 118The first form just performs the operation without any explicit barriers. 119The second form uses a read memory barrier, and the third variant uses a write 120memory barrier. 121.Pp 122The second variant of each operation includes a read memory barrier. 123This barrier ensures that the effects of this operation are completed before the 124effects of any later data accesses. 125As a result, the operation is said to have acquire semantics as it acquires a 126pseudo-lock requiring further operations to wait until it has completed. 127To denote this, the suffix 128.Dq Li _acq 129is inserted into the function name immediately prior to the 130.Dq Li _ Ns Aq Fa type 131suffix. 132For example, to subtract two integers ensuring that any later writes will 133happen after the subtraction is performed, use 134.Fn atomic_subtract_acq_int . 135.Pp 136The third variant of each operation includes a write memory barrier. 137This ensures that all effects of all previous data accesses are completed 138before this operation takes place. 139As a result, the operation is said to have release semantics as it releases 140any pending data accesses to be completed before its operation is performed. 141To denote this, the suffix 142.Dq Li _rel 143is inserted into the function name immediately prior to the 144.Dq Li _ Ns Aq Fa type 145suffix. 146For example, to add two long integers ensuring that all previous 147writes will happen first, use 148.Fn atomic_add_rel_long . 149.Pp 150A practical example of using memory barriers is to ensure that data accesses 151that are protected by a lock are all performed while the lock is held. 152To achieve this, one would use a read barrier when acquiring the lock to 153guarantee that the lock is held before any protected operations are performed. 154Finally, one would use a write barrier when releasing the lock to ensure that 155all of the protected operations are completed before the lock is released. 156.Ss Multiple Processors 157The current set of atomic operations do not necessarily guarantee atomicity 158across multiple processors. 159To guarantee atomicity across processors, not only does the individual 160operation need to be atomic on the processor performing the operation, but 161the result of the operation needs to be pushed out to stable storage and the 162caches of all other processors on the system need to invalidate any cache 163lines that include the affected memory region. 164On the 165.Tn i386 166architecture, the cache coherency model requires that the hardware perform 167this task, thus the atomic operations are atomic across multiple processors. 168On the 169.Tn ia64 170architecture, coherency is only guaranteed for pages that are configured to 171using a caching policy of either uncached or write back. 172.Ss Semantics 173This section describes the semantics of each operation using a C like notation. 174.Bl -hang 175.It Fn atomic_add p v 176.Bd -literal -compact 177*p += v; 178.Ed 179.It Fn atomic_clear p v 180.Bd -literal -compact 181*p &= ~v; 182.Ed 183.It Fn atomic_cmpset dst old new 184.Bd -literal -compact 185if (*dst == old) { 186 *dst = new; 187 return 1; 188} else 189 return 0; 190.Ed 191.El 192.Pp 193The 194.Fn atomic_cmpset 195functions are not implemented for the types 196.Dq Li char , 197.Dq Li short , 198.Dq Li 8 , 199and 200.Dq Li 16 . 201.Bl -hang 202.It Fn atomic_fetchadd p v 203.Bd -literal -compact 204tmp = *p; 205*p += v; 206return tmp; 207.Ed 208.El 209.Pp 210The 211.Fn atomic_fetchadd 212functions are only implemented for the types 213.Dq Li int , 214.Dq Li long 215and 216.Dq Li 32 217and do not have any variants with memory barriers at this time. 218.Bl -hang 219.It Fn atomic_load addr 220.Bd -literal -compact 221return (*addr) 222.Ed 223.El 224.Pp 225The 226.Fn atomic_load 227functions are only provided with acquire memory barriers. 228.Bl -hang 229.It Fn atomic_readandclear addr 230.Bd -literal -compact 231temp = *addr; 232*addr = 0; 233return (temp); 234.Ed 235.El 236.Pp 237The 238.Fn atomic_readandclear 239functions are not implemented for the types 240.Dq Li char , 241.Dq Li short , 242.Dq Li ptr , 243.Dq Li 8 , 244and 245.Dq Li 16 246and do 247not have any variants with memory barriers at this time. 248.Bl -hang 249.It Fn atomic_set p v 250.Bd -literal -compact 251*p |= v; 252.Ed 253.It Fn atomic_subtract p v 254.Bd -literal -compact 255*p -= v; 256.Ed 257.It Fn atomic_store p v 258.Bd -literal -compact 259*p = v; 260.Ed 261.El 262.Pp 263The 264.Fn atomic_store 265functions are only provided with release memory barriers. 266.Pp 267The type 268.Dq Li 64 269is currently not implemented for any of the atomic operations on the 270.Tn arm , 271.Tn i386 , 272and 273.Tn powerpc 274architectures. 275.Sh RETURN VALUES 276The 277.Fn atomic_cmpset 278function 279returns the result of the compare operation. 280The 281.Fn atomic_fetchadd , 282.Fn atomic_load , 283and 284.Fn atomic_readandclear 285functions 286return the value at the specified address. 287.Sh EXAMPLES 288This example uses the 289.Fn atomic_cmpset_acq_ptr 290and 291.Fn atomic_set_ptr 292functions to obtain a sleep mutex and handle recursion. 293Since the 294.Va mtx_lock 295member of a 296.Vt "struct mtx" 297is a pointer, the 298.Dq Li ptr 299type is used. 300.Bd -literal 301/* Try to obtain mtx_lock once. */ 302#define _obtain_lock(mp, tid) \\ 303 atomic_cmpset_acq_ptr(&(mp)->mtx_lock, MTX_UNOWNED, (tid)) 304 305/* 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); \\ 308 \\ 309 if (!_obtain_lock(mp, tid)) { \\ 310 if (((mp)->mtx_lock & MTX_FLAGMASK) != _tid) \\ 311 _mtx_lock_sleep((mp), _tid, (opts), (file), (line));\\ 312 else { \\ 313 atomic_set_ptr(&(mp)->mtx_lock, MTX_RECURSE); \\ 314 (mp)->mtx_recurse++; \\ 315 } \\ 316 } \\ 317} while (0) 318.Ed 319.Sh HISTORY 320The 321.Fn atomic_add , 322.Fn atomic_clear , 323.Fn atomic_set , 324and 325.Fn atomic_subtract 326operations were first introduced in 327.Fx 3.0 . 328This first set only supported the types 329.Dq Li char , 330.Dq Li short , 331.Dq Li int , 332and 333.Dq Li long . 334The 335.Fn atomic_cmpset , 336.Fn atomic_load , 337.Fn atomic_readandclear , 338and 339.Fn atomic_store 340operations were added in 341.Fx 5.0 . 342The types 343.Dq Li 8 , 344.Dq Li 16 , 345.Dq Li 32 , 346.Dq Li 64 , 347and 348.Dq Li ptr 349and all of the acquire and release variants 350were added in 351.Fx 5.0 352as well. 353The 354.Fn atomic_fetchadd 355operations were added in 356.Fx 6.0 . 357