1 This document contains brief definitions of LKMM-related terms. Like most
2 glossaries, it is not intended to be read front to back (except perhaps
3 as a way of confirming a diagnosis of OCD), but rather to be searched
7 Address Dependency: When the address of a later memory access is computed
8 based on the value returned by an earlier load, an "address
9 dependency" extends from that load extending to the later access.
10 Address dependencies are quite common in RCU read-side critical
14 2 p = rcu_dereference(gp);
18 In this case, because the address of "p->a" on line 3 is computed
19 from the value returned by the rcu_dereference() on line 2, the
20 address dependency extends from that rcu_dereference() to that
21 "p->a". In rare cases, optimizing compilers can destroy address
22 dependencies. Please see Documentation/RCU/rcu_dereference.txt
25 See also "Control Dependency" and "Data Dependency".
27 Acquire: With respect to a lock, acquiring that lock, for example,
28 using spin_lock(). With respect to a non-lock shared variable,
29 a special operation that includes a load and which orders that
30 load before later memory references running on that same CPU.
31 An example special acquire operation is smp_load_acquire(),
32 but atomic_read_acquire() and atomic_xchg_acquire() also include
35 When an acquire load returns the value stored by a release store
36 to that same variable, then all operations preceding that store
37 happen before any operations following that load acquire.
39 See also "Relaxed" and "Release".
41 Coherence (co): When one CPU's store to a given variable overwrites
42 either the value from another CPU's store or some later value,
43 there is said to be a coherence link from the second CPU to
46 It is also possible to have a coherence link within a CPU, which
47 is a "coherence internal" (coi) link. The term "coherence
48 external" (coe) link is used when it is necessary to exclude
51 See also "From-reads" and "Reads-from".
53 Control Dependency: When a later store's execution depends on a test
54 of a value computed from a value returned by an earlier load,
55 a "control dependency" extends from that load to that store.
61 Here, the control dependency extends from the READ_ONCE() on
62 line 1 to the WRITE_ONCE() on line 2. Control dependencies are
63 fragile, and can be easily destroyed by optimizing compilers.
64 Please see control-dependencies.txt for more information.
66 See also "Address Dependency" and "Data Dependency".
68 Cycle: Memory-barrier pairing is restricted to a pair of CPUs, as the
69 name suggests. And in a great many cases, a pair of CPUs is all
70 that is required. In other cases, the notion of pairing must be
71 extended to additional CPUs, and the result is called a "cycle".
72 In a cycle, each CPU's ordering interacts with that of the next:
75 WRITE_ONCE(x, 1); WRITE_ONCE(y, 1); WRITE_ONCE(z, 1);
76 smp_mb(); smp_mb(); smp_mb();
77 r0 = READ_ONCE(y); r1 = READ_ONCE(z); r2 = READ_ONCE(x);
79 CPU 0's smp_mb() interacts with that of CPU 1, which interacts
80 with that of CPU 2, which in turn interacts with that of CPU 0
81 to complete the cycle. Because of the smp_mb() calls between
82 each pair of memory accesses, the outcome where r0, r1, and r2
83 are all equal to zero is forbidden by LKMM.
87 Data Dependency: When the data written by a later store is computed based
88 on the value returned by an earlier load, a "data dependency"
89 extends from that load to that later store. For example:
92 2 WRITE_ONCE(y, r1 + 1);
94 In this case, the data dependency extends from the READ_ONCE()
95 on line 1 to the WRITE_ONCE() on line 2. Data dependencies are
96 fragile and can be easily destroyed by optimizing compilers.
97 Because optimizing compilers put a great deal of effort into
98 working out what values integer variables might have, this is
99 especially true in cases where the dependency is carried through
102 See also "Address Dependency" and "Control Dependency".
104 From-Reads (fr): When one CPU's store to a given variable happened
105 too late to affect the value returned by another CPU's
106 load from that same variable, there is said to be a from-reads
107 link from the load to the store.
109 It is also possible to have a from-reads link within a CPU, which
110 is a "from-reads internal" (fri) link. The term "from-reads
111 external" (fre) link is used when it is necessary to exclude
114 See also "Coherence" and "Reads-from".
116 Fully Ordered: An operation such as smp_mb() that orders all of
117 its CPU's prior accesses with all of that CPU's subsequent
118 accesses, or a marked access such as atomic_add_return()
119 that orders all of its CPU's prior accesses, itself, and
120 all of its CPU's subsequent accesses.
122 Marked Access: An access to a variable that uses an special function or
123 macro such as "r1 = READ_ONCE(x)" or "smp_store_release(&a, 1)".
125 See also "Unmarked Access".
127 Pairing: "Memory-barrier pairing" reflects the fact that synchronizing
128 data between two CPUs requires that both CPUs their accesses.
129 Memory barriers thus tend to come in pairs, one executed by
130 one of the CPUs and the other by the other CPU. Of course,
131 pairing also occurs with other types of operations, so that a
132 smp_store_release() pairs with an smp_load_acquire() that reads
137 Reads-From (rf): When one CPU's load returns the value stored by some other
138 CPU, there is said to be a reads-from link from the second
139 CPU's store to the first CPU's load. Reads-from links have the
140 nice property that time must advance from the store to the load,
141 which means that algorithms using reads-from links can use lighter
142 weight ordering and synchronization compared to algorithms using
143 coherence and from-reads links.
145 It is also possible to have a reads-from link within a CPU, which
146 is a "reads-from internal" (rfi) link. The term "reads-from
147 external" (rfe) link is used when it is necessary to exclude
150 See also Coherence" and "From-reads".
152 Relaxed: A marked access that does not imply ordering, for example, a
153 READ_ONCE(), WRITE_ONCE(), a non-value-returning read-modify-write
154 operation, or a value-returning read-modify-write operation whose
155 name ends in "_relaxed".
157 See also "Acquire" and "Release".
159 Release: With respect to a lock, releasing that lock, for example,
160 using spin_unlock(). With respect to a non-lock shared variable,
161 a special operation that includes a store and which orders that
162 store after earlier memory references that ran on that same CPU.
163 An example special release store is smp_store_release(), but
164 atomic_set_release() and atomic_cmpxchg_release() also include
167 See also "Acquire" and "Relaxed".
169 Unmarked Access: An access to a variable that uses normal C-language
170 syntax, for example, "a = b[2]";
172 See also "Marked Access".