1SET_MEMPOLICY(2) Linux Programmer's Manual SET_MEMPOLICY(2)
2
6 set_mempolicy - set default NUMA memory policy for a thread and its
7 children
8
10 #include <numaif.h>
11 long set_mempolicy(int mode, const unsigned long *nodemask,
13 unsigned long maxnode);
14 Link with -lnuma.
16
18 set_mempolicy() sets the NUMA memory policy of the calling thread,
19 which consists of a policy mode and zero or more nodes, to the values
20 specified by the mode, nodemask, and maxnode arguments.
21 A NUMA machine has different memory controllers with different dis‐
23 tances to specific CPUs. The memory policy defines from which node
24 memory is allocated for the thread.
25 This system call defines the default policy for the thread. The thread
27 policy governs allocation of pages in the process's address space out‐
28 side of memory ranges controlled by a more specific policy set by
29 mbind(2). The thread default policy also controls allocation of any
30 pages for memory-mapped files mapped using the mmap(2) call with the
31 MAP_PRIVATE flag and that are only read (loaded) from by the thread and
32 of memory-mapped files mapped using the mmap(2) call with the
33 MAP_SHARED flag, regardless of the access type. The policy is applied
34 only when a new page is allocated for the thread. For anonymous memory
35 this is when the page is first touched by the thread.
36 The mode argument must specify one of MPOL_DEFAULT, MPOL_BIND, MPOL_IN‐
38 TERLEAVE, MPOL_PREFERRED, or MPOL_LOCAL (which are described in detail
39 below). All modes except MPOL_DEFAULT require the caller to specify
40 the node or nodes to which the mode applies, via the nodemask argument.
41 The mode argument may also include an optional mode flag. The sup‐
43 ported mode flags are:
44 MPOL_F_NUMA_BALANCING (since Linux 5.12)
46 When mode is MPOL_BIND, enable the kernel NUMA balancing for the
47 task if it is supported by the kernel. If the flag isn't sup‐
48 ported by the kernel, or is used with mode other than MPOL_BIND,
49 -1 is returned and errno is set to EINVAL.
50 MPOL_F_RELATIVE_NODES (since Linux 2.6.26)
52 A nonempty nodemask specifies node IDs that are relative to the
53 set of node IDs allowed by the process's current cpuset.
54 MPOL_F_STATIC_NODES (since Linux 2.6.26)
56 A nonempty nodemask specifies physical node IDs. Linux will not
57 remap the nodemask when the process moves to a different cpuset
58 context, nor when the set of nodes allowed by the process's cur‐
59 rent cpuset context changes.
60 nodemask points to a bit mask of node IDs that contains up to maxnode
62 bits. The bit mask size is rounded to the next multiple of sizeof(un‐
63 signed long), but the kernel will use bits only up to maxnode. A NULL
64 value of nodemask or a maxnode value of zero specifies the empty set of
65 nodes. If the value of maxnode is zero, the nodemask argument is ig‐
66 nored.
67 Where a nodemask is required, it must contain at least one node that is
69 on-line, allowed by the process's current cpuset context, (unless the
70 MPOL_F_STATIC_NODES mode flag is specified), and contains memory. If
71 the MPOL_F_STATIC_NODES is set in mode and a required nodemask contains
72 no nodes that are allowed by the process's current cpuset context, the
73 memory policy reverts to local allocation. This effectively overrides
74 the specified policy until the process's cpuset context includes one or
75 more of the nodes specified by nodemask.
76 The mode argument must include one of the following values:
78 MPOL_DEFAULT
80 This mode specifies that any nondefault thread memory policy be
81 removed, so that the memory policy "falls back" to the system
82 default policy. The system default policy is "local alloca‐
83 tion"—that is, allocate memory on the node of the CPU that trig‐
84 gered the allocation. nodemask must be specified as NULL. If
85 the "local node" contains no free memory, the system will at‐
86 tempt to allocate memory from a "near by" node.
87 MPOL_BIND
89 This mode defines a strict policy that restricts memory alloca‐
90 tion to the nodes specified in nodemask. If nodemask specifies
91 more than one node, page allocations will come from the node
92 with the lowest numeric node ID first, until that node contains
93 no free memory. Allocations will then come from the node with
94 the next highest node ID specified in nodemask and so forth, un‐
95 til none of the specified nodes contain free memory. Pages will
96 not be allocated from any node not specified in the nodemask.
97 MPOL_INTERLEAVE
99 This mode interleaves page allocations across the nodes speci‐
100 fied in nodemask in numeric node ID order. This optimizes for
101 bandwidth instead of latency by spreading out pages and memory
102 accesses to those pages across multiple nodes. However, ac‐
103 cesses to a single page will still be limited to the memory
104 bandwidth of a single node.
105 MPOL_PREFERRED
107 This mode sets the preferred node for allocation. The kernel
108 will try to allocate pages from this node first and fall back to
109 "near by" nodes if the preferred node is low on free memory. If
110 nodemask specifies more than one node ID, the first node in the
111 mask will be selected as the preferred node. If the nodemask
112 and maxnode arguments specify the empty set, then the policy
113 specifies "local allocation" (like the system default policy
114 discussed above).
115 MPOL_LOCAL (since Linux 3.8)
117 This mode specifies "local allocation"; the memory is allocated
118 on the node of the CPU that triggered the allocation (the "local
119 node"). The nodemask and maxnode arguments must specify the
120 empty set. If the "local node" is low on free memory, the ker‐
121 nel will try to allocate memory from other nodes. The kernel
122 will allocate memory from the "local node" whenever memory for
123 this node is available. If the "local node" is not allowed by
124 the process's current cpuset context, the kernel will try to al‐
125 locate memory from other nodes. The kernel will allocate memory
126 from the "local node" whenever it becomes allowed by the
127 process's current cpuset context.
128 The thread memory policy is preserved across an execve(2), and is in‐
130 herited by child threads created using fork(2) or clone(2).
131
133 On success, set_mempolicy() returns 0; on error, -1 is returned and er‐
134 rno is set to indicate the error.
135
137 EFAULT Part of all of the memory range specified by nodemask and maxn‐
138 ode points outside your accessible address space.
139 EINVAL mode is invalid. Or, mode is MPOL_DEFAULT and nodemask is
141 nonempty, or mode is MPOL_BIND or MPOL_INTERLEAVE and nodemask
142 is empty. Or, maxnode specifies more than a page worth of bits.
143 Or, nodemask specifies one or more node IDs that are greater
144 than the maximum supported node ID. Or, none of the node IDs
145 specified by nodemask are on-line and allowed by the process's
146 current cpuset context, or none of the specified nodes contain
147 memory. Or, the mode argument specified both
148 MPOL_F_STATIC_NODES and MPOL_F_RELATIVE_NODES. Or, the
149 MPOL_F_NUMA_BALANCING isn't supported by the kernel, or is used
150 with mode other than MPOL_BIND.
151 ENOMEM Insufficient kernel memory was available.
153
155 The set_mempolicy() system call was added to the Linux kernel in ver‐
156 sion 2.6.7.
157
159 This system call is Linux-specific.
160
162 Memory policy is not remembered if the page is swapped out. When such
163 a page is paged back in, it will use the policy of the thread or memory
164 range that is in effect at the time the page is allocated.
165 For information on library support, see numa(7).
167
169 get_mempolicy(2), getcpu(2), mbind(2), mmap(2), numa(3), cpuset(7),
170 numa(7), numactl(8)
171
173 This page is part of release 5.13 of the Linux man-pages project. A
174 description of the project, information about reporting bugs, and the
175 latest version of this page, can be found at
176 https://www.kernel.org/doc/man-pages/.
177Linux 2021-06-20 SET_MEMPOLICY(2)