1set_mempolicy(2) System Calls Manual set_mempolicy(2)
2
6 set_mempolicy - set default NUMA memory policy for a thread and its
7 children
8
10 NUMA (Non-Uniform Memory Access) policy library (libnuma, -lnuma)
11
13 #include <numaif.h>
14 long set_mempolicy(int mode, const unsigned long *nodemask,
16 unsigned long maxnode);
17
19 set_mempolicy() sets the NUMA memory policy of the calling thread,
20 which consists of a policy mode and zero or more nodes, to the values
21 specified by the mode, nodemask, and maxnode arguments.
22 A NUMA machine has different memory controllers with different dis‐
24 tances to specific CPUs. The memory policy defines from which node
25 memory is allocated for the thread.
26 This system call defines the default policy for the thread. The thread
28 policy governs allocation of pages in the process's address space out‐
29 side of memory ranges controlled by a more specific policy set by
30 mbind(2). The thread default policy also controls allocation of any
31 pages for memory-mapped files mapped using the mmap(2) call with the
32 MAP_PRIVATE flag and that are only read (loaded) from by the thread and
33 of memory-mapped files mapped using the mmap(2) call with the
34 MAP_SHARED flag, regardless of the access type. The policy is applied
35 only when a new page is allocated for the thread. For anonymous memory
36 this is when the page is first touched by the thread.
37 The mode argument must specify one of MPOL_DEFAULT, MPOL_BIND, MPOL_IN‐
39 TERLEAVE, MPOL_PREFERRED, or MPOL_LOCAL (which are described in detail
40 below). All modes except MPOL_DEFAULT require the caller to specify
41 the node or nodes to which the mode applies, via the nodemask argument.
42 The mode argument may also include an optional mode flag. The sup‐
44 ported mode flags are:
45 MPOL_F_NUMA_BALANCING (since Linux 5.12)
47 When mode is MPOL_BIND, enable the kernel NUMA balancing for the
48 task if it is supported by the kernel. If the flag isn't sup‐
49 ported by the kernel, or is used with mode other than MPOL_BIND,
50 -1 is returned and errno is set to EINVAL.
51 MPOL_F_RELATIVE_NODES (since Linux 2.6.26)
53 A nonempty nodemask specifies node IDs that are relative to the
54 set of node IDs allowed by the process's current cpuset.
55 MPOL_F_STATIC_NODES (since Linux 2.6.26)
57 A nonempty nodemask specifies physical node IDs. Linux will not
58 remap the nodemask when the process moves to a different cpuset
59 context, nor when the set of nodes allowed by the process's cur‐
60 rent cpuset context changes.
61 nodemask points to a bit mask of node IDs that contains up to maxnode
63 bits. The bit mask size is rounded to the next multiple of sizeof(un‐
64 signed long), but the kernel will use bits only up to maxnode. A NULL
65 value of nodemask or a maxnode value of zero specifies the empty set of
66 nodes. If the value of maxnode is zero, the nodemask argument is ig‐
67 nored.
68 Where a nodemask is required, it must contain at least one node that is
70 on-line, allowed by the process's current cpuset context, (unless the
71 MPOL_F_STATIC_NODES mode flag is specified), and contains memory. If
72 the MPOL_F_STATIC_NODES is set in mode and a required nodemask contains
73 no nodes that are allowed by the process's current cpuset context, the
74 memory policy reverts to local allocation. This effectively overrides
75 the specified policy until the process's cpuset context includes one or
76 more of the nodes specified by nodemask.
77 The mode argument must include one of the following values:
79 MPOL_DEFAULT
81 This mode specifies that any nondefault thread memory policy be
82 removed, so that the memory policy "falls back" to the system
83 default policy. The system default policy is "local alloca‐
84 tion"—that is, allocate memory on the node of the CPU that trig‐
85 gered the allocation. nodemask must be specified as NULL. If
86 the "local node" contains no free memory, the system will at‐
87 tempt to allocate memory from a "near by" node.
88 MPOL_BIND
90 This mode defines a strict policy that restricts memory alloca‐
91 tion to the nodes specified in nodemask. If nodemask specifies
92 more than one node, page allocations will come from the node
93 with the lowest numeric node ID first, until that node contains
94 no free memory. Allocations will then come from the node with
95 the next highest node ID specified in nodemask and so forth, un‐
96 til none of the specified nodes contain free memory. Pages will
97 not be allocated from any node not specified in the nodemask.
98 MPOL_INTERLEAVE
100 This mode interleaves page allocations across the nodes speci‐
101 fied in nodemask in numeric node ID order. This optimizes for
102 bandwidth instead of latency by spreading out pages and memory
103 accesses to those pages across multiple nodes. However, ac‐
104 cesses to a single page will still be limited to the memory
105 bandwidth of a single node.
106 MPOL_PREFERRED
108 This mode sets the preferred node for allocation. The kernel
109 will try to allocate pages from this node first and fall back to
110 "near by" nodes if the preferred node is low on free memory. If
111 nodemask specifies more than one node ID, the first node in the
112 mask will be selected as the preferred node. If the nodemask
113 and maxnode arguments specify the empty set, then the policy
114 specifies "local allocation" (like the system default policy
115 discussed above).
116 MPOL_LOCAL (since Linux 3.8)
118 This mode specifies "local allocation"; the memory is allocated
119 on the node of the CPU that triggered the allocation (the "local
120 node"). The nodemask and maxnode arguments must specify the
121 empty set. If the "local node" is low on free memory, the ker‐
122 nel will try to allocate memory from other nodes. The kernel
123 will allocate memory from the "local node" whenever memory for
124 this node is available. If the "local node" is not allowed by
125 the process's current cpuset context, the kernel will try to al‐
126 locate memory from other nodes. The kernel will allocate memory
127 from the "local node" whenever it becomes allowed by the
128 process's current cpuset context.
129 The thread memory policy is preserved across an execve(2), and is in‐
131 herited by child threads created using fork(2) or clone(2).
132
134 On success, set_mempolicy() returns 0; on error, -1 is returned and er‐
135 rno is set to indicate the error.
136
138 EFAULT Part of all of the memory range specified by nodemask and maxn‐
139 ode points outside your accessible address space.
140 EINVAL mode is invalid. Or, mode is MPOL_DEFAULT and nodemask is
142 nonempty, or mode is MPOL_BIND or MPOL_INTERLEAVE and nodemask
143 is empty. Or, maxnode specifies more than a page worth of bits.
144 Or, nodemask specifies one or more node IDs that are greater
145 than the maximum supported node ID. Or, none of the node IDs
146 specified by nodemask are on-line and allowed by the process's
147 current cpuset context, or none of the specified nodes contain
148 memory. Or, the mode argument specified both
149 MPOL_F_STATIC_NODES and MPOL_F_RELATIVE_NODES. Or, the
150 MPOL_F_NUMA_BALANCING isn't supported by the kernel, or is used
151 with mode other than MPOL_BIND.
152 ENOMEM Insufficient kernel memory was available.
154
156 Linux.
157
159 Linux 2.6.7.
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)
171Linux man-pages 6.05 2023-07-16 set_mempolicy(2)