1mbind(2) System Calls Manual mbind(2)
2
6 mbind - set memory policy for a memory range
7
9 NUMA (Non-Uniform Memory Access) policy library (libnuma, -lnuma)
10
12 #include <numaif.h>
13 long mbind(void addr[.len], unsigned long len, int mode,
15 const unsigned long nodemask[(.maxnode + ULONG_WIDTH - 1)
16 / ULONG_WIDTH],
17 unsigned long maxnode, unsigned int flags);
18
20 mbind() sets the NUMA memory policy, which consists of a policy mode
21 and zero or more nodes, for the memory range starting with addr and
22 continuing for len bytes. The memory policy defines from which node
23 memory is allocated.
24 If the memory range specified by the addr and len arguments includes an
26 "anonymous" region of memory—that is a region of memory created using
27 the mmap(2) system call with the MAP_ANONYMOUS—or a memory-mapped file,
28 mapped using the mmap(2) system call with the MAP_PRIVATE flag, pages
29 will be allocated only according to the specified policy when the ap‐
30 plication writes (stores) to the page. For anonymous regions, an ini‐
31 tial read access will use a shared page in the kernel containing all
32 zeros. For a file mapped with MAP_PRIVATE, an initial read access will
33 allocate pages according to the memory policy of the thread that causes
34 the page to be allocated. This may not be the thread that called
35 mbind().
36 The specified policy will be ignored for any MAP_SHARED mappings in the
38 specified memory range. Rather the pages will be allocated according
39 to the memory policy of the thread that caused the page to be allo‐
40 cated. Again, this may not be the thread that called mbind().
41 If the specified memory range includes a shared memory region created
43 using the shmget(2) system call and attached using the shmat(2) system
44 call, pages allocated for the anonymous or shared memory region will be
45 allocated according to the policy specified, regardless of which
46 process attached to the shared memory segment causes the allocation.
47 If, however, the shared memory region was created with the SHM_HUGETLB
48 flag, the huge pages will be allocated according to the policy speci‐
49 fied only if the page allocation is caused by the process that calls
50 mbind() for that region.
51 By default, mbind() has an effect only for new allocations; if the
53 pages inside the range have been already touched before setting the
54 policy, then the policy has no effect. This default behavior may be
55 overridden by the MPOL_MF_MOVE and MPOL_MF_MOVE_ALL flags described be‐
56 low.
57 The mode argument must specify one of MPOL_DEFAULT, MPOL_BIND, MPOL_IN‐
59 TERLEAVE, MPOL_PREFERRED, or MPOL_LOCAL (which are described in detail
60 below). All policy modes except MPOL_DEFAULT require the caller to
61 specify the node or nodes to which the mode applies, via the nodemask
62 argument.
63 The mode argument may also include an optional mode flag. The sup‐
65 ported mode flags are:
66 MPOL_F_STATIC_NODES (since Linux-2.6.26)
68 A nonempty nodemask specifies physical node IDs. Linux does not
69 remap the nodemask when the thread moves to a different cpuset
70 context, nor when the set of nodes allowed by the thread's cur‐
71 rent cpuset context changes.
72 MPOL_F_RELATIVE_NODES (since Linux-2.6.26)
74 A nonempty nodemask specifies node IDs that are relative to the
75 set of node IDs allowed by the thread's current cpuset.
76 nodemask points to a bit mask of nodes containing up to maxnode bits.
78 The bit mask size is rounded to the next multiple of sizeof(unsigned
79 long), but the kernel will use bits only up to maxnode. A NULL value
80 of nodemask or a maxnode value of zero specifies the empty set of
81 nodes. If the value of maxnode is zero, the nodemask argument is ig‐
82 nored. Where a nodemask is required, it must contain at least one node
83 that is on-line, allowed by the thread's current cpuset context (unless
84 the MPOL_F_STATIC_NODES mode flag is specified), and contains memory.
85 The mode argument must include one of the following values:
87 MPOL_DEFAULT
89 This mode requests that any nondefault policy be removed,
90 restoring default behavior. When applied to a range of memory
91 via mbind(), this means to use the thread memory policy, which
92 may have been set with set_mempolicy(2). If the mode of the
93 thread memory policy is also MPOL_DEFAULT, the system-wide de‐
94 fault policy will be used. The system-wide default policy allo‐
95 cates pages on the node of the CPU that triggers the allocation.
96 For MPOL_DEFAULT, the nodemask and maxnode arguments must be
97 specify the empty set of nodes.
98 MPOL_BIND
100 This mode specifies a strict policy that restricts memory allo‐
101 cation to the nodes specified in nodemask. If nodemask speci‐
102 fies more than one node, page allocations will come from the
103 node with sufficient free memory that is closest to the node
104 where the allocation takes place. Pages will not be allocated
105 from any node not specified in the IR nodemask . (Before Linux
106 2.6.26, page allocations came from the node with the lowest nu‐
107 meric node ID first, until that node contained no free memory.
108 Allocations then came from the node with the next highest node
109 ID specified in nodemask and so forth, until none of the speci‐
110 fied nodes contained free memory.)
111 MPOL_INTERLEAVE
113 This mode specifies that page allocations be interleaved across
114 the set of nodes specified in nodemask. This optimizes for
115 bandwidth instead of latency by spreading out pages and memory
116 accesses to those pages across multiple nodes. To be effective
117 the memory area should be fairly large, at least 1 MB or bigger
118 with a fairly uniform access pattern. Accesses to a single page
119 of the area will still be limited to the memory bandwidth of a
120 single node.
121 MPOL_PREFERRED
123 This mode sets the preferred node for allocation. The kernel
124 will try to allocate pages from this node first and fall back to
125 other nodes if the preferred nodes is low on free memory. If
126 nodemask specifies more than one node ID, the first node in the
127 mask will be selected as the preferred node. If the nodemask
128 and maxnode arguments specify the empty set, then the memory is
129 allocated on the node of the CPU that triggered the allocation.
130 MPOL_LOCAL (since Linux 3.8)
132 This mode specifies "local allocation"; the memory is allocated
133 on the node of the CPU that triggered the allocation (the "local
134 node"). The nodemask and maxnode arguments must specify the
135 empty set. If the "local node" is low on free memory, the ker‐
136 nel will try to allocate memory from other nodes. The kernel
137 will allocate memory from the "local node" whenever memory for
138 this node is available. If the "local node" is not allowed by
139 the thread's current cpuset context, the kernel will try to al‐
140 locate memory from other nodes. The kernel will allocate memory
141 from the "local node" whenever it becomes allowed by the
142 thread's current cpuset context. By contrast, MPOL_DEFAULT re‐
143 verts to the memory policy of the thread (which may be set via
144 set_mempolicy(2)); that policy may be something other than "lo‐
145 cal allocation".
146 If MPOL_MF_STRICT is passed in flags and mode is not MPOL_DEFAULT, then
148 the call fails with the error EIO if the existing pages in the memory
149 range don't follow the policy.
150 If MPOL_MF_MOVE is specified in flags, then the kernel will attempt to
152 move all the existing pages in the memory range so that they follow the
153 policy. Pages that are shared with other processes will not be moved.
154 If MPOL_MF_STRICT is also specified, then the call fails with the error
155 EIO if some pages could not be moved.
156 If MPOL_MF_MOVE_ALL is passed in flags, then the kernel will attempt to
158 move all existing pages in the memory range regardless of whether other
159 processes use the pages. The calling thread must be privileged
160 (CAP_SYS_NICE) to use this flag. If MPOL_MF_STRICT is also specified,
161 then the call fails with the error EIO if some pages could not be
162 moved.
163
165 On success, mbind() returns 0; on error, -1 is returned and errno is
166 set to indicate the error.
167
169 EFAULT Part or all of the memory range specified by nodemask and maxn‐
170 ode points outside your accessible address space. Or, there was
171 an unmapped hole in the specified memory range specified by addr
172 and len.
173 EINVAL An invalid value was specified for flags or mode; or addr + len
175 was less than addr; or addr is not a multiple of the system page
176 size. Or, mode is MPOL_DEFAULT and nodemask specified a
177 nonempty set; or mode is MPOL_BIND or MPOL_INTERLEAVE and node‐
178 mask is empty. Or, maxnode exceeds a kernel-imposed limit. Or,
179 nodemask specifies one or more node IDs that are greater than
180 the maximum supported node ID. Or, none of the node IDs speci‐
181 fied by nodemask are on-line and allowed by the thread's current
182 cpuset context, or none of the specified nodes contain memory.
183 Or, the mode argument specified both MPOL_F_STATIC_NODES and
184 MPOL_F_RELATIVE_NODES.
185 EIO MPOL_MF_STRICT was specified and an existing page was already on
187 a node that does not follow the policy; or MPOL_MF_MOVE or
188 MPOL_MF_MOVE_ALL was specified and the kernel was unable to move
189 all existing pages in the range.
190 ENOMEM Insufficient kernel memory was available.
192 EPERM The flags argument included the MPOL_MF_MOVE_ALL flag and the
194 caller does not have the CAP_SYS_NICE privilege.
195
197 Linux.
198
200 Linux 2.6.7.
201 Support for huge page policy was added with Linux 2.6.16. For inter‐
203 leave policy to be effective on huge page mappings the policied memory
204 needs to be tens of megabytes or larger.
205 Before Linux 5.7. MPOL_MF_STRICT was ignored on huge page mappings.
207 MPOL_MF_MOVE and MPOL_MF_MOVE_ALL are available only on Linux 2.6.16
209 and later.
210
212 For information on library support, see numa(7).
213 NUMA policy is not supported on a memory-mapped file range that was
215 mapped with the MAP_SHARED flag.
216 The MPOL_DEFAULT mode can have different effects for mbind() and
218 set_mempolicy(2). When MPOL_DEFAULT is specified for set_mempolicy(2),
219 the thread's memory policy reverts to the system default policy or lo‐
220 cal allocation. When MPOL_DEFAULT is specified for a range of memory
221 using mbind(), any pages subsequently allocated for that range will use
222 the thread's memory policy, as set by set_mempolicy(2). This effec‐
223 tively removes the explicit policy from the specified range, "falling
224 back" to a possibly nondefault policy. To select explicit "local allo‐
225 cation" for a memory range, specify a mode of MPOL_LOCAL or MPOL_PRE‐
226 FERRED with an empty set of nodes. This method will work for set_mem‐
227 policy(2), as well.
228
230 get_mempolicy(2), getcpu(2), mmap(2), set_mempolicy(2), shmat(2),
231 shmget(2), numa(3), cpuset(7), numa(7), numactl(8)
232Linux man-pages 6.04 2023-03-30 mbind(2)