man2/set_mempolicy.2

   1 .\" Copyright 2003,2004 Andi Kleen, SuSE Labs.
   2 .\" and Copyright 2007 Lee Schermerhorn, Hewlett Packard
   3 .\"
   4 .\" Permission is granted to make and distribute verbatim copies of this
   5 .\" manual provided the copyright notice and this permission notice are
   6 .\" preserved on all copies.
   7 .\"
   8 .\" Permission is granted to copy and distribute modified versions of this
   9 .\" manual under the conditions for verbatim copying, provided that the
  10 .\" entire resulting derived work is distributed under the terms of a
  11 .\" permission notice identical to this one.
  12 .\"
  13 .\" Since the Linux kernel and libraries are constantly changing, this
  14 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  15 .\" responsibility for errors or omissions, or for damages resulting from
  16 .\" the use of the information contained herein.
  17 .\"
  18 .\" Formatted or processed versions of this manual, if unaccompanied by
  19 .\" the source, must acknowledge the copyright and authors of this work.
  20 .\"
  21 .\" 2006-02-03, mtk, substantial wording changes and other improvements
  22 .\" 2007-08-27, Lee Schermerhorn <Lee.Schermerhorn@hp.com>
  23 .\"     more precise specification of behavior.
  24 .\"
  25 .TH SET_MEMPOLICY 2 2007-08-27 Linux "Linux Programmer's Manual"
  26 .SH NAME
  27 set_mempolicy \- set default NUMA memory policy for a process and its children
  28 .SH SYNOPSIS
  29 .nf
  30 .B "#include <numaif.h>"
  31 .sp
  32 .BI "int set_mempolicy(int " mode ", unsigned long *" nodemask ,
  33 .BI "                  unsigned long " maxnode );
  34 .sp
  35 Link with \fI\-lnuma\fP
  36 .fi
  37 .SH DESCRIPTION
  38 .BR set_mempolicy ()
  39 sets the NUMA memory policy of the calling process,
  40 which consists of a policy mode and zero or more nodes,
  41 to the values specified by the
  42 .IR mode ,
  43 .I nodemask
  44 and
  45 .I maxnode
  46 arguments.
  47
  48 A NUMA machine has different
  49 memory controllers with different distances to specific CPUs.
  50 The memory policy defines from which node memory is allocated for
  51 the process.
  52
  53 This system call defines the default policy for the process.
  54 The process policy governs allocation of pages in the process'
  55 address space outside of memory ranges
  56 controlled by a more specific policy set by
  57 .BR mbind (2).
  58 The process default policy also controls allocation of any pages for
  59 memory mapped files mapped using the
  60 .BR mmap (2)
  61 call with the
  62 .B MAP_PRIVATE
  63 flag and that are only read [loaded] from by the task
  64 and of memory mapped files mapped using the
  65 .BR mmap (2)
  66 call with the
  67 .B MAP_SHARED
  68 flag, regardless of the access type.
  69 The policy is only applied when a new page is allocated
  70 for the process.
  71 For anonymous memory this is when the page is first
  72 touched by the application.
  73
  74 The
  75 .I mode
  76 argument must specify one of
  77 .BR MPOL_DEFAULT ,
  78 .BR MPOL_BIND ,
  79 .B MPOL_INTERLEAVE
  80 or
  81 .BR MPOL_PREFERRED .
  82 All modes except
  83 .B MPOL_DEFAULT
  84 require the caller to specify via the
  85 .I nodemask
  86 parameter
  87 one or more nodes.
  88
  89 .I nodemask
  90 points to a bit mask of node IDs that contains up to
  91 .I maxnode
  92 bits.
  93 The bit mask size is rounded to the next multiple of
  94 .IR "sizeof(unsigned long)" ,
  95 but the kernel will only use bits up to
  96 .IR maxnode .
  97 A NULL value of
  98 .I nodemask
  99 or a
 100 .I maxnode
 101 value of zero specifies the empty set of nodes.
 102 If the value of
 103 .I maxnode
 104 is zero,
 105 the
 106 .I nodemask
 107 argument is ignored.
 108
 109 The
 110 .B MPOL_DEFAULT
 111 mode is the default and means to allocate memory locally,
 112 i.e., on the node of the CPU that triggered the allocation.
 113 .I nodemask
 114 must be specified as NULL.
 115 If the "local node" contains no free memory, the system will
 116 attempt to allocate memory from a "near by" node.
 117
 118 The
 119 .B MPOL_BIND
 120 mode defines a strict policy that restricts memory allocation to the
 121 nodes specified in
 122 .IR nodemask .
 123 If
 124 .I nodemask
 125 specifies more than one node, page allocations will come from
 126 the node with the lowest numeric node ID first, until that node
 127 contains no free memory.
 128 Allocations will then come from the node with the next highest
 129 node ID specified in
 130 .I nodemask
 131 and so forth, until none of the specified nodes contain free memory.
 132 Pages will not be allocated from any node not specified in the
 133 .IR nodemask .
 134
 135 .B MPOL_INTERLEAVE
 136 interleaves page allocations across the nodes specified in
 137 .I nodemask
 138 in numeric node ID order.
 139 This optimizes for bandwidth instead of latency
 140 by spreading out pages and memory accesses to those pages across
 141 multiple nodes.
 142 However, accesses to a single page will still be limited to
 143 the memory bandwidth of a single node.
 144 .\" NOTE:  the following sentence doesn't make sense in the context
 145 .\" of set_mempolicy() -- no memory area specified.
 146 .\" To be effective the memory area should be fairly large,
 147 .\" at least 1MB or bigger.
 148
 149 .B MPOL_PREFERRED
 150 sets the preferred node for allocation.
 151 The kernel will try to allocate pages from this node first
 152 and fall back to "near by" nodes if the preferred node is low on free
 153 memory.
 154 If
 155 .I nodemask
 156 specifies more than one node ID, the first node in the
 157 mask will be selected as the preferred node.
 158 If the
 159 .I nodemask
 160 and
 161 .I maxnode
 162 arguments specify the empty set, then the memory is allocated on
 163 the node of the CPU that triggered the allocation (like
 164 .BR MPOL_DEFAULT ).
 165
 166 The process memory policy is preserved across an
 167 .BR execve (2),
 168 and is inherited by child processes created using
 169 .BR fork (2)
 170 or
 171 .BR clone (2).
 172 .SH RETURN VALUE
 173 On success,
 174 .BR set_mempolicy ()
 175 returns 0;
 176 on error, \-1 is returned and
 177 .I errno
 178 is set to indicate the error.
 179 .SH ERRORS
 180 .TP
 181 .B EFAULT
 182 Part of all of the memory range specified by
 183 .I nodemask
 184 and
 185 .I maxnode
 186 points outside your accessible address space.
 187 .TP
 188 .B EINVAL
 189 .I mode
 190 is invalid.
 191 Or,
 192 .I mode
 193 is
 194 .B MPOL_DEFAULT
 195 and
 196 .I nodemask
 197 is non-empty,
 198 or
 199 .I mode
 200 is
 201 .B MPOL_BIND
 202 or
 203 .B MPOL_INTERLEAVE
 204 and
 205 .I nodemask
 206 is empty.
 207 Or,
 208 .I maxnode
 209 specifies more than a page worth of bits.
 210 Or,
 211 .I nodemask
 212 specifies one or more node IDs that are
 213 greater than the maximum supported node ID,
 214 or are not allowed in the calling task's context.
 215 .\" "calling task's context" refers to cpusets.
 216 .\" No man page avail to ref. --Lee Schermerhorn
 217 Or, none of the node IDs specified by
 218 .I nodemask
 219 are on-line, or none of the specified nodes contain memory.
 220 .TP
 221 .B ENOMEM
 222 Insufficient kernel memory was available.
 223 .SH CONFORMING TO
 224 This system call is Linux-specific.
 225 .SH NOTES
 226 Process policy is not remembered if the page is swapped out.
 227 When such a page is paged back in, it will use the policy of
 228 the process or memory range that is in effect at the time the
 229 page is allocated.
 230 .SS "Versions and Library Support"
 231 See
 232 .BR mbind (2).
 233 .SH SEE ALSO
 234 .BR mbind (2),
 235 .BR mmap (2),
 236 .BR get_mempolicy (2),
 237 .BR numactl (8),
 238 .BR numa (3)