man2/move_pages.2

   1 .\" This manpage is Copyright (C) 2006 Silicon Graphics, Inc.
   2 .\"                               Christoph Lameter
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM_TWO_PARA)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\" %%%LICENSE_END
  14 .\"
  15 .\" FIXME Should programs normally be using move_pages() directly, or should
  16 .\" they rather be using interfaces in the numactl package?
  17 .\" (e.g., compare with recommendation in mbind(2)).
  18 .\" Does this page need to give advice on this topic?
  19 .\"
  20 .TH move_pages 2 (date) "Linux man-pages (unreleased)"
  21 .SH NAME
  22 move_pages \- move individual pages of a process to another node
  23 .SH LIBRARY
  24 NUMA (Non-Uniform Memory Access) policy library
  25 .RI ( libnuma ", " \-lnuma )
  26 .SH SYNOPSIS
  27 .nf
  28 .B #include <numaif.h>
  29 .PP
  30 .BI "long move_pages(int " pid ", unsigned long " count ", void **" pages ,
  31 .BI "                const int *" nodes ", int *" status ", int " flags );
  32 .fi
  33 .SH DESCRIPTION
  34 .BR move_pages ()
  35 moves the specified
  36 .I pages
  37 of the process
  38 .I pid
  39 to the memory nodes specified by
  40 .IR nodes .
  41 The result of the move is reflected in
  42 .IR status .
  43 The
  44 .I flags
  45 indicate constraints on the pages to be moved.
  46 .PP
  47 .I pid
  48 is the ID of the process in which pages are to be moved.
  49 If
  50 .I pid
  51 is 0, then
  52 .BR move_pages ()
  53 moves pages of the calling process.
  54 .PP
  55 To move pages in another process requires the following privileges:
  56 .IP \(bu 3
  57 In kernels up to and including Linux 4.12:
  58 the caller must be privileged
  59 .RB ( CAP_SYS_NICE )
  60 or the real or effective user ID of the calling process must match the
  61 real or saved-set user ID of the target process.
  62 .IP \(bu
  63 The older rules allowed the caller to discover various
  64 virtual address choices made by the kernel that could lead
  65 to the defeat of address-space-layout randomization
  66 for a process owned by the same UID as the caller,
  67 the rules were changed starting with Linux 4.13.
  68 Since Linux 4.13,
  69 .\" commit 197e7e521384a23b9e585178f3f11c9fa08274b9
  70 permission is governed by a ptrace access mode
  71 .B PTRACE_MODE_READ_REALCREDS
  72 check with respect to the target process; see
  73 .BR ptrace (2).
  74 .PP
  75 .I count
  76 is the number of pages to move.
  77 It defines the size of the three arrays
  78 .IR pages ,
  79 .IR nodes ,
  80 and
  81 .IR status .
  82 .PP
  83 .I pages
  84 is an array of pointers to the pages that should be moved.
  85 These are pointers that should be aligned to page boundaries.
  86 .\" FIXME Describe the result if pointers in the 'pages' array are
  87 .\" not aligned to page boundaries
  88 Addresses are specified as seen by the process specified by
  89 .IR pid .
  90 .PP
  91 .I nodes
  92 is an array of integers that specify the desired location for each page.
  93 Each element in the array is a node number.
  94 .I nodes
  95 can also be NULL, in which case
  96 .BR move_pages ()
  97 does not move any pages but instead will return the node
  98 where each page currently resides, in the
  99 .I status
 100 array.
 101 Obtaining the status of each page may be necessary to determine
 102 pages that need to be moved.
 103 .PP
 104 .I status
 105 is an array of integers that return the status of each page.
 106 The array contains valid values only if
 107 .BR move_pages ()
 108 did not return an error.
 109 Preinitialization of the array to a value
 110 which cannot represent a real numa node or valid error of status array
 111 could help to identify pages that have been migrated.
 112 .PP
 113 .I flags
 114 specify what types of pages to move.
 115 .B MPOL_MF_MOVE
 116 means that only pages that are in exclusive use by the process
 117 are to be moved.
 118 .B MPOL_MF_MOVE_ALL
 119 means that pages shared between multiple processes can also be moved.
 120 The process must be privileged
 121 .RB ( CAP_SYS_NICE )
 122 to use
 123 .BR MPOL_MF_MOVE_ALL .
 124 .SS Page states in the status array
 125 The following values can be returned in each element of the
 126 .I status
 127 array.
 128 .TP
 129 .B 0..MAX_NUMNODES
 130 Identifies the node on which the page resides.
 131 .TP
 132 .B \-EACCES
 133 The page is mapped by multiple processes and can be moved only if
 134 .B MPOL_MF_MOVE_ALL
 135 is specified.
 136 .TP
 137 .B \-EBUSY
 138 The page is currently busy and cannot be moved.
 139 Try again later.
 140 This occurs if a page is undergoing I/O or another kernel subsystem
 141 is holding a reference to the page.
 142 .TP
 143 .B \-EFAULT
 144 This is a zero page or the memory area is not mapped by the process.
 145 .TP
 146 .B \-EIO
 147 Unable to write back a page.
 148 The page has to be written back
 149 in order to move it since the page is dirty and the filesystem
 150 does not provide a migration function that would allow the move
 151 of dirty pages.
 152 .TP
 153 .B \-EINVAL
 154 A dirty page cannot be moved.
 155 The filesystem does not
 156 provide a migration function and has no ability to write back pages.
 157 .TP
 158 .B \-ENOENT
 159 The page is not present.
 160 .TP
 161 .B \-ENOMEM
 162 Unable to allocate memory on target node.
 163 .SH RETURN VALUE
 164 On success
 165 .BR move_pages ()
 166 returns zero.
 167 .\" FIXME . Is the following quite true: does the wrapper in numactl
 168 .\" do the right thing?
 169 On error, it returns \-1, and sets
 170 .I errno
 171 to indicate the error.
 172 If positive value is returned, it is the number of
 173 nonmigrated pages.
 174 .SH ERRORS
 175 .TP
 176 .B Positive value
 177 The number of nonmigrated pages if they were the result of nonfatal
 178 reasons (since
 179 .\" commit a49bd4d7163707de377aee062f17befef6da891b
 180 Linux 4.17).
 181 .TP
 182 .B E2BIG
 183 Too many pages to move.
 184 Since Linux 2.6.29,
 185 .\" commit 3140a2273009c01c27d316f35ab76a37e105fdd8
 186 the kernel no longer generates this error.
 187 .TP
 188 .B EACCES
 189 .\" FIXME Clarify "current cpuset" in the description of the EACCES error.
 190 .\" Is that the cpuset of the caller or the target?
 191 One of the target nodes is not allowed by the current cpuset.
 192 .TP
 193 .B EFAULT
 194 Parameter array could not be accessed.
 195 .TP
 196 .B EINVAL
 197 Flags other than
 198 .B MPOL_MF_MOVE
 199 and
 200 .B MPOL_MF_MOVE_ALL
 201 was specified or an attempt was made to migrate pages of a kernel thread.
 202 .TP
 203 .B ENODEV
 204 One of the target nodes is not online.
 205 .TP
 206 .B EPERM
 207 The caller specified
 208 .B MPOL_MF_MOVE_ALL
 209 without sufficient privileges
 210 .RB ( CAP_SYS_NICE ).
 211 Or, the caller attempted to move pages of a process belonging
 212 to another user but did not have privilege to do so
 213 .RB ( CAP_SYS_NICE ).
 214 .TP
 215 .B ESRCH
 216 Process does not exist.
 217 .SH VERSIONS
 218 .BR move_pages ()
 219 first appeared on Linux in version 2.6.18.
 220 .SH STANDARDS
 221 This system call is Linux-specific.
 222 .SH NOTES
 223 For information on library support, see
 224 .BR numa (7).
 225 .PP
 226 Use
 227 .BR get_mempolicy (2)
 228 with the
 229 .B MPOL_F_MEMS_ALLOWED
 230 flag to obtain the set of nodes that are allowed by
 231 .\" FIXME Clarify "current cpuset".  Is that the cpuset of the caller
 232 .\" or the target?
 233 the current cpuset.
 234 Note that this information is subject to change at any
 235 time by manual or automatic reconfiguration of the cpuset.
 236 .PP
 237 Use of this function may result in pages whose location
 238 (node) violates the memory policy established for the
 239 specified addresses (See
 240 .BR mbind (2))
 241 and/or the specified process (See
 242 .BR set_mempolicy (2)).
 243 That is, memory policy does not constrain the destination
 244 nodes used by
 245 .BR move_pages ().
 246 .PP
 247 The
 248 .I <numaif.h>
 249 header is not included with glibc, but requires installing
 250 .I libnuma\-devel
 251 or a similar package.
 252 .SH SEE ALSO
 253 .BR get_mempolicy (2),
 254 .BR mbind (2),
 255 .BR set_mempolicy (2),
 256 .BR numa (3),
 257 .BR numa_maps (5),
 258 .BR cpuset (7),
 259 .BR numa (7),
 260 .BR migratepages (8),
 261 .BR numastat (8)