Documentation/vm/hugetlbpage.txt

   1
   2 The intent of this file is to give a brief summary of hugetlbpage support in
   3 the Linux kernel.  This support is built on top of multiple page size support
   4 that is provided by most modern architectures.  For example, i386
   5 architecture supports 4K and 4M (2M in PAE mode) page sizes, ia64
   6 architecture supports multiple page sizes 4K, 8K, 64K, 256K, 1M, 4M, 16M,
   7 256M and ppc64 supports 4K and 16M.  A TLB is a cache of virtual-to-physical
   8 translations.  Typically this is a very scarce resource on processor.
   9 Operating systems try to make best use of limited number of TLB resources.
  10 This optimization is more critical now as bigger and bigger physical memories
  11 (several GBs) are more readily available.
  12
  13 Users can use the huge page support in Linux kernel by either using the mmap
  14 system call or standard SYSV shared memory system calls (shmget, shmat).
  15
  16 First the Linux kernel needs to be built with the CONFIG_HUGETLBFS
  17 (present under "File systems") and CONFIG_HUGETLB_PAGE (selected
  18 automatically when CONFIG_HUGETLBFS is selected) configuration
  19 options.
  20
  21 The /proc/meminfo file provides information about the total number of
  22 persistent hugetlb pages in the kernel's huge page pool.  It also displays
  23 information about the number of free, reserved and surplus huge pages and the
  24 default huge page size.  The huge page size is needed for generating the
  25 proper alignment and size of the arguments to system calls that map huge page
  26 regions.
  27
  28 The output of "cat /proc/meminfo" will include lines like:
  29
  30 .....
  31 HugePages_Total: vvv
  32 HugePages_Free:  www
  33 HugePages_Rsvd:  xxx
  34 HugePages_Surp:  yyy
  35 Hugepagesize:    zzz kB
  36
  37 where:
  38 HugePages_Total is the size of the pool of huge pages.
  39 HugePages_Free  is the number of huge pages in the pool that are not yet
  40                 allocated.
  41 HugePages_Rsvd  is short for "reserved," and is the number of huge pages for
  42                 which a commitment to allocate from the pool has been made,
  43                 but no allocation has yet been made.  Reserved huge pages
  44                 guarantee that an application will be able to allocate a
  45                 huge page from the pool of huge pages at fault time.
  46 HugePages_Surp  is short for "surplus," and is the number of huge pages in
  47                 the pool above the value in /proc/sys/vm/nr_hugepages. The
  48                 maximum number of surplus huge pages is controlled by
  49                 /proc/sys/vm/nr_overcommit_hugepages.
  50
  51 /proc/filesystems should also show a filesystem of type "hugetlbfs" configured
  52 in the kernel.
  53
  54 /proc/sys/vm/nr_hugepages indicates the current number of "persistent" huge
  55 pages in the kernel's huge page pool.  "Persistent" huge pages will be
  56 returned to the huge page pool when freed by a task.  A user with root
  57 privileges can dynamically allocate more or free some persistent huge pages
  58 by increasing or decreasing the value of 'nr_hugepages'.
  59
  60 Pages that are used as huge pages are reserved inside the kernel and cannot
  61 be used for other purposes.  Huge pages cannot be swapped out under
  62 memory pressure.
  63
  64 Once a number of huge pages have been pre-allocated to the kernel huge page
  65 pool, a user with appropriate privilege can use either the mmap system call
  66 or shared memory system calls to use the huge pages.  See the discussion of
  67 Using Huge Pages, below.
  68
  69 The administrator can allocate persistent huge pages on the kernel boot
  70 command line by specifying the "hugepages=N" parameter, where 'N' = the
  71 number of huge pages requested.  This is the most reliable method of
  72 allocating huge pages as memory has not yet become fragmented.
  73
  74 Some platforms support multiple huge page sizes.  To allocate huge pages
  75 of a specific size, one must precede the huge pages boot command parameters
  76 with a huge page size selection parameter "hugepagesz=<size>".  <size> must
  77 be specified in bytes with optional scale suffix [kKmMgG].  The default huge
  78 page size may be selected with the "default_hugepagesz=<size>" boot parameter.
  79
  80 When multiple huge page sizes are supported, /proc/sys/vm/nr_hugepages
  81 indicates the current number of pre-allocated huge pages of the default size.
  82 Thus, one can use the following command to dynamically allocate/deallocate
  83 default sized persistent huge pages:
  84
  85         echo 20 > /proc/sys/vm/nr_hugepages
  86
  87 This command will try to adjust the number of default sized huge pages in the
  88 huge page pool to 20, allocating or freeing huge pages, as required.
  89
  90 On a NUMA platform, the kernel will attempt to distribute the huge page pool
  91 over all the set of allowed nodes specified by the NUMA memory policy of the
  92 task that modifies nr_hugepages.  The default for the allowed nodes--when the
  93 task has default memory policy--is all on-line nodes with memory.  Allowed
  94 nodes with insufficient available, contiguous memory for a huge page will be
  95 silently skipped when allocating persistent huge pages.  See the discussion
  96 below of the interaction of task memory policy, cpusets and per node attributes
  97 with the allocation and freeing of persistent huge pages.
  98
  99 The success or failure of huge page allocation depends on the amount of
 100 physically contiguous memory that is present in system at the time of the
 101 allocation attempt.  If the kernel is unable to allocate huge pages from
 102 some nodes in a NUMA system, it will attempt to make up the difference by
 103 allocating extra pages on other nodes with sufficient available contiguous
 104 memory, if any.
 105
 106 System administrators may want to put this command in one of the local rc
 107 init files.  This will enable the kernel to allocate huge pages early in
 108 the boot process when the possibility of getting physical contiguous pages
 109 is still very high.  Administrators can verify the number of huge pages
 110 actually allocated by checking the sysctl or meminfo.  To check the per node
 111 distribution of huge pages in a NUMA system, use:
 112
 113         cat /sys/devices/system/node/node*/meminfo | fgrep Huge
 114
 115 /proc/sys/vm/nr_overcommit_hugepages specifies how large the pool of
 116 huge pages can grow, if more huge pages than /proc/sys/vm/nr_hugepages are
 117 requested by applications.  Writing any non-zero value into this file
 118 indicates that the hugetlb subsystem is allowed to try to obtain that
 119 number of "surplus" huge pages from the kernel's normal page pool, when the
 120 persistent huge page pool is exhausted. As these surplus huge pages become
 121 unused, they are freed back to the kernel's normal page pool.
 122
 123 When increasing the huge page pool size via nr_hugepages, any existing surplus
 124 pages will first be promoted to persistent huge pages.  Then, additional
 125 huge pages will be allocated, if necessary and if possible, to fulfill
 126 the new persistent huge page pool size.
 127
 128 The administrator may shrink the pool of persistent huge pages for
 129 the default huge page size by setting the nr_hugepages sysctl to a
 130 smaller value.  The kernel will attempt to balance the freeing of huge pages
 131 across all nodes in the memory policy of the task modifying nr_hugepages.
 132 Any free huge pages on the selected nodes will be freed back to the kernel's
 133 normal page pool.
 134
 135 Caveat: Shrinking the persistent huge page pool via nr_hugepages such that
 136 it becomes less than the number of huge pages in use will convert the balance
 137 of the in-use huge pages to surplus huge pages.  This will occur even if
 138 the number of surplus pages it would exceed the overcommit value.  As long as
 139 this condition holds--that is, until nr_hugepages+nr_overcommit_hugepages is
 140 increased sufficiently, or the surplus huge pages go out of use and are freed--
 141 no more surplus huge pages will be allowed to be allocated.
 142
 143 With support for multiple huge page pools at run-time available, much of
 144 the huge page userspace interface in /proc/sys/vm has been duplicated in sysfs.
 145 The /proc interfaces discussed above have been retained for backwards
 146 compatibility. The root huge page control directory in sysfs is:
 147
 148         /sys/kernel/mm/hugepages
 149
 150 For each huge page size supported by the running kernel, a subdirectory
 151 will exist, of the form:
 152
 153         hugepages-${size}kB
 154
 155 Inside each of these directories, the same set of files will exist:
 156
 157         nr_hugepages
 158         nr_hugepages_mempolicy
 159         nr_overcommit_hugepages
 160         free_hugepages
 161         resv_hugepages
 162         surplus_hugepages
 163
 164 which function as described above for the default huge page-sized case.
 165
 166
 167 Interaction of Task Memory Policy with Huge Page Allocation/Freeing
 168
 169 Whether huge pages are allocated and freed via the /proc interface or
 170 the /sysfs interface using the nr_hugepages_mempolicy attribute, the NUMA
 171 nodes from which huge pages are allocated or freed are controlled by the
 172 NUMA memory policy of the task that modifies the nr_hugepages_mempolicy
 173 sysctl or attribute.  When the nr_hugepages attribute is used, mempolicy
 174 is ignored.
 175
 176 The recommended method to allocate or free huge pages to/from the kernel
 177 huge page pool, using the nr_hugepages example above, is:
 178
 179     numactl --interleave <node-list> echo 20 \
 180                                 >/proc/sys/vm/nr_hugepages_mempolicy
 181
 182 or, more succinctly:
 183
 184     numactl -m <node-list> echo 20 >/proc/sys/vm/nr_hugepages_mempolicy
 185
 186 This will allocate or free abs(20 - nr_hugepages) to or from the nodes
 187 specified in <node-list>, depending on whether number of persistent huge pages
 188 is initially less than or greater than 20, respectively.  No huge pages will be
 189 allocated nor freed on any node not included in the specified <node-list>.
 190
 191 When adjusting the persistent hugepage count via nr_hugepages_mempolicy, any
 192 memory policy mode--bind, preferred, local or interleave--may be used.  The
 193 resulting effect on persistent huge page allocation is as follows:
 194
 195 1) Regardless of mempolicy mode [see Documentation/vm/numa_memory_policy.txt],
 196    persistent huge pages will be distributed across the node or nodes
 197    specified in the mempolicy as if "interleave" had been specified.
 198    However, if a node in the policy does not contain sufficient contiguous
 199    memory for a huge page, the allocation will not "fallback" to the nearest
 200    neighbor node with sufficient contiguous memory.  To do this would cause
 201    undesirable imbalance in the distribution of the huge page pool, or
 202    possibly, allocation of persistent huge pages on nodes not allowed by
 203    the task's memory policy.
 204
 205 2) One or more nodes may be specified with the bind or interleave policy.
 206    If more than one node is specified with the preferred policy, only the
 207    lowest numeric id will be used.  Local policy will select the node where
 208    the task is running at the time the nodes_allowed mask is constructed.
 209    For local policy to be deterministic, the task must be bound to a cpu or
 210    cpus in a single node.  Otherwise, the task could be migrated to some
 211    other node at any time after launch and the resulting node will be
 212    indeterminate.  Thus, local policy is not very useful for this purpose.
 213    Any of the other mempolicy modes may be used to specify a single node.
 214
 215 3) The nodes allowed mask will be derived from any non-default task mempolicy,
 216    whether this policy was set explicitly by the task itself or one of its
 217    ancestors, such as numactl.  This means that if the task is invoked from a
 218    shell with non-default policy, that policy will be used.  One can specify a
 219    node list of "all" with numactl --interleave or --membind [-m] to achieve
 220    interleaving over all nodes in the system or cpuset.
 221
 222 4) Any task mempolicy specifed--e.g., using numactl--will be constrained by
 223    the resource limits of any cpuset in which the task runs.  Thus, there will
 224    be no way for a task with non-default policy running in a cpuset with a
 225    subset of the system nodes to allocate huge pages outside the cpuset
 226    without first moving to a cpuset that contains all of the desired nodes.
 227
 228 5) Boot-time huge page allocation attempts to distribute the requested number
 229    of huge pages over all on-lines nodes with memory.
 230
 231 Per Node Hugepages Attributes
 232
 233 A subset of the contents of the root huge page control directory in sysfs,
 234 described above, will be replicated under each the system device of each
 235 NUMA node with memory in:
 236
 237         /sys/devices/system/node/node[0-9]*/hugepages/
 238
 239 Under this directory, the subdirectory for each supported huge page size
 240 contains the following attribute files:
 241
 242         nr_hugepages
 243         free_hugepages
 244         surplus_hugepages
 245
 246 The free_' and surplus_' attribute files are read-only.  They return the number
 247 of free and surplus [overcommitted] huge pages, respectively, on the parent
 248 node.
 249
 250 The nr_hugepages attribute returns the total number of huge pages on the
 251 specified node.  When this attribute is written, the number of persistent huge
 252 pages on the parent node will be adjusted to the specified value, if sufficient
 253 resources exist, regardless of the task's mempolicy or cpuset constraints.
 254
 255 Note that the number of overcommit and reserve pages remain global quantities,
 256 as we don't know until fault time, when the faulting task's mempolicy is
 257 applied, from which node the huge page allocation will be attempted.
 258
 259
 260 Using Huge Pages
 261
 262 If the user applications are going to request huge pages using mmap system
 263 call, then it is required that system administrator mount a file system of
 264 type hugetlbfs:
 265
 266   mount -t hugetlbfs \
 267         -o uid=<value>,gid=<value>,mode=<value>,size=<value>,nr_inodes=<value> \
 268         none /mnt/huge
 269
 270 This command mounts a (pseudo) filesystem of type hugetlbfs on the directory
 271 /mnt/huge.  Any files created on /mnt/huge uses huge pages.  The uid and gid
 272 options sets the owner and group of the root of the file system.  By default
 273 the uid and gid of the current process are taken.  The mode option sets the
 274 mode of root of file system to value & 0777.  This value is given in octal.
 275 By default the value 0755 is picked. The size option sets the maximum value of
 276 memory (huge pages) allowed for that filesystem (/mnt/huge). The size is
 277 rounded down to HPAGE_SIZE.  The option nr_inodes sets the maximum number of
 278 inodes that /mnt/huge can use.  If the size or nr_inodes option is not
 279 provided on command line then no limits are set.  For size and nr_inodes
 280 options, you can use [G|g]/[M|m]/[K|k] to represent giga/mega/kilo. For
 281 example, size=2K has the same meaning as size=2048.
 282
 283 While read system calls are supported on files that reside on hugetlb
 284 file systems, write system calls are not.
 285
 286 Regular chown, chgrp, and chmod commands (with right permissions) could be
 287 used to change the file attributes on hugetlbfs.
 288
 289 Also, it is important to note that no such mount command is required if the
 290 applications are going to use only shmat/shmget system calls or mmap with
 291 MAP_HUGETLB.  Users who wish to use hugetlb page via shared memory segment
 292 should be a member of a supplementary group and system admin needs to
 293 configure that gid into /proc/sys/vm/hugetlb_shm_group.  It is possible for
 294 same or different applications to use any combination of mmaps and shm*
 295 calls, though the mount of filesystem will be required for using mmap calls
 296 without MAP_HUGETLB.  For an example of how to use mmap with MAP_HUGETLB see
 297 map_hugetlb.c.
 298
 299 *******************************************************************
 300
 301 /*
 302  * hugepage-shm:  see Documentation/vm/hugepage-shm.c
 303  */
 304
 305 *******************************************************************
 306
 307 /*
 308  * hugepage-mmap:  see Documentation/vm/hugepage-mmap.c
 309  */