10 The intent of this file is to give a brief summary of hugetlbpage support in
11 the Linux kernel. This support is built on top of multiple page size support
12 that is provided by most modern architectures. For example, x86 CPUs normally
13 support 4K and 2M (1G if architecturally supported) page sizes, ia64
14 architecture supports multiple page sizes 4K, 8K, 64K, 256K, 1M, 4M, 16M,
15 256M and ppc64 supports 4K and 16M. A TLB is a cache of virtual-to-physical
16 translations. Typically this is a very scarce resource on processor.
17 Operating systems try to make best use of limited number of TLB resources.
18 This optimization is more critical now as bigger and bigger physical memories
19 (several GBs) are more readily available.
21 Users can use the huge page support in Linux kernel by either using the mmap
22 system call or standard SYSV shared memory system calls (shmget, shmat).
24 First the Linux kernel needs to be built with the CONFIG_HUGETLBFS
25 (present under "File systems") and CONFIG_HUGETLB_PAGE (selected
26 automatically when CONFIG_HUGETLBFS is selected) configuration
29 The ``/proc/meminfo`` file provides information about the total number of
30 persistent hugetlb pages in the kernel's huge page pool. It also displays
31 default huge page size and information about the number of free, reserved
32 and surplus huge pages in the pool of huge pages of default size.
33 The huge page size is needed for generating the proper alignment and
34 size of the arguments to system calls that map huge page regions.
36 The output of ``cat /proc/meminfo`` will include lines like::
48 is the size of the pool of huge pages.
50 is the number of huge pages in the pool that are not yet
53 is short for "reserved," and is the number of huge pages for
54 which a commitment to allocate from the pool has been made,
55 but no allocation has yet been made. Reserved huge pages
56 guarantee that an application will be able to allocate a
57 huge page from the pool of huge pages at fault time.
59 is short for "surplus," and is the number of huge pages in
60 the pool above the value in ``/proc/sys/vm/nr_hugepages``. The
61 maximum number of surplus huge pages is controlled by
62 ``/proc/sys/vm/nr_overcommit_hugepages``.
63 Note: When the feature of freeing unused vmemmap pages associated
64 with each hugetlb page is enabled, the number of surplus huge pages
65 may be temporarily larger than the maximum number of surplus huge
66 pages when the system is under memory pressure.
68 is the default hugepage size (in kB).
70 is the total amount of memory (in kB), consumed by huge
72 If huge pages of different sizes are in use, this number
73 will exceed HugePages_Total \* Hugepagesize. To get more
74 detailed information, please, refer to
75 ``/sys/kernel/mm/hugepages`` (described below).
78 ``/proc/filesystems`` should also show a filesystem of type "hugetlbfs"
79 configured in the kernel.
81 ``/proc/sys/vm/nr_hugepages`` indicates the current number of "persistent" huge
82 pages in the kernel's huge page pool. "Persistent" huge pages will be
83 returned to the huge page pool when freed by a task. A user with root
84 privileges can dynamically allocate more or free some persistent huge pages
85 by increasing or decreasing the value of ``nr_hugepages``.
87 Note: When the feature of freeing unused vmemmap pages associated with each
88 hugetlb page is enabled, we can fail to free the huge pages triggered by
89 the user when ths system is under memory pressure. Please try again later.
91 Pages that are used as huge pages are reserved inside the kernel and cannot
92 be used for other purposes. Huge pages cannot be swapped out under
95 Once a number of huge pages have been pre-allocated to the kernel huge page
96 pool, a user with appropriate privilege can use either the mmap system call
97 or shared memory system calls to use the huge pages. See the discussion of
98 :ref:`Using Huge Pages <using_huge_pages>`, below.
100 The administrator can allocate persistent huge pages on the kernel boot
101 command line by specifying the "hugepages=N" parameter, where 'N' = the
102 number of huge pages requested. This is the most reliable method of
103 allocating huge pages as memory has not yet become fragmented.
105 Some platforms support multiple huge page sizes. To allocate huge pages
106 of a specific size, one must precede the huge pages boot command parameters
107 with a huge page size selection parameter "hugepagesz=<size>". <size> must
108 be specified in bytes with optional scale suffix [kKmMgG]. The default huge
109 page size may be selected with the "default_hugepagesz=<size>" boot parameter.
111 Hugetlb boot command line parameter semantics
114 Specify a huge page size. Used in conjunction with hugepages
115 parameter to preallocate a number of huge pages of the specified
116 size. Hence, hugepagesz and hugepages are typically specified in
119 hugepagesz=2M hugepages=512
121 hugepagesz can only be specified once on the command line for a
122 specific huge page size. Valid huge page sizes are architecture
125 Specify the number of huge pages to preallocate. This typically
126 follows a valid hugepagesz or default_hugepagesz parameter. However,
127 if hugepages is the first or only hugetlb command line parameter it
128 implicitly specifies the number of huge pages of default size to
129 allocate. If the number of huge pages of default size is implicitly
130 specified, it can not be overwritten by a hugepagesz,hugepages
131 parameter pair for the default size. This parameter also has a
132 node format. The node format specifies the number of huge pages
133 to allocate on specific nodes.
135 For example, on an architecture with 2M default huge page size::
137 hugepages=256 hugepagesz=2M hugepages=512
139 will result in 256 2M huge pages being allocated and a warning message
140 indicating that the hugepages=512 parameter is ignored. If a hugepages
141 parameter is preceded by an invalid hugepagesz parameter, it will
144 Node format example::
146 hugepagesz=2M hugepages=0:1,1:2
148 It will allocate 1 2M hugepage on node0 and 2 2M hugepages on node1.
149 If the node number is invalid, the parameter will be ignored.
152 Specify the default huge page size. This parameter can
153 only be specified once on the command line. default_hugepagesz can
154 optionally be followed by the hugepages parameter to preallocate a
155 specific number of huge pages of default size. The number of default
156 sized huge pages to preallocate can also be implicitly specified as
157 mentioned in the hugepages section above. Therefore, on an
158 architecture with 2M default huge page size::
161 default_hugepagesz=2M hugepages=256
162 hugepages=256 default_hugepagesz=2M
164 will all result in 256 2M huge pages being allocated. Valid default
165 huge page size is architecture dependent.
167 When CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP is set, this enables HugeTLB
168 Vmemmap Optimization (HVO).
170 When multiple huge page sizes are supported, ``/proc/sys/vm/nr_hugepages``
171 indicates the current number of pre-allocated huge pages of the default size.
172 Thus, one can use the following command to dynamically allocate/deallocate
173 default sized persistent huge pages::
175 echo 20 > /proc/sys/vm/nr_hugepages
177 This command will try to adjust the number of default sized huge pages in the
178 huge page pool to 20, allocating or freeing huge pages, as required.
180 On a NUMA platform, the kernel will attempt to distribute the huge page pool
181 over all the set of allowed nodes specified by the NUMA memory policy of the
182 task that modifies ``nr_hugepages``. The default for the allowed nodes--when the
183 task has default memory policy--is all on-line nodes with memory. Allowed
184 nodes with insufficient available, contiguous memory for a huge page will be
185 silently skipped when allocating persistent huge pages. See the
186 :ref:`discussion below <mem_policy_and_hp_alloc>`
187 of the interaction of task memory policy, cpusets and per node attributes
188 with the allocation and freeing of persistent huge pages.
190 The success or failure of huge page allocation depends on the amount of
191 physically contiguous memory that is present in system at the time of the
192 allocation attempt. If the kernel is unable to allocate huge pages from
193 some nodes in a NUMA system, it will attempt to make up the difference by
194 allocating extra pages on other nodes with sufficient available contiguous
197 System administrators may want to put this command in one of the local rc
198 init files. This will enable the kernel to allocate huge pages early in
199 the boot process when the possibility of getting physical contiguous pages
200 is still very high. Administrators can verify the number of huge pages
201 actually allocated by checking the sysctl or meminfo. To check the per node
202 distribution of huge pages in a NUMA system, use::
204 cat /sys/devices/system/node/node*/meminfo | fgrep Huge
206 ``/proc/sys/vm/nr_overcommit_hugepages`` specifies how large the pool of
207 huge pages can grow, if more huge pages than ``/proc/sys/vm/nr_hugepages`` are
208 requested by applications. Writing any non-zero value into this file
209 indicates that the hugetlb subsystem is allowed to try to obtain that
210 number of "surplus" huge pages from the kernel's normal page pool, when the
211 persistent huge page pool is exhausted. As these surplus huge pages become
212 unused, they are freed back to the kernel's normal page pool.
214 When increasing the huge page pool size via ``nr_hugepages``, any existing
215 surplus pages will first be promoted to persistent huge pages. Then, additional
216 huge pages will be allocated, if necessary and if possible, to fulfill
217 the new persistent huge page pool size.
219 The administrator may shrink the pool of persistent huge pages for
220 the default huge page size by setting the ``nr_hugepages`` sysctl to a
221 smaller value. The kernel will attempt to balance the freeing of huge pages
222 across all nodes in the memory policy of the task modifying ``nr_hugepages``.
223 Any free huge pages on the selected nodes will be freed back to the kernel's
226 Caveat: Shrinking the persistent huge page pool via ``nr_hugepages`` such that
227 it becomes less than the number of huge pages in use will convert the balance
228 of the in-use huge pages to surplus huge pages. This will occur even if
229 the number of surplus pages would exceed the overcommit value. As long as
230 this condition holds--that is, until ``nr_hugepages+nr_overcommit_hugepages`` is
231 increased sufficiently, or the surplus huge pages go out of use and are freed--
232 no more surplus huge pages will be allowed to be allocated.
234 With support for multiple huge page pools at run-time available, much of
235 the huge page userspace interface in ``/proc/sys/vm`` has been duplicated in
237 The ``/proc`` interfaces discussed above have been retained for backwards
238 compatibility. The root huge page control directory in sysfs is::
240 /sys/kernel/mm/hugepages
242 For each huge page size supported by the running kernel, a subdirectory
243 will exist, of the form::
247 Inside each of these directories, the set of files contained in ``/proc``
248 will exist. In addition, two additional interfaces for demoting huge
254 nr_hugepages_mempolicy
255 nr_overcommit_hugepages
260 The demote interfaces provide the ability to split a huge page into
261 smaller huge pages. For example, the x86 architecture supports both
262 1GB and 2MB huge pages sizes. A 1GB huge page can be split into 512
263 2MB huge pages. Demote interfaces are not available for the smallest
264 huge page size. The demote interfaces are:
267 is the size of demoted pages. When a page is demoted a corresponding
268 number of huge pages of demote_size will be created. By default,
269 demote_size is set to the next smaller huge page size. If there are
270 multiple smaller huge page sizes, demote_size can be set to any of
271 these smaller sizes. Only huge page sizes less than the current huge
272 pages size are allowed.
275 is used to demote a number of huge pages. A user with root privileges
276 can write to this file. It may not be possible to demote the
277 requested number of huge pages. To determine how many pages were
278 actually demoted, compare the value of nr_hugepages before and after
279 writing to the demote interface. demote is a write only interface.
281 The interfaces which are the same as in ``/proc`` (all except demote and
282 demote_size) function as described above for the default huge page-sized case.
284 .. _mem_policy_and_hp_alloc:
286 Interaction of Task Memory Policy with Huge Page Allocation/Freeing
287 ===================================================================
289 Whether huge pages are allocated and freed via the ``/proc`` interface or
290 the ``/sysfs`` interface using the ``nr_hugepages_mempolicy`` attribute, the
291 NUMA nodes from which huge pages are allocated or freed are controlled by the
292 NUMA memory policy of the task that modifies the ``nr_hugepages_mempolicy``
293 sysctl or attribute. When the ``nr_hugepages`` attribute is used, mempolicy
296 The recommended method to allocate or free huge pages to/from the kernel
297 huge page pool, using the ``nr_hugepages`` example above, is::
299 numactl --interleave <node-list> echo 20 \
300 >/proc/sys/vm/nr_hugepages_mempolicy
302 or, more succinctly::
304 numactl -m <node-list> echo 20 >/proc/sys/vm/nr_hugepages_mempolicy
306 This will allocate or free ``abs(20 - nr_hugepages)`` to or from the nodes
307 specified in <node-list>, depending on whether number of persistent huge pages
308 is initially less than or greater than 20, respectively. No huge pages will be
309 allocated nor freed on any node not included in the specified <node-list>.
311 When adjusting the persistent hugepage count via ``nr_hugepages_mempolicy``, any
312 memory policy mode--bind, preferred, local or interleave--may be used. The
313 resulting effect on persistent huge page allocation is as follows:
315 #. Regardless of mempolicy mode [see
316 :ref:`Documentation/admin-guide/mm/numa_memory_policy.rst <numa_memory_policy>`],
317 persistent huge pages will be distributed across the node or nodes
318 specified in the mempolicy as if "interleave" had been specified.
319 However, if a node in the policy does not contain sufficient contiguous
320 memory for a huge page, the allocation will not "fallback" to the nearest
321 neighbor node with sufficient contiguous memory. To do this would cause
322 undesirable imbalance in the distribution of the huge page pool, or
323 possibly, allocation of persistent huge pages on nodes not allowed by
324 the task's memory policy.
326 #. One or more nodes may be specified with the bind or interleave policy.
327 If more than one node is specified with the preferred policy, only the
328 lowest numeric id will be used. Local policy will select the node where
329 the task is running at the time the nodes_allowed mask is constructed.
330 For local policy to be deterministic, the task must be bound to a cpu or
331 cpus in a single node. Otherwise, the task could be migrated to some
332 other node at any time after launch and the resulting node will be
333 indeterminate. Thus, local policy is not very useful for this purpose.
334 Any of the other mempolicy modes may be used to specify a single node.
336 #. The nodes allowed mask will be derived from any non-default task mempolicy,
337 whether this policy was set explicitly by the task itself or one of its
338 ancestors, such as numactl. This means that if the task is invoked from a
339 shell with non-default policy, that policy will be used. One can specify a
340 node list of "all" with numactl --interleave or --membind [-m] to achieve
341 interleaving over all nodes in the system or cpuset.
343 #. Any task mempolicy specified--e.g., using numactl--will be constrained by
344 the resource limits of any cpuset in which the task runs. Thus, there will
345 be no way for a task with non-default policy running in a cpuset with a
346 subset of the system nodes to allocate huge pages outside the cpuset
347 without first moving to a cpuset that contains all of the desired nodes.
349 #. Boot-time huge page allocation attempts to distribute the requested number
350 of huge pages over all on-lines nodes with memory.
352 Per Node Hugepages Attributes
353 =============================
355 A subset of the contents of the root huge page control directory in sysfs,
356 described above, will be replicated under each the system device of each
357 NUMA node with memory in::
359 /sys/devices/system/node/node[0-9]*/hugepages/
361 Under this directory, the subdirectory for each supported huge page size
362 contains the following attribute files::
368 The free\_' and surplus\_' attribute files are read-only. They return the number
369 of free and surplus [overcommitted] huge pages, respectively, on the parent
372 The ``nr_hugepages`` attribute returns the total number of huge pages on the
373 specified node. When this attribute is written, the number of persistent huge
374 pages on the parent node will be adjusted to the specified value, if sufficient
375 resources exist, regardless of the task's mempolicy or cpuset constraints.
377 Note that the number of overcommit and reserve pages remain global quantities,
378 as we don't know until fault time, when the faulting task's mempolicy is
379 applied, from which node the huge page allocation will be attempted.
381 .. _using_huge_pages:
386 If the user applications are going to request huge pages using mmap system
387 call, then it is required that system administrator mount a file system of
391 -o uid=<value>,gid=<value>,mode=<value>,pagesize=<value>,size=<value>,\
392 min_size=<value>,nr_inodes=<value> none /mnt/huge
394 This command mounts a (pseudo) filesystem of type hugetlbfs on the directory
395 ``/mnt/huge``. Any file created on ``/mnt/huge`` uses huge pages.
397 The ``uid`` and ``gid`` options sets the owner and group of the root of the
398 file system. By default the ``uid`` and ``gid`` of the current process
401 The ``mode`` option sets the mode of root of file system to value & 01777.
402 This value is given in octal. By default the value 0755 is picked.
404 If the platform supports multiple huge page sizes, the ``pagesize`` option can
405 be used to specify the huge page size and associated pool. ``pagesize``
406 is specified in bytes. If ``pagesize`` is not specified the platform's
407 default huge page size and associated pool will be used.
409 The ``size`` option sets the maximum value of memory (huge pages) allowed
410 for that filesystem (``/mnt/huge``). The ``size`` option can be specified
411 in bytes, or as a percentage of the specified huge page pool (``nr_hugepages``).
412 The size is rounded down to HPAGE_SIZE boundary.
414 The ``min_size`` option sets the minimum value of memory (huge pages) allowed
415 for the filesystem. ``min_size`` can be specified in the same way as ``size``,
416 either bytes or a percentage of the huge page pool.
417 At mount time, the number of huge pages specified by ``min_size`` are reserved
418 for use by the filesystem.
419 If there are not enough free huge pages available, the mount will fail.
420 As huge pages are allocated to the filesystem and freed, the reserve count
421 is adjusted so that the sum of allocated and reserved huge pages is always
422 at least ``min_size``.
424 The option ``nr_inodes`` sets the maximum number of inodes that ``/mnt/huge``
427 If the ``size``, ``min_size`` or ``nr_inodes`` option is not provided on
428 command line then no limits are set.
430 For ``pagesize``, ``size``, ``min_size`` and ``nr_inodes`` options, you can
431 use [G|g]/[M|m]/[K|k] to represent giga/mega/kilo.
432 For example, size=2K has the same meaning as size=2048.
434 While read system calls are supported on files that reside on hugetlb
435 file systems, write system calls are not.
437 Regular chown, chgrp, and chmod commands (with right permissions) could be
438 used to change the file attributes on hugetlbfs.
440 Also, it is important to note that no such mount command is required if
441 applications are going to use only shmat/shmget system calls or mmap with
442 MAP_HUGETLB. For an example of how to use mmap with MAP_HUGETLB see
443 :ref:`map_hugetlb <map_hugetlb>` below.
445 Users who wish to use hugetlb memory via shared memory segment should be
446 members of a supplementary group and system admin needs to configure that gid
447 into ``/proc/sys/vm/hugetlb_shm_group``. It is possible for same or different
448 applications to use any combination of mmaps and shm* calls, though the mount of
449 filesystem will be required for using mmap calls without MAP_HUGETLB.
451 Syscalls that operate on memory backed by hugetlb pages only have their lengths
452 aligned to the native page size of the processor; they will normally fail with
453 errno set to EINVAL or exclude hugetlb pages that extend beyond the length if
454 not hugepage aligned. For example, munmap(2) will fail if memory is backed by
455 a hugetlb page and the length is smaller than the hugepage size.
464 see tools/testing/selftests/vm/map_hugetlb.c
467 see tools/testing/selftests/vm/hugepage-shm.c
470 see tools/testing/selftests/vm/hugepage-mmap.c
472 The `libhugetlbfs`_ library provides a wide range of userspace tools
473 to help with huge page usability, environment setup, and control.
475 .. _libhugetlbfs: https://github.com/libhugetlbfs/libhugetlbfs