1
2The intent of this file is to give a brief summary of hugetlbpage support in
3the Linux kernel.  This support is built on top of multiple page size support
4that is provided by most modern architectures.  For example, x86 CPUs normally
5support 4K and 2M (1G if architecturally supported) page sizes, ia64
6architecture supports multiple page sizes 4K, 8K, 64K, 256K, 1M, 4M, 16M,
7256M and ppc64 supports 4K and 16M.  A TLB is a cache of virtual-to-physical
8translations.  Typically this is a very scarce resource on processor.
9Operating systems try to make best use of limited number of TLB resources.
10This optimization is more critical now as bigger and bigger physical memories
11(several GBs) are more readily available.
12
13Users can use the huge page support in Linux kernel by either using the mmap
14system call or standard SYSV shared memory system calls (shmget, shmat).
15
16First the Linux kernel needs to be built with the CONFIG_HUGETLBFS
17(present under "File systems") and CONFIG_HUGETLB_PAGE (selected
18automatically when CONFIG_HUGETLBFS is selected) configuration
19options.
20
21The /proc/meminfo file provides information about the total number of
22persistent hugetlb pages in the kernel's huge page pool.  It also displays
23information about the number of free, reserved and surplus huge pages and the
24default huge page size.  The huge page size is needed for generating the
25proper alignment and size of the arguments to system calls that map huge page
26regions.
27
28The output of "cat /proc/meminfo" will include lines like:
29
30.....
31HugePages_Total: vvv
32HugePages_Free:  www
33HugePages_Rsvd:  xxx
34HugePages_Surp:  yyy
35Hugepagesize:    zzz kB
36
37where:
38HugePages_Total is the size of the pool of huge pages.
39HugePages_Free  is the number of huge pages in the pool that are not yet
40                allocated.
41HugePages_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.
46HugePages_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
52in the kernel.
53
54/proc/sys/vm/nr_hugepages indicates the current number of "persistent" huge
55pages in the kernel's huge page pool.  "Persistent" huge pages will be
56returned to the huge page pool when freed by a task.  A user with root
57privileges can dynamically allocate more or free some persistent huge pages
58by increasing or decreasing the value of 'nr_hugepages'.
59
60Pages that are used as huge pages are reserved inside the kernel and cannot
61be used for other purposes.  Huge pages cannot be swapped out under
62memory pressure.
63
64Once a number of huge pages have been pre-allocated to the kernel huge page
65pool, a user with appropriate privilege can use either the mmap system call
66or shared memory system calls to use the huge pages.  See the discussion of
67Using Huge Pages, below.
68
69The administrator can allocate persistent huge pages on the kernel boot
70command line by specifying the "hugepages=N" parameter, where 'N' = the
71number of huge pages requested.  This is the most reliable method of
72allocating huge pages as memory has not yet become fragmented.
73
74Some platforms support multiple huge page sizes.  To allocate huge pages
75of a specific size, one must precede the huge pages boot command parameters
76with a huge page size selection parameter "hugepagesz=<size>".  <size> must
77be specified in bytes with optional scale suffix [kKmMgG].  The default huge
78page size may be selected with the "default_hugepagesz=<size>" boot parameter.
79
80When multiple huge page sizes are supported, /proc/sys/vm/nr_hugepages
81indicates the current number of pre-allocated huge pages of the default size.
82Thus, one can use the following command to dynamically allocate/deallocate
83default sized persistent huge pages:
84
85	echo 20 > /proc/sys/vm/nr_hugepages
86
87This command will try to adjust the number of default sized huge pages in the
88huge page pool to 20, allocating or freeing huge pages, as required.
89
90On a NUMA platform, the kernel will attempt to distribute the huge page pool
91over all the set of allowed nodes specified by the NUMA memory policy of the
92task that modifies nr_hugepages.  The default for the allowed nodes--when the
93task has default memory policy--is all on-line nodes with memory.  Allowed
94nodes with insufficient available, contiguous memory for a huge page will be
95silently skipped when allocating persistent huge pages.  See the discussion
96below of the interaction of task memory policy, cpusets and per node attributes
97with the allocation and freeing of persistent huge pages.
98
99The success or failure of huge page allocation depends on the amount of
100physically contiguous memory that is present in system at the time of the
101allocation attempt.  If the kernel is unable to allocate huge pages from
102some nodes in a NUMA system, it will attempt to make up the difference by
103allocating extra pages on other nodes with sufficient available contiguous
104memory, if any.
105
106System administrators may want to put this command in one of the local rc
107init files.  This will enable the kernel to allocate huge pages early in
108the boot process when the possibility of getting physical contiguous pages
109is still very high.  Administrators can verify the number of huge pages
110actually allocated by checking the sysctl or meminfo.  To check the per node
111distribution 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
116huge pages can grow, if more huge pages than /proc/sys/vm/nr_hugepages are
117requested by applications.  Writing any non-zero value into this file
118indicates that the hugetlb subsystem is allowed to try to obtain that
119number of "surplus" huge pages from the kernel's normal page pool, when the
120persistent huge page pool is exhausted. As these surplus huge pages become
121unused, they are freed back to the kernel's normal page pool.
122
123When increasing the huge page pool size via nr_hugepages, any existing surplus
124pages will first be promoted to persistent huge pages.  Then, additional
125huge pages will be allocated, if necessary and if possible, to fulfill
126the new persistent huge page pool size.
127
128The administrator may shrink the pool of persistent huge pages for
129the default huge page size by setting the nr_hugepages sysctl to a
130smaller value.  The kernel will attempt to balance the freeing of huge pages
131across all nodes in the memory policy of the task modifying nr_hugepages.
132Any free huge pages on the selected nodes will be freed back to the kernel's
133normal page pool.
134
135Caveat: Shrinking the persistent huge page pool via nr_hugepages such that
136it becomes less than the number of huge pages in use will convert the balance
137of the in-use huge pages to surplus huge pages.  This will occur even if
138the number of surplus pages it would exceed the overcommit value.  As long as
139this condition holds--that is, until nr_hugepages+nr_overcommit_hugepages is
140increased sufficiently, or the surplus huge pages go out of use and are freed--
141no more surplus huge pages will be allowed to be allocated.
142
143With support for multiple huge page pools at run-time available, much of
144the huge page userspace interface in /proc/sys/vm has been duplicated in sysfs.
145The /proc interfaces discussed above have been retained for backwards
146compatibility. The root huge page control directory in sysfs is:
147
148	/sys/kernel/mm/hugepages
149
150For each huge page size supported by the running kernel, a subdirectory
151will exist, of the form:
152
153	hugepages-${size}kB
154
155Inside 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
164which function as described above for the default huge page-sized case.
165
166
167Interaction of Task Memory Policy with Huge Page Allocation/Freeing
168===================================================================
169
170Whether huge pages are allocated and freed via the /proc interface or
171the /sysfs interface using the nr_hugepages_mempolicy attribute, the NUMA
172nodes from which huge pages are allocated or freed are controlled by the
173NUMA memory policy of the task that modifies the nr_hugepages_mempolicy
174sysctl or attribute.  When the nr_hugepages attribute is used, mempolicy
175is ignored.
176
177The recommended method to allocate or free huge pages to/from the kernel
178huge page pool, using the nr_hugepages example above, is:
179
180    numactl --interleave <node-list> echo 20 \
181				>/proc/sys/vm/nr_hugepages_mempolicy
182
183or, more succinctly:
184
185    numactl -m <node-list> echo 20 >/proc/sys/vm/nr_hugepages_mempolicy
186
187This will allocate or free abs(20 - nr_hugepages) to or from the nodes
188specified in <node-list>, depending on whether number of persistent huge pages
189is initially less than or greater than 20, respectively.  No huge pages will be
190allocated nor freed on any node not included in the specified <node-list>.
191
192When adjusting the persistent hugepage count via nr_hugepages_mempolicy, any
193memory policy mode--bind, preferred, local or interleave--may be used.  The
194resulting effect on persistent huge page allocation is as follows:
195
1961) Regardless of mempolicy mode [see Documentation/vm/numa_memory_policy.txt],
197   persistent huge pages will be distributed across the node or nodes
198   specified in the mempolicy as if "interleave" had been specified.
199   However, if a node in the policy does not contain sufficient contiguous
200   memory for a huge page, the allocation will not "fallback" to the nearest
201   neighbor node with sufficient contiguous memory.  To do this would cause
202   undesirable imbalance in the distribution of the huge page pool, or
203   possibly, allocation of persistent huge pages on nodes not allowed by
204   the task's memory policy.
205
2062) One or more nodes may be specified with the bind or interleave policy.
207   If more than one node is specified with the preferred policy, only the
208   lowest numeric id will be used.  Local policy will select the node where
209   the task is running at the time the nodes_allowed mask is constructed.
210   For local policy to be deterministic, the task must be bound to a cpu or
211   cpus in a single node.  Otherwise, the task could be migrated to some
212   other node at any time after launch and the resulting node will be
213   indeterminate.  Thus, local policy is not very useful for this purpose.
214   Any of the other mempolicy modes may be used to specify a single node.
215
2163) The nodes allowed mask will be derived from any non-default task mempolicy,
217   whether this policy was set explicitly by the task itself or one of its
218   ancestors, such as numactl.  This means that if the task is invoked from a
219   shell with non-default policy, that policy will be used.  One can specify a
220   node list of "all" with numactl --interleave or --membind [-m] to achieve
221   interleaving over all nodes in the system or cpuset.
222
2234) Any task mempolicy specifed--e.g., using numactl--will be constrained by
224   the resource limits of any cpuset in which the task runs.  Thus, there will
225   be no way for a task with non-default policy running in a cpuset with a
226   subset of the system nodes to allocate huge pages outside the cpuset
227   without first moving to a cpuset that contains all of the desired nodes.
228
2295) Boot-time huge page allocation attempts to distribute the requested number
230   of huge pages over all on-lines nodes with memory.
231
232Per Node Hugepages Attributes
233=============================
234
235A subset of the contents of the root huge page control directory in sysfs,
236described above, will be replicated under each the system device of each
237NUMA node with memory in:
238
239	/sys/devices/system/node/node[0-9]*/hugepages/
240
241Under this directory, the subdirectory for each supported huge page size
242contains the following attribute files:
243
244	nr_hugepages
245	free_hugepages
246	surplus_hugepages
247
248The free_' and surplus_' attribute files are read-only.  They return the number
249of free and surplus [overcommitted] huge pages, respectively, on the parent
250node.
251
252The nr_hugepages attribute returns the total number of huge pages on the
253specified node.  When this attribute is written, the number of persistent huge
254pages on the parent node will be adjusted to the specified value, if sufficient
255resources exist, regardless of the task's mempolicy or cpuset constraints.
256
257Note that the number of overcommit and reserve pages remain global quantities,
258as we don't know until fault time, when the faulting task's mempolicy is
259applied, from which node the huge page allocation will be attempted.
260
261
262Using Huge Pages
263================
264
265If the user applications are going to request huge pages using mmap system
266call, then it is required that system administrator mount a file system of
267type hugetlbfs:
268
269  mount -t hugetlbfs \
270	-o uid=<value>,gid=<value>,mode=<value>,pagesize=<value>,size=<value>,\
271	min_size=<value>,nr_inodes=<value> none /mnt/huge
272
273This command mounts a (pseudo) filesystem of type hugetlbfs on the directory
274/mnt/huge.  Any files created on /mnt/huge uses huge pages.  The uid and gid
275options sets the owner and group of the root of the file system.  By default
276the uid and gid of the current process are taken.  The mode option sets the
277mode of root of file system to value & 01777.  This value is given in octal.
278By default the value 0755 is picked. If the paltform supports multiple huge
279page sizes, the pagesize option can be used to specify the huge page size and
280associated pool.  pagesize is specified in bytes.  If pagesize is not specified
281the paltform's default huge page size and associated pool will be used. The
282size option sets the maximum value of memory (huge pages) allowed for that
283filesystem (/mnt/huge).  The size option can be specified in bytes, or as a
284percentage of the specified huge page pool (nr_hugepages).  The size is
285rounded down to HPAGE_SIZE boundary.  The min_size option sets the minimum
286value of memory (huge pages) allowed for the filesystem.  min_size can be
287specified in the same way as size, either bytes or a percentage of the
288huge page pool.  At mount time, the number of huge pages specified by
289min_size are reserved for use by the filesystem.  If there are not enough
290free huge pages available, the mount will fail.  As huge pages are allocated
291to the filesystem and freed, the reserve count is adjusted so that the sum
292of allocated and reserved huge pages is always at least min_size.  The option
293nr_inodes sets the maximum number of inodes that /mnt/huge can use.  If the
294size, min_size or nr_inodes option is not provided on command line then
295no limits are set.  For pagesize, size, min_size and nr_inodes options, you
296can use [G|g]/[M|m]/[K|k] to represent giga/mega/kilo. For example, size=2K
297has the same meaning as size=2048.
298
299While read system calls are supported on files that reside on hugetlb
300file systems, write system calls are not.
301
302Regular chown, chgrp, and chmod commands (with right permissions) could be
303used to change the file attributes on hugetlbfs.
304
305Also, it is important to note that no such mount command is required if
306applications are going to use only shmat/shmget system calls or mmap with
307MAP_HUGETLB.  For an example of how to use mmap with MAP_HUGETLB see map_hugetlb
308below.
309
310Users who wish to use hugetlb memory via shared memory segment should be a
311member of a supplementary group and system admin needs to configure that gid
312into /proc/sys/vm/hugetlb_shm_group.  It is possible for same or different
313applications to use any combination of mmaps and shm* calls, though the mount of
314filesystem will be required for using mmap calls without MAP_HUGETLB.
315
316Syscalls that operate on memory backed by hugetlb pages only have their lengths
317aligned to the native page size of the processor; they will normally fail with
318errno set to EINVAL or exclude hugetlb pages that extend beyond the length if
319not hugepage aligned.  For example, munmap(2) will fail if memory is backed by
320a hugetlb page and the length is smaller than the hugepage size.
321
322
323Examples
324========
325
3261) map_hugetlb: see tools/testing/selftests/vm/map_hugetlb.c
327
3282) hugepage-shm:  see tools/testing/selftests/vm/hugepage-shm.c
329
3303) hugepage-mmap:  see tools/testing/selftests/vm/hugepage-mmap.c
331
3324) The libhugetlbfs (http://libhugetlbfs.sourceforge.net) library provides a
333   wide range of userspace tools to help with huge page usability, environment
334   setup, and control. Furthermore it provides useful test cases that should be
335   used when modifying code to ensure no regressions are introduced.
336