About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / vm / hugetlbpage.txt


Based on kernel version 4.16.1. Page generated on 2018-04-09 11:53 EST.

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, x86 CPUs normally
5	support 4K and 2M (1G if architecturally supported) 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	default huge page size and information about the number of free, reserved
24	and surplus huge pages in the pool of huge pages of default size.
25	The huge page size is needed for generating the proper alignment and
26	size of the arguments to system calls that map huge page regions.
27	
28	The output of "cat /proc/meminfo" will include lines like:
29	
30	.....
31	HugePages_Total: uuu
32	HugePages_Free:  vvv
33	HugePages_Rsvd:  www
34	HugePages_Surp:  xxx
35	Hugepagesize:    yyy kB
36	Hugetlb:         zzz kB
37	
38	where:
39	HugePages_Total is the size of the pool of huge pages.
40	HugePages_Free  is the number of huge pages in the pool that are not yet
41	                allocated.
42	HugePages_Rsvd  is short for "reserved," and is the number of huge pages for
43	                which a commitment to allocate from the pool has been made,
44	                but no allocation has yet been made.  Reserved huge pages
45	                guarantee that an application will be able to allocate a
46	                huge page from the pool of huge pages at fault time.
47	HugePages_Surp  is short for "surplus," and is the number of huge pages in
48	                the pool above the value in /proc/sys/vm/nr_hugepages. The
49	                maximum number of surplus huge pages is controlled by
50	                /proc/sys/vm/nr_overcommit_hugepages.
51	Hugepagesize    is the default hugepage size (in Kb).
52	Hugetlb         is the total amount of memory (in kB), consumed by huge
53	                pages of all sizes.
54	                If huge pages of different sizes are in use, this number
55	                will exceed HugePages_Total * Hugepagesize. To get more
56	                detailed information, please, refer to
57	                /sys/kernel/mm/hugepages (described below).
58	
59	
60	/proc/filesystems should also show a filesystem of type "hugetlbfs" configured
61	in the kernel.
62	
63	/proc/sys/vm/nr_hugepages indicates the current number of "persistent" huge
64	pages in the kernel's huge page pool.  "Persistent" huge pages will be
65	returned to the huge page pool when freed by a task.  A user with root
66	privileges can dynamically allocate more or free some persistent huge pages
67	by increasing or decreasing the value of 'nr_hugepages'.
68	
69	Pages that are used as huge pages are reserved inside the kernel and cannot
70	be used for other purposes.  Huge pages cannot be swapped out under
71	memory pressure.
72	
73	Once a number of huge pages have been pre-allocated to the kernel huge page
74	pool, a user with appropriate privilege can use either the mmap system call
75	or shared memory system calls to use the huge pages.  See the discussion of
76	Using Huge Pages, below.
77	
78	The administrator can allocate persistent huge pages on the kernel boot
79	command line by specifying the "hugepages=N" parameter, where 'N' = the
80	number of huge pages requested.  This is the most reliable method of
81	allocating huge pages as memory has not yet become fragmented.
82	
83	Some platforms support multiple huge page sizes.  To allocate huge pages
84	of a specific size, one must precede the huge pages boot command parameters
85	with a huge page size selection parameter "hugepagesz=<size>".  <size> must
86	be specified in bytes with optional scale suffix [kKmMgG].  The default huge
87	page size may be selected with the "default_hugepagesz=<size>" boot parameter.
88	
89	When multiple huge page sizes are supported, /proc/sys/vm/nr_hugepages
90	indicates the current number of pre-allocated huge pages of the default size.
91	Thus, one can use the following command to dynamically allocate/deallocate
92	default sized persistent huge pages:
93	
94		echo 20 > /proc/sys/vm/nr_hugepages
95	
96	This command will try to adjust the number of default sized huge pages in the
97	huge page pool to 20, allocating or freeing huge pages, as required.
98	
99	On a NUMA platform, the kernel will attempt to distribute the huge page pool
100	over all the set of allowed nodes specified by the NUMA memory policy of the
101	task that modifies nr_hugepages.  The default for the allowed nodes--when the
102	task has default memory policy--is all on-line nodes with memory.  Allowed
103	nodes with insufficient available, contiguous memory for a huge page will be
104	silently skipped when allocating persistent huge pages.  See the discussion
105	below of the interaction of task memory policy, cpusets and per node attributes
106	with the allocation and freeing of persistent huge pages.
107	
108	The success or failure of huge page allocation depends on the amount of
109	physically contiguous memory that is present in system at the time of the
110	allocation attempt.  If the kernel is unable to allocate huge pages from
111	some nodes in a NUMA system, it will attempt to make up the difference by
112	allocating extra pages on other nodes with sufficient available contiguous
113	memory, if any.
114	
115	System administrators may want to put this command in one of the local rc
116	init files.  This will enable the kernel to allocate huge pages early in
117	the boot process when the possibility of getting physical contiguous pages
118	is still very high.  Administrators can verify the number of huge pages
119	actually allocated by checking the sysctl or meminfo.  To check the per node
120	distribution of huge pages in a NUMA system, use:
121	
122		cat /sys/devices/system/node/node*/meminfo | fgrep Huge
123	
124	/proc/sys/vm/nr_overcommit_hugepages specifies how large the pool of
125	huge pages can grow, if more huge pages than /proc/sys/vm/nr_hugepages are
126	requested by applications.  Writing any non-zero value into this file
127	indicates that the hugetlb subsystem is allowed to try to obtain that
128	number of "surplus" huge pages from the kernel's normal page pool, when the
129	persistent huge page pool is exhausted. As these surplus huge pages become
130	unused, they are freed back to the kernel's normal page pool.
131	
132	When increasing the huge page pool size via nr_hugepages, any existing surplus
133	pages will first be promoted to persistent huge pages.  Then, additional
134	huge pages will be allocated, if necessary and if possible, to fulfill
135	the new persistent huge page pool size.
136	
137	The administrator may shrink the pool of persistent huge pages for
138	the default huge page size by setting the nr_hugepages sysctl to a
139	smaller value.  The kernel will attempt to balance the freeing of huge pages
140	across all nodes in the memory policy of the task modifying nr_hugepages.
141	Any free huge pages on the selected nodes will be freed back to the kernel's
142	normal page pool.
143	
144	Caveat: Shrinking the persistent huge page pool via nr_hugepages such that
145	it becomes less than the number of huge pages in use will convert the balance
146	of the in-use huge pages to surplus huge pages.  This will occur even if
147	the number of surplus pages it would exceed the overcommit value.  As long as
148	this condition holds--that is, until nr_hugepages+nr_overcommit_hugepages is
149	increased sufficiently, or the surplus huge pages go out of use and are freed--
150	no more surplus huge pages will be allowed to be allocated.
151	
152	With support for multiple huge page pools at run-time available, much of
153	the huge page userspace interface in /proc/sys/vm has been duplicated in sysfs.
154	The /proc interfaces discussed above have been retained for backwards
155	compatibility. The root huge page control directory in sysfs is:
156	
157		/sys/kernel/mm/hugepages
158	
159	For each huge page size supported by the running kernel, a subdirectory
160	will exist, of the form:
161	
162		hugepages-${size}kB
163	
164	Inside each of these directories, the same set of files will exist:
165	
166		nr_hugepages
167		nr_hugepages_mempolicy
168		nr_overcommit_hugepages
169		free_hugepages
170		resv_hugepages
171		surplus_hugepages
172	
173	which function as described above for the default huge page-sized case.
174	
175	
176	Interaction of Task Memory Policy with Huge Page Allocation/Freeing
177	===================================================================
178	
179	Whether huge pages are allocated and freed via the /proc interface or
180	the /sysfs interface using the nr_hugepages_mempolicy attribute, the NUMA
181	nodes from which huge pages are allocated or freed are controlled by the
182	NUMA memory policy of the task that modifies the nr_hugepages_mempolicy
183	sysctl or attribute.  When the nr_hugepages attribute is used, mempolicy
184	is ignored.
185	
186	The recommended method to allocate or free huge pages to/from the kernel
187	huge page pool, using the nr_hugepages example above, is:
188	
189	    numactl --interleave <node-list> echo 20 \
190					>/proc/sys/vm/nr_hugepages_mempolicy
191	
192	or, more succinctly:
193	
194	    numactl -m <node-list> echo 20 >/proc/sys/vm/nr_hugepages_mempolicy
195	
196	This will allocate or free abs(20 - nr_hugepages) to or from the nodes
197	specified in <node-list>, depending on whether number of persistent huge pages
198	is initially less than or greater than 20, respectively.  No huge pages will be
199	allocated nor freed on any node not included in the specified <node-list>.
200	
201	When adjusting the persistent hugepage count via nr_hugepages_mempolicy, any
202	memory policy mode--bind, preferred, local or interleave--may be used.  The
203	resulting effect on persistent huge page allocation is as follows:
204	
205	1) Regardless of mempolicy mode [see Documentation/vm/numa_memory_policy.txt],
206	   persistent huge pages will be distributed across the node or nodes
207	   specified in the mempolicy as if "interleave" had been specified.
208	   However, if a node in the policy does not contain sufficient contiguous
209	   memory for a huge page, the allocation will not "fallback" to the nearest
210	   neighbor node with sufficient contiguous memory.  To do this would cause
211	   undesirable imbalance in the distribution of the huge page pool, or
212	   possibly, allocation of persistent huge pages on nodes not allowed by
213	   the task's memory policy.
214	
215	2) One or more nodes may be specified with the bind or interleave policy.
216	   If more than one node is specified with the preferred policy, only the
217	   lowest numeric id will be used.  Local policy will select the node where
218	   the task is running at the time the nodes_allowed mask is constructed.
219	   For local policy to be deterministic, the task must be bound to a cpu or
220	   cpus in a single node.  Otherwise, the task could be migrated to some
221	   other node at any time after launch and the resulting node will be
222	   indeterminate.  Thus, local policy is not very useful for this purpose.
223	   Any of the other mempolicy modes may be used to specify a single node.
224	
225	3) The nodes allowed mask will be derived from any non-default task mempolicy,
226	   whether this policy was set explicitly by the task itself or one of its
227	   ancestors, such as numactl.  This means that if the task is invoked from a
228	   shell with non-default policy, that policy will be used.  One can specify a
229	   node list of "all" with numactl --interleave or --membind [-m] to achieve
230	   interleaving over all nodes in the system or cpuset.
231	
232	4) Any task mempolicy specified--e.g., using numactl--will be constrained by
233	   the resource limits of any cpuset in which the task runs.  Thus, there will
234	   be no way for a task with non-default policy running in a cpuset with a
235	   subset of the system nodes to allocate huge pages outside the cpuset
236	   without first moving to a cpuset that contains all of the desired nodes.
237	
238	5) Boot-time huge page allocation attempts to distribute the requested number
239	   of huge pages over all on-lines nodes with memory.
240	
241	Per Node Hugepages Attributes
242	=============================
243	
244	A subset of the contents of the root huge page control directory in sysfs,
245	described above, will be replicated under each the system device of each
246	NUMA node with memory in:
247	
248		/sys/devices/system/node/node[0-9]*/hugepages/
249	
250	Under this directory, the subdirectory for each supported huge page size
251	contains the following attribute files:
252	
253		nr_hugepages
254		free_hugepages
255		surplus_hugepages
256	
257	The free_' and surplus_' attribute files are read-only.  They return the number
258	of free and surplus [overcommitted] huge pages, respectively, on the parent
259	node.
260	
261	The nr_hugepages attribute returns the total number of huge pages on the
262	specified node.  When this attribute is written, the number of persistent huge
263	pages on the parent node will be adjusted to the specified value, if sufficient
264	resources exist, regardless of the task's mempolicy or cpuset constraints.
265	
266	Note that the number of overcommit and reserve pages remain global quantities,
267	as we don't know until fault time, when the faulting task's mempolicy is
268	applied, from which node the huge page allocation will be attempted.
269	
270	
271	Using Huge Pages
272	================
273	
274	If the user applications are going to request huge pages using mmap system
275	call, then it is required that system administrator mount a file system of
276	type hugetlbfs:
277	
278	  mount -t hugetlbfs \
279		-o uid=<value>,gid=<value>,mode=<value>,pagesize=<value>,size=<value>,\
280		min_size=<value>,nr_inodes=<value> none /mnt/huge
281	
282	This command mounts a (pseudo) filesystem of type hugetlbfs on the directory
283	/mnt/huge.  Any files created on /mnt/huge uses huge pages.  The uid and gid
284	options sets the owner and group of the root of the file system.  By default
285	the uid and gid of the current process are taken.  The mode option sets the
286	mode of root of file system to value & 01777.  This value is given in octal.
287	By default the value 0755 is picked. If the platform supports multiple huge
288	page sizes, the pagesize option can be used to specify the huge page size and
289	associated pool.  pagesize is specified in bytes.  If pagesize is not specified
290	the platform's default huge page size and associated pool will be used. The
291	size option sets the maximum value of memory (huge pages) allowed for that
292	filesystem (/mnt/huge).  The size option can be specified in bytes, or as a
293	percentage of the specified huge page pool (nr_hugepages).  The size is
294	rounded down to HPAGE_SIZE boundary.  The min_size option sets the minimum
295	value of memory (huge pages) allowed for the filesystem.  min_size can be
296	specified in the same way as size, either bytes or a percentage of the
297	huge page pool.  At mount time, the number of huge pages specified by
298	min_size are reserved for use by the filesystem.  If there are not enough
299	free huge pages available, the mount will fail.  As huge pages are allocated
300	to the filesystem and freed, the reserve count is adjusted so that the sum
301	of allocated and reserved huge pages is always at least min_size.  The option
302	nr_inodes sets the maximum number of inodes that /mnt/huge can use.  If the
303	size, min_size or nr_inodes option is not provided on command line then
304	no limits are set.  For pagesize, size, min_size and nr_inodes options, you
305	can use [G|g]/[M|m]/[K|k] to represent giga/mega/kilo. For example, size=2K
306	has the same meaning as size=2048.
307	
308	While read system calls are supported on files that reside on hugetlb
309	file systems, write system calls are not.
310	
311	Regular chown, chgrp, and chmod commands (with right permissions) could be
312	used to change the file attributes on hugetlbfs.
313	
314	Also, it is important to note that no such mount command is required if
315	applications are going to use only shmat/shmget system calls or mmap with
316	MAP_HUGETLB.  For an example of how to use mmap with MAP_HUGETLB see map_hugetlb
317	below.
318	
319	Users who wish to use hugetlb memory via shared memory segment should be a
320	member of a supplementary group and system admin needs to configure that gid
321	into /proc/sys/vm/hugetlb_shm_group.  It is possible for same or different
322	applications to use any combination of mmaps and shm* calls, though the mount of
323	filesystem will be required for using mmap calls without MAP_HUGETLB.
324	
325	Syscalls that operate on memory backed by hugetlb pages only have their lengths
326	aligned to the native page size of the processor; they will normally fail with
327	errno set to EINVAL or exclude hugetlb pages that extend beyond the length if
328	not hugepage aligned.  For example, munmap(2) will fail if memory is backed by
329	a hugetlb page and the length is smaller than the hugepage size.
330	
331	
332	Examples
333	========
334	
335	1) map_hugetlb: see tools/testing/selftests/vm/map_hugetlb.c
336	
337	2) hugepage-shm:  see tools/testing/selftests/vm/hugepage-shm.c
338	
339	3) hugepage-mmap:  see tools/testing/selftests/vm/hugepage-mmap.c
340	
341	4) The libhugetlbfs (https://github.com/libhugetlbfs/libhugetlbfs) library
342	   provides a wide range of userspace tools to help with huge page usability,
343	   environment setup, and control.
344	
345	Kernel development regression testing
346	=====================================
347	
348	The most complete set of hugetlb tests are in the libhugetlbfs repository.
349	If you modify any hugetlb related code, use the libhugetlbfs test suite
350	to check for regressions.  In addition, if you add any new hugetlb
351	functionality, please add appropriate tests to libhugetlbfs.
Hide Line Numbers


About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog