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.
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).
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
21 The kernel built with huge page support should show the number of configured
22 huge pages in the system by running the "cat /proc/meminfo" command.
24 /proc/meminfo also provides information about the total number of hugetlb
25 pages configured in the kernel. It also displays information about the
26 number of free hugetlb pages at any time. It also displays information about
27 the configured huge page size - this is needed for generating the proper
28 alignment and size of the arguments to the above system calls.
30 The output of "cat /proc/meminfo" will have lines like:
40 HugePages_Total is the size of the pool of huge pages.
41 HugePages_Free is the number of huge pages in the pool that are not yet
43 HugePages_Rsvd is short for "reserved," and is the number of huge pages for
44 which a commitment to allocate from the pool has been made,
45 but no allocation has yet been made. Reserved huge pages
46 guarantee that an application will be able to allocate a
47 huge page from the pool of huge pages at fault time.
48 HugePages_Surp is short for "surplus," and is the number of huge pages in
49 the pool above the value in /proc/sys/vm/nr_hugepages. The
50 maximum number of surplus huge pages is controlled by
51 /proc/sys/vm/nr_overcommit_hugepages.
53 /proc/filesystems should also show a filesystem of type "hugetlbfs" configured
56 /proc/sys/vm/nr_hugepages indicates the current number of configured hugetlb
57 pages in the kernel. Super user can dynamically request more (or free some
58 pre-configured) huge pages.
59 The allocation (or deallocation) of hugetlb pages is possible only if there are
60 enough physically contiguous free pages in system (freeing of huge pages is
61 possible only if there are enough hugetlb pages free that can be transferred
62 back to regular memory pool).
64 Pages that are used as hugetlb pages are reserved inside the kernel and cannot
65 be used for other purposes.
67 Once the kernel with Hugetlb page support is built and running, a user can
68 use either the mmap system call or shared memory system calls to start using
69 the huge pages. It is required that the system administrator preallocate
70 enough memory for huge page purposes.
72 The administrator can preallocate huge pages on the kernel boot command line by
73 specifying the "hugepages=N" parameter, where 'N' = the number of huge pages
74 requested. This is the most reliable method for preallocating huge pages as
75 memory has not yet become fragmented.
77 Some platforms support multiple huge page sizes. To preallocate huge pages
78 of a specific size, one must preceed the huge pages boot command parameters
79 with a huge page size selection parameter "hugepagesz=<size>". <size> must
80 be specified in bytes with optional scale suffix [kKmMgG]. The default huge
81 page size may be selected with the "default_hugepagesz=<size>" boot parameter.
83 /proc/sys/vm/nr_hugepages indicates the current number of configured [default
84 size] hugetlb pages in the kernel. Super user can dynamically request more
85 (or free some pre-configured) huge pages.
87 Use the following command to dynamically allocate/deallocate default sized
90 echo 20 > /proc/sys/vm/nr_hugepages
92 This command will try to configure 20 default sized huge pages in the system.
93 On a NUMA platform, the kernel will attempt to distribute the huge page pool
94 over the all on-line nodes. These huge pages, allocated when nr_hugepages
95 is increased, are called "persistent huge pages".
97 The success or failure of huge page allocation depends on the amount of
98 physically contiguous memory that is preset in system at the time of the
99 allocation attempt. If the kernel is unable to allocate huge pages from
100 some nodes in a NUMA system, it will attempt to make up the difference by
101 allocating extra pages on other nodes with sufficient available contiguous
104 System administrators may want to put this command in one of the local rc init
105 files. This will enable the kernel to request huge pages early in the boot
106 process when the possibility of getting physical contiguous pages is still
107 very high. Administrators can verify the number of huge pages actually
108 allocated by checking the sysctl or meminfo. To check the per node
109 distribution of huge pages in a NUMA system, use:
111 cat /sys/devices/system/node/node*/meminfo | fgrep Huge
113 /proc/sys/vm/nr_overcommit_hugepages specifies how large the pool of
114 huge pages can grow, if more huge pages than /proc/sys/vm/nr_hugepages are
115 requested by applications. Writing any non-zero value into this file
116 indicates that the hugetlb subsystem is allowed to try to obtain "surplus"
117 huge pages from the buddy allocator, when the normal pool is exhausted. As
118 these surplus huge pages go out of use, they are freed back to the buddy
121 When increasing the huge page pool size via nr_hugepages, any surplus
122 pages will first be promoted to persistent huge pages. Then, additional
123 huge pages will be allocated, if necessary and if possible, to fulfill
124 the new huge page pool size.
126 The administrator may shrink the pool of preallocated huge pages for
127 the default huge page size by setting the nr_hugepages sysctl to a
128 smaller value. The kernel will attempt to balance the freeing of huge pages
129 across all on-line nodes. Any free huge pages on the selected nodes will
130 be freed back to the buddy allocator.
132 Caveat: Shrinking the pool via nr_hugepages such that it becomes less
133 than the number of huge pages in use will convert the balance to surplus
134 huge pages even if it would exceed the overcommit value. As long as
135 this condition holds, however, no more surplus huge pages will be
136 allowed on the system until one of the two sysctls are increased
137 sufficiently, or the surplus huge pages go out of use and are freed.
139 With support for multiple huge page pools at run-time available, much of
140 the huge page userspace interface has been duplicated in sysfs. The above
141 information applies to the default huge page size which will be
142 controlled by the /proc interfaces for backwards compatibility. The root
143 huge page control directory in sysfs is:
145 /sys/kernel/mm/hugepages
147 For each huge page size supported by the running kernel, a subdirectory
148 will exist, of the form
152 Inside each of these directories, the same set of files will exist:
155 nr_overcommit_hugepages
160 which function as described above for the default huge page-sized case.
162 If the user applications are going to request huge pages using mmap system
163 call, then it is required that system administrator mount a file system of
167 -o uid=<value>,gid=<value>,mode=<value>,size=<value>,nr_inodes=<value> \
170 This command mounts a (pseudo) filesystem of type hugetlbfs on the directory
171 /mnt/huge. Any files created on /mnt/huge uses huge pages. The uid and gid
172 options sets the owner and group of the root of the file system. By default
173 the uid and gid of the current process are taken. The mode option sets the
174 mode of root of file system to value & 0777. This value is given in octal.
175 By default the value 0755 is picked. The size option sets the maximum value of
176 memory (huge pages) allowed for that filesystem (/mnt/huge). The size is
177 rounded down to HPAGE_SIZE. The option nr_inodes sets the maximum number of
178 inodes that /mnt/huge can use. If the size or nr_inodes option is not
179 provided on command line then no limits are set. For size and nr_inodes
180 options, you can use [G|g]/[M|m]/[K|k] to represent giga/mega/kilo. For
181 example, size=2K has the same meaning as size=2048.
183 While read system calls are supported on files that reside on hugetlb
184 file systems, write system calls are not.
186 Regular chown, chgrp, and chmod commands (with right permissions) could be
187 used to change the file attributes on hugetlbfs.
189 Also, it is important to note that no such mount command is required if the
190 applications are going to use only shmat/shmget system calls. Users who
191 wish to use hugetlb page via shared memory segment should be a member of
192 a supplementary group and system admin needs to configure that gid into
193 /proc/sys/vm/hugetlb_shm_group. It is possible for same or different
194 applications to use any combination of mmaps and shm* calls, though the
195 mount of filesystem will be required for using mmap calls.
197 *******************************************************************
200 * Example of using huge page memory in a user application using Sys V shared
201 * memory system calls. In this example the app is requesting 256MB of
202 * memory that is backed by huge pages. The application uses the flag
203 * SHM_HUGETLB in the shmget system call to inform the kernel that it is
204 * requesting huge pages.
206 * For the ia64 architecture, the Linux kernel reserves Region number 4 for
207 * huge pages. That means the addresses starting with 0x800000... will need
208 * to be specified. Specifying a fixed address is not required on ppc64,
211 * Note: The default shared memory limit is quite low on many kernels,
212 * you may need to increase it via:
214 * echo 268435456 > /proc/sys/kernel/shmmax
216 * This will increase the maximum size per shared memory segment to 256MB.
217 * The other limit that you will hit eventually is shmall which is the
218 * total amount of shared memory in pages. To set it to 16GB on a system
219 * with a 4kB pagesize do:
221 * echo 4194304 > /proc/sys/kernel/shmall
225 #include <sys/types.h>
228 #include <sys/mman.h>
231 #define SHM_HUGETLB 04000
234 #define LENGTH (256UL*1024*1024)
236 #define dprintf(x) printf(x)
238 /* Only ia64 requires this */
240 #define ADDR (void *)(0x8000000000000000UL)
241 #define SHMAT_FLAGS (SHM_RND)
243 #define ADDR (void *)(0x0UL)
244 #define SHMAT_FLAGS (0)
253 if ((shmid = shmget(2, LENGTH,
254 SHM_HUGETLB | IPC_CREAT | SHM_R | SHM_W)) < 0) {
258 printf("shmid: 0x%x\n", shmid);
260 shmaddr = shmat(shmid, ADDR, SHMAT_FLAGS);
261 if (shmaddr == (char *)-1) {
262 perror("Shared memory attach failure");
263 shmctl(shmid, IPC_RMID, NULL);
266 printf("shmaddr: %p\n", shmaddr);
268 dprintf("Starting the writes:\n");
269 for (i = 0; i < LENGTH; i++) {
270 shmaddr[i] = (char)(i);
271 if (!(i % (1024 * 1024)))
276 dprintf("Starting the Check...");
277 for (i = 0; i < LENGTH; i++)
278 if (shmaddr[i] != (char)i)
279 printf("\nIndex %lu mismatched\n", i);
282 if (shmdt((const void *)shmaddr) != 0) {
283 perror("Detach failure");
284 shmctl(shmid, IPC_RMID, NULL);
288 shmctl(shmid, IPC_RMID, NULL);
293 *******************************************************************
296 * Example of using huge page memory in a user application using the mmap
297 * system call. Before running this application, make sure that the
298 * administrator has mounted the hugetlbfs filesystem (on some directory
299 * like /mnt) using the command mount -t hugetlbfs nodev /mnt. In this
300 * example, the app is requesting memory of size 256MB that is backed by
303 * For ia64 architecture, Linux kernel reserves Region number 4 for huge pages.
304 * That means the addresses starting with 0x800000... will need to be
305 * specified. Specifying a fixed address is not required on ppc64, i386
311 #include <sys/mman.h>
314 #define FILE_NAME "/mnt/hugepagefile"
315 #define LENGTH (256UL*1024*1024)
316 #define PROTECTION (PROT_READ | PROT_WRITE)
318 /* Only ia64 requires this */
320 #define ADDR (void *)(0x8000000000000000UL)
321 #define FLAGS (MAP_SHARED | MAP_FIXED)
323 #define ADDR (void *)(0x0UL)
324 #define FLAGS (MAP_SHARED)
327 void check_bytes(char *addr)
329 printf("First hex is %x\n", *((unsigned int *)addr));
332 void write_bytes(char *addr)
336 for (i = 0; i < LENGTH; i++)
337 *(addr + i) = (char)i;
340 void read_bytes(char *addr)
345 for (i = 0; i < LENGTH; i++)
346 if (*(addr + i) != (char)i) {
347 printf("Mismatch at %lu\n", i);
357 fd = open(FILE_NAME, O_CREAT | O_RDWR, 0755);
359 perror("Open failed");
363 addr = mmap(ADDR, LENGTH, PROTECTION, FLAGS, fd, 0);
364 if (addr == MAP_FAILED) {
370 printf("Returned address is %p\n", addr);
375 munmap(addr, LENGTH);