1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
|
.\" Copyright 2003 Andries E. Brouwer (aeb@cwi.nl)
.\"
.\" SPDX-License-Identifier: Linux-man-pages-copyleft
.\"
.TH alloc_hugepages 2 2023-03-30 "Linux man-pages 6.04"
.SH NAME
alloc_hugepages, free_hugepages \- allocate or free huge pages
.SH SYNOPSIS
.nf
.BI "void *syscall(SYS_alloc_hugepages, int " key ", void " addr [. len "], \
size_t " len ,
.BI " int " prot ", int " flag );
.\" asmlinkage unsigned long sys_alloc_hugepages(int key, unsigned long addr,
.\" unsigned long len, int prot, int flag);
.BI "int syscall(SYS_free_hugepages, void *" addr );
.\" asmlinkage int sys_free_hugepages(unsigned long addr);
.fi
.PP
.IR Note :
glibc provides no wrappers for these system calls,
necessitating the use of
.BR syscall (2).
.SH DESCRIPTION
The system calls
.BR alloc_hugepages ()
and
.BR free_hugepages ()
were introduced in Linux 2.5.36 and removed again in Linux 2.5.54.
They existed only on i386 and ia64 (when built with
.BR CONFIG_HUGETLB_PAGE ).
In Linux 2.4.20, the syscall numbers exist,
but the calls fail with the error
.BR ENOSYS .
.PP
On i386 the memory management hardware knows about ordinary pages (4\ KiB)
and huge pages (2 or 4\ MiB).
Similarly ia64 knows about huge pages of
several sizes.
These system calls serve to map huge pages into the
process's memory or to free them again.
Huge pages are locked into memory, and are not swapped.
.PP
The
.I key
argument is an identifier.
When zero the pages are private, and
not inherited by children.
When positive the pages are shared with other applications using the same
.IR key ,
and inherited by child processes.
.PP
The
.I addr
argument of
.BR free_hugepages ()
tells which page is being freed: it was the return value of a
call to
.BR alloc_hugepages ().
(The memory is first actually freed when all users have released it.)
The
.I addr
argument of
.BR alloc_hugepages ()
is a hint, that the kernel may or may not follow.
Addresses must be properly aligned.
.PP
The
.I len
argument is the length of the required segment.
It must be a multiple of the huge page size.
.PP
The
.I prot
argument specifies the memory protection of the segment.
It is one of
.BR PROT_READ ,
.BR PROT_WRITE ,
.BR PROT_EXEC .
.PP
The
.I flag
argument is ignored, unless
.I key
is positive.
In that case, if
.I flag
is
.BR IPC_CREAT ,
then a new huge page segment is created when none
with the given key existed.
If this flag is not set, then
.B ENOENT
is returned when no segment with the given key exists.
.SH RETURN VALUE
On success,
.BR alloc_hugepages ()
returns the allocated virtual address, and
.BR free_hugepages ()
returns zero.
On error, \-1 is returned, and
.I errno
is set to indicate the error.
.SH ERRORS
.TP
.B ENOSYS
The system call is not supported on this kernel.
.SH FILES
.TP
.I /proc/sys/vm/nr_hugepages
Number of configured hugetlb pages.
This can be read and written.
.TP
.I /proc/meminfo
Gives info on the number of configured hugetlb pages and on their size
in the three variables HugePages_Total, HugePages_Free, Hugepagesize.
.SH STANDARDS
Linux on Intel processors.
.SH HISTORY
These system calls are gone;
they existed only in Linux 2.5.36 through to Linux 2.5.54.
.SH NOTES
Now the hugetlbfs filesystem can be used instead.
Memory backed by huge pages (if the CPU supports them) is obtained by
using
.BR mmap (2)
to map files in this virtual filesystem.
.PP
The maximal number of huge pages can be specified using the
.B hugepages=
boot parameter.
.\".PP
.\" requires CONFIG_HUGETLB_PAGE (under "Processor type and features")
.\" and CONFIG_HUGETLBFS (under "Filesystems").
.\" mount \-t hugetlbfs hugetlbfs /huge
.\" SHM_HUGETLB
|
