man2/getcpu.2

   1 .\" This man page is Copyright (C) 2006 Andi Kleen <ak@muc.de>.
   2 .\"
   3 .\" %%%LICENSE_START(VERBATIM_ONE_PARA)
   4 .\" Permission is granted to distribute possibly modified copies
   5 .\" of this page provided the header is included verbatim,
   6 .\" and in case of nontrivial modification author and date
   7 .\" of the modification is added to the header.
   8 .\" %%%LICENSE_END
   9 .\"
  10 .\" 2008, mtk, various edits
  11 .\"
  12 .TH getcpu 2 (date) "Linux man-pages (unreleased)"
  13 .SH NAME
  14 getcpu \- determine CPU and NUMA node on which the calling thread is running
  15 .SH LIBRARY
  16 Standard C library
  17 .RI ( libc ", " \-lc )
  18 .SH SYNOPSIS
  19 .nf
  20 .BR "#define _GNU_SOURCE" "             /* See feature_test_macros(7) */"
  21 .B #include <sched.h>
  22 .PP
  23 .BI "int getcpu(unsigned int *" cpu ", unsigned int *" node );
  24 .fi
  25 .SH DESCRIPTION
  26 The
  27 .BR getcpu ()
  28 system call identifies the processor and node on which the calling
  29 thread or process is currently running and writes them into the
  30 integers pointed to by the
  31 .I cpu
  32 and
  33 .I node
  34 arguments.
  35 The processor is a unique small integer identifying a CPU.
  36 The node is a unique small identifier identifying a NUMA node.
  37 When either
  38 .I cpu
  39 or
  40 .I node
  41 is NULL nothing is written to the respective pointer.
  42 .PP
  43 The information placed in
  44 .I cpu
  45 is guaranteed to be current only at the time of the call:
  46 unless the CPU affinity has been fixed using
  47 .BR sched_setaffinity (2),
  48 the kernel might change the CPU at any time.
  49 (Normally this does not happen
  50 because the scheduler tries to minimize movements between CPUs to
  51 keep caches hot, but it is possible.)
  52 The caller must allow for the possibility that the information returned in
  53 .I cpu
  54 and
  55 .I node
  56 is no longer current by the time the call returns.
  57 .SH RETURN VALUE
  58 On success, 0 is returned.
  59 On error, \-1 is returned, and
  60 .I errno
  61 is set to indicate the error.
  62 .SH ERRORS
  63 .TP
  64 .B EFAULT
  65 Arguments point outside the calling process's address space.
  66 .SH VERSIONS
  67 .BR getcpu ()
  68 was added in kernel 2.6.19 for x86-64 and i386.
  69 Library support was added in glibc 2.29
  70 (Earlier glibc versions did not provide a wrapper for this system call,
  71 necessitating the use of
  72 .BR syscall (2).)
  73 .SH STANDARDS
  74 .BR getcpu ()
  75 is Linux-specific.
  76 .SH NOTES
  77 Linux makes a best effort to make this call as fast as possible.
  78 (On some architectures, this is done via an implementation in the
  79 .BR vdso (7).)
  80 The intention of
  81 .BR getcpu ()
  82 is to allow programs to make optimizations with per-CPU data
  83 or for NUMA optimization.
  84 .\"
  85 .SS C library/kernel differences
  86 The kernel system call has a third argument:
  87 .PP
  88 .in +4n
  89 .nf
  90 .BI "int getcpu(unsigned int *" cpu ", unsigned int *" node ,
  91 .BI "           struct getcpu_cache *" tcache );
  92 .fi
  93 .in
  94 .PP
  95 The
  96 .I tcache
  97 argument is unused since Linux 2.6.24,
  98 and (when invoking the system call directly)
  99 should be specified as NULL,
 100 unless portability to Linux 2.6.23 or earlier is required.
 101 .PP
 102 .\" commit 4307d1e5ada595c87f9a4d16db16ba5edb70dcb1
 103 .\" Author: Ingo Molnar <mingo@elte.hu>
 104 .\" Date:   Wed Nov 7 18:37:48 2007 +0100
 105 .\" x86: ignore the sys_getcpu() tcache parameter
 106 In Linux 2.6.23 and earlier, if the
 107 .I tcache
 108 argument was non-NULL,
 109 then it specified a pointer to a caller-allocated buffer in thread-local
 110 storage that was used to provide a caching mechanism for
 111 .BR getcpu ().
 112 Use of the cache could speed
 113 .BR getcpu ()
 114 calls, at the cost that there was a very small chance that
 115 the returned information would be out of date.
 116 The caching mechanism was considered to cause problems when
 117 migrating threads between CPUs, and so the argument is now ignored.
 118 .\"
 119 .\" ===== Before kernel 2.6.24: =====
 120 .\" .I tcache
 121 .\" is a pointer to a
 122 .\" .IR "struct getcpu_cache"
 123 .\" that is used as a cache by
 124 .\" .BR getcpu ().
 125 .\" The caller should put the cache into a thread-local variable
 126 .\" if the process is multithreaded,
 127 .\" because the cache cannot be shared between different threads.
 128 .\" .I tcache
 129 .\" can be NULL.
 130 .\" If it is not NULL
 131 .\" .BR getcpu ()
 132 .\" will use it to speed up operation.
 133 .\" The information inside the cache is private to the system call
 134 .\" and should not be accessed by the user program.
 135 .\" The information placed in the cache can change between kernel releases.
 136 .\"
 137 .\" When no cache is specified
 138 .\" .BR getcpu ()
 139 .\" will be slower,
 140 .\" but always retrieve the current CPU and node information.
 141 .\" With a cache
 142 .\" .BR getcpu ()
 143 .\" is faster.
 144 .\" However, the cached information is updated only once per jiffy (see
 145 .\" .BR time (7)).
 146 .\" This means that the information could theoretically be out of date,
 147 .\" although in practice the scheduler's attempt to maintain
 148 .\" soft CPU affinity means that the information is unlikely to change
 149 .\" over the course of the caching interval.
 150 .SH SEE ALSO
 151 .BR mbind (2),
 152 .BR sched_setaffinity (2),
 153 .BR set_mempolicy (2),
 154 .BR sched_getcpu (3),
 155 .BR cpuset (7),
 156 .BR vdso (7)