man2/getcpu.2

   1 .\" This man page is Copyright (C) 2006 Andi Kleen <ak@muc.de>.
   2 .\" %%%LICENSE_START(VERBATIM_ONE_PARA)
   3 .\" Permission is granted to distribute possibly modified copies
   4 .\" of this page provided the header is included verbatim,
   5 .\" and in case of nontrivial modification author and date
   6 .\" of the modification is added to the header.
   7 .\" %%%END_LICENSE
   8 .\" 2008, mtk, various edits
   9 .TH GETCPU 2 2012-07-13 "Linux" "Linux Programmer's Manual"
  10 .SH NAME
  11 getcpu \- determine CPU and NUMA node on which the calling thread is running
  12 .SH SYNOPSIS
  13 .nf
  14 .B #include <linux/getcpu.h>
  15 .sp
  16 .BI "int getcpu(unsigned *" cpu ", unsigned *" node \
  17 ", struct getcpu_cache *" tcache );
  18 .fi
  19
  20 .IR Note :
  21 There is no glibc wrapper for this system call; see NOTES.
  22 .SH DESCRIPTION
  23 The
  24 .BR getcpu ()
  25 system call identifies the processor and node on which the calling
  26 thread or process is currently running and writes them into the
  27 integers pointed to by the
  28 .I cpu
  29 and
  30 .I node
  31 arguments.
  32 The processor is a unique small integer identifying a CPU.
  33 The node is a unique small identifier identifying a NUMA node.
  34 When either
  35 .I cpu
  36 or
  37 .I node
  38 is NULL nothing is written to the respective pointer.
  39
  40 The third argument to this system call is nowadays unused (see NOTES).
  41
  42 The information placed in
  43 .I cpu
  44 is only guaranteed to be current at the time of the call:
  45 unless the CPU affinity has been fixed using
  46 .BR sched_setaffinity (2),
  47 the kernel might change the CPU at any time.
  48 (Normally this does not happen
  49 because the scheduler tries to minimize movements between CPUs to
  50 keep caches hot, but it is possible.)
  51 The caller must allow for the possibility that the information returned in
  52 .I cpu
  53 and
  54 .I node
  55 is no longer current by the time the call returns.
  56 .SH RETURN VALUE
  57 On success, 0 is returned.
  58 On error, \-1 is returned, and
  59 .I errno
  60 is set appropriately.
  61 .SH ERRORS
  62 .TP
  63 .B EFAULT
  64 Arguments point outside the calling process's address space.
  65 .SH VERSIONS
  66 .BR getcpu ()
  67 was added in kernel 2.6.19 for x86_64 and i386.
  68 .SH CONFORMING TO
  69 .BR getcpu ()
  70 is Linux specific.
  71 .SH NOTES
  72 Linux makes a best effort to make this call as fast possible.
  73 The intention of
  74 .BR getcpu ()
  75 is to allow programs to make optimizations with per-CPU data
  76 or for NUMA optimization.
  77
  78 Glibc does not provide a wrapper for this system call; call it using
  79 .BR syscall (2);
  80 or use
  81 .BR sched_getcpu (3)
  82 instead.
  83
  84 The
  85 .I tcache
  86 argument is unused since Linux 2.6.24.
  87 .\" commit 4307d1e5ada595c87f9a4d16db16ba5edb70dcb1
  88 .\" Author: Ingo Molnar <mingo@elte.hu>
  89 .\" Date:   Wed Nov 7 18:37:48 2007 +0100
  90 .\" x86: ignore the sys_getcpu() tcache parameter
  91 In earlier kernels,
  92 if this argument was non-NULL,
  93 then it specified a pointer to a caller-allocated buffer in thread-local
  94 storage that was used to provide a caching mechanism for
  95 .BR getcpu ().
  96 Use of the cache could speed
  97 .BR getcpu ()
  98 calls, at the cost that there was a very small chance that
  99 the returned information would be out of date.
 100 The caching mechanism was considered to cause problems when
 101 migrating threads between CPUs, and so the argument is now ignored.
 102 .\"
 103 .\" ===== Before kernel 2.6.24: =====
 104 .\" .I tcache
 105 .\" is a pointer to a
 106 .\" .IR "struct getcpu_cache"
 107 .\" that is used as a cache by
 108 .\" .BR getcpu ().
 109 .\" The caller should put the cache into a thread-local variable
 110 .\" if the process is multithreaded,
 111 .\" because the cache cannot be shared between different threads.
 112 .\" .I tcache
 113 .\" can be NULL.
 114 .\" If it is not NULL
 115 .\" .BR getcpu ()
 116 .\" will use it to speed up operation.
 117 .\" The information inside the cache is private to the system call
 118 .\" and should not be accessed by the user program.
 119 .\" The information placed in the cache can change between kernel releases.
 120 .\"
 121 .\" When no cache is specified
 122 .\" .BR getcpu ()
 123 .\" will be slower,
 124 .\" but always retrieve the current CPU and node information.
 125 .\" With a cache
 126 .\" .BR getcpu ()
 127 .\" is faster.
 128 .\" However, the cached information is only updated once per jiffy (see
 129 .\" .BR time (7)).
 130 .\" This means that the information could theoretically be out of date,
 131 .\" although in practice the scheduler's attempt to maintain
 132 .\" soft CPU affinity means that the information is unlikely to change
 133 .\" over the course of the caching interval.
 134 .SH SEE ALSO
 135 .BR mbind (2),
 136 .BR sched_setaffinity (2),
 137 .BR set_mempolicy (2),
 138 .BR sched_getcpu (3),
 139 .BR cpuset (7)