1.\" Copyright (c) 2015 Adrian Chadd <adrian@FreeBSD.org> 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 13.\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23.\" SUCH DAMAGE. 24.\" 25.\" $FreeBSD$ 26.\" 27.Dd October 22, 2018 28.Dt NUMA 4 29.Os 30.Sh NAME 31.Nm NUMA 32.Nd Non-Uniform Memory Access 33.Sh SYNOPSIS 34.Cd options MAXMEMDOM 35.Cd options NUMA 36.Sh DESCRIPTION 37Non-Uniform Memory Access is a computer architecture design which 38involves unequal costs between processors, memory and IO devices 39in a given system. 40.Pp 41In a 42.Nm 43architecture, the latency to access specific memory or IO devices 44depends upon which processor the memory or device is attached to. 45Accessing memory local to a processor is faster than accessing memory 46that is connected to one of the other processors. 47.Fx 48implements NUMA-aware memory allocation policies. 49By default it attempts to ensure that allocations are balanced across 50each domain. 51Users may override the default domain selection policy using 52.Xr cpuset 1 . 53.Pp 54.Nm 55support is enabled when the 56.Cd NUMA 57option is specified in the kernel configuration file. 58Each platform defines the 59.Cd MAXMEMDOM 60constant, which specifies the maximum number of supported NUMA domains. 61This constant may be specified in the kernel configuration file. 62.Nm 63support can be disabled at boot time by setting the 64.Va vm.numa.disabled 65tunable to 1. 66Other values for this tunable are currently ignored. 67.Pp 68Thread and process 69.Nm 70policies are controlled with the 71.Xr cpuset_getdomain 2 72and 73.Xr cpuset_setdomain 2 74syscalls. 75The 76.Xr cpuset 1 77tool is available for starting processes with a non-default 78policy, or to change the policy of an existing thread or process. 79See 80.Xr SMP 4 81for information about CPU to domain mapping. 82.Pp 83Systems with non-uniform access to I/O devices may mark those devices 84with the local VM domain identifier. 85Drivers can find out their local domain information by calling 86.Xr bus_get_domain 9 . 87.Ss MIB Variables 88The operation of 89.Nm 90is controlled and exposes information with these 91.Xr sysctl 8 92MIB variables: 93.Pp 94.Bl -tag -width indent -compact 95.It Va vm.ndomains 96The number of VM domains which have been detected. 97.Pp 98.It Va vm.phys_locality 99A table indicating the relative cost of each VM domain to each other. 100A value of 10 indicates equal cost. 101A value of -1 means the locality map is not available or no 102locality information is available. 103.Pp 104.It Va vm.phys_segs 105The map of physical memory, grouped by VM domain. 106.El 107.Sh IMPLEMENTATION NOTES 108The current 109.Nm 110implementation is VM-focused. 111The hardware 112.Nm 113domains are mapped into a contiguous, non-sparse 114VM domain space, starting from 0. 115Thus, VM domain information (for example, the domain identifier) is not 116necessarily the same as is found in the hardware specific information. 117Policy information is available in both struct thread and struct proc. 118.Sh SEE ALSO 119.Xr cpuset 1 , 120.Xr cpuset_getaffinity 2 , 121.Xr cpuset_setaffinity 2 , 122.Xr SMP 4 , 123.Xr bus_get_domain 9 124.Sh HISTORY 125.Nm 126first appeared in 127.Fx 9.0 128as a first-touch allocation policy with a fail-over to round-robin allocation 129and was not configurable. 130It was then modified in 131.Fx 10.0 132to implement a round-robin allocation policy and was also not configurable. 133.Pp 134The 135.Xr numa_getaffinity 2 136and 137.Xr numa_setaffinity 2 138syscalls and the 139.Xr numactl 1 140tool first appeared in 141.Fx 11.0 142and were removed in 143.Fx 12.0 . 144The current implementation appeared in 145.Fx 12.0 . 146.Sh AUTHORS 147This manual page written by 148.An Adrian Chadd Aq Mt adrian@FreeBSD.org . 149.Sh NOTES 150No statistics are kept to indicate how often 151.Nm 152allocation policies succeed or fail. 153