man/man4/smp.4

.\" Copyright (c) 1997
.\"	Steve Passe <fsmp@FreeBSD.org>.  All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. The name of the developer may NOT be used to endorse or promote products
.\"    derived from this software without specific prior written permission.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.Dd January 4, 2019
.Dt SMP 4
.Os
.Sh NAME
.Nm SMP
.Nd description of the FreeBSD Symmetric Multi-Processor kernel
.Sh SYNOPSIS
.Cd options SMP
.Sh DESCRIPTION
The
.Nm
kernel implements symmetric multi-processor support.
.Pp
.Nm
support can be disabled by setting the loader tunable
.Va kern.smp.disabled
to 1.
.Pp
The number of CPUs detected by the system is available in
the read-only sysctl variable
.Va hw.ncpu .
.Pp
The number of online threads per CPU core is available in the read-only sysctl
variable
.Va kern.smp.threads_per_core .
The number of physical CPU cores detected by the system is available in the
read-only sysctl variable
.Va kern.smp.cores .
.Pp
.Fx
allows specific CPUs on a multi-processor system to be disabled.
This can be done using the
.Va hint.lapic.X.disabled
tunable, where X is the APIC ID of a CPU.
Setting this tunable to 1 will result in the corresponding CPU being
disabled.
.Pp
.Fx
supports simultaneous multithreading on x86 and powerpc platforms.
On x86, the logical CPUs can be disabled by setting the
.Va machdep.hyperthreading_allowed
tunable to zero.
.Pp
The
.Xr sched_ule 4
scheduler implements CPU topology detection and adjusts the scheduling
algorithms to make better use of modern multi-core CPUs.
The sysctl variable
.Va kern.sched.topology_spec
reflects the detected CPU hardware in a parsable XML format.
The top level XML tag is <groups>, which encloses one or more <group> tags
containing data about individual CPU groups.
A CPU group contains CPUs that are detected to be "close" together, usually
by being cores in a single multi-core processor.
Attributes available in a <group> tag are "level", corresponding to the
nesting level of the CPU group and "cache-level", corresponding to the
level of CPU caches shared by the CPUs in the group.
The <group> tag contains the <cpu> and <flags> tags.
The <cpu> tag describes CPUs in the group.
Its attributes are "count", corresponding to the number of CPUs in the
group and "mask", corresponding to the integer binary mask in which
each bit position set to 1 signifies a CPU belonging to the group.
The contents (CDATA) of the <cpu> tag is the comma-delimited list
of CPU indexes (derived from the "mask" attribute).
The <flags> tag contains special tags (if any) describing the relation
of the CPUs in the group.
The possible flags are currently "HTT" and "SMT", corresponding to
the various implementations of hardware multithreading.
An example topology_spec output for a system consisting of
two quad-core processors is:
.Bd -literal
<groups>
  <group level="1" cache-level="0">
    <cpu count="8" mask="0xff">0, 1, 2, 3, 4, 5, 6, 7</cpu>
    <flags></flags>
    <children>
      <group level="2" cache-level="0">
        <cpu count="4" mask="0xf">0, 1, 2, 3</cpu>
        <flags></flags>
      </group>
      <group level="2" cache-level="0">
        <cpu count="4" mask="0xf0">4, 5, 6, 7</cpu>
        <flags></flags>
      </group>
    </children>
  </group>
</groups>
.Ed
.Pp
This information is used internally by the kernel to schedule related
tasks on CPUs that are closely grouped together.
.Sh COMPATIBILITY
Support for multi-processor systems is present for all Tier-1 and Tier-2
architectures on
.Fx .
Currently, this includes x86, powerpc, mips, arm and arm64.
Support is enabled using
.Cd options SMP .
It is permissible to use the SMP kernel configuration on non-SMP hardware.
.Sh I386 NOTES
For i386 systems, the
.Nm
kernel supports motherboards that follow the Intel MP specification,
version 1.4.
In addition to
.Cd options SMP ,
i386 also requires
.Cd device apic .
The
.Xr mptable 1
command may be used to view the status of multi-processor support.
.Sh SEE ALSO
.Xr cpuset 1 ,
.Xr mptable 1 ,
.Xr sched_4bsd 4 ,
.Xr sched_ule 4 ,
.Xr loader 8 ,
.Xr sysctl 8 ,
.Xr condvar 9 ,
.Xr msleep 9 ,
.Xr mtx_pool 9 ,
.Xr mutex 9 ,
.Xr rwlock 9 ,
.Xr sema 9 ,
.Xr sx 9
.Sh HISTORY
The
.Nm
kernel's early history is not (properly) recorded.
It was developed
in a separate CVS branch until April 26, 1997, at which point it was
merged into 3.0-current.
By this date 3.0-current had already been
merged with Lite2 kernel code.
.Pp
.Fx 5.0
introduced support for a host of new synchronization primitives, and
a move towards fine-grained kernel locking rather than reliance on
a Giant kernel lock.
The SMPng Project relied heavily on the support of BSDi, who provided
reference source code from the fine-grained SMP implementation found
in
.Bsx .
.Pp
.Fx 5.0
also introduced support for SMP on the sparc64 architecture.
.Sh AUTHORS
.An Steve Passe Aq Mt fsmp@FreeBSD.org
.Sh CAVEATS
The
.Va kern.smp.threads_per_core
and
.Va kern.smp.cores
sysctl variables are provided as a best-effort guess.
If an architecture or platform adds SMT and
.Fx
has not yet implemented detection, the reported values may be inaccurate.
In this case,
.Va kern.smp.threads_per_core
will report
.Dv 1
and
.Va kern.smp.cores
will report the same value as
.Va hw.ncpu .