Add an initial NUMA affinity/policy configuration for threads and processes.

This is based on work done by jeff@ and jhb@, as well as the numa.diff
patch that has been circulating when someone asks for first-touch NUMA
on -10 or -11.

* Introduce a simple set of VM policy and iterator types.
* tie the policy types into the vm_phys path for now, mirroring how
  the initial first-touch allocation work was enabled.
* add syscalls to control changing thread and process defaults.
* add a global NUMA VM domain policy.
* implement a simple cascade policy order - if a thread policy exists, use it;
  if a process policy exists, use it; use the default policy.
* processes inherit policies from their parent processes, threads inherit
  policies from their parent threads.
* add a simple tool (numactl) to query and modify default thread/process
  policities.
* add documentation for the new syscalls, for numa and for numactl.
* re-enable first touch NUMA again by default, as now policies can be
  set in a variety of methods.

This is only relevant for very specific workloads.

This doesn't pretend to be a final NUMA solution.

The previous defaults in -HEAD (with MAXMEMDOM set) can be achieved by
'sysctl vm.default_policy=rr'.

This is only relevant if MAXMEMDOM is set to something other than 1.
Ie, if you're using GENERIC or a modified kernel with non-NUMA, then
this is a glorified no-op for you.

Thank you to Norse Corp for giving me access to rather large
(for FreeBSD!) NUMA machines in order to develop and verify this.

Thank you to Dell for providing me with dual socket sandybridge
and westmere v3 hardware to do NUMA development with.

Thank you to Scott Long at Netflix for providing me with access
to the two-socket, four-domain haswell v3 hardware.

Thank you to Peter Holm for running the stress testing suite
against the NUMA branch during various stages of development!

Tested:

* MIPS (regression testing; non-NUMA)
* i386 (regression testing; non-NUMA GENERIC)
* amd64 (regression testing; non-NUMA GENERIC)
* westmere, 2 socket (thankyou norse!)
* sandy bridge, 2 socket (thankyou dell!)
* ivy bridge, 2 socket (thankyou norse!)
* westmere-EX, 4 socket / 1TB RAM (thankyou norse!)
* haswell, 2 socket (thankyou norse!)
* haswell v3, 2 socket (thankyou dell)
* haswell v3, 2x18 core (thankyou scott long / netflix!)

* Peter Holm ran a stress test suite on this work and found one
  issue, but has not been able to verify it (it doesn't look NUMA
  related, and he only saw it once over many testing runs.)

* I've tested bhyve instances running in fixed NUMA domains and cpusets;
  all seems to work correctly.

Verified:

* intel-pcm - pcm-numa.x and pcm-memory.x, whilst selecting different
  NUMA policies for processes under test.

Review:

This was reviewed through phabricator (https://reviews.freebsd.org/D2559)
as well as privately and via emails to freebsd-arch@.  The git history
with specific attributes is available at https://github.com/erikarn/freebsd/
in the NUMA branch (https://github.com/erikarn/freebsd/compare/local/adrian_numa_policy).

This has been reviewed by a number of people (stas, rpaulo, kib, ngie,
wblock) but not achieved a clear consensus.  My hope is that with further
exposure and testing more functionality can be implemented and evaluated.

Notes:

* The VM doesn't handle unbalanced domains very well, and if you have an overly
  unbalanced memory setup whilst under high memory pressure, VM page allocation
  may fail leading to a kernel panic.  This was a problem in the past, but it's
  much more easily triggered now with these tools.

* This work only controls the path through vm_phys; it doesn't yet strongly/predictably
  affect contigmalloc, KVA placement, UMA, etc.  So, driver placement of memory
  isn't really guaranteed in any way.  That's next on my plate.

Sponsored by:	Norse Corp, Inc.; Dell

2 files changed, 173 insertions, 0 deletions

diff --git a/share/man/man4/Makefile b/share/man/man4/Makefile
index 498f905..cd4c91b 100644
--- a/share/man/man4/Makefile
+++ b/share/man/man4/Makefile
@@ -364,6 +364,7 @@ MAN=	aac.4 \
 	nsp.4 \
 	${_ntb.4} \
 	null.4 \
+	numa.4 \
 	${_nvd.4} \
 	${_nvme.4} \
 	${_nvram.4} \
diff --git a/share/man/man4/numa.4 b/share/man/man4/numa.4
new file mode 100644
index 0000000..984c464
--- /dev/null
+++ b/share/man/man4/numa.4
@@ -0,0 +1,172 @@
+.\" Copyright (c) 2015 Adrian Chadd <adrian@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. Redistributions in binary form must reproduce the above copyright
+.\"    notice, this list of conditions and the following disclaimer in the
+.\"    documentation and/or other materials provided with the distribution.
+.\"
+.\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS 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 AUTHORS 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.
+.\"
+.\" $FreeBSD$
+.\"
+.Dd May 10, 2015
+.Dt NUMA 4
+.Os
+.Sh NAME
+.Nm NUMA
+.Nd Non-Uniform Memory Access
+.Sh SYNOPSIS
+.Cd options SMP
+.Cd options MAXMEMDOM=16
+.Pp
+.In sys/numa.h
+.In sys/cpuset.h
+.In sys/bus.h
+.Sh DESCRIPTION
+Non-Uniform Memory Access is a computer architecture design which
+involves unequal costs between processors, memory and IO devices
+in a given system.
+.Pp
+In a
+.Nm
+architecture, the latency to access specific memory or IO devices
+depends upon which processor the memory or device is attached to.
+Accessing memory local to a processor is faster than accessing memory
+that is connected to one of the other processors.
+.Pp
+.Nm
+is enabled when the
+.Cd MAXMEMDOM
+option is used in a kernel configuration
+file and is set to a value greater than 1.
+.Pp
+Thread and process
+.Nm
+policies are controlled with the
+.Xr numa_setaffinity 2
+and
+.Xr numa_getaffinity 2
+syscalls.
+.Pp
+The
+.Xr numactl 1
+tool is available for starting processes with a non-default
+policy, or to change the policy of an existing thread or process.
+.Pp
+Systems with non-uniform access to I/O devices may mark those devices
+with the local VM domain identifier.
+Drivers can find out their local domain information by calling
+.Xr bus_get_domain 9 .
+.Ss MIB Variables
+The operation of
+.Nm
+is controlled and exposes information with these
+.Xr sysctl 8
+MIB variables:
+.Pp
+.Bl -tag -width indent -compact
+.It Va vm.ndomains
+The number of VM domains which have been detected.
+.Pp
+.It Va vm.default_policy
+The default VM domain allocation policy.
+Defaults to "first-touch-rr".
+The valid values are "first-touch", "first-touch-rr",
+"rr", where "rr" is a short-hand for "round-robin."
+See
+.Xr numa_setaffinity 2
+for more information about the available policies.
+.Pp
+.It Va vm.phys_locality
+A table indicating the relative cost of each VM domain to each other.
+A value of 10 indicates equal cost.
+A value of -1 means the locality map is not available or no
+locality information is available.
+.Pp
+.It Va vm.phys_segs
+The map of physical memory, grouped by VM domain.
+.El
+.Sh IMPLEMENTATION NOTES
+The current
+.Nm
+implementation is VM-focused.
+The hardware
+.Nm
+domains are mapped into a contiguous, non-sparse
+VM domain space, starting from 0.
+Thus, VM domain information (for example, the domain identifier) is not
+necessarily the same as is found in the hardware specific information.
+.Pp
+The
+.Nm
+allocation policies are implemented as a policy and iterator in
+.Pa sys/vm/vm_domain.c
+and
+.Pa sys/vm/vm_domain.h .
+Policy information is available in both struct thread and struct proc.
+Processes inherit
+.Nm
+policy from parent processes and threads inherit
+.Nm
+policy from parent threads.
+Note that threads do not explicitly inherit their
+.Nm
+policy from processes.
+Instead, if no thread policy is set, the system
+will fall back to the process policy.
+.Pp
+For now,
+.Nm
+domain policies only influence physical page allocation in
+.Pa sys/vm/vm_phys.c .
+This is useful for userland memory allocation, but not for kernel
+and driver memory allocation.
+These features will be implemented in future work.
+.Sh SEE ALSO
+.Xr numactl 1 ,
+.Xr numa_getaffinity 2 ,
+.Xr numa_setaffinity 2 ,
+.Xr bus_get_domain 9
+.Sh HISTORY
+.Nm
+first appeared in
+.Fx 9.0
+as a first-touch allocation policy with a fail-over to round-robin allocation
+and was not configurable.
+It was then modified in
+.Fx 10.0
+to implement a round-robin allocation policy and was also not configurable.
+.Pp
+The
+.Xr numa_getaffinity 2
+and
+.Xr numa_setaffinity 2
+syscalls first appeared in
+.Fx 11.0 .
+.Pp
+The
+.Xr numactl 1
+tool first appeared in
+.Fx 11.0 .
+.Sh AUTHORS
+This manual page written by
+.An Adrian Chadd Aq Mt adrian@FreeBSD.org .
+.Sh NOTES
+No statistics are kept to indicate how often
+.Nm
+allocation policies succeed or fail.