VFIO/start-win10

#!/bin/bash
#
# script to create cpusets to isolate CPU threads for KVM/libvirt VMs
# requires 'cset'
# available on Fedora through my COPR:
#  - https://copr.fedorainfracloud.org/coprs/jlay/cpuset/
# review comments below regarding CPUs and XML entries
#
# this script:
# - isolates host tasks from the VM CPUs
# - applies some generic tweaks
# - resets host isolation on VM shutdown
#
# define the CPUs (cores/threads) to isolate for the VM
VM_CORES='0-20,24-44'
# Tip: refer to 'virsh capabilities | grep siblings'
# this will show CPU topology information for the host
# allows one to pick optimal layout for the VM
#
# the pinned CPUs are definend like so in the VM libvirt XML:
#    <vcpupin vcpu='0' cpuset='0'/>
#    <vcpupin vcpu='1' cpuset='24'/>
#    <vcpupin vcpu='2' cpuset='1'/>
#    <vcpupin vcpu='3' cpuset='25'/>
# [...]
#     <emulatorpin cpuset='10-11'/>
#
# NOTE: the pinned emulator CPUs must be included in VM_CORES
# probably any iothreads too
#
# example above is based on a threadripper 3960x:
#  - one NUMA node
#  - SMT enabled
#
# reduce kernel jitter
sudo sh -c 'sysctl vm.stat_interval=120 ; sysctl kernel.watchdog=0'

# attempt to reset shielding, we want to create it with current params
# and remove any existing slices from previous VM boots
# we want it to be created with current params, such as possibly-changed pinned CPUs
sudo cset shield --reset --sysset=host.slice --userset=windows10.slice || echo "something happened resetting shielding"

# destroy old cpusets
sudo cset set -d windows10.slice || echo "ignoring failure to destroy VM cpuset"
sudo cset set -d host.slice || echo "ignoring failure to destroy host cpuset"

# shield cores
# userset slice must match the partition defined in libvirt xml, eg:
#
#  </cputune>
#  <resource>
#    <partition>/windows10</partition>
#  </resource>
#  <os>
sudo cset shield -c $VM_CORES --sysset=host.slice --userset=windows10.slice

# start the VM
virsh start windows10

# determine QEMU vCPU PIDs
QEMU_PIDS=$(sudo egrep 'vcpu.*pid' /var/run/libvirt/qemu/windows10.xml | cut -d\' -f4)

# set higher priority for vCPU PIDs
sudo renice -20 -p ${QEMU_PIDS}
# either of these attempts at real time seem to freeze on boot
# try a different kernel (not 5.8.11)
#for CPUPID in ${QEMU_PIDS}; do 
##	sudo chrt -f -p 99 ${CPUPID}
#	sudo chrt -r -p 99 ${CPUPID}
#done

echo "Waiting for windows10 VM to stop before resetting cpusets"
while true; do
	# chill a bit
	sleep 10
	# get vm state, check if it's off
	if test -f "/var/run/libvirt/qemu/windows10.xml"; then
		VM_STATE="on" # doesn't do much (yet?), just here to fill a spot
	else
		VM_STATE="off"
		echo -e "windows10 VM shut down, setting cpusets back to normal\n"
		sudo cset shield --reset --sysset=host.slice --userset=windows10.slice
		exit 0
	fi
done