a backup of the instructions used to set up a win10 kvm on linux

about_kvm.txt

`~/kvm` contains resourses used to set up kvm, gpu passthrough, virtio and bridged
networking. 

`/etc/default/grub`: 
added 'intel_iommu=on iommu=pt kvm.ignore_msrs=1 video=efifb:off' to GRUB_CMDLINE_LINUX_DEFAULT
regenerated grub config using `sudo update-grub`

added `softdep nouveau pre: vfio-pci` to /etc/modprobe.d/vfio.conf so that vfio
would load before nouveau.

IOMMU configured according to https://wiki.archlinux.org/title/PCI_passthrough_via_OVMF
run ./iommu.sh to see IOMMU groups

graphics card on the host is isolated by binding it to vfio-pci
(also by editing kernel parameters in /etc/default/grub)

current kernel parameter for vfio
vfio-pic.ids=10de:1c81,10de:0fb9,1033:0194

current virtual machine configuration:
sudo virt-install --boot uefi --os-variant win10 \
	--name win10 --ram=16384 --vcpus 4 \
	--cpu host,-hypervisor --features kvm_hidden=on --hvm \
	--disk path=/media/hdstorage/win10-gaming.img,size=600 \
	--cdrom /media/hdstorage/win10_1909.iso \
	--graphics vnc

kvm_hidden enabled to avoid virtual machine detection
`--cpu host,-hypervisor` hides the CPUID bit, also for avoiding virtual machine detection

port forward 127.0.0.1:5900 to connect to vnc

see https://wiki.libvirt.org/page/Networking for network configuations.

a bridged network `br0` is set up by following:
  https://fabianlee.org/2019/04/01/kvm-creating-a-bridged-network-with-netplan-on-ubuntu-bionic/
for use by the kvm guest windows10, as well as enp3s0 (host ethernet)
see `/etc/netplan/`

!!IMPORTANT: disable iptables and multicast processing for bridges
```
echo 0 > /proc/sys/net/bridge/bridge-nf-call-iptables
echo 0 > /sys/devices/virtual/net/br0/bridge/multicast_querier
echo 0 > /sys/devices/virtual/net/br0/bridge/multicast_snooping
```
otherwise, the bridge won't relay ARP and DHCP requests.
this server is configured to do this by default,
by adding `net/bridge/bridge-nf-call-iptables = 0` to `/etc/ufw/sysctl.conf`
(note it doesn't work on this system, strange. In the end I set iptables to ACCEPT for all bridges)
final solution: `iptables -I FORWARD -m physdev --physdev-is-bridged -j ACCEPT`

stuffs added by `virsh edit win10` after vm creation.
 * <cputune> for pinning vcpu to physical cpu (performance)
 * <topology> for enabling all 4 cpu cores
 * added virtio-scsi controller: <controller type='scsi' index='0' model='virtio-scsi'>
 * installed `virtio-win-guest-tools.exe` inside the guest os
   (do this strictly after adding virtio controller, but before switching drive bus mode,
     otherwise BSOD.)
 * changed <target dev='sda' bus='sata'/> to <target dev='sda' bus='scsi'/>  for virtio
 * changed network configuration to: 
 * <interface type='bridge'>
      <mac address='52:54:00:14:0e:1f'/>
      <source bridge='br0'/>
      <model type='virtio'/>
   </interface>
   (for bridged networking and virtio network card)
 * used <hostdev> to passthrough a NVIDIA GPU, GPU audio and a NEC USB 3.0 Controller
   (paths given by lspci are 01:00.0 and 01:00.1 and 04:00.0, respectively)


MSI enabled on gfxcard following this guide: http://vfio.blogspot.com/2014/09/vfio-interrupts-and-how-to-coax-windows.html
(for fixing distorted hdmi audio)

Uploaded to GitHub Gist in case of a drive failure.