[These are some not-exhaustive notes based on an IRC chat with some of
the QEMU/libvirt developers -- Thomas Huth, Andrea Bolognani, Dan
Berrangé, Philippe Mathieu-Daudé.  Mistakes are my own.  This should
actually be captured in an upstream document on "input devices", but
until then we have to make do with this.] — 07 Oct 2020

(Virtual) Input Devices 101 for those in a hurry
================================================

For x86, currently you can't disable the PS/2 mouse & keyboard, as
they're baked into the machine type.

Non-x86 architectures
---------------------

Other architectures don't have those [PS/2 devices], and generally you
want to use a 'qemu-xhci' USB controller and then plug usb-tablet +
usb-keyboard in them.

'ppc64' uses USB, so 'usb-mouse', 'usb-keyboard', 'usb-YouNameIt' are
all supported.  These USB devices are pluggable, but are not built-in
[i.e. they _can_ be disabled if need be.]

In general, most non-x86 architectures, you can start guests without
input devices by using QEMU's '-nodefaults' command-line switch.


Quirk on 'ppc64'
----------------

On 'ppc64', if graphics are present libvirt will automatically add USB
keyboard and mouse too.  If graphics are present *and* a USB tablet is
also present, however, the USB mouse will be skipped to avoid having two
pointing devices.


Headless and headful guests; a word on s390x
---------------------------------------------

If you want a headless guest, don't add graphics or input devices and
you'll get as little of that as QEMU allows.

If you want a guest with GUI, add graphics as well as a 'qemu-xhci' USB
controller, a USB keyboard and a USB tablet for the best possible user
experience

This will work for all architectures except s390x, where you have to use
'virtio-input' instead of USB.


What is QEMU's '-nodefaults'?
-----------------------------

The idea of '-nodefaults' is to run minimal hardware, bare minimum,
nothing plugged in.   So if your motherboard comes with PS/2 and USB
buses built-in, then the buses "merely exist".  However you won't have
any device automatically created to use the said buses.

                - - -

Use `virsh qemu-monitor-command vm1 --hmp info qom-tree` to get a
fine-grained view of devices.