d/p/target-s390x-Fix-translation-exception-on-illegal-in.patch: avoid segfaults by uretprobes (LP 1929926)

This is upstream with 6.1 and can then be dropped.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

d/control*, d/rules: disable xen by default, but provide universe package qemu-system-x86-xen as alternative

xen went to universe in Ubuntu and in general we wanted to split
support/bug handling of the much less exercised xen from kvm use cases.

Therefore in Ubuntu the qemu-system-x86 in main can not depend on xen
anymore, but we provide a separate qemu-system-x86-xen which fulfills
the xen-hipervisor needs and at the same time keeps the more commonly
used qemu-system-x86 free of xen code.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

Same reasoning as with qemu-img, there are many common cases which
just should work out of the box.

Note: this was discussed with Debian back then but they were not
interested. Our reasoning was very strong around the Ubuntu OpenStack
usage which isn't that important for Debian. Therefore this change is
Ubuntu-only.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

qemu-utils are most of the time used to inspect/interact with images
of some sort and on the server side a common one is ceph.
  https://docs.ceph.com/en/latest/rbd/qemu-rbd/



Adding qemu-block-extra as a depedency just helps to always be able
to handle all device types and isn't too much of an extra footprint.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

qemu-system can load (block for now) modules after startup
by means of device hotplug mechanism. When qemu and modules
are upgraded, but running instances are not, running instances
(which are older qemu) can't load new modules from disk.
For this to work, during upgrade we copy old modules to
/run/qemu/$version, and running (old) qemu can load modules
from there.  But in order to be able to do that, this directory
should be mounted without -o noexec. By default on debian whole
/run is mounted -o noexec, so this does not work by default.
qemu-block-extra provides a systemd mount unit run-qemu.mount
which mounts /run/qemu as a separate tmpfs without -o noexec.
On debian we do not enable this mount unit by default since
apparently this is against policy. But ubuntu wants to be
more user-friendly and enables it by default. Do this by
running dh_installsystemd with arguments depending on $VENDOR.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

Fix saving modules on upgrades (LP: #1913421)

See merge request qemu-team/qemu!20

enable SDL as secondary UI backend

See merge request qemu-team/qemu!24

(#979679, CVE-2020-35504, #984455, CVE-2020-35505, #984454, CVE-2020-35506)

usb-redir-avoid-dynamic-stack-allocation-CVE-2021-3527.patch
(#988157)

fix various crashes in ppc system emulation.
Thanks to Christian Ehrhardt for pointing this out

On apt install --reinstall those would fire in an unexpected way.

Preparing to unpack .../qemu-block-extra_1%3a6.0+dfsg-1~ubuntu1_amd64.deb ...
+ set -e
+ [ -d /run/qemu ]
+ exit
dpkg: warning: old qemu-block-extra package pre-removal script subprocess returned error exit status 1
dpkg: trying script from the new package instead ...
dpkg: ... it looks like that went OK

Making the RC explicit resolves this issue.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

An admin can make /run/qemu non noexec in any way he likes. But providing
a (default disabled) mount unit is an easy way to make that easy as admin
(and all sort of configuration frameworks) are used to enable/start
systemd units.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

As discussed there is no point in saving them as long as /run/qemu is
mounted as noexec - also this can for the time being serve as a opt-in.
Therefore only save old modules when the underlying mount point of
/run/qemu is not noexec.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

After upgrading through multiple versions of the package there might
be more than just one set of modules backed up.

We don't want to be hard on remove, but on purge clear the whole
directory.

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

d/control: qemu 6.0 broke libvirt <7.2 add a breaks to avoid partial upgrade issues (LP: #1932264)

See merge request qemu-team/qemu!23

  - d/control: add build dependency libsdl2-dev
  - d/control: enable sdl on build
  - d/control: enable sdl audio on build
  - d/qemu-system-gui.install: add ui-sdl.so and audio-sdl.so
  - d/control: add runtime dependency to libgl1
  - d/p/build-most-modules-statically-hack.diff: build sdl audio as module

Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>

With qemu 6.0 you'll see issues with hotplug on attach-device (there are more
cases than the examples show)

virsh -c qemu:///system attach-device impish-2nd-hotplug hotplug-rng.xml
error: Failed to attach device from hotplug-rng.xml
error: internal error: unable to execute QEMU command
  'object-add': Parameter 'props' is unexpected

virsh -c qemu:///system attach-device impish-2nd-hotplug hotplug-mem64.xml
error: Failed to attach device from hotplug-mem64.xml
error: internal error: unable to execute QEMU command
  'object-add': Parameter 'size' is missing

This is due to qemu
https://git.qemu.org/?p=qemu.git;a=commit;h=50243407457a
It was reported and discussed in
https://bugzilla.redhat.com/show_bug.cgi?id=1944985
It is intentional and will stay that way, so higher layers need to adapt.
https://libvirt.org/git/?p=libvirt.git;a=commit;h=e0eeb2cc67d9
https://libvirt.org/git/?p=libvirt.git;a=commit;h=96850dfe216f
https://libvirt.org/git/?p=libvirt.git;a=commit;h=4f33b817b292

That means >=7.2 is needed for qemu 6.0 and to avoid any potential
issues on e.g. partial upgrades adding a breaks in d/control will help.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>