IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Most existing disk images expect UEFI; those that don't, require
either using direct kernel boot or manually providing a firmware
image, both of which imply user intervention.
Using UEFI by default means that at least images belonging to
the first group, of which more are going to pop up as the
ecosystem matures further, can work out of the box.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
With libvirt 9.8.0, the test driver now has a stub getDomainCapabilities
implementation. But we still have some code that needs to handle
a driver with missing domcaps.
Make our magicuri mock return NO_SUPPORT for domcaps, when the URI
doesn't have any domcaps XML passed in. This is enough for our test
purposes.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
When specifying the socket.source option for filesystem devices, like
this:
--filesystem type=mount,driver.type=virtiofs,source.socket=/xyz.sock,target.dir=tag1
virt-install is writing the xml as:
<filesystem type="mount">
<source>
<socket>/xyz.sock</socket>
</source>
<target dir="tag1"/>
<driver type="virtiofs"/>
</filesystem>
This produces an error such as:
ERROR missing source information for device mount_tag1
But the socket should be an attribute of source rather than a child
element. After this patch, the same command results in the following XML
and no error is produced:
<filesystem type="mount">
<source socket="/xyz.sock"/>
<target dir="tag1"/>
<driver type="virtiofs"/>
</filesystem>
Resolves: RHEL-1126
Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
When shim in the guest sees unpopulated EFI NVRAM, like when
we create a new UEFI VM, it invokes fallback.efi to populate
initial NVRAM boot entries. When the guest also has a TPM device,
shim will do a one time VM reset. This reset throws off the
reboot detection that is central to virt-install's install
process.
The main install case that this will usually be relevant is
the combo of UEFI and --cloud-init. The latter usually implies
use of a distro cloud image, which will be using shim, and the
--cloud-init process requires a multi stage install compared
to just a plain import install.
For that case, we disable the default TPM device for the first
boot.
https://bugzilla.redhat.com/show_bug.cgi?id=2133525
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Commit f107e39989 ("Switch to more traditional logging structure",
2019-06-17) replaced "logging.exception" with "log.exception", effectively
shifting the argument lists 4 characters to the left. The second and
further lines of multiline invocations were not accordingly unindented,
however, which ended up setting a suboptimal precedent as well. Unindent
those lines now.
Reported-by: Cole Robinson <crobinso@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Just to be sure, this patch implements the second approach (described in
the previous patch) as well.
Note that there is precedent for suppressing "guest_lookup" exceptions:
refer to the "Error determining machine list" branch from commit
ae7ebc220b ("details: Properly limit machine type list by guests
arch/type", 2013-09-01).
(
In fact, that branch gets activated when opening the details window for a
domain that uses a non-default emulator; the "virt-manager --debug" log
contains:
> ERROR (details:613) Error determining machine list
> Traceback (most recent call last):
> File "virtManager/details/details.py", line 605, in _init_details
> capsinfo = caps.guest_lookup(
> File "virtinst/capabilities.py", line 319, in guest_lookup
> raise ValueError(msg)
> ValueError: Host does not support domain type kvm with machine
> 'pc-q35-8.1' for virtualization type 'hvm' with architecture 'x86_64'
)
Fixes: #539
Fixes: 05fcc7410e ("virtinst: fix caching of domain capabilities", 2022-07-27)
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
When I try to open the details window for a domain that does not use the
system default emulator, I get the following exception:
> Traceback (most recent call last):
> File "virtManager/vmwindow.py", line 40, in get_instance
> cls._instances[key] = vmmVMWindow(vm)
> File "virtManager/vmwindow.py", line 83, in __init__
> self._details = vmmDetails(self.vm, self.builder, self.topwin,
> File "virtManager/details/details.py", line 389, in __init__
> self._init_details()
> File "virtManager/details/details.py", line 807, in _init_details
> vmmAddHardware.build_video_combo(self.vm, video_dev)
> File "virtManager/addhardware.py", line 816, in build_video_combo
> default = DeviceVideo.default_model(vm.xmlobj)
> File "virtinst/devices/video.py", line 47, in default_model
> if (guest.lookup_domcaps().supports_video_virtio() and
> File "virtinst/guest.py", line 656, in lookup_domcaps
> if not self._domcaps or not _compare(self._domcaps):
> File "virtinst/guest.py", line 646, in _compare
> if self.os.machine and not _compare_machine(domcaps):
> File "virtinst/guest.py", line 633, in _compare_machine
> capsinfo = self.lookup_capsinfo()
> File "virtinst/guest.py", line 674, in lookup_capsinfo
> self._capsinfo = self.conn.caps.guest_lookup(
> File "virtinst/capabilities.py", line 319, in guest_lookup
> raise ValueError(msg)
> ValueError: Host does not support domain type kvm with machine
> 'pc-q35-8.1' for virtualization type 'hvm' with architecture 'x86_64'
This is a regression; according to git-bisect, it was introduced in commit
05fcc7410e ("virtinst: fix caching of domain capabilities", 2022-07-27).
"lookup_capsinfo" (and "guest_lookup" called by it) are unsuitable for
machine type alias checking (or for anything else) if the domain uses an
emulator that differs from the system default emulator. The information
returned by virConnectGetCapabilities() pertains to the system default
emulator. Thus, when using a non-default emulator, we should either not
call "lookup_capsinfo" for machine type alias checking, *or* we should
suppress the exception, and pretend that the alias check was a mismatch.
It turns out that we can avoid the "lookup_capsinfo" call (and thereby the
exception) in practice if we just delay the call until after the direct
(non-alias) comparison.
Fixes: #539
Fixes: 05fcc7410e ("virtinst: fix caching of domain capabilities", 2022-07-27)
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Before this, on clone the serial file would remain the same for the cloned
domain. This doesn't make much sense as the output would be an intermix of two
unrelated output sequences.
Here we apply the the same filename changing algorithm, as with disk files.
Signed-off-by: Oleg Vasilev <oleg.vasilev@virtuozzo.com>
virt-install prints the total_size value to the progress bar even if it
is meaningless.
This value can be confusing to user, so for execute prosess that doesn't
copy files (total_size = 0B), we hide the total_size value.
For example, 'Creating domain...' doesn't need to print the total_size
value.
Signed-off-by: Toshiki Sonoda <sonoda.toshiki@fujitsu.com>
Signed-off-by: Haruka Ohata <ohata.haruka@fujitsu.com>
When a sparse file is created during a disk allocation,
virt-install prints not the created disk size but a sparse file size.
Therefore, we fix to print the created disk size during disk allocation
instead of the size of the sparse file by updating the meter with the
self.capacity.
Signed-off-by: Toshiki Sonoda <sonoda.toshiki@fujitsu.com>
Signed-off-by: Haruka Ohata <ohata.haruka@fujitsu.com>
Reproducer:
Reproducer:
./virt-install --connect test:///default \
--location tests/data/fakemedia/fake-f26-netinst.iso
Before:
Starting install...
Retrieving 'vmlinuz' | 0 B 00:00:00 ...
Retrieving 'initrd.img' | 0 B 00:00:00 ...
After:
Starting install...
Retrieving 'vmlinuz' | 9 B 00:00:00 ...
Retrieving 'initrd.img' | 9 B 00:00:00 ...
progress.end() currently only reports the total amount of bytes
that were last written to the UI. It should report the total amount
that's been passed to update().
Reported-by: Toshiki Sonoda <sonoda.toshiki@fujitsu.com>
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The sheepdog project is no longer actively developed, Libvirt removed
the support for sheepdog storage backend since v8.8.0, Let's drop it.
Signed-off-by: Lin Ma <lma@suse.com>
Nowadays it could be as simple as `virt-install --install fedora36`.
Trying to represent the interdepencies here is not worth it, but
let's keep a simple string around to avoid the default parser
usage string, which is huge
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This commit added support for cpu physical address bits control, It's
useful for VMs with huge amount of ram.
E.g.
--cpu Cascadelake-Server,maxphysaddr.mode=emulate,maxphysaddr.bits=46
Signed-off-by: Lin Ma <lma@suse.com>
Anything passed to --boot should imply --install no_install=yes
in the absence of other --install options. This is historically
what we've done but we regressed in 4.1.0
Resolves: https://github.com/virt-manager/virt-manager/issues/426
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Each bit here is part of the CLI API, we need to be sure we are
covering each one. Extend the test suite to hit one case we are missing
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Properly setting the metadata cache size can provide better performance
in case of using big qcow2 images.
This patch introduces two driver options:
* driver.metadata_cache.max_size
* driver.metadata_cache.max_size.unit
E.g. --disk ...,driver.type=qcow2,\
driver.metadata_cache.max_size=2,\
driver.metadata_cache.max_size.unit=MiB
BTW, Metadata cache size control is currently supported only for qcow2.
Regarding how to properly caluclate the cache size of qcow2, Please refer
to qemu's documentation.
Signed-off-by: Lin Ma <lma@suse.com>
We need to change the flow from
* parse all the strings
* set capabilities defaults
* build installer
* fill in all guest defaults
To
* parse boot and metadata strings
* set capabilities defaults
* build installer
* set --name default
* parse all the remaining strings
* fill in all guest defaults
Because --disk parsing depends on --name for some path generation.
So this fixes --disk names when --name is implicitly specified by
--install or --osinfo
Signed-off-by: Cole Robinson <crobinso@redhat.com>
It allows to set the thread pool size to optimize spawning worker threads
for the default event loop in real time environment. For example:
--iothreads defaultiothread.thread_pool_min=8,\
defaultiothread.thread_pool_max=16
Signed-off-by: Lin Ma <lma@suse.com>
These two optional attributes allow setting lower and upper boundary for
number of worker threads for given IOThread. For example:
--iothreads iothreads=2,\
iothreadids.iothread0.id=1,\
iothreadids.iothread1.id=2,\
iothreadids.iothread1.thread_pool_min=8,\
iothreadids.iothread1.thread_pool_max=16
Signed-off-by: Lin Ma <lma@suse.com>
Otherwise going host-passthrough -> custom can cause libvirt validation
error due to libvirt fills the default value(migratable='on') for the
host-passthrough in domain XML.
Signed-off-by: Lin Ma <lma@suse.com>
Remove the open coded version logging in cli.py and virt-manager
connection.py, and move it into virtinst connection open
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The various version numbers reported by libvirt are all relevant pieces
of information when debugging problem reports.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Knowledge of the version number is not always available from the user
bug report. They may merely have a log file from an automated system.
Including the version number in the log message is key information.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The domain capabilities XML always contains the canonical machine name.
When the selected machine type for a guest is an alias, the check
comparing it to the domain capabilities machine will always fail. This
prevents the domain capabilities result from being cached.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
This allows support for host/guest clipboard sharing when using vnc
guests (and possibly other graphics types in the future). This channel
is similar to the spicevmc channel, but it contains a couple additional
options to enable/disable clipboard sharing and specify the mouse mode.
In the case of spice, these settings are specified on the 'graphics'
element, but for qemu-vdagent, they are specified on the channel. For
example:
--channel=qemu-vdagent,source.clipboard.copypaste=on,source.mouse.mode=client
Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Since there's a release pending, this is a bit of future proofing
if win11 pops up in a future osinfo-db without accompanying
libosinfo firmware APIs landing
Signed-off-by: Cole Robinson <crobinso@redhat.com>
* Check it for test:/// URI too
* Move the whole logic into osdict.requires_firmware_efi
* Wrap it all in an exception handler the final API breaks us
* Add some coverage exclusions since this is currently untestable
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Windows 11 only supports EFI for installation, legacy BIOS is
discontinued as an option for new installs, unless you switch into
the console and run various regedit hacks in the installer.
This is not something we expect users to do, so osinfo-db will
report:
<firmware arch="x86_64" type="efi"/>
<firmware arch="x86_64" type="bios" supported="false"/>
for Windows 11.
When seeing 'bios' with supported=false, we need to prefer EFI
firmware no matter what.
For other OS we should stil prefer BIOS, because EFI blocks the usage of
internal qcow2 snapshots until some libvirt work is done to use the new
QMP snapshot commands.
Note this relies on a new API osinfo_os_get_complete_firmware_list
just added to libosinfo in version 1.10, so for most existing OS
distros this will be dormant functionality until they rebase to
latest libosinfo APIs.
The Windows 11 entry for osinfo-db will be the first that exposes
the metadata indicating lack of BIOS support.
Related: https://github.com/virt-manager/virt-manager/issues/310
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
`sudo` throws errors and is not even required to write the `disable`
file. Also add a message to the file that it was created by us
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Originally we thought it would be for the clouduser, but then
we changed it, and now it's ambiguous. Rename it to make the
usage clear, and add an alias to keep any users working
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The Guest object passed in by the user is no longer altered in place,
instead we act on copies of the original XML. We need to rework how
unattended/cloudinit device altering works a bit to handle this.
There's some XML churn because working on parsexml gives different
output ordering, but this is expected.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
* libvirt fills in cbitpos and reducedPhysBits for us
* libvirt errors if type is missing
* libvirt errors if host/qemu doesn't support sev
So drop it all. This simplifies testing because we don't need
sev domcaps in place just to generate the XML
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Via the virt-manager UI we aren't converting relative path to
absolute path, even though we do it internally when needed.
We were benefiting from this in the test suite in some ways, so we
need to adjust tests to strip out the dev dir on XML comparison
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This adds the power of --xml to individual device options. For example
this makes it easier to make custom XML changes for a single --disk
device from both virt-install and virt-xml
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This is a virt-xml option to refresh a VM to use the latest machine
type version for the machine type it's currently using. Ex:
pseries-2.11 -> pseries
pc-q35-5.0 -> q35
This is useful for when qemu deprecates and removes the machine type
out from under you, or to pick up bug fixes.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
We enable default graphics for all of ppc64, but really the only
thing we have ever tested is pseries, so make that clear.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This reverts commit 20d3bf9484.
Indeed graphics does work on aarch64, but for example, fedora
server aarch64 installs don't output any graphical output and only
put data on the serial console. So we would need to be more specific
about how we enable this if we make the change.
qemu 6.1, fairly new when we added this option, has an unfortunate
bug with >= 15 root ports, so we choose 14 instead of our original 16
https://gitlab.com/qemu-project/qemu/-/issues/641
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Libvirt defaults to PCIe for arm32/aarch64 and riscv -M virt too.
Rename q35_pcie_root_ports to num_pcie_root_ports and extend the
logic to those archs too
Signed-off-by: Cole Robinson <crobinso@redhat.com>
It's generally not as valuable for non-x86 where we don't have the
history of supporting non-virtio OSes, but as time goes on it will
likely become more relevant for non-x86 arches, so let's make this
change now to get ahead of it.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Otherwise going host-passthrough -> host-model -> host-passthrough
can cause libvirt validation errors or leave unintended bits in
the XML
Signed-off-by: Cole Robinson <crobinso@redhat.com>
- Move most xml suboption testing to many-devices test
- Clarify every specific bit we are testing in the singleton tests
- Consolidate/drop/reduce a lot of tests
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The code previously was just encoding the same defaults as libvirt,
which doesn't really add anything.
Instead, let's prefer type='emulator' model='tpm-crb', which
gives the most modern virtualization friendly config. When we don't
know if that will work, we mostly leave things up to libvirt to fill
in.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Break out the CPU secure stuff from being inline in the class, most
of the logic there is not relevant to understanding what the
domcapabilities API provides
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Add extra PCIe root ports to enable q35 device hotplug to work out
of the box. A typical modern linux guest has 7-8 PCI devices added
by default, so this gives plenty of wiggle room.
The smart thing to do would be to count the attached PCI devices
and add 4-5 extra, but that takes more work and isn't trivial.
The number can be overridden on the cli with:
--controller q35_pcie_root_ports=X
Use =0 to go back to the old behavior.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
When libvirt and qemu are new enough, use host-passthrough for the
CPU default. Nowadays this is recommended over host-model for most
end user usage where migration isn't a critical feature.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
These have been used in the past, but no current code touches them,
so drop the parsing infrastructure
Signed-off-by: Cole Robinson <crobinso@redhat.com>
host-copy was the old default, but it's fundamentally flawed. Since
we switched to host-model default a few years back, it's not advertised
in the docs or selectable via virt-manager any more.
Have it print a warning and invoke host-model-only
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This was previously discussed here:
https://listman.redhat.com/archives/virt-tools-list/2020-September/msg00017.html
For the x86 + hvm case, failure to specify an --osinfo/--os-variant
OS, and failure to detect an OS from install media, will now throw
a big error:
```
--os-variant/--osinfo OS name is required, but no value was
set or detected.
This is now a fatal error. Specifying an OS name is required
for modern, performant, and secure virtual machine defaults.
If you expected virt-install to detect an OS name from the
install media, you can set a fallback OS name with:
--osinfo detect=on,name=OSNAME
You can see a full list of possible OS name values with:
virt-install --osinfo list
If your Linux distro is not listed, try one of generic values
such as: linux2020, linux2018, linux2016
If you just need to get the old behavior back, you can use:
--osinfo detect=on,require=off
Or export VIRTINSTALL_OSINFO_DISABLE_REQUIRE=1
```
The thread goes into more detail, but basically, for x86 VMs at least,
it's unlikely you will _ever_ want the default 'generic' behavior,
which gives gives no virtio, no PCIe, no usb3, IDE disks, slow
network devices, etc.
Many people use virt-install in scripts and CI, and this may now
cause breakage. The environment variable is there to help them
get things back to normal as quick as possible, but it will still
noisy up their logs with the warning to hopefully get them to make
a useful change to their virt-install invocations.
This is limited to x86, since that's where most of our defaults
historically differ, and where we can depend on libosinfo to give
the most accurate device info. This may be relevant to change for
other KVM architectures in the future.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Historically over time it's been more likely that an issue is reported
for osinfo-db missing the usb-tablet annotation. So for example we
always enable it for the default 'generic' case. We also want to
enable it for osinfo's 'unknown' case too.
Rather than add another check for that, let's just drop the osinfo
checking entirely. Some very old OS don't support usb-tablet, but
specifying it for those cases doesn't cause issues AFAIK, and users
can override it with `--input none` if they want.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Currently we skip the warning if the user explicitly requests
`--osinfo generic`. Upcoming changes to the defaults here will make
it tempting to specify `--osinfo generic` just to make things work,
and we want to dissuade that, so enable the warning for that case too.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
We are about to change the some defaults around os handling. Let's
start recommending the nicer named --osinfo more, since new error
messages are going to promote it a bit as well
Signed-off-by: Cole Robinson <crobinso@redhat.com>
I'm still seeing blog posts that recommend using
--os-type linux --os-variant XXX
Which has been a no op for a long time but is mostly harmless.
Current git would make this an error condition, but that's too
disruptive IMO. Just print a warning
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Have the UI push users more towards better defaults, by discouraging
the 'generic' entry and offering the 'linuxXXXX' entries when their
distro or OS version is not in the list.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Previously we tried to use a combination of distro class and version
number to produce a correct ordering that was independent of the
osinfo short ID. The original intent was to have correct ordering
for Windows entries in the virt-manager UI, since the short ID
values are all over the place.
Nowadays that doesn't really matter, since we weed out old
unsupported entries by default. And in the mean time, our current
sort method gives some weird results like interspersing silverblue
entries with fedora entries.
Using a natural/human sort is simpler and handles things pretty well.
Change the UI to sort by the OS label too which preserves some of
the good behavior of original method
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The canonical tool for this is `osinfo-query os`, which we still
reference in the man pages and in the list output.
However, we are about to make missing --os-variant fatal for common
usage, and I don't want to force users to install an extra tool just
to figure out what an acceptable --os-variant value is.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
After checking with qemu devs, this option is not really recommended
for common usage and doesn't get used much in practice. So I don't
think it is suitable for the UI
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This recommendation came from an internal discussion. The cases are
* For block storage. This means guest requests are passed through
to the host device, which seems a more reasonable default than
ignoring them
* For sparse disk images we will create. discard=unmap helps preserve
the sparseness of the disk image. If a user requests non-sparse, they
are likely more concerned with performance than saving disk space,
so we leave the default as is. We limit this to disk images we will
create, since that's the easiest case to check, and it's less clear
if we should change the behavior here for an arbitrary existing
disk image.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This is from Gerd's suggestions here:
https://www.kraxel.org/blog/2019/09/display-devices-in-qemu/
When the guest supports it, we should use virtio. qxl is on the way
out, and the benefits are marginal and add a security and maintenance
burden.
While here, check domcaps that qxl or virtio are actually available.
Modern qemu has device modules, so device support may not be installed.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Let users choose libvirt's os.firmware=efi setting in the UI, putting
it about the firmware path list, since it's the preferred default
these days.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
<os firmware='efi'> is the libvirt official way to do what we
historically implement with `--boot uefi`, and UEFI setup in
virt-manager.
Let's prefer libvirt's official method if the support is advertised
in domcapabilities.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
This replaces the pattern:
Guest.set_uefi_path(Guest.get_uefi_path())
With a single entrypoint
Guest.enable_uefi()
to immediately change the guest config to use UEFI, using our
default logic.
This will make it easier to change that logic in the future, like
using <os firmware='efi'> instead of hardcoded paths
Signed-off-by: Cole Robinson <crobinso@redhat.com>
We were not correctly accounting for the internal representation of
some fields, and just trying to a string comparison. We need to be
a bit smarter than that
Fixes: https://github.com/virt-manager/virt-manager/issues/356
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Strip back the logic to:
* Only try to toggle source_type=memfd and access_mode=shared
* Disable the field if guest has any <numa> config
* Disable the field if domcaps does not report virtiofs and memfd
This is the simplest future proof case, though it will exclude some
legit guest configs and some libvirt+qemu back compat.
My feeling is the <numa> stuff in particular is pretty advanced, so if
users have it configured they can toggle shared memory via the XML
without too much trouble.
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Linux memfd memory backend doesn't require any host setup, We prefer to
use it as the simplest memory XML adjustments to make virtiofs works.
Signed-off-by: Lin Ma <lma@suse.com>
Check whether virtiofs is exposed in domcapabilities, We can use it as a
proxy for 'libvirt is new enough to allow bare memory access mode=shared'
as well.
Signed-off-by: Lin Ma <lma@suse.com>
Both these windows versions are now longer supported, and UEFI isn't
the default, so I don't think this hack is much needed anymore
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The bare metal world is moving to a situation where UEFI is going to be
the only supported firmware and there will be a strong expectation for
TPM and SecureBoot support.
With this in mind, if we're enabling UEFI on a VM, it makes sense to
also provide a TPM alongside it.
Since this requires swtpm to be installed we can't do this
unconditionally. The forthcoming libvirt release expands the domain
capabilities to report whether TPMs are supported, so we check that.
The user can disable the default TPM by requesting --tpm none
https://github.com/virt-manager/virt-manager/issues/310
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
This address string decomposing is strictly and virt-* cli feature.
Move it to cli.py to make that explicit
Signed-off-by: Cole Robinson <crobinso@redhat.com>
The sync_vcpus_topology method will sometimes set the self.vcpus prop,
but other times leave it unset. This is confusing an unhelpful
behaviour. Both callers have logic to set the self.vcpus prop
to a default value of sync_vcpus_topology failed to do so. It makes
more sense to just pass this default value in.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
When setting CPU defaults we want to force create the topology even if
the user has not specified anything. In particular this allows for
overriding the QEMU defaults, to expose vCPUs as cores instead of
sockets which is a much saner default for Windows.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
In real world silicon though it is rare to have high socket/die counts,
but common to have huge core counts.
Some OS will even refuse to use sockets over a certain count.
Thus we prefer to expose cores to the guest rather than sockets as the
default for missing fields.
This matches a recent change made in QEMU for new machine types
commit 4a0af2930a4e4f64ce551152fdb4b9e7be106408
Author: Yanan Wang <wangyanan55@huawei.com>
Date: Wed Sep 29 10:58:09 2021 +0800
machine: Prefer cores over sockets in smp parsing since 6.2
Closes: https://github.com/virt-manager/virt-manager/issues/155
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The product of sockets * dies * cores * threads must be equal to the
vCPU count. While libvirt and QEMU will report this error scenario,
it makes sense to catch it in virt-install, so we can test our local
logic for setting defaults for topology.
This exposes some inconsistent configurations in the test suite.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Any missing values in the topology need to be calculated based on the
other values which are set.
We can take account of fact that 'total_vcpus' treats any unset values
as being 1 to simplify the way we set topology defaults.
This ensures that topology defaulting takes account of dies.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
It is always permitted to set dies==1 regardless of architecture or
machine type. The only constraint is around setting values > 1, for
archs/machines that don't support the dies concept.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Although using --cpu topology.XXX is the preferred way to set topology,
it is still possible via the --vcpus parameter. For consistency, this
should support the full set of parameters, so dies needs to be added.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
We should only be returning a driver_type value for volumes that
report support_format(), meaning they support file type formats like
qcow2. Any other reported format should be ignored
Dropping the check for 'unknown' value changes one test case a bit,
but it hardcodes raw which is what libvirt gives us anyways, so it's
okay
Signed-off-by: Cole Robinson <crobinso@redhat.com>
Usually, when storage volume is attached as a disk and disk xml is filled with
default values, the "<driver type=...>" value is copied from volume's
"<format type=...>". This makes sense for volumes of storage pool of type
"dir", where format types include "raw, qcow2...".
However, the same approach cannot be used for the storage pool of type "disk".
In that case, format types include "none, linux, fat16, fat32...". Such formats
cannot be used for disk's "<driver type=...>".
Therefore, when generating disk XML for volume of storage pool type "disk",
driver type should always be "raw".
