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>
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>
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>
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>
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>
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>
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>
Shuffling bits of code around, adding comments and grouping CLI options
to make the code easier to read and understand at a glance.
Brings the ordering of XML options in line with libvirt's own output as
implemented in `src/conf/cpu_conf.c` and `src/conf/numa_conf.c`.
This includes support for the following options:
* numa.interconnects.latency[0-9]*.initiator
* numa.interconnects.latency[0-9]*.target
* numa.interconnects.latency[0-9]*.cache
* numa.interconnects.latency[0-9]*.type
* numa.interconnects.latency[0-9]*.value
* numa.interconnects.latency[0-9]*.unit
* The same suboptions for `numa.interconnects.bandwith[0-9]*`
Note that the cache= attribute is only explicitly defined for <latency>
nodes in the documentation. However, since <latency> and <bandwidth>
nodes are otherwise identical, the docs also don't explicitly forbid it
for <bandwidth> nodes, and libvirt happily accepts XML that does specify
it for for <bandwidth> nodes, this implements the cache= attribute for
<bandwidth> elements as well.
This includes support for the following options:
* numa.cell[0-9]*.cache[0-9]*.level
* numa.cell[0-9]*.cache[0-9]*.associativity
* numa.cell[0-9]*.cache[0-9]*.policy
* numa.cell[0-9]*.cache[0-9]*.size.value
* numa.cell[0-9]*.cache[0-9]*.size.unit
* numa.cell[0-9]*.cache[0-9]*.line.value
* numa.cell[0-9]*.cache[0-9]*.line.unit
Init a shared log instance in virtinst/logger.py, and use that
throughout the code base, so we aren't calling directly into
'logging'. This helps protect our logging output from being
cluttered with other library output, as happens with some
'requests' usage
For CPU modes other then "custom" there is no model so we should not
check the suffix of model name.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
VM configured with mode="host-model" will have the CPU definition
expanded once the VM is started. Libvirt will try to use the closest
CPU model with some features enabled/disabled.
The issue is that there are some models that include spec-ctrl or ibpb
features and they will not appear in the explicit list of features and
virt-manager will not correctly detect if all security features are
enabled or not. As a workaround we can check the suffix of CPU model to
figure out which security features are enabled by the model itself.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
By default we copy CPU security features to the guest if specific CPU
model is selected. However, this may break migration and will affect
performance of the guest. This adds an option to disable this default
behavior.
The checkbox is clickable only on x86 and only on host where we can
detect any CPU security features, otherwise a tooltip is set to notify
users that there is nothing to copy.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
This will allow users to override the default behavior of virt-install
which copies CPU security features available on the host to the guest
XML if specific CPU model is configured.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
If user selects specific CPU model to be configured for guest we will
automatically add CPU security features to make sure that the guest is
not vulnerable to these CPU HW bugs.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1582667
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
We will need to pass another variable into the setter so we cannot use
the property setter.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
It is not True/False attribute, it can contain 'unknown' value as well
which means libvirt cannot figure out whether the CPU model is usable
with that specific QEMU binary.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
The copyright headers in every file were chjanged in this previous commit
commit b6dcee8eb7
Author: Cole Robinson <crobinso@redhat.com>
Date: Tue Mar 20 15:00:02 2018 -0400
Use consistent and minimal license header for every file
Where before this they said "
"either version 2 of the License, or (at your option) any later version."
Now they just say
"GNU GPLv2"
This fixes it to say "GNU GPLv2 or later" again.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Lin Ma