[pve-devel] [PATCH qemu-server 4/8] machine: correctly select pve machine version for non pinned windows guests

[Dominik Csapak]

On 3/6/25 14:10, Fiona Ebner wrote:
> Am 06.03.25 um 11:44 schrieb Dominik Csapak:
>> when we don't have a specific machine version on a windows guest, we use
>> the creation meta info to pin the machine version. Currently we always
>> append the pve machine version from the current installed kvm version,
>> which is not necessarily the version we pinned the guest to.
>>
>> Instead, use either the info from the creation meta info if it exists,
>> or use 'pve0'.
>>
>> For non-windows machines, we used the current QEMU machine version so we
>> should use the pve machine version from that too.
>>
>> Signed-off-by: Dominik Csapak <d.csapak at proxmox.com>
>> ---
>>   PVE/QemuServer/Machine.pm | 13 +++++++++----
>>   1 file changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/PVE/QemuServer/Machine.pm b/PVE/QemuServer/Machine.pm
>> index f1acde8f..e3da8e21 100644
>> --- a/PVE/QemuServer/Machine.pm
>> +++ b/PVE/QemuServer/Machine.pm
>> @@ -237,14 +237,19 @@ sub get_vm_machine {
>>   		if (PVE::QemuServer::Helpers::min_version($meta->{'creation-qemu'}, 9, 1)) {
>>   		    # need only major.minor
>>   		    ($base_version) = ($meta->{'creation-qemu'} =~ m/^(\d+.\d+)/);
>> +		    # append pve machine version if we have one
>> +		    if (my $pvever = $meta->{'creation-pve-machine'}) {
>> +			$base_version .= "+pve$pvever"
> 
> Since this is only the fallback handling for the rare edge case where no
> explicit machine version is set for a Windows guest, not sure if it's
> even worth doing this. I.e. can we just avoid the additional meta
> property and always use pve0 here? Did you intend any other use for the
> creation-pve-machine?
> 

I though the intention of the code was to pin the guest to that version where it
was created. If we omit the machine version, this is not true anymore,
since I'd then create a windows vm with e.g. pc-q35-9.2+pve1 then set it
to q35, and would get pc-q35-9.2+pve0

and while i don't have anymore use cases for the current case, wouldn't
this approach here correct also when we decide to bump again in the future?

also recording with which pve version the vm was created could help in debugging
issues at some point.. (my patch only adds the version in the meta info if it's > 0 anyway)

> Further below we already have:
> 
>> 	# for version-pinned machines that do not include a pve-version (e.g.
>> 	# pc-q35-4.1), we assume 0 to keep them stable in case we bump
>> 	$machine .= '+pve0';
> 
> so let's just follow this here too?

this here make sense IMHO, since the user wanted a specific version but without
pve versions info, so default to pve0 sounds fine to me?

> 
> 
>> +		    }
>>   		}
>>   	    }
>>   	    $machine = windows_get_pinned_machine_version($machine, $base_version, $kvmversion);
>> +	} else{
>> +	    $arch //= 'x86_64';
>> +	    $machine ||= default_machine_for_arch($arch);
>> +	    my $pvever = get_pve_version($kvmversion);
>> +	    $machine .= "+pve$pvever";
>>   	}
>> -	$arch //= 'x86_64';
>> -	$machine ||= default_machine_for_arch($arch);
>> -	my $pvever = get_pve_version($kvmversion);
>> -	$machine .= "+pve$pvever";
>>       }
>>   
>>       if ($machine !~ m/\+pve\d+?(?:\.pxe)?$/) {
>