Re: [libvirt] [PATCH 06/30] conf: move virt type / os type / arch validation to post-parse

On 12/10/19 5:57 AM, Daniel P. Berrangé wrote: > >> Another issue: other drivers use the machine= setting too, libxl at >> least. This appears to drop filling in a default machine type for them, >> at least at XML parse time. > > I thought that was the case too, but I couln't find any functional use > of def->os.machine in any of the other drivers: > > $ git grep os.machine | grep -v -E '(qemu|conf)/' > libvirt-domain.c: * "os.machine" - the machine hardware name as a string > libxl/libxl_conf.c: def->os.machine); > libxl/xen_common.c: def->os.machine = g_strdup(capsdata->machinetype); > vbox/vbox_common.c: VIR_DEBUG("def->os.machine %s", def->os.machine); > vz/vz_sdk.c: if (def->os.machine != NULL || def->os.bootmenu != 0 || > > > that libxl_conf.c usage is just an error message > > The xen_common.c usage is just when parsing XM/XL config files, > so output only. > > The vz_sdk.c usage reports error for any non-NULL machine > > > I guess there is a difference in that if the user does dumpxml for a > libxl guest we've no longer filled in the machine. I can fix that, > but it doesn't look like it'll have a functional effect. Check grep for 'xenfv' and 'xenpv', domaincapabilities does some machine value handling, and virt-manager will use the reported VM machine value and feed it back to domaincapabilities, so it could be used in a round about way.