Re: [libvirt] [PATCH v4 3/8] libxl: error out on not supported CPU mode, instead of silently ignoring

On 02/15/2018 02:47 PM, Marek Marczykowski-Górecki wrote: > On Tue, Feb 13, 2018 at 09:02:35AM -0700, Jim Fehlig wrote: > > It looks like we never answered my question from V3, i.e. should we change > > the default mode in the post-parse callback if NUMA or CPU features are > > defined within <cpu>? > > Hmm, but this means changing the config behind user's back, no? Well, sort of. But it is really just making the implicit explicit. > You have disabled nested HVM, upgrade, now you have enabled. > Global switch is some consolation here, but is it enough? I _think_ so. With a global default that disables nesting, are there any scenarios you envision where a previously disabled nesting becomes enabled after upgrade?

BTW, I'm fine with this patch, just playing devil's advocate with handling such things post-parse vs domain start. It's not the only place where default xen config is not reflected in libvirt :-/. Regards, Jim > > > It sure feels like such configuration tweaks and error checks should be > > handled in the parsing phase, similar to qemuDomainDefVcpusPostParse() and > > qemuDomainDefCPUPostParse() in the qemu driver. I think danpb is vaguely > > familiar with this series and can help answer that question. Daniel, do you > > have an opinion? Or a general comment about guidelines for checking/tweaking > > config in parse phase vs domain start phase? >