On Tue, Dec 11, 2012 at 12:21:10AM +0100, Andreas Färber wrote: > Am 10.12.2012 19:55, schrieb Igor Mammedov: > > On Mon, 10 Dec 2012 10:46:13 -0200 > > Eduardo Habkost <ehabk...@redhat.com> wrote: > > > >> On Sun, Dec 09, 2012 at 08:45:50PM +0100, Andreas Färber wrote: > >>> Simplifies the upcoming cleanup of cpu_x86_find_by_name(). > >> > >> ...by making cpu_x86_register() more complicated, and having CPU model > >> name lookup spread into different parts of the code. > >> > >> The CPU model lookup is a bit complex because of the "host" exception, > >> but at least the complexity was hidden inside cpu_x86_find_by_name() > >> (making it very easy to replace that logic by a CPU subclass lookup, > >> later). > > (Somehow I didn't get this message ... yet) > > > +1 > > Could you guys re-review this in light of the subclasses patch? The > issue I was facing is that I did not see a reliable way to register the > host class depending on kvm_enabled(). Therefore ..._find_by_name() > becomes the class lookup whereas the host check remains a special check > in cpu_x86_init() [a faulty one I now see, ignoring "host-x86_64-cpu"]. > cpu_x86_register() is problematic in that it gets the X86CPU served on a > silver plate, so it's too late to choose subclasses in that function.
I have a patch[1] to change that in my APIC-ID-topology work in progress branch[2]. I can submit this (and maybe a few other code movements) as PATCH tomorrow. I have a branch for CPU subclasses using an approach similar to the one you used, but that includes "host" as well[3]. But I was waiting until the properties series is resubmitted, so I could use it as base and make the subclass code simpler. I will take a deeper look at your RFC tomorrow, and probably send additional comments. [1] https://github.com/ehabkost/qemu-hacks/commit/850c5531c84839919d650b2dac6e98fa4c2529c4 [2] https://github.com/ehabkost/qemu-hacks/commits/work/apicid [3] https://github.com/ehabkost/qemu-hacks/commits/work/cpu-model-classes -- Eduardo
