Re: [Qemu-devel] CPU code roadmap (was Re: [PATCH 0/7] x86 CPU patches that didn't get into 1.2)

[Eduardo Habkost]

On Mon, Sep 10, 2012 at 12:46:18PM -0300, Eduardo Habkost wrote:
> On Mon, Sep 10, 2012 at 05:17:03PM +0200, Andreas Färber wrote:
> > Hi Eduardo,
> > 
> > Am 05.09.2012 22:41, schrieb Eduardo Habkost:
> > > This is a small queue of patches that I consider "ready to go", that 
> > > didn't
> > > enter QEMU 1.2.
> > 
> > There's so many x86 CPU series on the list again... Am I understanding
> > correctly that this is the start, applying directly to master? Or does
> > it have further prerequisites?
> 
> Correct. This one does not have any prerequisites and used 'master' as
> base. But it is a prerequisite for many other series.
> 
> 
> > 
> > Would be great to get one batch reviewed and applied, so that we can all
> > test and review from the same base. ;)
> 
> True. At the same time, I wouldn't like to hold from submitting patches
> for review because others are still pending.
> 
> Let's try to be very clear about dependencies on each series, though, to
> avoid confusion (I could have been more explicit and mentioned that this
> can be applied directly to master).
> 
> I am aware of three "properly ordered" series on the list, by now:
> 
> 1) This one ("x86 CPU patches that didn't get into 1.2")
> 2) "unduplicate feature names" series, from myself, explicitly depending
>    on (1)
> 3) "CPU properties" series, from Igor, explicitly depending on (1) and (2)

New item:

4) [PATCH v2 0/4] Allow changing of Hypervisor CPUIDs
   http://article.gmane.org/gmane.comp.emulators.kvm.devel/97991

I am keeping the current queue at:
https://github.com/ehabkost/qemu/commits/cpu-queue

Note that "cpu-queue" is going to be a rebased-often branch (with
frequent non-fast-forward updates), as it is just a place where I am
storing the work in progress while it's being reviewed in the list.


> 
> And I am aware of the following series, that still need to be
> rebased/resubmitted:
> 
> - "Allow changing of Hypervisor CPUIDs" from Don Slutz
> - CPU model classes, from myself
> - APIC ID / topology fixes, from myself (previously sent as RFC)
> - Make X86CPU child of DeviceState (there are 3 proposals today, and I
>   don't know which one will be adopted
> 
> Also, I have plans to submit series for the following:
> 
> - Call GET_SUPPORTED_CPUID only once and store the results
> - General kvm_arch_get_supported_cpuid() refactor/cleanup
> - General refactoring of feature flag handling, to use a generic
>   "feature words" structure
> - Fix -cpu check/enforce to properly use GET_SUPPORTED_CPUID
> - Fix -cpu host to properly use GET_SUPPORTED_CPUID
> - Addition of feature flag names for CPUID leaf 7
> 
> Anything else I am missing?
> 
> -- 
> Eduardo
> 

-- 
Eduardo