Re: [Qemu-devel] v2: KVM: support for 'count' in CPUID functions 4, 0xb and 0xd.

From: Amit Shah <amit.shah@redhat.com>
To: Alexander Graf <agraf@suse.de>
Cc: qemu-devel@nongnu.org, aliguori@us.ibm.com, avi@redhat.com,
	kvm@vger.kernel.org
Subject: Re: [Qemu-devel] v2:  KVM: support for 'count' in CPUID functions 4, 0xb and 0xd.
Date: Tue, 13 Jan 2009 14:40:53 +0530	[thread overview]
Message-ID: <20090113091053.GA22533@amit-x200.pnq.redhat.com> (raw)
In-Reply-To: <122E97F4-926C-44D4-B595-257AFBB08966@suse.de>

On Tue, Jan 13, 2009 at 09:25:32AM +0100, Alexander Graf wrote:
>
> On 13.01.2009, at 08:09, Amit Shah wrote:
>
>> These two patches were sent separately earlier. There are a couple of
>> small changes in this version, including removal of a debug message  
>> that
>> was left behind.
>
> Are you sure this actually implements the CPUID 4 leaf? I bet it simply 
> submits multiple KVM_SET_CPUID ioctls, from which the kernel module picks 

I've also sent kernel patches to the kvm list.

Why do you think we submit multiple KVM_SET_CUPID calls, btw?

> the last one (0), not exposing _any_ cache information to the guest.

Please also see the kernel patches.

> For exactly this purpose KVM_SET_CPUID2 was introduced quite a while  
> ago, enabling support for all CPUID leaf flavors on the kernel side.  
> Merely the userspace call is missing.

We can do without using SET_CPUID2 (and hence not needing to do any
trickery for kernel-userspace version mismatches).

> Nevertheless - thanks for taking onto this. Cpuid4 enabling is one of  
> the (pretty low-prio) items on my TODO list for Mac OS X guest support.

Guess it'll work now. With these patches, not only does the "core2duo"
cpu type work with kvm, it also helps us get closer to use the "host"
cpu type. (Also, a simple userspace program in the guest that dumps
cpuid values indeed shows the correct values for all the sub-leaves for
function 4).

Amit