Re: [RFC] allow multi-core guests: introduce cores= option to -cpu

From: Gleb Natapov <gleb@redhat.com>
To: Avi Kivity <avi@redhat.com>
Cc: Alexander Graf <agraf@suse.de>,
	Andre Przywara <andre.przywara@amd.com>,
	anthony@codemonkey.ws, qemu-devel@nongnu.org,
	kvm@vger.kernel.org
Subject: Re: [RFC] allow multi-core guests: introduce cores= option to -cpu
Date: Sun, 5 Jul 2009 18:04:07 +0300	[thread overview]
Message-ID: <20090705150407.GK881@redhat.com> (raw)
In-Reply-To: <4A50BE5D.5010005@redhat.com>

On Sun, Jul 05, 2009 at 05:53:17PM +0300, Avi Kivity wrote:
> On 07/05/2009 04:23 PM, Alexander Graf wrote:
>>> I thought of using -smp [processors=]2,cores=4,threads=2 (for a total 
>>> of 16 threads), but I think it makes more sense with -cpu.
>>
>>
>> I actually think putting this in -smp makes more sense. -cpu really  
>> shouldn't need to be touched by normal users and as long as you can  
>> either -cpu host or -cpu safe that should be enough.
>
> Maybe.  But in that case -cpu core2duo should imply cores=2 and -smp 2  
> -cpu core2duo will bring up 4 cores spread across two sockets.
>
core2duo does not imply 2 cores. OSes use cpuid to discover this
information.

>> But then again maybe we should replace -smp with something more useful  
>> like -numa where you'd then specify #CPUs, #cores, mem-cpu connection,  
>> etc.
>
> I'd prefer -numa to specify the memory topology (and connections of  
> sockets to memory nodes), and -smp or -cpu to specify the intra-socket  
> topology.
>
I agree. -numa is a different story. It is possible to have zillion
socket setup without numa at all.

--
			Gleb.