Re: [PATCH 3/4] Move KVMState to common header

From: Glauber Costa <glommer@redhat.com>
To: Jan Kiszka <jan.kiszka@web.de>
Cc: kvm@vger.kernel.org, avi@redhat.com,
	Anthony Liguori <anthony@codemonkey.ws>
Subject: Re: [PATCH 3/4] Move KVMState to common header
Date: Mon, 8 Jun 2009 17:14:49 -0300	[thread overview]
Message-ID: <20090608201449.GL11966@poweredge.glommer> (raw)
In-Reply-To: <4A2D6C7A.4080008@web.de>

On Mon, Jun 08, 2009 at 09:54:34PM +0200, Jan Kiszka wrote:
> Glauber Costa wrote:
> > This patch should be applied to main qemu, but I'll
> > first post it here for appreciation. In this patch,
> > we move KVMState definition to kvm.h header. With this
> > done, we can also use its definition in our files, until
> > there is no more such thing as "our" files. This is too
> > selfish anyway.
> > 
> > Later on, we'll move our internal state inside it.
> 
> Well, in upstream no one outside kvm-all.c needs to (and likely should
> be allowed to) access fields from struct KVMState & KVMSlot directly.
> That avoids misuse outside the KVM layer and enforces KVM arch code to
> properly call into the generic layer.
> 
> But I see the problem for qemu-kvm's transition time, so let's try to
> find an intermediate solution until its code layout is aligned (I don't
> see any blockers for this). Suggestion: Replicate the relevant
> structures into a new, temporary header. If upstream may extend its
> original structures, this should from now on have happened *first*
> inside qemu-kvm, so no inconsistency can arise unless downstream messed
> it up already. At some point (hopefully not too far away), no user of
> that header will remain and we will be able to drop it again.
I'm fine with whatever anthony wants.

> 
> Jan
>