From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from eggs.gnu.org ([2001:4830:134:3::10]:49169)
	by lists.gnu.org with esmtp (Exim 4.71)
	(envelope-from <akong@redhat.com>) id 1VyWft-0005OE-I1
	for qemu-devel@nongnu.org; Wed, 01 Jan 2014 20:02:19 -0500
Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71)
	(envelope-from <akong@redhat.com>) id 1VyWfn-0008Hz-If
	for qemu-devel@nongnu.org; Wed, 01 Jan 2014 20:02:13 -0500
Received: from mx1.redhat.com ([209.132.183.28]:49819)
	by eggs.gnu.org with esmtp (Exim 4.71)
	(envelope-from <akong@redhat.com>) id 1VyWfn-0008Hr-AI
	for qemu-devel@nongnu.org; Wed, 01 Jan 2014 20:02:07 -0500
Date: Thu, 2 Jan 2014 09:02:01 +0800
From: Amos Kong <akong@redhat.com>
Message-ID: <20140102010201.GB2187@amosk.info>
References: <1388477217-23491-1-git-send-email-akong@redhat.com>
	<52C2EB94.4050906@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <52C2EB94.4050906@redhat.com>
Subject: Re: [Qemu-devel] [PATCH] qdev: unparent device when fails to set
	properties
List-Id: <qemu-devel.nongnu.org>
List-Unsubscribe: <https://lists.nongnu.org/mailman/options/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=unsubscribe>
List-Archive: <http://lists.nongnu.org/archive/html/qemu-devel>
List-Post: <mailto:qemu-devel@nongnu.org>
List-Help: <mailto:qemu-devel-request@nongnu.org?subject=help>
List-Subscribe: <https://lists.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=subscribe>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org, aliguori@amazon.com, afaerber@suse.de

On Tue, Dec 31, 2013 at 05:06:44PM +0100, Paolo Bonzini wrote:
> Il 31/12/2013 09:06, Amos Kong ha scritto:
> > When it fails to set properties, qdev's parent is already set
> 
> Do not confuse the QOM parent (which is /machine/peripheral, of which
> the new device is a child) with the qdev parent bus (which has a link to
> the new device)!
> 
> In general, you should add the device to the QOM tree before using it to
> set a link.  So I believe that object_property_add_child should be
> called before qdev_set_parent_bus.  This is the root cause of the bug;
> the fix then could be one of the following:
> 
> 1) move qdev_set_parent_bus later;
> 
> 2) move object_property_add_child before the setting of properties.
 
I agree to fix this problem by adjust the initialization order.
Thanks for the detail explain.
 
> I slightly prefer the first, so that initialization happens in this order:
> 
> 1) create object
> 
> 2) set properties -- if it fails, you can just unref the object
> 
> 3) add child -- if it fails due to duplicate ID, you can again just
> unref the object
> 
> 4) set parent bus (cannot fail)
> 
> 5) realize -- if it fails, you need to unparent the object and unref the
> object
> 
> 6) drop the reference now that the object is kept solidly alive by
> QOM/qdev, and return.
> 
> This matches a bit more closely what happens with object-add, too.
> 
> Paolo

-- 
			Amos.