Re: [PATCH RFC] qemu: fix hot remove assigned device

From: Gerd Hoffmann <kraxel@redhat.com>
To: "Han, Weidong" <weidong.han@intel.com>
Cc: "'Avi Kivity'" <avi@redhat.com>,
	"'Paul Brook'" <paul@codesourcery.com>,
	"'kvm@vger.kernel.org'" <kvm@vger.kernel.org>
Subject: Re: [PATCH RFC] qemu: fix hot remove assigned device
Date: Wed, 10 Jun 2009 10:49:27 +0200	[thread overview]
Message-ID: <4A2F7397.1010909@redhat.com> (raw)
In-Reply-To: <715D42877B251141A38726ABF5CABF2C054596AA31@pdsmsx503.ccr.corp.intel.com>

On 06/10/09 10:31, Han, Weidong wrote:
> Avi Kivity wrote:
>> Can you check pci_dev->qdev instead of assigned?  A little less
>> ugly.
>
> I tried to find an easy and clean way to check it, but I found the
> members of struct PCIDevice and DeviceState were not suitable for
> this checking, and qdev is not pointer in struct PCIDevice.

Well, certain DeviceState pointers (type for example) are set only in 
case the device was created using the qdev api.  I think you certainly 
should get away without adding a new parameter to pci_unregister_device.

Also I've just sent a patch to the qemu-devel fixing the qdev register 
API for pci devices, allowing to register config space callbacks.  Once 
this is merged you can switch pci passthrough to the new qdev API.

cheers,
   Gerd