On Wed, Feb 2, 2011 at 5:37 PM, Eduardo Habkost <ehabk...@redhat.com> wrote:
> On Wed, Feb 02, 2011 at 05:16:23PM +0000, Blue Swirl wrote:
>> On Wed, Feb 2, 2011 at 7:55 AM, Paolo Bonzini <pbonz...@redhat.com> wrote:
>> > On 02/01/2011 07:10 PM, Blue Swirl wrote:
>> >>
>> >> One way to solve this which would preserve the device model would be
>> >> to add stub devices. For example, hw/vmmouse-stub.c would be:
>> >> void *vmmouse_init(void *m)
>> >> {
>> >> return NULL;
>> >> }
>> >
>> > This is the wrong direction, unless you can somehow automatically generate
>> > the stub file.
>> >
>> > The only other solution I can think of is weak symbols. There are subtle
>> > differences between Windows and Linux weak symbols, but luckily we do not
>> > care about it. See http://cygwin.com/ml/cygwin/2010-04/msg00281.html
>>
>> Boards should create optional devices so that failure in the device
>> creation is handled more intelligently. For example, pci_vmsvga_init()
>> is just a wrapper (a bit useless one also) to pci_create_simple(bus,
>> -1, "vmware-svga"). pci_create_simple() in turn uses
>> qdev_init_nofail() which aborts if the device can't be created.
>> Instead, non-aborting version should be used so that if 'vmware-svga'
>> device can't be initialized, a fallback device (VGA) could be used or
>> error message printed.
>>
>> In this case, vmmouse and vmport are not converted to qdev but that
>> should be done anyway.
>
> In the meantime, what should we do? Keep the CONFIG_VMWARE_VGA option
> broken and useless just to avoid an #ifdef?
Please try the patch set I just sent.
