Re: [Qemu-devel] [PATCH V3] virtio: do not require IOMMU to be created in advance

[Peter Xu]

On Wed, Mar 08, 2017 at 05:09:45PM +0800, Peter Xu wrote:
> On Wed, Mar 08, 2017 at 10:17:09AM +0200, Marcel Apfelbaum wrote:
> 
> [...]
> 
> > 
> > I am sorry I was not clear enough:
> > If a device is added after the system is up (hotplug), we cannot depend on 
> > the "machine_done"
> > event to enable "bus master".
> > This is why we have
> >    if (qdev_hotplug)
> >         pci_init_bus_master(pci_dev);
> > 
> > The code you proposed changes the order, so this call is done *after* 
> > realize.
> > 
> > My question was: What if any other device may require the bus_master_as
> > at realize time (and can be hot-plugged) ?
> > For example: hcd-ehci/hcd-ohci devices call pci_get_address_space()
> > and caches the bus_master_as.
> 
> Oh, I didn't notice that there are other devices that used
> bus_master_as during realization. If so... Would this really work even
> without hot plug? Considering that bus_master_as won't be inited until
> machine done phase?

Please ignore my question... I think the answer is that these devices
are only caching the pointer of bus_master_as. So it won't really use
the address space before machine_done.

If so, IMHO moving pci_init_bus_master() after device specific
realize() is okay as well then, right?

Thanks,

-- peterx