RE: [Xen-devel] [PATCH][VTD] enable integrated graphics passthrough for Calpella and Sandybridge

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Tue, 8 Jun 2010, Kay, Allen M wrote:
> Isaku,
> 
> Thanks for the feedback.
> 
> > pci_{read, write}_block() would be better than
> > switch(len) case 1: case 2: case4:...
> 
> Done!
> 
> > Is it really necessary to move PCIBus and PCIBridge to the header file?
> > Doesn't pci_bridge_init() work?
> 
> I changed to code to utilize pci_bridge_init().  However, I still need to move
> PCIBus and PCIBridge defines to the header file.  The alternative is to 
> pollute
> pc.c with graphics passthrough specific code.
> 
> > Overriding pci config read/write methods of i440fx would be much cleaner
> > than hooking pci_data_read/write. (pass igd_pci_read/write to 
> > pci_register_device() in i440fx_init() in hw/piix_pci.c instead of NULL)
> 
> Doing this resulted in a lot of duplicated code and also force code path to 
> change
> even when IGD passthrough is not used.  To do it correctly, I also need to
> put in IGD or PASSTHROGH awareness in piix_pci.c.  For now, I'm going to keep
> the original patch.
> 

I see.

Even though the patch is an improvement over what we currently have in
qemu-xen, it is still not acceptable in upstream qemu as it is.
Considering that we are pretty close to have the merge complete, I think
it worth trying to come up with something cleaner.

Isaku, you are the latest one to touch the pci_host code in qemu, do you
have any other suggestion?

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel