[Ksummit-2012-discuss] PCI breakout session
bhelgaas at google.com
Mon Jun 18 17:39:13 UTC 2012
On Mon, Jun 18, 2012 at 11:02 AM, H. Peter Anvin <hpa at zytor.com> wrote:
> On 06/18/2012 03:28 AM, Michael S. Tsirkin wrote:
>>> I'd say don't do it :-) Or rather try to give each device in the VM its
>>> own dedicated virtual host bridge & bus :-)
>> This only makes the problem of resource allocation worse.
>> For example for IO this means you waste 4K of space per device. Since
>> x86 only uses 16 bit IO addresses, this allows up to 16 devices.
> OK, maybe this is a crazy idea, but we could do an extension to the PCI
> standard to allow I/O address space to be subdivided finer than 4K. We
> could even try to push it at the PCI-SIG; it might be useful for some
> PCI-Express configurations, since PCI-Express has exactly the same issue.
You mean like "en1k" (currently badly implemented because it's a quirk
and can't work with any reassignment)?
I'm not that familiar with x86 hardware, but on ia64 it's typical to
have each host bridge in its own domain, with 64K I/O space per host
bridge instead of 64K per system. Maybe something similar could be
done on x86.
More information about the Ksummit-2012-discuss