Hello!
I have experimented a bit further with the emulation of the PCI layer.
The attached patches allow to hide a PCI device from domain 0 by passing a
''hide='' argument to the grub module line of the domain 0
kernel (i.e.,
module /vmlinuz-2.6.12-xen0 root=/dev/hda2 ro hide=02:09.0). For example
to hide the second ethernet card in my system, I pass
''hide=02:09.0'',
where ''2'' indicates the bus number, ''9'' the
device number and ''0'' the
function. ''lspci'' or ''cat /proc/pci'' can
tell which devices are available
in a system. The hidden device becomes automatically visible to the first
user domain that is started or to any subsequent domain that is started
after the previously owning domain is killed (this is hardcoded for now).
That PCI device''s config space is not emulated, but its config space
''shines through'', meaning that all inb/w/l and outb/w/l from
the kernel
are first intercepted by Xen and then issued again in the PCI emulation
code. The Host and ISA bridge PCI entries are emulated by the QEMU code,
though (see below).
If you want to try the patches, apply them *after* one full XEN compile -
the reason being that the Linux patch applies against the
linux-2.6.12-xen0 directory. The .config file is my domU configuration
file.
A ''cat /proc/pci'' in the user domain shows the 2nd PCI card of
the system:
PCI devices found:
Bus 0, device 0, function 0:
Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 2).
Bus 0, device 1, function 0:
ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II]
(rev 0).
Bus 0, device 9, function 0:
Ethernet controller: Intel Corporation 82557/8/9 [Ethernet Pro 100]
(rev 8).
IRQ 9.
Master Capable. Latency=66. Min Gnt=8.Max Lat=56.
Non-prefetchable 32 bit memory at 0xc0101000 [0xc0101fff].
I/O at 0x2040 [0x207f].
Non-prefetchable 32 bit memory at 0xc0200000 [0xc02fffff].
Although the device appears on bus ''0'', it is internally
correctly handled
as being on ''bus 2''.
One problem is certainly that the IRQ number is not correct for the device
to function - it should be ''21'' on that machine.
The positive side I think is that so far the changes to Linux are minimal.
The hiding of PCI devices could be done in Xen if also domain 0 received
an emulated PCI layer, which would completely eliminate changes to Linux -
at least for making PCI devices visible.
Signed-off-by: Stefan Berger <stefanb@us.ibm.com>
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel