Re: [Xen-users] Re: About NIC passthrough to the guest system

[Paul Schulze <avlex@xxxxxxx>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Christian and Alex,

On 27 Apr 2009, at 12:36, Christian Tramnitz wrote:

> When running hvm you need vt-d support to pass-through devices.
> Apart from that I doubt that you will be able to passthrough one  
> port of a dual-port nic, you may have to pass-through the whole card.
If the DomU is starting without displaying any error, I think it is  
safe to assume, that it is running on hardware with VT-d support,  
since in my experience, a HVM DomU does not start at all, if the guest  
config sets up a device for PCI passthrough (I could be wrong here,  
haven't tried it lately).
That depends on the NICs design. I own a D-Link DFE-580TX (4 ports)  
and on that one, each port is more or less stand-alone, not sharing  
anything with the other chips on the board. My current setup (2 ports  
in Dom0, 2 ports passed through to my IPCop DomU) works just fine with  
that. That Intel NIC looks like a similar setup though, so it should  
work somehow. However, I am currently running my DomU in PV mode, but  
with only a few modifications, Alex should be able to do that too, if  
necessary. It is also important to take a good look at the IRQs (lspci  
- -v) and keep the sharing of IRQs between all Xen instances to a  
minimum. Oh yeah and never pass a device to a DomU that shares an IRQ  
with your HDD controller, be it SCSI, SATA or IDE... that can really  
lead to "unexpected" results.
> Alex Chan wrote:
> [...]
> >  [root@hm02 xen]# cat /etc/modprobe.conf
> > options pciback hide=(0a:00.1)
> > alias eth0 bnx2
> > alias eth1 bnx2
> > alias eth2 bnx2
> > alias eth3 e1000e
> > alias eth4 e1000e
> > alias eth5 e1000e
> > alias eth6 e1000e
> > alias scsi_hostadapter megaraid_sas
> > alias scsi_hostadapter1 ata_piix
> > alias scsi_hostadapter2 usb-storage
> [...]

And finally, I looked over your config, Alex, and it looks ok, but I  
am wondering, why you define pciback.hide in /etc/modprobe.conf  
instead of the kernel command-line. Since I assume, the device driver  
for your NIC is built as a module, and /etc/modprobe.conf should be  
accounted on loading it, this might be correct. But pciback is built  
into your kernel anyway so to avoid any doubt on whether it is or not,  
I would suggest putting it on the command-line (simply add  
pciback.hide=(0a:00.1)). This might not help at all, but at least then  
you can be sure that nothing else happens to the PCI device besides  
being seized by pciback.
I hope this helps


Paul.

- --
Paul Schulze
Mail: avlex82@xxxxxxxxx

Why can't a programmer tell the difference between Halloween and  
Christmas?
Because OCT31 = DEC25.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.10 (Darwin)

iEYEARECAAYFAkn1m9YACgkQj2zIQLJNnKManQCfVhQlcLQFUsyxhwbwRqBXrota
S54An2sMkqxVCUdAW4CSv5WQqNE/Ir79
=fozf
-----END PGP SIGNATURE-----

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