[virt-tools-list] virt-install --pxe REVISIT

[Phil Meyer]

Pleas guide me as to where and/or who would work this bug:

Fedora 12+ and RHEL6 both suffer this issue.

I would speculate that all systems that ship a reasonably new version of 
qemu featuring the seabios suffer from this issue.

Here is a sample virt-install that works on servers with older (not 
seabios) versions of qemu.

virt-install \
--noreboot \
--hvm \
--virt-type=kvm  \
--name=a008474 \
--ram=1024 \
--vcpus=1 \
--arch=x86_64 \
--uuid=ff1507df-2f77-4618-b9d6-8c9b27e6528f \
--os-type=linux \
--os-variant=rhel6 \
--disk path=/dev/mapper/008474,bus=virtio,cache=none 
--network=bridge=br1,model=virtio,mac=52:54:ac:14:21:04 \
--pxe \
--vnc \
--vnclisten=0.0.0.0 \
--noautoconsole \
--wait=-1

We are really only interested in the affect of calling: --pxe

If the libvirt provided NAT interface is used, it works.

If a bridge is set up and (as in the above example) is used for the VM, 
the dhcp request shows up on the vnet assigned to the guest, but never 
to the bridge.

ie:

# tshark --i vnet0
      Always shows the dhcp request.

# tshark -i br0
      Never shows any packets from the guest.

However:

If you press <CNTRL>b to drop into the gpxe BIOS before it aborts, 
manual processing does indeed work.

gpxe> dhcp net0
     WORKS!

By entering imgload, etc. manually, the PXE boot can work.

Very strange.

Any advice is welcome.  This is a show stopper for us.

Thanks!