Re: [Xen-devel] [edk2] Question about PEX boot on Xen with OVMF as bios

[Wei Liu <wei.liu2@xxxxxxxxxx>]

Hi

Please CC me and Anthony (CC'ed) in the future if you have questions
regarding OVMF on Xen.

On Tue, May 26, 2015 at 12:02:22PM +0200, Laszlo Ersek wrote:
> On 05/25/15 03:50, lidonglin wrote:
> > Hi all:
> 
> > Recentlly, I want to use PXE boot on Xen with OVMF as bios. At
> > beginning, I just add rtl8139 as guest nic device, and I compile a
> > release ovmf. When I enter into uefi, I can't find network boot menu.
> > According to edk2/OvmfPkg/README file, I know there is a virtio-net
> > driver build into ovmf. So I replace rtl8139 with virtio-net. When I
> > enter into uefi boot menu again. I see the network boot entry, I'm
> > very happy. But after I choose network boot entry, uefi can't display
> > pxe boot process, only black sceen with one while dot at the
> > beginning of the screen. I try e1000 according to edk2/OvmfPkg/README
> > file, and I get the same result. Except e1000 and virtio-net, I can't
> > see the network boot entry if I use any other nics. Who can tell me
> > how to use pxe boot on Xen with ovmf. Thanks.
> 
> (1) The dots that you mention are progress info from the higher level
> edk2 network / PXE stack. I can't recall the details without looking at
> the code, but if you see at least one dot, that's a sign that PXE boot
> is being attempted.
> 
> (2) The only SNP (Simple Network Protocol) driver that OVMF contains,
> when built purely from the upstream repo, is for the virtio-net device
> of QEMU.
> 
> I doubt that you can use virtio devices on a Xen host.
> 

That's correct. Xen HVM doesn't enable virtio device.

> (3) Spelling out a special case of (2) -- there's no xen-netfront driver
> in OVMF.
> 

This is also correct. IIRC there is only Xen PV disk driver in OVMF.
I think the infrastructure needed to port netfront driver is already
there. Just that the driver is not yet there.

Anthony, am I right?

> (4) One possibility I can see for what you want to do is:
> - Configure the e1000 device model for your guest
> - Build OVMF as described in OvmfPkg/README, near "E1000_ENABLE".
>   This requires downloading a proprietary, non-redistributable driver
>   binary from Intel.
> 

You also need to specify model=e1000 in your guest configuration file.
The default model is 8139.

Wei.

> (5) Another possibility is to use the iPXE drivers (PCI oproms) bundled
> with upstream qemu. You'd have to build a recent qemu checkout, and even
> so, I'm unsure how PCI oproms are supposed to work on Xen.
> 
> Since the iPXE bundle in QEMU provides a UEFI driver for rtl8139, and
> you tried that, and it failed, I'm thinking that you either used a too
> old qemu, or that this approach is not viable on Xen.
> 
> (6) Assuming you get the guest set up alright (correct device model
> selected, appropriate driver available etc), it's still 99% certain that
> your initial host network config will not be suitable for PXE booting a
> guest with the edk2 stack.
> 
> In my experience the biggest hurdle in PXE-booting with edk2 has
> consistently been: firewall rules, bridge setup, DHCP setup, PXE/TFTP
> setup, etc. Tcpdump and the relevant networking specs usually help.
> 
> Thanks
> Laszlo
> 
> ------------------------------------------------------------------------------
> One dashboard for servers and applications across Physical-Virtual-Cloud 
> Widest out-of-the-box monitoring support with 50+ applications
> Performance metrics, stats and reports that give you Actionable Insights
> Deep dive visibility with transaction tracing using APM Insight.
> http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/edk2-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel