On Mon, Apr 11, 2011 at 9:10 PM, Alex Williamson <alex.william...@redhat.com> wrote: > On Mon, 2011-04-11 at 13:57 -0600, Alex Williamson wrote: >> On Mon, 2011-04-11 at 14:48 -0500, Anthony Liguori wrote: >> > On 04/11/2011 02:35 PM, Alex Williamson wrote: >> > > This series replaces our current gPXE based PXE ROMs with iPXE >> > > versions from the iPXE project (http://ipxe.org). This version >> > > adds ipxe to our submodules so it can be easily included in >> > > releases. I'm still including a script for updating these, >> > > perhaps someone better with Makefiles can eventually adopt this >> > > to a build target. >> > > >> > > This email series is mainly for reference, there's too much >> > > renaming and replacing binary files to send out to the mailing >> > > list. I'll strip out the binaries here so the rest can be >> > > reviewed. For the real code, please pull: >> > > >> > > git://github.com/awilliam/qemu.git (ipxe branch) >> > > >> > > Thanks to Anthony for already setting up an ipxe mirror. >> > > Thanks, >> > >> > Looks good to me. How different is this from what we've been shipping? >> > Have you tested PXE boot from the builtin TFTP server and from an >> > external one (like dnsmasq)? >> >> We were shipping v0.9.9, which was tagged 10/2009. There's been a gpxe >> v1.0.0 release since then, plus the split between ipxe and gpxe. I >> think Michael is hoping to have a release soon, but the code feels >> pretty stable to me as is. >> >> I've tested external booting from dhcp/tftp server for all the NICs. >> I'll make a pass through testing with the builtin server and report >> back. Thanks, > > Check, tested virtio-net-pci, e1000, rtl8139, ne2k_pci, pcnet, and > i82550 using builtin server, with bootfile and tftp loaded > kernel/initrd. All work. Thanks,
Whenever QEMU wants to merge a new iPXE build it would be nice to run the tests that Alex performed, plus: * PXELINUX (to test the PXE interface) * undionly.kpxe (to test the UNDI interface) It's fairly easy to do using the builtin DHCP/TFTP servers, but not that hard with dnsmasq either. The guest could be a Linux kernel and initramfs that boots and writes "PASS" to the serial port (or over a TCP connection). We can launch this with a timeout to catch failures or hangs. Or maybe iPXE has automated tests that we can run? Stefan