PXE boot fail with Intel E1000 => Fixed in SVN

[trevort]

I've been working with Windows Deployment Services, and IF I set a host to use an Intel Pro adaptor it will not load the PXE boot program. It will get an IP, and start to load the file sent to it from the server, but when it tries to load it will crash. If I use the PCNet adapter, it will load, but win2k8 doesn't have a driver for that in the PXE boot environment.

Attached is a screenshot of the failed boot screen, a txt file of my guest environment, and the log file.

[trevort]

vm info

[trevort]

log file

[trevort]

Also, see ticket #2536, which seems related.

[trevort]

screenshot

[trevort]

Just a couple more bits of info:

I'm running Ubuntu 9.1 with the latest updates (kernal 2.6.31-20), 64bit. I don't have a virtualization feature on my processor though, so that feature is 'disabled' (it's not an option). I did remove and re-add the boot image and that didn't help. When I use any of the intelpro nic adapters the VM will either freeze at 100% CPU utilization or it will error out (as shown in the screen shot). If I use the PCNet adapter I don't have any problems (but I the boot fails once windows PE loads since windows PE doesn't have a driver for it).

[trevort]

I tested with a windows host (Vista 32bit) using the same VMs. The problem persists - windows and linux hosts both have the same problem.

Is there any additional files/logs/info I should upload / produce?

[runout]

There is a temporary solution. Add two NIC: the first one is PCnet card, the second - Intel card. PXE will use the first;than, when WinPE boots, i will configure Intel card with DHCP-client and will use it during installation. It works form me and i suppose for you too. Good luck!

[aleksey]

Recent versions of WDS (based on Windows Server 2008 and above) use TFTP window of eight packets to send data back to PXE client. When two data packets arrive to our e1000 UNDI driver in short time frame they are erroneously combined to form a single packet. The problem is now resolved and the fix will be available in the next maintenance release.

[frank]

Fixed in VBox 4.0.2.

[jars99]

I'm still having this problem in Ubuntu 10.10 with VirtualBox 4.0.2 r69518. If I change the adapter to anything besides intel everything seems to work fine.

[aleksey]

Could you enable the internal packet capture with

VBoxManage modifyvm <vm_name> --nictrace1 on --nictracefile1 <name_of_choice>

reproduce the problem and send the capture file to aleksey dot ilyushin at oracle dot com? Actually I am only interested in the part of the file where the problem manifests itself. Please attach vbox.log as well.

[paperdiesel]

I'm also having this problem with 4.0.4 r70112. Selecting any of the 3 Intel adapter types causes "FATAL: Could not read from the boot medium! System halted." PXE boot works fine, however, when using either of the 2 PCnet adapters. I have emailed aleksey dot ilyushin at oracle dot com with the capture and VBox.log files.

[frank]

paperdiesel, do you have the extension pack installed? This is necessary to use the Intel PXE code which supports the Intel adapters. In case you already installed the extension pack then please attach a VBox log of such a VM session where booting from PXE does not work.

[jars99]

aleksey - I sent those files to you - did you get them? Any idea why the packet capture wasn't getting any real data?

Any other updates on this? I'm anxious to move all my systems to 4.0, but I want to wait until this is fixed.

[aleksey]

jars99,

I got the files, thanks! Judging by the empty packet capture file (it is binary btw) and missing LanBootRom in [/Devices/pcbios/0/Config/] it is very likely that you do not have the extension pack installed. Can you check the output of the following command?

VBoxManage list extpacks

Oracle VM VirtualBox Extension Pack contains PXE ROM with E1000 support.