On Tue, Feb 13, 2018 at 02:17:40PM -0800, Kilian Cavalotti wrote:
> I'm trying to deploy PE servers (Rx30s and C6320s) via PXE in UEFI
> boot mode, without success.
> The servers PXE-boot just fine when configured in BIOS boot mode, but
> switching to UEFI apparently makes the initial DHCP request timeout,
> and generate a PXE-E18 Server Response Timeout error.
> Looking at the network traffic, I see just 2 DHCP frames in the
> initial exchange:
> 1. DHCP Discovery from the server
> 2. DHCP Offer from the DHCP/TFTP server
> And then nothing else, as if the PXE-booting server never received the
> DHCP Offer, or at least didn't act on it.
> Again, the exact same setup works flawlessly when the PXE-booting node
> is in BIOS boot mode.
> Both the Ethernet switches and DHCP servers are configured to use
> Jumbo Frames (MTU 9000). I know that other vendors have released fixes
> in their UEFI network stack in the past to fix issues related to that,
> but I don't have any evidence that jumbo frames are involved here.
> Anyway, has anybody ever observed the same behavior?
I have been using UEFI and PXE on older generation of dell (from Rx10).
This worked fine to boot iPXE. For some reason my firmware was not able
to handle large payload (like >10MB initrd). I used iPXE to lift that
I have been beaten by jumbo frames. PXE is using tftp, a udp based protocol.
The booting node was using 1500 MTU, while the tftp server was using 9000.
Because the protocol is udp based, there is no mss to negociate the mtu,
and the tftp server was sending too large segments for the booting node
which it didn't received. You may confirm that using tcpdump on the tftp
The workaround was setting the tftp server with 1500bytes mtu.
Linux-PowerEdge mailing list