[ipxe-devel] ipxe hangs on boot when receiving a jumbo frame

ahmed ahmed at gandi.net
Tue Apr 21 15:36:14 UTC 2015


Hi Michael,
Thank you for you response.

ifstat indicate that we are using undionly
"net0: XX:XX:XX:XX:XX:XX using undionly on UNDI-PCI82:00.1 (closed)

  [Link:up, TX:0 TXE:0 RX:0 RXE:0]"

seems that this one does not support jumbo

Ahmed

Le 21/04/2015 16:57, Michael Brown a écrit :
> On 20/04/15 09:31, ahmed wrote:
>> We have spent some days trying to debug this problem with no success.
>> we hope to find some help here.
>> As indicated on the mail subject, we have three servers A, B, and C on
>> the same broadcast domain.
>> A and B communicate via a multi-cast protocol with frame of size 9000
>> (real packet size less than 256 but they are padded with 0 to MTU size).
>>
>> When booting C receive those packets  as a result it timeout on boot, we
>> have on debug message with unknown protocol 0000.
>> We suppose that only part of the payload is read and then either an
>> overflow screw up everything or the remaining part of the payload is
>> considered as a new packet and mess up things.
>> We have also tried decreasing MTU from 9000 (in A,B) to 4000 and 2000
>> but still the same error.
>> C boots only when we bring back A,B MTU to 1500
> 
> Which driver are you using (according to "ifstat")?
> 
> Michael
> 



More information about the ipxe-devel mailing list