[lwip-devel] ARP queueing

[Russell Hind]

I’ve been investigating an issue with ARP queueing.  We’re not using LWIPs memp 
or malloc, just using the standard malloc that comes with our compiler.

We allocate all our pbufs upfront and don’t want them copied.  We only send UDP 
packets.  We’ve disabled ARP_QUEUEING as we’re happy to drop the first packet 
that requires an ARP request as we’ll retry our packet a little later.

In 1.3.2, we’ve disabled ARP_QUEUEING and that has the desired behaviour.  The 
ARP request is dynamically allocated but our packet is dropped rather than 
copied.

I see that in the latest version, even with ARP_QUEUEING disabled, the outgoing 
packet will be copied and stored on the ARP entry (etharp.c line 1186).  Also, 
#34681 limits ARP queue lengths to a defined maximum.

It would seem more logical that if you want one outgoing packet queued then 
defining ARP_QUEUEING and setting ARP_QUEUE_LENGTH=1 would be the better way to 
do it.  That way if ARP_QUEUEING is disabled, then no outgoing packets will be 
copied.

This would allow for both behaviours, but the current change prevents us from 
stopping our packets being copied.

Is there a reason this couldn’t be changed before the next release?  We’d like 
to move to it but I think this behaviour is wrong and will adversely affect our 
system so we’ll either have to have a custom patch for LWIP in our source or 
stick with 1.3.2.

Thanks

Russell