I'd be happy to provide any useful debugs to get behind this. The matching sequence numbers of the broken data is what drives me mad and kills all speculations about lost data on the PPP line (also the PPP trace proves that just the 2x1024 byte packets are sent on it, not the smaller second fragments of the packets)
I'd love to know that I did something stupid, but after a debugging day by a colleague and me we don't see that ...
I think this situation can only be seen on slow netif's. On Ethernet the packet is up and away even before that can occur.
Again, it can't happen, lwIP is mostly sequential, the lwIP stack is (or
should be!) waiting while you are sending data over the uart port.
Sylvain