The debugging output is great! Thanks. I'll try that on the SAMA5D36 too. With CONFIG_SIM_WALLTIME=y the loopback tcpblaster seems to chunk along pretty quickly: Sent 50 buffers: 200.0 Kb (avg 4.0 Kb) in 7823.46 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.6 Kb (avg 1.3 Kb) in 2575.50 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.9 Kb (avg 1.3 Kb) in 2575.50 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.6 Kb (avg 1.3 Kb) in 2575.50 Sec ( 0.0 Kb/Sec) Sent 50 buffers: 200.0 Kb (avg 4.0 Kb) in 7665.90 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.6 Kb (avg 1.3 Kb) in 2499.75 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.9 Kb (avg 1.3 Kb) in 2424.00 Sec ( 0.0 Kb/Sec) Received 50 buffers: 66.6 Kb (avg 1.3 Kb) in 2424.00 Sec ( 0.0 Kb/Sec) So it appears to be working fine in the simulator. Now if I could get the simulator to network with the linux host, I could test out non-loopback mode. I am still not sure what's wrong there.
The timing still looks really bad. Without CONFIG_NET_WALLTIME, the simulation just runs as fast as it can, but it should still mostly preserve fidelity in "relative" time. So, although the output is generated fast, it is still saying it took about 2500 seconds to send 1.3 Kb at a rate of zero Kb/sec. Still looks bad to me.
So there still looks like a timing problem. I think that test excludes about all other sources of problems except: (1) your configuration, and (2) your Ethernet driver. Timing could still be the culprit.
Timestamping the debug output will still be a good idea.
