We need some simple way to determine available CPU resources when doing something like serial tests. Presently, all I can say is that at <baudrate X> it starts falling apart. Would be nice if I could say that, at <baudrate X> CPU utilization is >100% explaining the boards inability to run at this baudrate. At <baudrate X-1> the CPU utilization is 70% leaving little room to application processing... You get the idea. Would be nice to have in other situations too - such as recently when I worked on playmepeg stuff and thought the board was really slow. Turned out the demo was holding back in attempt to match sync. A CPU load readout would have saved me some time in that situation.
Maybe the techniques used for the network tests (look at nc6_test_slave) would be useful here. They establish a background load level using some number of computationally intensive threads, along with being able to measure the actual idle-ness of the processor.
More relevant is the stuff that Andrew Lunn wrote for CPU usage as per http://sources.redhat.com/ml/ecos-discuss/2000-02/msg00047.html
This bug has moved to http://bugs.ecos.sourceware.org/show_bug.cgi?id=63986