gprof2dot.py fails with KeyError exception about jrfonseca HOT 3 CLOSED

Mathieu, thanks for the report.

I've commited a fix to this issue, and gprof2dot now works.

But I found another issue with the profile you sent -- the resulting dot file 
is huge
and dot takes forever to produce a graph. I'll need some more time to analyze 
this,
but my guess is that due to IRQs usage or kernel preemption the call graph 
turns into
a dense spaghetti ball. I'll either need to provide a mechanism to mark IRQs, 
or a
heuristic to prune nodes and edges inside call cycles.

Mathieu, I've finally managed to come up with an heuristic to propagate time 
inside
recursive function cycles. The generated dot file is much smaller, and dot can 
now
produce result in a finite time. Resulting output attached.

It is now possible to make some sense out of the kernel-side call graph -- there
appears to be a lot of time spend waiting for buffers --, but it still looks a 
bit
like spaghetti. I think this is due to oprofile limitations (call stack is not 
100%
exact), and I can't see anything else that gprof2dot can do to improve the 
situation.

Jose, thank you very much for your time.