[ltt-dev] UST Instrumenting function entries and exits

[Mathieu Desnoyers]

* Michel Dagenais (michel.dagenais at polymtl.ca) wrote:
> 
> > We could do a compromise and save the string the when there is a new
> > address. There are many caching schemes that can work.
> 
> You normally save the address (or the caller return address if you are
> in a hook function like mcount). Then you use the debugging info (e.g.
> using libdwarf) to map to the function name. Indeed, you also need to
> instrument library loading to know which shared library is loaded at
> which offset.

I fully support the message above. :) The first step would be to dump
just the function addresses into the trace, and later on we can add
instrumentation of library load/unload to be able to map these addresses
to actual function names. Meanwhile, users still have the ability to use
objdump to find the mapping from addresses to functions by hand.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com