Re: [Valgrind-developers] (Skin writing) Linker interference with detection of function calls?

[Julian S. <js...@ac...>]

> I've been trying to get a valgrind skin i'm tinkering with to get a
> program to report whenever it calls or returns from a function. 

Simple as it sounds, this is a real swamp.  For one thing, tail-calls from
one function to another just look like plain Jmps, not JmpCall, and so
you can't reliably use the mechanism you proposed.  Similarly, there are
99 ways to return from a function (98 of them admittedly really stupid),
such as   popl %edx; jmp *%edx

Josef Wiedendorfer is the real guru here as I think kcachegrind/calltree
manages to track calls/returns.  I'm sure he can elaborate.


>  Am I misunderstanding something about UCode or what? I'm sure this is a
> solved problem,
> since I imagine every single skin would depend on the behavior that I am
> hoping for (one
> call of the instrumentation code for every call/ret).

Well, actually, no.  As far as I know only calltree does this.  For the 
most part skins like memcheck just see the client program moving values
on/off the stack and jumping round the place, but have no concept of
calls/returns happening.  The stack backtraces you see in error messages
are made by walking the stack when an error occurs, but nobody keeps
track of when functions are entered/exited, basically because it's nearly
impossible to do so reliably.

Having said all that ... what is it you are trying to achieve with
your skin?

J