profiling conundrum

[Robert Collins]

Ok, the profiling patch I sent is somewhat wrong - the gcrt0.o should
not be linked with cygwin1.dll. However: this call
monstartup ((u_long) &eprol, (u_long) &etext);
is wrong. 

The result being that the gmon data fields
p->lowpc
p->highpc
p->textsize 

are wrong, and the in-cygwin calls to mcount() return without logging
the call arc because of the test for frompc >  p->textsize.

So, if anyone has some idea, on how the gcrt0.o code can be tweaked such
that the monstartup call is given the real low and high limits for the
object, that'd be great.

Rob