Re: [patch/rfa] hppa-linux target, 2nd try

[Andrew Cagney <cagney@gnu.org>]

>>> I actually started looking at that about 10 days ago, but it is hard for
>>> me to figure out what is going wrong, because I don't know too well the
>>> machinery for calling functions inside shared libraries. And then I got
>>> pre-empted by something else.
> 
> 
> ah, ok :)
> 
> the problem seems to be that gdb wants to jump directly to the malloc()
> (or whatever) shared lib function, whereas normally to do a shlib call
> you have to go through the import stub/plt (so that the r19/dp are
> set correctly).... not sure how to do that in gdb. 

Two^D^D^Dthree^D^D^D^Dfour FYIs,

On some GNU/Linux systems, GDB is know to call the wrong "malloc", 
there's apparently one in crt0.  The bug is in the shlib symbol lookup 
code not mimicing the shlib symbol lookup semantics.

There's NAME_OF_MALLOC in the architecture vector, it might help.

There was talk of parameterizing the inferior function call code with 
the full function symbol (and not just it's parameters) so that 
alternative abi's could be handled.  That doesn't seem to have got very 
far :-(

The 64-bit SVR4 PPC ABI code sets the DP register, see the tail end of 
ppc64_sysv_abi_push_dummy_call.

Andrew


>>> It's something that's on my reasonably-near-future list.
> 
> 
> cool, thanks.
> 
> randolph
> -- Randolph Chung Debian GNU/Linux Developer, hppa/ia64 ports http://www.tausq.org/