Re: PATCH: MIPS support for libjava (version 2).

[David Daney <ddaney at avtrex dot com>]

Anthony Green wrote:

> On Wed, 2003-09-10 at 14:46, David Daney wrote:
>  
>
> > +  mips*el*-linux*)
> > +       can_unwind_signal=yes
> > +       if test "x${with_libgcj_mips_xgot}" = "xyes"; then
> > +               libgcj_flags="${libgcj_flags} -Wa,-xgot"
> > +       fi
> > +       ;;
> >    
>
> Maybe you should use 
> mipsel*-linux* | mipsisa32el*-linux*)
Probably a good Idea.  Should I revise the patch to reflect this change?

> You can build a mips64el-linux-gnu toolchain that supports each of the
> o32, n32 and n64 ABIs simulateously, and I believe all of your changes
> are only for the o32 ABI.  Is this true? 
As posted I suspect that it will only work for o32 ABI systems.

There are really two parts to the patch.  The signal unwinding things, 
and the backtrace code.

I think the unwinding should work under all ABIs as it is agnostic to 
register size and usage.

The main problem I can think of for n32 and n64 is the backtrace code. 
There are assumptions that pointers and ints are the same width in the 
code.  Also there are assumptions about function prologs that may not 
hold under n32 and n64.  On the plus side I don't think it cares about 
register usage.

Didn't someone once suggest that the exception unwinder code should be 
hijacked and used to generate stack traces?

David.