Re: PATCH for 64-bit MIPS ELF buglets

[Mark Mitchell <mark@codesourcery.com>]

>>>>> "Ian" == Ian Lance Taylor <ian@zembu.com> writes:

    Ian> However, I admit that the existing code does not work that
    Ian> way.  It does a normal 32 bit relocation, and overwrites the
    Ian> upper 32 bits with 32 bits of sign extension.  It should
    Ian> probably add the sign extension to the upper 32 bits of the
    Ian> 64 bit value that it finds in the object file.  I don't think
    Ian> that making that change would break any existing object
    Ian> files.

I assume by "existing" you mean "existing until recently trashed by
N32 work." :-) Your suggestion about how to implement R_MIPS_64 is
well-taken; I'll do as you suggest.  Pedantically, though, it still
requires different handling: it particular the src_mask and size are
different.  That's really what I was complaining about in terms of
using R_MIPS_64 to mean two different things.  For example, the
N32/64-bit ABIs can do *exactly* the same thing with the relocations
that they share, assuming all values are sign-extended to the proper
range.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com