to be soft-float. the floating-point usage in ld.elf_so is only
because of stdarg functions like printf(), and gcc 4.5 is good
about not actually executing the FP instructions in such functions
unless FP values are passed to them, which ld.elf_so doesn't do.