libffi: fix the build for MIPS soft-float with recent compilers

libffi: fix the build for MIPS soft-float with recent compilers

Recent MIPS compilers have an stricter agreement between the compiler
and the assembler on the ABI. The compiler now passes -msoft-float to
the assembler where it previously did not.

libffi uses MIPS hard-float instructions so when we try to build it for
soft-float using a recent compiler it fails because of the explained
above. This package should be fixed in order to add support for
soft-float.

In the meantime we can add a little hack in order to be able to build it
for soft-float. Basically we add ".set hardfloat" so the assembler will
not disable the hard-float instructions. The build will of course
include some hard-float bits which shouldn't exist in a soft-float
build, and that may cause runtime problems. But that's what we already
had before and we had lived with it.

We choose to apply this temporary fix because it only affects MIPS
soft-float builds. Otherwise we would need to disable libffi and
recursively all the packages depending on it (python, libglib2...) for
MIPS soft-float, which would be a bit overkill.

Fixes:

  http://autobuild.buildroot.net/results/6b0607b7cb0ac5525c2e47e819301f38bd2d6b30/

Signed-off-by: Vicente Olivert Riera <Vincent.Riera@imgtec.com>
Tested-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>