[Qemu-devel] broken floating point on mips32

[Vince Weaver]

Hello

so I spent an entire day trying to figure out why floating point programs
run with mips-linux-user were losing the top halves of their values, thus
giving wrong results.

In the end it turns out to be because even when running 32-bit binaries on
the 32-bit Mips emulation, the "FR" (ie, 64-bit FPU support) bit is being
set in the CP0 status register by default.  This means programs compiled
for 32-bit MIPS are breaking because they assume the old-style hybrid
64/32 floating point register layout.

>From reading manuals, it seems like the FR bit should not be set by
default for 32-bit MIPS processors.  Is this a bug in the emulation that
should be fixed?

For the time being I have a patch that adds a new machine type with the FR
bit turned off and things work properly.  I was wondering if any MIPS
experts out there could tell me what the proper fix is.

Thanks,

Vince