Re: floating point used in kernel

[Peter Bergner <peter@bergner.org>]

:: oana :: wrote:
> i get this in dmesg:
>
> floating point used in kernel (task=c01a9420, pc=c0004028)

We had a similar problem with the ppc64 kernel (which inherited
the ppc kernel's FP unavailable exception handler).  Our problem
was 3rd party modules which were not built with -msoft-float.
We were actually seeing memory corruption in userland applications
due to this, since (IIRC) the userland's FP registers were not being
flushed to its thread_struct when the kernel takes a FP unavailable
exception.

Note, the userland app doesn't have to explicitly have FP code for
there to be FP instructions in your binary, since gcc will sometimes
use the FP registers for copying memory.

Peter