Enforcing soft-float.

[Juli Mallett]

On Tue, Mar 13, 2012 at 06:35, Andrew Duane <aduane at juniper.net> wrote:
> If it can't be done safely/correctly, it can't be done. I would think
> killing the program would be appropriate.
>
> Getting a wrong answer is catastrophic.

My greatest concern is that any user program can use floating point
instructions to panic the kernel by getting it to truncate its own
pointers.  Heck, an even slightly well-crafted program could probably
predict likely values of the kernel stack pointer and set up many
malicious userspace stacks that could match the truncated address, and
get its own code running in kernel mode.