Currently, all new thread created with zero fpu states (512byte buffer of
fxsave area is filled up with 0s), then when a new thread triggers "fpu
device not avialbe" exception (lazy fpu algorithm) at the first time, the
handler just blindly call fxrstor to fetch all those zero states from fxsave
area buffer. This patch is just to guarantee each new thread to have correct
inital fpu states.
Note that x86 fpu code isn't changed in this patch because the 32/64 bit
fpu code are 99% duplicated. I assume that they should be merged in future.
Signed-off-by: Zhu, Bing <bing.zhu@intel.com>
9f086cb53a