> Can anyone verify if they can reproduce this problem?

(sigh)  Yes, I can reproduce it. The problem is not in the NFS code, but
rather in gcc's code generation for the sparc.

gcc thinks it can use a FP register to store a `u_quad_t' value into
memory. Apparently, it is only inclined to do so if it runs out of
available registers, e.g. when compiling into leaf routines.

This is not a smart thing to do as it completely defeats the optimization
(if the program does not use the FP otherwise), because of the "FP processor
disabled" trap you incur. The kernel is one such program..


-pk