> There are reasons I do not want to do this with libc - basically, a libc
> approach can be bypassed with static bins, which I cant allow in this case.
In which case you'll need to modify the actual kernel. The syscall map is
static and its a very dangerous thing to play with as you shouldnt assume
its just going to be a vector table. We pull tricks already - eg the sparc
uses the low bit of the vector to indicate a call it can skip the expensive
window saving on.
Although Alan's points are valid, I'm getting rid of that low-bit
"feature" because after optimizing the trap entry/exit code last night
I decided the complexity/performance-gain ratio didn't warrant the
optimization.
Later,
David S. Miller
davem@caip.rutgers.edu