On Tue, 8 Aug 2000, Greg Kritsch wrote:

> Actually, I have slightly greater confidence in NetBSD getting it "right" in
> the kernel header files than arbitrary developer X remembering to do that to
> his LKM.  My confidence goes way down when we're depending on someone
> developing on a port that doesn't need this trick.  Though I suppose the
> only impact is poor performance of the LKM in that case.  Also, there's the
> fact that the short call attribute doesn't seem to exist.

The thing is that an lkm can conceivably call ANY function in the kernel.
While vfs and driver modules call only a few routines, the system call
modules can call anything. :-) So the entire source tree would need this
change. :-( If we leave one off, then that means that you can write an LKM
which doesn't work on powerpc, just because the call doesn't have the
fixup on the prototype. If we add them all, then that means the source
tree bloats because of how we compile LKMs on powerpc.

> You'll have to pardon my performance driven real time background.

That's fine. :-)

Take care,

Bill