Matt,

I think we've reached agreement on the key points; and I was never
proposing to make these changes pre-2.0.


>However, I do think that passing in curproc is wrong.  If you are
>running off a softintr event, the supplied proc pointer must be 0.

I agree about not passing curproc. As someone else said when I
mentined this back in January, ``curproc should die''. I thought I'd
said I passed curproc n from the NFS-boot code solely to preserve,
exactly and completely, the existing semantics (where sosend() picks
up curproc right after NFS calls it).

I have private uses which call into sosend() from kcont context, which
pass some _other_ struct proc * down the callchain; but thats a
different story.

>I have set of changes I'll commit after the 2.0 branch.

As do I, so I'd appreciate the chance ot see yours for comment before
they get committed.  (I have a curproc-free version of the patch I
posted in a parallel tree, which might also reasonably be committed.)