jdolecek@netbsd.org said:
> I think you can keep using curlwp, it should be always TRT in
> selrecord() context. Add KASSERT(selector == curproc) to be sure tho
> :) 

OK, but this would be ugly. Passing a proc pointer to many functions
seems superfluous sometimes; only for MP systems it is probably less
expensive to pass it all the way than looking up curcpu() or so
everytime.
But then, looking up curlwp is equally expensive.
Anyway, before I'll get serious about this I want to have a chance
to do performance comparisions. For now, I don't want to interfere
with efoorts to eliminate more serious pthread problems.

best regards
Matthias