nathanw@wasabisystems.com said:
> My best guess was that the signal code was written that way to make it
> possible for signals to still be delivered even if the program was
> doing some USER_LDT hackery. 

Ah yes, this might have been the intention.
I've just modified the regress/.../ldt program to kill descriptor #2
(LUCODE_SEL). With the existing code, the program just hangs. It doesn't
get a signal, so there must be a bug somewhere which thwarts the good
intentions behind GUCODE_SEL.
With my modifications, it is much worse - the box just hangs solidly,
no DDB, nothing.
It seems there is an unhandled case somewhere in trap handling...

best regards
Matthias