Re: Returning to userland via a signal handler

[Martin Husemann <martin%duskware.de@localhost>]

On Thu, May 21, 2009 at 01:43:47PM +0000, Martin Husemann wrote:
> Maybe the following analyzis is wrong, or there is something wrong in
> sparc64 machdep code...

<uwe> set me straight and it looks indeed like a problem in the sparc64 MD
code.

When the new lwp is getting ready to run, we pass via lwp_trampoline, call
lwp_startup (yay!) and end up in rft_user, where we notice that there is
a signal pending and reenter trap handling, posting the signal and returning
from the handler via the ucontext.

This means there either is a difference between %tstate handling in the
"reenter trap handling" path vs. the straight return to userland, or the 
signal delivery is not realy needed and we just have been lucky. I'm
investigating.

Martin