> Now, the trampoline is used only to return from the signal.

Have you a scheme to avoid executing the stack [1] on return
from a signal handler?

I had a cunning thought - which avoids requiring any special
code in the program.
Why not make the signal handler return to an illegal PC?
This would fault into the kernel which could act appropriately...
(probably checking the user SP to be the expected one.)

If gcc is writing code to the stack in order to return from
nested c++ functions it is being very lazy.  It would be just
as easy to write data to the stack and return to code that
processed it (clearly possibly since a cpu emulater would work).

[1] I presume this is code to restore the progam's registers
and to re-enable signal handlers for sigset().

	David

-- 
David Laight: david@l8s.co.uk