Signal Stack

问题:

I did read that signals need to have a separate stack, why and how do you think it is implemented ?

Are they dynamically allocated or statically allocated ? How is memory allocation done ? Is it the same for all signals ?

回答1:

The reason that signals need a separate stack is that, if the normal stack gets corrupted or overflows, the signal can still execute. I think the signal stack is usually allocated dynamically, but it could implemented be either way. You can set a new signal stack with sigaltstack. It is the same for all signals.

回答2:

Another, more esoteric reason is to create user-level threads. See the paper, "Portable Multithreading" by Ralf Engelschall: http://www.gnu.org/software/pth/rse-pmt.ps. It describes how the "signal stack trampoline trick" is used to effect userspace threading.