Switching stacks efficiently

Question

Due to some reason, I switch the stack for calling some functions in my application. I use makecontext/getcontext/swapcontext for that purpose. However, I find it to be too slow. I tried to use custom made code for that purpose, which saves the stack pointer and other registers and then assign the stack pointer the value of the new memory which I want to use as a stack. However, I keep getting stack smashing detected error.

Are there some special permissions set for the stack by the OS or else what is the matter here? How to circumvent the problem.

Answer 1

The excellent GNU Pth library makes heavy use of these techniques. It's very well documented, and determines the most efficient context switching mechanism at compile time. edit: at configure time actually.

The author's paper: rse-pmt.ps gives a technical account of user-space context switching and related issues - alternative signal stacks, etc.

Answer 2

You could look at other software doing the same dirty tricks as you do. In particular Chicken Scheme. You might perhaps consider using longjmp after manually doing dirty things on the target jmp_buf. Of course, none of this is portable.

But please explain more your overall goal. Your questions are generally too mysterious.... (and that is repulsive to some)

Answer 3

The makecontext()/getcontext()/setcontext()/swapcontext() are quite efficient as they merely save the current values of the processor registers. However, in Linux/GLIBC at least, setcontext() and getcontext() invoke the rt_sigprocmask() system call to save/restore the signal mask of the calling thread. This may be the reason why you face some performance issues as this triggers a context switch into the kernel.