Pushing the Return Address

To return to the caller a subroutine must have the correct return address in $ra when the jr instruction is performed. But this address does not have to remain in $ra all the time the subroutine is running. It works fine to save the value of $ra and then to restore it when needed.

In the picture, the operating system calls main. The return address to the operating system is in $ra. As soon as it gets control, main pushes the return address on the stack (step 1). The return address that main should use when it returns to the operating system is now on the top of the stack.

The "push" and "pop" instructions in the picture are conceptual. Actual code does these operations in the usual way.

main computes for a bit, and then calls subC using a jal instruction (step 2). This puts the return address to main in $ra. Luckily, it does not matter that $ra has been changed because the return address that main will use is safely on the stack.

Because subC does not call another subroutine, the return address in $ra will not be altered. When subC returns to main it uses a jr $ra instruction (step 3).

Control returns to main, which computes for a while longer. It returns to the OS by popping the stack into $ra and executing a jr $ra instruction (step 4).

Trap handler service 10 is another way to return to the OS. For this example let us not do that.