• Introduction
• Initialization
• Other Architecture Comparison
• Saving Registers in Function Calls

Introduction

The stack pointer is defined by the ABI.

• In the standard ABI and embedded ABI it is register x2 AKA sp.
• The stack pointer is a software convention, it has no implementation in hardware or specific instructions.
• The compiler implements the stack. It is possible to program in assembler with no stack.

Initialization

The start-up code must initialize the stack before the compiler generated code is executed. The linker might define a region for the stack, or the region could be allocated at runtime. For example in SiFive’s example linker script .

    .stack (NOLOAD) : ALIGN(16) {
        PROVIDE(metal_segment_stack_begin = .);
        . += __stack_size; /* Hart 0 */
        PROVIDE( _sp = . );
        PROVIDE(metal_segment_stack_end = .);
    } >ram :ram

Once SP is initialized a C function can be called.

    __asm__ volatile  ("la    sp, _sp;"
                      "jal   zero, _start;"
                      :  /* output: none %0 */
                      : /* input: none */
                      : /* clobbers: none */); 
    // This point will not be executed, _start() will be called with no return.

Other Architecture Comparison

Compared to ARM Cortex:

• NO automatic stack saving on interrupt.
• NO special instructions for multiple register push/pop.
• NO shadow stack register. (MSP/PSP)

Saving Registers in Function Calls

By default the only registers defined for saving are written to the stack on function calls.

• A function with the ‘naked’ attribute will save no registers.
• A function with the ‘interrupt’ attribute will save all registers.