[ECOS]exception stack

["james chen" <james_ch1@sina.com>]

Hello,

    In vectors.S (hal/arm/arch/current/src/vectors.S), sp register in each
ARM state had been initialized during rest vector, but in
undefined_instruction code, the first instruction is initializing the sp
register again with different stack, the undef_exception_stack is larger
than exception_stack. Is it done for special reason or other? may anyone
tell me why, thanks!

========================================================
        .global start
start:
        // initialize interrupt/exception environments
        ldr     sp,.__startup_stack
        mov     r0,#(CPSR_IRQ_DISABLE|CPSR_FIQ_DISABLE|CPSR_IRQ_MODE)
        msr     cpsr,r0
        ldr     sp,.__exception_stack
        mov     r0,#(CPSR_IRQ_DISABLE|CPSR_FIQ_DISABLE|CPSR_UNDEF_MODE)
        msr     cpsr,r0
        ldr     sp,.__exception_stack

        // initialize CPSR (machine state register)
        mov     r0,#(CPSR_IRQ_DISABLE|CPSR_FIQ_DISABLE|CPSR_SUPERVISOR_MODE)
        msr     cpsr,r0

        // Note: some functions in LIBGCC1 will cause a "restore from
SPSR"!!
        msr     spsr,r0

        // initialize stack
        ldr     sp,.__startup_stack

       .................

//
// Exception handlers
// Assumption: get here from a Supervisor context [mode]
//
        .code   32
undefined_instruction:
        ldr     sp,.__undef_exception_stack     // get good stack
        stmfd   sp!,{r0,fp,ip,lr}
        mrs     r0,spsr
        stmfd   sp!,{r0}
        mov     ip,sp                   // save SP which will vanish with

                                        //   mode switch
       .......

// -------------------------------------------------------------------------
// Temporary interrupt stack

        .section ".bss"

// Small stacks, only used for saving information between CPU modes
__exception_stack_base:
        .rept   32
        .long   0
        .endr
__exception_stack:
        .rept   32
        .long   0
        .endr
__undef_exception_stack:
     .........