TTL	C:\pocket\test\asm.cpp

	AREA	|.drectve|, DRECTVE
	DCB	"-defaultlib:coredll.lib "
	DCB	"-defaultlib:corelibc.lib "

	EXPORT	|?do_it@@YAXXZ|				; do_it
	AREA	|.pdata|, PDATA
|$T222|	DCD	|?do_it@@YAXXZ|
	DCD	0x40000100
	AREA	|.text|, CODE
|?do_it@@YAXXZ| PROC					; do_it
|$M220|
	
	mcr     p15, 0, r0, c7, c5, 0 		;/* invalidate i cache & BTB */
										;	CPWAIT  r0
	mrc	p15, 0, r0, c2, c0, 0
	mov	r0, r0
	sub 	pc, pc, #4		
										;	ldr	r9, =0xa1300100

		mov		r9, r1					;	add		r9, r1, #0x100	
		ldr		r5, =0xa0000100			;	ldr		r5, =0xa0000100
		ldr		r7, =0xa0100000
label   ldr     r8, [r9]
        str     r8, [r5]
		add     r9, r9, #4 
		add     r5, r5, #4 
		cmp		r5, r7
		blt		label

crash	b crash

	MOV		r4, #0xA0000000
	add		r2,r4,#0x8000
	
	mov 	r11,r2
	ldr		r10,=337

;	mcr     p15, 0, r0, c7, c10, 4 ;/* drain the write buffer*/

;	CPWAIT  r0
;	mrc	p15, 0, r0, c2, c0, 0
;	mov	r0, r0
;	sub 	pc, pc, #4		


	mcr	p15, 0, r0, c8, c7, 0x00 	;/* invalidate I+D TLB */
;	CPWAIT  r0
	mrc	p15, 0, r0, c2, c0, 0
	mov	r0, r0
	sub 	pc, pc, #4		

;	//; they skipped this, unnecessary? seems like we need to do this
	mcr     p15, 0, r0, c7, c5, 0 		;/* invalidate i cache & BTB */
	
; S bit set, p and d bit set (no 26 bit mode)
;	mov     r3, #0x120 	; xscale says p needs to be 0 ???
;	mcr     p15, 0, r3, c1, c0, 0   ;/* disable the MMU */

;	CPWAIT  r0
;	mrc	p15, 0, r0, c2, c0, 0
;	mov	r0, r0
;	sub 	pc, pc, #4		


;        /*; make sure the pipeline is emptied*/
        mov     r0,#0
        mov     r0,r0
        mov     r0,r0
        mov     r0,r0

;	/* zero PID in Process ID Virtual Address Map register. */
;	mov	r0, #0
	mcr	p15, 0, r0, c13, c0, 0


;	CPWAIT  r0
	mrc	p15, 0, r0, c2, c0, 0
	mov	r0, r0
	sub 	pc, pc, #4		

;	ldr	r5, =0xA00512F5
;	ldr	r8, =0xefef0000
;	str     r8, [r5] 


	mov     r0, #0
    mov     r1, r10
	ldr     r2, [r2, #0]
	mov     r2, #0
	mov	pc, r11	


|$M221|

	ENDP  ; |?do_it@@YAXXZ|, do_it

	END