.686
.model flat, stdcall
option casemap:none

include windows.inc
include kernel32.inc
include user32.inc

includelib user32.lib
includelib kernel32.lib

.data
stack dword NULL

.code
start:
invoke VirtualAlloc, NULL, 16380d, MEM_COMMIT, PAGE_EXECUTE_READWRITE    ; allocating 16kb of memory
assume fs:nothing                                                        ; assume FS:nothing so masm doesnt err
mov dword ptr fs:[8],0                                                   ; set TIB's stack base to 0
mov dword ptr fs:[4],7fff0000h                                           ; set TIB's stack limit to the end of user land memory minus 64kb                                                            ; store esp in evx
xchg esp, eax                                                            ; swap the stacks  ( EAX holds the address to allocated memory )
add esp, 16384                                                           ; set esp to point to the end of allocated memory
mov ebp, esp                                                             ; mov the stack pointer into the base pointer
sub ebp, 4                                                               ; and sub the base pointer by 4 ( EBP is usually ESP-4 )
push eax                                                                 ; push our old stack address onto the new stack                                
invoke MessageBox,NULL, NULL, NULL, NULL                                 ; call some stuff on the new stack
pop esp                                                                  ; yep ;) you can pop esp, in this example, its the best way 
mov ebp, esp                                                             ; to the original stack,  now me mov esp into ebp
sub ebp, 4                                                               ; and sub the base pointer by 4 to get it back to its correct value
invoke ExitProcess, NULL                                                 ; and we push data onto our old stack, then exit.
end start