On Mon, 10 May 2004, Alex Fedosov wrote:
>
> To: Allan Cruse <cruse@euclid.math.usfca.edu>
> Subject: bug in try32bit.s (and derivatives)
>
 
I spotted a bug in your try32bit.s demo (and all subsequent demos 
and projects that are based on it) in the compute_vram_offset 
procedure. Originally, it looks like this:
 
compute_vram_offset:
                                                                                
        imul    edi, ebx, #4096         ; EDI = offset to page
 
        movzx   ebx, dh                 ; extended row-number
        imul    ebx, #160               ;  times row-length
        add     edi, ebx                ; added to page-origin
 
        movzx   ebx, dl                 ; extended col-number
        imul    ebx, #2                 ;  times col-width
        add     edi, ebx                ; added to row-origin
 
        ret
 
 
Code that runs after this procedure (specifically, scroll_vpage_up
procedure) expects to find the video RAM page number in the EBX 
register but this function uses the EBX register to temporarily
store the offset within VRAM on the current page. Because of this,
scroll_vpage_up would crash whenever called. (And we never noticed 
before because only for the current project our Linux executables 
print more than a whole screen of text, requiring scrolling.)
  
Pushing EBX on the stack at the beginning of the procedure and popping
before returning seems work fine as a fix.
 
Alex
 

-------------------------------------------------------------------
Note added by Instructor: The list of posted demo-programs derived
from 'try32bit.s' and thus affected by this bug-fix is shown below.

		'tryexec.s'		posted 29 MAR 2004
		'ondemand.s'		posted 05 APR 2004
		'trydebug.s'		posted 07 APR 2004
		'trydebug2.s'		posted 08 APR 2004
		'trydebug3.s'		posted 12 APR 2004

Thanks to Alex Fedosov for pointing out this previously undetected 
error, and for proposing a very simple way of correcting it. --ABC
-------------------------------------------------------------------