Subj : Interrupts
To   : Vitus Jensen
From : Lee Aroner
Date : Sat May 26 2001 03:29 am

 LA>>    Not to niggle the point, but of course OS/2 uses interrupts. Each    
 LA>> and every one of those API calls is a wrapper around an interrupt    
 LA>> call.

 VJ>> Well, to be exact: OS/2 uses callgates to jump into kernel code. 
 VJ>> Callbacks  are similar to interrupt gates in that 
 VJ>> they switch  priviledge
 VJ>> levels and have a predefined entry point into  the kernel but they
 VJ>> additionally copy bytes from lower to  higher level stacks while
 VJ>> swichting (byte count defined in  the callgate).  Something what a
 VJ>> interrupt gate won't do.
 VJ>> It is a common design of OSses to do this switching via a software
 VJ>> interrupt but OS/2 is different.

 LA>    Ah, but are not those callgates accessed through an interrupt?    
 LA> It's irrelevant that it is an invalid interrupt, a spade 's a    
 LA> spade after all...

VJ> It's no INT xx instruction but really a call far.  You do a 
  > *CALL FAR XXXX:0000*.
  > The offset doesn't matter (as the callgate defines the 
  > entry point) but the XXXX (selector) points to an entry in 
  > the GDT or LDT (LDT should be possible but I've never seen 
  > a callgate in it in OS/2) which is a callgate (descriptor 
  > type 4).  The descriptor defines the destination 
  > (selector:offset) and how many words are copied from the 
  > lower ring stack to the new stack.

VJ> You can examine how this works by installing _Theseus_ on 
  > your machine.  Debug your program using any debugger which 
  > can display assembler instructions and trace until you find 
  > a "CALL XXX:0000" instruction, write down the XXXX.
  > Now start Theseus and have a look at the GDT.  Lookup XXXX 
  > (ignore bit 0-2), you find a descriptor with /Descriptor 
  > Priviledge Level/ 3 (because it's callable from a ring 3 
  > application), some offset and a selector which is probably 
  > 0168 (perhaps dependent on OS/2 version).
  > Lookup 0168 in the GDT, you find the code selector for the 
  > kernel, ring 0, limit FFFFFFFF.

VJ> I would like to have an Intel or AMD manual about the i386 
  > because this stuff IMHO is best explained by the processor 
  > vendor but I have only a bad german short reference.  One 
  > of these days I should "lend" one from the library...

VJ> Bye,
  >    Vitus

VJ> PS: I'm maintaining an OS for embedded systems where we use 
  > an interrupt, a structure pointed to by register XWA and a 
  > software dispatcher.  This is probably what you have in 
  > mind.


   Vitus, thanks for the detailed description, and for setting me 
   straight. I had assumed that OS/2 used the same method of forcing 
   a processor exception to jump to ring zero as MonopolySoft uses, 
   apparently I was wrong.

                                       LRA


 -- SPEED 2.01 #2720: Between two evils, pick the one you never tried before

--- Maximus/2 3.01
 * Origin: Top Hat 2 BBS (1:343/41)

.