> segment (8K segments).  On the 386 the base address of each of the
 > segments is a 32 bit number, so 8K segments of size 64K gives 512 meg
 > of addressable memory.

 dn> Only 4K entries in each descriptor table is possible. See below.

 >    Each task has a LDT (local descriptor table) that is the same size
 > as the GDT, so in theory a 286 protected mode task running on a 386
 > could access 1gig.  Perhaps David Nugent will be able to explain
 > what's wrong with my theory.

 dn> There are only 8K unique descriptors available to any 
 dn> task in total. Each desciptor is 16 bits (obviously), 
 dn> less 2 for priviledge level, less 1 for the LDT/GDT 
 dn> select bit, leaving only 13 bits remaining for the 
 dn> actual descriptor table index.

    So each descriptor table can have 2^13=8192 descriptors.  Therefore
between both tables there are 16384 descriptors of which 16383 are
available (one entry in the GDT is taken for the LDT).  That makes the
total memory addressable by a 286 program running on a 386 just 64K less
than 1 gig.  Of course if multiple programs are going to use the same GDT
(IE the GDT is really global) then the amount of memory addressable per
task is 1 gig - 64K* number of tasks.
   Is there any problem in my logic?


   Also I've found a mention in the text book I referred to which states
that a 286 mode program can address 1 gig of virtual memory (which is
obviously wrong as other information given in the same chapter gives a
total of 1gig-64K), but it's just a generalisation I guess.


   cya


--- Maximus/2 2.01wb




SD®¯Øê(