I have gotten some complaints  about our choice of the "modified" REAL
address memory model as being  unpalatable, since it violates the soul
of  x86.  I  agree with  that  sentiment.  So,  what do  we do.   Dave
Armstrong and I have looked at  a lot of options.  We are motivated by
the following:

a. Give you  something to do that *feels* consistent  with the soul of
   x86.

b. Don't  expand the requirements  of the project  unnecessarily this
   late.  in the semester.

We came up with the following:

We will move  from Real-address mode to a  somewhat simplified version
of  Segmented  mode.   Recall  from  lecture and  discussion  that  in
segmented mode, an  address is computed by using  the segment selector
as an index into one of  the two descriptor tables (GDT or LDT).  [Bit
[2] of the  segment selector tells us which of GDT  or LDT to access.]
This  gives you  the  8 byte  descriptor,  from which  you can  easily
extract the 32-bit Base Address.   Finally, that base address is added
to the 32  bit offset to form the 32bit linear  address.  [The rest of
the world uses the term virtual address instead of linear address.]

From  there the  linear/physical mapping  is  performed as  we did  in
class.

You  should  also  know  that   a  hardware  speedup  exists  in  many
implementations, as  follows: The first  time the segment  selector is
used to  compute a linear address,  the Base Address  contained in the
descriptor  is  (in addition  to  being  used  to compute  the  linear
address)  saved in  a hidden  32  bit register  alongside the  segment
selector.  Subsequent accesses via  the same segment selector need not
therefore access  the GDT  or LDT.  Instead  they can simply  read the
base  address directly from  the hidden  32bit register  alongside the
segment selector.

So, what does this have to do with you?

We will assume (for purposes of  this course only) that the entries of
the GDT  and LDT  have the following  *remarkable* property:  The base
address part of each descriptor  is simply the 16 bit segment selector
shifted left by 16 bits.  This "magical" assumption means you need not
implement or access the GDT or LDT in your project.  Do you see why?

When you load a segment  selector, you can immediately load its hidden
Base address register  with the 16 bit segment  selector, shifted left
by 16 bits.

Then when you need to perform  an address computation, you can add the
proper Base address  (obtained from the hidden 32bit  register) to the
32 bit offset computed from other information in the ISA.

Yale Patt