On Thursday, 31 December 1998,
     HERBERT ROSENAU wrote to IVAN TODOROSKI about Some weird addressing...

 IT>>    Do you get a single descritor with a 1000000 byte limit, or a
 IT>> bunch    of small descriptors with 65536 limits?

 HR> In a 32 bit process you will become 245 descriptors. Each
 HR> describes a single page of 4 KB. Then you can acces all Bytes in a
 HR> range of 245 * 4KB.

 HR> Lets look at the dimenson:

 HR> 1000000 / 4096 = 245 pages

 HR> x[1003519] is functional altough it's is illegal from your view.
 HR>            The processor (and the operatingsystem) can bounds check only
 HR>     on boundary of a single page = 4096 Bytes.
 HR> x[1003520] will crash with access violation. It will address the
 HR> unallocated
 HR>            page 246.

  From the little assembly protected mode programming I did, I would say
  that assigning a separate new descriptor to every 4KB of memory would
  be extremely inefficient and a terrible waste of address space, since
  the descriptor tables in 80386 and later processors can hold only 8191
  descriptors, which would limit the address space of every single
  process to little less than 32 megs. (8191*4096)

  I think that you might have mistaken the descriptors for page table
  entries. Descriptors can describe memory chunks of any size up to 4GB
  (well almost any size, because of granularity issues), while page
  table entries can describe only fixed 4KB chunks of memory.

 HR> You'll become a logical continuous block of memory. It my or my
 HR> not physical continous. From view as application programmer it
 HR> dosn't matter if all pages bounded physically or not. Because the
 HR> processor works with virtual addressspace.

  Yes, the page mapping is transparent from programmers (even assembly
  programmers) point of view. But chunking up an allocated piece of
  memory in multiple decriptors is NOT totally transparent. I think it
  can BE MADE transparent in software (or the OS) by installing
  appropriate handlers for exceptions 11,12,13 and 14, but it would be
  messy and would require additional CPU processing. Not efficient in
  any case, especially since you already have a good, transparent and
  efficient hardware mechanism (arbitrary sized descriptors coupled with
  page mapping) to alocate chunks of arbitrary size.

     Any more info is appreciated!

                                                            - Ivan -

.!. Cannot understand Tagline: bort, etry, augh anyway?
--- Terminate 5.00/Pro [OS/2]
 þ TerMail/QWK þ