textfiles/computers/aboutems.txt

This document was exported from WP 5.0 and some of the formatting codes
were lost (especially footnotery).  Please pardon the occasional anomalies
especially out-of-place "1"s which were originally superscripted (footnote).





             MEMORY EXPANSION IN 80x86-BASED COMPUTERS UNDER MS-DOS

	     			prepared by
	     		John Wilson, Hyperdyne, Inc.
				Annandale, VA

IN THE BEGINNING ...

        In the beginning, the was the 8080 microprocessor.  The 8080
was, defendably, the "first" microprocessor.1  When the 8080 was
invented, memory for computers was very expensive.  The 8080 could
directly address 64 thousand bytes of information, which was a very
great deal in those days.  Few systems (or more correctly, few
system owners) could afford to actually place a full 64 K of memory
in their systems.  But things changed quickly.  The Silicon Valley
guys learned to make transistors smaller and better and it became
much more economically feasible to talk about fully populated 64K
machines.  On the software front, things like spreadsheets and word
processors were being invented.  These programs were hungry for
memory and the microcomputer user rapidly outgrew the 64K box.

        Before going on with the story, a little digression into the
origins of the 64K limit.  This limit was the direct result of the
fact that the CPU chip had only 16 address lines.  The address
lines are a set of wires coming out of the CPU which allow the CPU
to indicate what item of memory it wants to read or write.  In most
computers, the size of the "pieces" are bytes, or 8-bit characters. 
These 16 wires are called the Address Bus. The voltages on the 16
address lines are interpreted as a binary number (with the first
pin representing the 1's place, the second pin representing the 2's
place, the third pin representing the 4's place, etc.) the
resultant number is the Address being addressed by the CPU.  The
number of distinct patterns of 16 things, each of which can have
two values, is 2 raised to the 16th power, or 65536.  This number
is 64 times the quantity "1K" which is 1024.  (Early on,
programmers engineers decided to use the "K" as a unit of counting
things like memory because it was a nice power of two and figured
it was close enough to a "real" thousand that nobody would notice). 
The point of this story is that the original microprocessors could
only address 64K because of the simple reason that they only had
16 address lines.  Why didn't they just build one with more address
lines?  Well its not that simple.  Those address lines had to have
circuitry driving them, and other circuitry to actually generate
the 16-bit addresses, and other circuitry to decode the new
instructions that would use more than 16 bits, and so on, and so
on.  And this circuitry was made of transistors.  And,
unfortunately, the state of the art of chip manufacturing did not
allow chips to be built with many more transistors.  So, mankind
waited for the silicon boys to learn how to fit more transistors
on one raisin-sized chip.

        Meanwhile, back the ranch, The Intel Corporation, in a stroke
of electromarketing genius took the basic 8080 architecture and
doubled up a small portion of the internal workings of the CPU's
address circuitry.  They basically duplicated the address register
(the transistors that hold the pattern of bits to place on the
address bus), slid it left 4 bits, and added some simple circuitry
to add it arithmetically to the "old" address register.  An so was
born the infamous Intel Segment Register.  By making this simple
kluge to the 8080, Intel created the 8086 and 8088 microprocessors
which now effectively had TWENTY address lines and could therefore
address 2 to the 20th locations, a little over one million bytes. 
This new unit was called a Megabyte even those it was further away
from a million than a "K" was from a thousand, but then again you
bought the "K", didn't you?  The 20 bit address bus allowed the new
amazing spreadsheets to really do some amazing things. This, in
turn allowed Intel to beat competitor Motorola to the marketplace
(who were expanding the address bus the "right" way in their 68000
family).  This in turn led IBM to select the Intel family over
Motorola's as the basis of their new PC, and the rest is history.

        Using the power of the 8088, the Microsoft corporation adapted
the CP/M operating system to the new chip, and, using its new
power, created MS-DOS.  Because of a lot of reasons, the 8088 and
MS-DOS took over the world.  And everyone was happy.  Except .....
the programmers kept getting more bold in the ways they found to
use memory, and the users wanted still BIGGER spreadsheets, and
started playing with things like CAD/CAM, DBMSs, Artificial
intelligence, desktop publishing, etc, all of which had insatiable
appetites for memory.  The silicon boys kept up with the hunger by
developing bigger and better CPUs, the 80286, 80386, 80486 etc.). 
These 80286 had a 24-bit address bus and could therefore address
16 MB of memory directly.  No one could possibly want to put that
much actual memory in a PC, right?  In a pre-emptive first strike,
they also created 80386 which had a 32-bit address bus and could
therefore address 2 to the 32 bytes or over 4 billion
bytes!  (Pow! Bam! Take THAT, Power User).  The day had finally
come when the CPUs sitting on desktops could finally address more
memory than anyone could afford to buy.
-----------
1 Yes, there was an 8008, and a 4004 before that, but their only,
surviving significance today is that they were the predecessors of
the 8080.
----------

        End of problem, right?  Wrong.  Unfortunately, there were
millions and millions of the 20-bit machines out there now (in the
mid-1980's).  Probably more significantly there were hundreds of
millions of dollars invested in MS-DOS software that did not know
how to use the new 32-bit instructions and capabilities. 
Especially MS-DOS.  (Unix and OS/2 and a number of other less well-
known operating environments do use the 32-bit mode, but MS-DOS was
still king).  Because Intel wanted to sell more than 3 of these new
chips, they wisely decided to build "modes" into the new chips to
make them compatible with MS-DOS.  A Mode is essentially a switch,
inside the CPU that turns it into another chip, insofar a all
logical functionality is concerned.  So when you're running MS-DOS
on your shiny new 386 or 486, you're still running with only 20
bits of address and consequently a 1 MB address space.  This mode
is called real mode and the lower 1 MB of addressable memory in
real mode is called conventional memory.

        The solution? EMS or Expanded Memory Specification.


EXPANDED MEMORY

        Expanded memory is a way to allow more than 1 MB of memory to
be used by MS-DOS applications.  How can this be?  The CPU can only
address 1 million different addresses;  how can I have more than
one MB in my PC?  The answer is that the CPU re-uses the same
address to get to more than one byte of data.  It does this by
allowing any one address to actually be used to reference several,
distinct, physically different storage locations - but only one at
a time, of course.

        EMS memory usually resides on special "EMS cards", like the
AST "RAMPage".  (I say usually because there are some clever ways
of implementing EMS on 80286 and 80386 machines that don't use
"special" EMS cards; this is discussed later in the section
entitled "Software Approaches to EMS").  The EMS card has memory
chips on it just like regular system memory cards.  The difference
is that the memory chips on the card are not connected directly to
the CPUs address bus.  These chips are actually wired to another
address bus, totally contained on the card, that has more than 20
address bits, usually 24 or so.  Where do these extra address bits
come from?  Well some of them are just passed-through systems
address bits.  The rest come from special storage locations onboard
the EMS card, called page registers.  How are these page registers
loaded?  These registers are themselves directly addressable by the
CPU.  To access a byte of data in EMS memory, the CPU first loads
the page register (itself simply another location accessible by the
CPU) and then makes a normal memory reference.  Some of the bits
come from the address bus, and the rest come from the bits
previously squirreled away in the page registers.  Thus, it can be
seen that a given address on the regular CPUs address bus forms
only part of the address needed to select a particular byte of
memory on the EMS card.  To uniquely identify a byte, you need to
specify the regular address plus the contents of the page register. 
It follows that one CPU address can correspond to several EMS
memory locations, each of which differs only by the contents of the
page register.  The CPU can thus access more than 1 MB of memory
on the EMS card by using its normal address bus augmented by the
page registers.

HOW IT ALL FITS TOGETHER

        In the early versions of EMS, all EMS memory was mapped to
appear to be in a special address range in the range of all
possible addresses accessible by the CPU.  This was usually at
addresses between D0000 and DFFFF (in hexadecimal notation).  This
includes exactly 64KB possible addresses.  This area of address
space is called the Page Frame, an analogy to a frame around a
picture.  The page frame is logically divided into 4 16KB "pages".1 
When we say that the EMS memory is "mapped" into this range, what
we mean is that the EMS card does not respond to any addresses
outside this range.  When an address is placed on the address bus
----------
1 For the sake of simplicity in the following discussion, the
multi-page nature of the page frame will not be mentioned further. 
The explanations apply to each page within the page frame
individually.  It should also be noted that, strictly speaking, the
4-page, 64 KB page frame applies only to EMS versions 3.2 and
earlier.  In EMS 4.0+, the page frame is not limited to just 4
pages and can, in fact, be all or partially located within the 640
KB address range normally occupied by conventional RAM.  This
feature is used by various multitasking overlays to DOS, such as
Desqview, which actually allow program code to be paged in and out.
----------
outside this range, the EMS board remains totally passive, just
like it was not plugged in at all.  When the CPU asserts an address
within this range, the EMS board comes to life and responds like
regular memory.  When the CPU references the EMS address space, the
CPUs address bits are used along with the page register bits to
actually specify which EMS byte to access.  The net effect is to
make the EMS card memory appear as a series of "banks" which can
be made (one at a time) to masquerade as "real" system memory at
a certain address in the range D0000-DFFFF hex.  These banks are
called EMS pages.  Within each page, the practically any byte
addressed is selected by the system address bus bits that were
passed through by the EMS circuitry.  The particular page selected
is determined by what value was previously written into the EMS
page registers.  All the physical EMS memory locations that respond
to a common value in the page registers are said to be in the same
page.  That is, once that special value is loaded into the page
register, any of those locations can be accessed by the CPU using
normal memory read/write instructions.  If another value is loaded
into the page registers, a totally different set of EMS memory
bytes (i.e, physically different transistors) are made to respond
to the same CPU addresses as before.

        The model that this behavior suggests is that the EMS memory,
divided up into 16K pages, exists out in limbo somewhere, and is
unable to be addressed by the CPU in any way.  The CPU can however
invoke the right magic to instantaneously plug any one of these
disembodied pages right into its addressable memory space.  The
magic consists of loading the Holy Page Register.  The CPU can,
with equal ease, banish that same page back into limbo, by putting
a different value in the page registers.  The thing that makes this
magic useful is that, any data stored in an EMS page is faultlessly
remembered even after it has been banished to zombie land.  And,
furthermore, that data can be read by the CPU just by remapping it
into the address space.

        This means that an MS-DOS program can juggle megabytes of
memory resident data using just 20 address bits in good old 8088
real mode.  Of course at any one given instant, most of that data
is in zombie land, but no matter, it can be called back from the
netherworld with a simple, hardware-assisted incantation in
microseconds.

EMS MEMORY MANAGERS OR DRIVERS

        Each manufacturer of an EMS board is free to actually design
the actual  circuitry of his EMS board to suit his whim, his
engineers, and his marketing plan.  Most boards are different in
a real, physical way from one another.  The magic incantations
necessary to shuffle EMS pages between here and Hoboken is
different for each one.  Does each application program need to know
which particular brand/model of board is plugged in and what its
religion is?  Fortunately not.  Enter the Enhanced Memory
Specification.   EMS is a specification of a standardized way that
applications interact with EMS hardware.  This interaction is via
the software interrupt feature of the 8088/MS-DOS.  All
applications that wish to use EMS memory call interrupt 67H the
same way with the same arguments, no matter who made the board. 
When the interrupt is issued, control passes to the memory resident
EMS management software, usually called either an EMS memory
manager or EMS driver (same thing).  This piece of software is
unique to each brand of board and is normally supplied by the
boards vendor.  It is the express purpose of this piece of software
to turn the standard EMS invocation arguments into the particular
set of hardware incantations understood by the board.  Beware
mixing boards and drivers from different sources!  This may work
in rare circumstances but will eventually lead to consumption of
excessive amounts of alcoholic beverages.

EXTENDED MEMORY

        OK. Now we know how EMS works: it expands a selected 64KB-
sized range of addresses in the CPUs address space to several
megabytes by paging-in one chunk at a time.  But what about
"Extended Memory"?.  Actually extended memory is a much simpler
concept.  Remember those unused address lines in the 80286 and
80386? (MS-DOS and other real-mode applications only use the first
20 of the 80286's 24 and the 80386's 32).  They were not put there
for decoration.  The CPU can be put in Protected Mode  and can then
use those extra address lines to address megabytes and megabytes
of memory without needing the help of the special EMS hardware like
page registers and private (EMS-card) address busses.  In protected
mode, the CPU           can address all physical memory in the same, natural
way.  In fact, the one megabyte boundary loses all significance,
except for the painful memory of what it used to be like back in
that awful 20-bit real mode.  Extended memory is thus just like
conventional memory, just extended up to higher addresses.  The
down side is, of course, that MS-DOS does not know how to switch
into protected mode, and wouldn't know what to do there if it did. 
Rectifying this shortcoming, and all its implications, is the sole
reason for OS/2.

SOFTWARE APPROACHES TO EMS IMPLEMENTATION

        The discussion of EMS so far has talked exclusively about
hardware approaches to EMS.  In the 8088, hardware must be employed
to supplement the deficiencies of that chip.  In the 80286 and
80386, however, there are software-only methods to give the same
functionality as EMS hardware.  Both approaches use extended memory
for the storage of EMS page data.

        In the 80286, EMS memory contents are brought into the 1 MB,
conventionally-addressable range by physically copying 16 KB blocks
of memory to and from extended memory.  The EMS "page" that the
application program sees is actually a block of conventional memory
that has been filled with the contents of a block of extended
memory.  The copying is done  by a piece of software known as an
EMS Emulator (driver) which is usually loaded like other DOS device
drivers in CONFIG.SYS.   Note that to access the extended memory,
the EMS Driver must switch into protected mode, copy the data, then
hightail it back into real-mode to keep DOS happy.  The extended
memory blocks, in this scheme, are emulating a block of memory that
would normally be physically resident on the EMS card.  Note that
these are not really "paged"-in in the same sense as true EMS
pages, nor is there any "mapping" going on.  All physical memory
contents retain their actual addresses as far as the CPU is
concerned, only there contents are copied back and forth.

        The advantage of this scheme is that EMS functionality can be
achieved in machines without actual EMS hardware.  A disadvantage
of this scheme is its performance.  Whole 16K blocks must be moved
to access a new page (which takes milliseconds), in contrast to
"true" EMS where just a page register must be loaded (which takes
microseconds).  Another disadvantage is the fact that some precious
conventional memory is consumed by the emulated page frame.

        In the 80386, the solution is much more elegant.  In true EMS,
the contents of the page registers can be thought of as a memory-
mapping table.  That is, the contents of the page register, in
essence, point to a particular block of EMS-card-resident memory -
 change the page register contents and a new physical page shows
up in the page frame.  The 386 was designed for operating systems
much more sophisticated than MS-DOS; these operating systems take
for granted the presence of memory mapping capability.  The 80386
has, in fact, an internal memory mapping facility much more
sophisticated than the crude, bank-oriented page registers of an
EMS card.  The 386's memory management unit allows any arbitrary-
sized chunk of physical memory to be mapped to anywhere in the
address space, including the lower 1 MB.  And, most importantly,
to the address were an EMS-aware application expects to find the
page frame and the EMS pages contained therein.  With the 80386,
hardware within the CPU performs the mapping previously done by EMS
hardware.  Programming of the CPUs mapping registers is performed
by a device driver usually known as an Expanded Memory Manager. 
Memory managers are written to run on the (standard) 80386 and not
some particular vendor's EMS hardware.  This allows third-party
vendors to produce EMS emulators for any 80386.  Examples are
"QEMM-386" by Quarterdeck Systems and "386-to-the-Max" by Qualitas.

        Finally, there is one more software approach to EMS that can
be used with any machine.  That approach is called Virtual EMS and
employs a system's hard disk as storage for EMS pages.  A device
driver intercepts EMS calls in much the same way as the approach
described above for the 80286, except that copying is done between
a page frame in conventional memory, and sectors of your hard disk. 
This is a clever approach, and allows EMS memory to be much greater
than the amount of memory in your machine, but, because disk is
thousands of times slower than semiconductor memory, this approach
should only be used by the terminally desperate.

	APPENDIX - SUMMARY FOR USERS

        EMS is the specification of a software technique for making
more than 640 KB (the normal DOS limit) available to your programs. 
Put simply, EMS reserves a block of memory space in your PC and
allows a block of RAM (usually resident on an EMS card) to be
switched into that address range.  There are generally many
identical blocks of RAM present on the EMS card, each and any one
of which can be "plugged" in -- only one at a time.  Your CPU can
use one of these blocks to store data in, and then switch in
another block, store data in that, switch in yet another block, and
so on, and so on.  Later, the CPU can recall these blocks in the
same or different order and read back the original data.  In many
ways, this performs the same function as your system's disk --
except that it's all done in solid-state memory and is thousands
of times faster.

        There are four approaches to actually implementing EMS,
depending to some extent, on what type of machine you have.  These
approaches are:

                -       an EMS memory card (like the AST "Rampage") [any DOS
                        computer, but usually 8088s or 80286s]

                -       an EMS emulator [80286 or 80386, but usually only
                        on 80286's]

                -       an Extended Memory Manager [386 only] (for example
                        Quarterdeck System's "QEMM-386 or Qualitas's "386-
                        to-the-Max" )

                -       a Virtual Memory Manager [any DOS machine]

        An EMS memory card is more than just a memory expansion card:
it contains special circuitry to perform the bank-switching
operation discussed above.  To use an EMS card, you will have to
perform two steps:
(1) Configure the EMS card hardware to match your computer's
configuration and (2) install a special EMS card driver for the
board in your CONFIG.SYS file.  Details and procedures differ for
different makes and models of EMS cards.  Consult your EMS card's
users manual for instructions.  Note that drivers are usually card-
specific; you cannot, in general, use Vendor A's driver with Vendor
B's card.

        Not all "memory" expansion cards are EMS cards.  There is
another type of memory called extended memory which is used by
other operating systems such as OS/2 and Unix.  It is also used by
a few DOS utilities, most notably IBM's VDISK RAM disk emulator. 
If your computer is advertised as having more than 640 KB of memory
installed, it's a good bet that it's extended memory and not EMS
memory.  Few applications can use extended memory, although by using a
software technique which will be discussed in a moment, extended
memory can be made to serve as EMS memory.  Before deciding on an
EMS strategy, determine the type of memory your computer already
has installed.  Be forewarned: IBM, as usual, has a different name
for extended memory (like everything else).  They call it (you
guessed it) expanded memory.  So, if you bought it from IBM, and
it's called expanded memory, it's extended memory.  Everyone else
pretty much sticks to standard nomenclature but to be sure, look
for the phrase "EMS x.x compliant" in the documentation, where x.x
is usually a number like 3.2 or 4.0.
----------
1 The 'S' in EMS stands for "Specification".  EMS is not a
particular way to build EMS memory, rather, it is the specification
of a software interface to it.  Different vendors can, and do,
implement EMS differently.  What is the same, however, is the way
that applications programs interface to this memory.
----------
Many EMS cards allow the memory contained on them to be
configured as all EMS memory, all extended memory, or a mixture. 
If you use VDISK or any other special programs that use extended
memory, you may wish to reserve part of the board's memory for use
as extended memory.  Otherwise, there's really no good reason for
not configuring all of your memory as EMS.  (Note that EMS boards
are generally more expensive than "plain" extended memory boards
because of the additional circuitry required).  Consult your
board's users manual for the proper switch settings or software
settings to give the mix you desire.  EMS cards can be used in any
machine, but are usually found in 8088s and 80286s because there
is a better and cheaper way to go in 80386s as will be discussed
below.

EMS EMULATORS

        A less common approach to implementing EMS in your computer
is a EMS Emulator.  This is a software-only approach that requires
no special hardware to use.  It essentially turns extended memory
into expanded memory.  Unfortunately, there is a price for this
magic - performance.  Because extended memory lacks the special
hardware to switch its address like the blocks of memory on an EMS
card, this software must copy whole blocks of data (16 KB's worth)
back and forth between your program and extended memory every time
a new block is required -- even if it's just to read a single byte. 
Depending on the nature of the program, this can be a few times
slower or hundreds of times slower than "true" EMS.
This is not a recommended solution for
that reason, however, if necessary, it can be used.  This approach
can only be used on 80286 and 80386 machines which have extended
memory.  Machines based on the 8088 (like the original PC and XT)
cannot accommodate extended memory.  On the 80386, a much better
solution is described below.

EXTENDED MEMORY MANAGERS

        Built into every 80386 is a special capability that can be
used to do an excellent job of providing EMS memory without the use
of EMS hardware.  This facility is called the paging unit or Memory
Management Unit (MMU).  The MMU consists of circuitry very much
like the switching circuitry onboard EMS cards, except much more
sophisticated.  It was actually included for use by advanced
operating systems but can be used quite nicely to emulate EMS in
80386-based DOS machines.  The MMU, like the EMS emulators, can
turn extended memory into expanded memory through software-only
means.  Unlike those emulators, the MMU, in conjunction with a
piece of software known as an Extended Memory Manager (EMM), does
not suffer any performance penalty.  In fact, it is usually faster
than true EMS cards because: the circuitry is onboard the CPU chip;
the 80386 is faster than lower-class machines that usually use EMS
cards; and the extended memory used is often fast, 32-bit system
memory rather than card-based memory which is slowed down by the
relatively slow I/O bus.  On 80386 systems, this is definitely the
way to go.  Excellent EMMs include "386-to-the-Max" by Qualitas,
Inc., and "QEMM-386" by Quarterdeck systems, Inc..  To use these
EMMs, you need to install them in your CONFIG.SYS file.  Like the
EMS cards, you will have to configure them to partition your
available system memory between extended and expanded memory. 
Consult the users manual for the package you are using.

VIRTUAL MEMORY MANAGERS

        Virtual Memory Managers are another software-only approach to
EMS.  These function almost identically to the EMS emulators
discussed above, except that they use the system disk rather than
extended memory as the storage medium for blocks of memory copied
out of your program.  As you can imagine, this is excruciatingly 
s-l-o-w.  Use this approach only as a last resort.