Inside The Macintosh Coprocessor Platform And A/ROSE
Inside The Macintosh Coprocessor Platform And A/ROSE
JOSEPH MAURER
The Macintosh® Coprocessor Platform TM provides a foundation for connectivity
products such as the Serial NB Card, the TokenTalk NB Card, and the Coax-
Twinax Card. Its operating system is A/ROSE, the Apple Real-time Operating
System Environment. This article introduces you to the Macintosh Coprocessor
Platform and A/ROSE, and gives you a taste of what is involved in developing a
connectivity product on this foundation.
The Macintosh Coprocessor Platform and A/ROSE together provide a hardware and software
foundation for developers who want to create NuBus TM add-on cards for the Macintosh II family of
computers. The developer's guide that comes with the kit is a hefty 400-page tome. If you're curious
about how NuBus cards are built but not curious enough to tackle the developer's guide, read on.
This article gives you an overview of the origins of the Macintosh Coprocessor Platform, its
architecture, and details of its real-time, multitasking, message-based operating system, A/ROSE. It
shows you some A/ROSE code. And it shows you how to experiment with some A/ROSE
applications included on the Developer Essentials disc.
HOW IT ALL BEGAN
When development of various networking and communications products for the Macintosh II
started at Apple, around 1987, it became obvious that the Macintosh Operating System didn't meet
these products' needs for processing power and operating system capability. After all, the Macintosh
OS was designed for human interaction rather than for connectivity to mainframe computers. It is
not real-time (interrupts can be disabled for longer than is acceptable for fast interrupt-driven
input/output), and it aims to provide a pleasant and efficient graphic user interface, rather than
processor-intensive I/O handling. The solution was to make an "intelligent"
NuBus card, with its own 68000 processor, its own
working space in RAM, and its own basic operating system services; and to design this card not only
as a basis for Apple's own products, but also as a tool for NuBus expansion card developers. The
result was the Macintosh Coprocessor Platform. Its operating system, A/ROSE, was designed to
respond to the needs of connectivity products, complement the capabilities of the Mac OS, and yet
be generic enough to become the foundation for a new breed of message-based, distributed software
architectures. The work on A/ROSE started in August 1987, and the first version was operational by
February 1988.
Today, developers can build on this platform in designing products for communications and
networking, data acquisition, signal processing, or any other heavy-duty processing. Time-
consuming and/or time-critical tasks can be offloaded from the main logic board to a dedicated
processor on the NuBus card. This increases the overall computational speed, of course, and allows
for faster response times in the foreground applications. Moreover, unlike the standard Mac OS,
A/ROSE provides the real-time and multitasking capabilities required for handling multiple
communications protocols.
Nevertheless, A/ROSE on a Macintosh Coprocessor Platform still depends on the Mac OS (and its
limitations--see Technical Note #221) for transferring large amounts of data across the NuBus
through a driver to a Macintosh application. This means that ample data buffering (and careful error
handling) should be provided on the card if the project requires high-performance data transfers. As
you'll see in the next section, the card provides plenty of room for large buffers.
THE MACINTOSH COPROCESSOR PLATFORM UP CLOSE
The most prominent feature of the
Macintosh Coprocessor Platform card is all the empty space on
it, inviting hardware developers to heat up their soldering irons and to put plenty of advanced
hardware on it. A complete master-slave NuBus interface comes for free, implemented by means of
two chunky Texas Instruments ASICs (application-specific integrated circuits), 2441 and 2425. This
interface manages to give the on-board MC68000 access to the whole 32-bit NuBus address space
(by means of an address extension register). Conversely, the 24-bit address space of the local
MC68000 can be accessed directly from across the NuBus. Custom hardware on the card can be
enabled to take over the 68000 bus and even go to the NuBus, but A/ROSE tasks usually take care
of servicing chips on the board, and communicate with the higher levels of the software design.
Figure 1 The Macintosh Coprocessor Platform Card
Figure 2 Memory Map of the Macintosh Coprocessor Platform With A/ROSE Running
The MC68000 on the Macintosh Coprocessor Platform card runs at 10 MHz (the NuBus clock
speed) without wait states. Standard 512K of dynamic RAM is expandable up to 4M. Two 32K
EPROMs contain the declaration ROM code needed to make the SlotManager happy, plus some
pieces of code to help the MC68000 out of a Reset and to provide low-level diagnostic routines. The
card also carries a programmable timer, used by A/ROSE for scheduling time-sliced tasks.
A/ROSE UP CLOSE
A/ROSE is a minimal, multitasking, distributed, message-based operating system. Here's what this
means, in real terms:
It's minimal: The module that provides basic A/ROSE functionality, the A/ROSE kernel, fits into 6K;
and a complete standard configuration of A/ROSE on a NuBus card amounts to only 23K of code
and takes up only about 48K of buffer space. This leaves more than 400K for your code on a
standard 512K RAM card. Still, as you will see, A/ROSE is a strong software platform to build on.
It's multitasking: A/ROSE does pre-emptive multitasking, with round-robin task scheduling (taking 32
priority levels into account).
It's real-time: A/ROSE offers 110 microseconds context switch time, with 20 microseconds of latency
(guaranteed interrupt response time).
It's distributed and message-based: The A/ROSE software can be present on several cards, and it is
completely autonomous and independent on each card. Tasks defined by users and by A/ROSE
communicate with each other, even across the NuBus to other slots or the Mac ® OS, by means of
messages. These messages can carry pointers to data buffers along with them. Thousands of such
messages can be passed per second (fastest from task to task within a card, and slower, of course,
between different slots).
The A/ROSE kernel is responsible for task scheduling, interprocess communication, and memory
management. The calls that correspond to these responsibilities are shown in Table 1. The standard
configuration also includes utilities for bookkeeping and timer services. These utility functions are
carried out by the A/ROSE managers: the Name Manager, the InterCard Communication Manager,
the Remote System Manager, the Echo Manager, the Timer Library, the Trace Manager, and the
Print Manager.
Table 1
The Ten A/ROSE Primitives
|
|
Name | Description
|
AROSEFreeMem() | Frees a block of memory*
|
FreeMsg() | Frees a message buffer*
|
AROSEGetMem() | Allocates a block of memory*
|
GetMsg() | Allocates a message buffer*
|
Receive() | Receives a message+
|
Reschedule() | Changes a task's scheduling mode
|
Send() | Sends a message*
|
SpI() | Sets the hardware priority level
|
StartTask() | Initiates a task
|
StopTask() | Stops a task
|
Notes:
* Implemented in A/ROSE Prep with the same parameters.
+ Implemented in A/ROSE Prep with a supplementary parameter.
The A/ROSE architecture, shown in Figure 3, is completed by A/ROSE Prep, a version of A/ROSE
that runs on the main CPU under the Macintosh Operating System and that is necessary to establish
communication between the Mac OS and A/ROSE. The A/ROSE Prep file has the file type INIT,
and contains among its numerous resources a DRVR named .IPC (for interprocess communication),
and an INIT that executes at INIT31 time and basically installs and opens the .IPC driver. The
.IPC driver takes care of the communication of Mac OS processes with A/ROSE tasks. Nothing can
be downloaded to the Macintosh Coprocessor Platform if the A/ROSE Prep file is not in the System
Folder: it contains card-dependent information needed for the download routines to succeed.
The programming interface to the .IPC driver (described in the A/ROSE header files arose.h, os.h,
managers.h, iccmDefs.h, ipcGDefs.h, and provided through the library IPCGlue.o) mimics that of
A/ROSE itself as closely as possible, providing the look and feel of A/ROSE even if there is no
A/ROSE around. More practically speaking, with the A/ROSE Prep file in your System Folder, you
can do a lot of interesting A/ROSE experiments even without a Macintosh Coprocessor Platform.
For your convenience, the A/ROSE Prep file is included in the A/ROSE folder on the
accompanying Developer Essentials disc.
Figure 3 The Architecture of A/ROSE
WHAT'S THIS ABOUT MESSAGES AND TASKS?
Interprocess communication in A/ROSE takes place by means of messages passed back and forth
between tasks. A typical example consists of a client/server relationship between A/ROSE program
modules, as illustrated in Figure 4 on the next page.
The client task needs to know that the required server task exists; thus, the server task is initialized
before the client task. Next, the client task issues a GetMsg()
call to request a message buffer
from a preallocated pool of message buffers that is maintained by A/ROSE and the size of which is
specified by the user. After the message is filled with addressing information, command codes, and
parameters, it is sent to the server task. At this point, the sending task loses rights to the message
buffer, and should not use it again until it comes back through a Receive()
call. On the other
side, the server task usually sits in an infinite loop, waiting for messages requesting a service,
handling these requests, and sending replies.
Figure 4 How Interprocess Communication Takes Place in A/ROSE
After receiving the reply, the client task can reuse the message buffer for subsequent requests, or
release the buffer by means of a FreeMsg() call and go ahead with other business.
A message provides up to 24 bytes of user data, and is fixed length and asynchronous. If the data to
be sent does not fit into the message proper, then it can be put anywhere in the sender's memory
and the address and size of the data area can be passed in the message.
Each message is identified by a message ID and a message code. The message code is defined by
agreement between the sender and the receiver. A convention followed in A/ROSE is for outgoing
messages to use an even-numbered code and for replies to those messages to set the code to the next
odd number.
The structure of an A/ROSE message is shown in Figure 5.
Figure 5 The Structure of an A/ROSE Message (54 Bytes) In C, the messsage structure is declared as follows:
struct mMessage {
struct mMessage *mNext;
/* Used to chain messages internally. */
long mId; /* Unique identifier for a message. */
short mCode; /* User-defined message code. */
short mStatus; /* Message return status. */
unsigned short mPriority; /* Range is 0 (low) to 31 (high). */
tid_type mFrom; /* Task ID of task sending message. */
tid_type mTo;
/* Task ID of task to which msg. is sent. */
unsigned long mSData[3];
/* Used for sender's private information. */
unsigned long mOData[3];
/* Used by receiver to send data back. */
long mDataSize;
/* Size of data to which mDataPtr points. */
char *mDataPtr; /* Pointer to variable length data. */
};
Tasks in A/ROSE accept and reply to messages. A task is identified to
A/ROSE by a task ID, which is a 32-bit field of type tid_type. Each
task also has an associated name and type that is readable by humans.
This is very close to the NameBinding protocol of AppleTalk in spirit
and implementation.
Tasks are started with a call to the A/ROSE StartTask()
primitive. Tasks have one of 32 priority levels, with level 31 as the
highest priority and level 0 as the lowest. Tasks run either in slice
mode or in run-to-block mode. In slice mode, a task runs for one
major tick (about 50 milliseconds), and then relinquishes control of
the CPU to a task of higher or equal priority, if one is available. In
run-to-block mode, a task runs until it is blocked or until it
completes. A task becomes blocked if it issues a Receive()
call for
a message that is not available. New tasks are scheduled for execution
in the order of priority; a task is run only if no eligible tasks of higher
priority are waiting.
THE A/ROSE MANAGERS
A manager in A/ROSE is just another task, which does its job in accepting and replying to messages
with predefined message codes. As mentioned earlier, the A/ROSE managers are the Name
Manager, the InterCard Communication Manager, the Remote System Manager, the Echo
Manager, the Timer Library, the Trace Manager, and the Print Manager. The first four are
discussed in greater detail here. Use of the Name Manager and the InterCard Communication
Manager is demonstrated in the sample program ShowTasks and in "Building a Download File,"
later in this article.
THE NAME MANAGER
The Name Manager maintains a cross-reference between task IDs and their associated name and
type. User tasks can register themselves with the Name Manager by specifying an object name and
an object type, and then other tasks that need to refer to this task can look up the task by name and
type by calling the A/ROSE Lookup_Task()
utility. Conversely, for a given task ID, the Name
Manager brings back the object name and object type if you send it a message with mCode =
NM_LOOKUP_NAME.
The Name Manager also provides notification services. These services include signaling when a
NuBus card is shut down or started up, checking to see if a task is present or not, and signaling when
a task terminates.
THE INTERCARD COMMUNICATION MANAGER
The InterCard Communication Manager (ICCM) enables user tasks to communicate with tasks on
other NuBus cards or on the main logic board. There are only three message codes a user task may
send to the ICCM: ICC_GETCARDS
, ICC_DETACH
, and ICC_ATTACH
.
ICC_GETCARDS
returns a long integer for each of the sixteen possible NuBus slots. A positive
number represents the task ID of the Name Manager running under A/ROSE on a Macintosh
Coprocessor Platform card. For slot = 0, this is the task ID of the Name Manager incorporated in
A/ROSE Prep, under the Macintosh OS. The task ID of a Name Manager is required to look up
specific tasks on any card on the NuBus.
The message codes ICC_DETACH
and ICC_ATTACH
are provided for NuBus cards that get
power from a source other than the NuBus, so that when the power to the Macintosh main logic
board is turned off, the NuBus card continues to function. With these message codes, you can delink
the NuBus card from the outside world, thus preventing access over the NuBus.
THE REMOTE SYSTEM MANAGER
The Remote System Manager running on a NuBus card enables tasks running on
any other NuBus card or the main processor to execute certain A/ROSE primitives remotely. The
A/ROSE primitivesA/ROSEGetMem()
, A/ROSEFreeMem()
, StartTask()
,
and StopTask()
are supported, enabling tasks to be downloaded, started, and stopped dynamically. The Remote System
Manager registers itself with the Name Manager with the name RSM and the type RSM.
THE ECHO MANAGER
The Echo Manager echoes all messages sent to it. This can be very useful in the initial stages of
testing A/ROSE applications.
DOWNLOADING TO THE CARD
All code running on a NuBus card is downloaded to the card's memory from the main logic board.
Code can be downloaded statically or dynamically, to one or multiple cards.
In static downloading, the user builds the entire memory image of the application to be run on the
card by linking the code with A/ROSE object files. The main program is user code; it callsosinit()
to initialize A/ROSE and osstart()
to start the operating system. Before starting
the operating system, the main program must start all the necessary managers and user tasks. The
memory image is downloaded onto the NuBus card using the static downloading facility, which halts
the card, downloads the code, and starts the card again.
In dynamic downloading, the user downloads a generic version of A/ROSE onto a NuBus card by
invoking StartAROSE()
. Once the A/ROSE kernel and requisite managers are up and running,
the user can download tasks using the dynamic downloading facility.
A/ROSE provides a number of ways to download the code. The MPW tool Download takes the
pathname of a file as parameter, and tries to download it to every Macintosh Coprocessor Platform
card it finds (if used without the optional slot parameter ). This is convenient during the
development cycle under MPW. Another possibility is to use the Macintosh application ndld.
Finally, you can use the NewDownload()
routine directly from within your own application. (See
the sidebar on the next page for a description of Download and ndld.)
SOME SAMPLES OF A/ROSE PROGRAMMING
You'll find some samples of A/ROSE programming in the A/ROSE folder on the
Developer Essentials disc.
You can run these applications under MultiFinder after booting with A/ROSE Prep in the
System Folder. With the exception of the downloading operation, all these applications will run
whether or not your machine has a Macintosh Coprocessor Platform card installed.
You can take a closer look at the complete source code on the Developer Essentials disc. I'll show and
discuss some fragments of it here.
TASKSAMPLE AND CLIENTAPPLI
The TaskSample application opens a window and waits for A/ROSE messages. The ClientAppli
application looks for a server named myTaskName and sends a message on each button-click. The
server TaskSample simply returns each message it receives to the sender, and ClientAppli displays the
number of messages it has sent and received.
To experiment with producing alerts or error messages, launch both applications, then quit
TaskSample and continue sending messages to it; restart it again and continue; or hit Command-Q
immediately after the Send button, so that ClientAppli has gone by the time TaskSample sends the
reply.
If you run the applications, you will notice a certain delay in messages being passed back and forth.
This has to do with the SleepTime value (selected in the SleepTime menu), which is passed to theWaitNextEvent()
call under MultiFinder. In the two sample programs, the A/ROSE PrepSend()
and Receive()
services are called only once at each tour through the event loop.
Depending on the SleepTime value, the background application more or less slows down, and this
explains the delay observed on the screen.
SHOWTASKS
ShowTasks is a tool that shows all the A/ROSE tasks that are "visible" in the machine (there might
be "invisible" A/ROSE tasks, too). It goes through all sixteen NuBus slots, looks for all visible tasks,
and displays them by task identifier, object name, and object type. Sample output of this program
might look like the following (which reflects the situation where TaskSample and ClientAppli are
running):
slot = $0 :
00000003: name "echo manager", type "echo manager"
00000004: name "myTaskName", type "myTaskType"
00000005: name "ClientApp", type "ClientType"
This indicates that there are no A/ROSE tasks running on a NuBus card at this time; slot $0
represents the good old main logic board where the A/ROSE Prep driver does its best to make us
believe that there is an instance of A/ROSE.
Now let's look at some of the source code. For the sake of clarity in the following fragments, error
handling is completely suppressed. Needless to say, nobody should ever try to compile this sort of
code! The source code on the CD gives a more realistic idea of A/ROSE programming.
Here are the outlines of main()
and the two basic subroutines AskICCM()
and NameLookup()
,
with explanatory text following the code:
static tid_type cards[16]; // Place for 4 bytes per slot.
main()
{
short slot, index;
tid_type tid;
(void) OpenQueue(nil); // Set up a message queue for me.
AskICCM();
// Request Name Manager TID for each slot, store in cards[].
for (slot=0; slot<16; slot++) {
if (cards[slot] > 0) { // Name Manager TID is OK.
printf("\nSlot = $%X :\n",slot);
index = 0;
while (tid = Lookup_Task("=", "=",
cards[slot], &index))
// Ask Name Manager for info about registered tasks.
NameLookup(cards[slot], tid);
}
}
CloseQueue(); // Be nice with A/ROSE Prep.
} // End main().
void AskICCM()
{
mMessage *m;
m = GetMsg();
m->mTo = GetICCTID();
m->mCode = ICC_GETCARDS;
m->mDataPtr = (char *) cards;
m->mDataSize = sizeof (tid_type) * 16;
Send(m);
m = Receive(OS_MATCH_ALL, OS_MATCH_ALL, ICC_GETCARDS+1,
OS_NO_TIMEOUT, 0);
// SlotInfo is now in cards[0..15] (if nothing failed!).
FreeMsg(m);
} // End AskICCM().
#define bufferSize 512
void NameLookup(tid_type ntid, tid_type tid)
// ntid = Name Manager TID.
// tid = ID of the task for which we request the name.
{
struct pb_lookup_name *lnam_ptr; // (See text.)
char buffer[bufferSize];
mMessage *m;
m = GetMsg();
m->mTo = ntid;
m->mCode = NM_LOOKUP_NAME;
m->mDataPtr = buffer;
m->mDataSize = bufferSize;
lnam_ptr = (pb_lookup_name *) &buffer;
lnam_ptr->lnm_index = 0;
lnam_ptr->lnm_tid = tid;
lnam_ptr->lnm_RAsize = bufferSize -
(sizeof(pb_lookup_name) - sizeof(ra_lnm));
Send(m);
m = Receive (OS_MATCH_ALL, OS_MATCH_ALL, NM_LOOKUP_NAME+1,
OS_NO_TIMEOUT, nil);
DisplayTaskInfo(lnam_ptr); // Lots of silly string handling.
FreeMsg(m);
} // End NameLookup().
The OpenQueue()
call is needed to make use of the A/ROSE
Prep services; it takes a procedure pointer as parameter. If a
procedure is specified, it gets called repeatedly during a blockingReceive()
request,
which avoids blocking the machine during
waiting. In our case, we don't use blocking receives, and don't need
this feature. By the way, OpenQueue()
returns a task identifier
that will be ours for the rest of the process.
We have to deal with the InterCard Communication Manager, inAskICCM()
,
and the Name Manager, indirectly inLookup_Task()
and directly in NameLookup()
.
First, we want to ask the InterCard Communication Manager what it knows about
the sixteen NuBus slots. Naturally, we send a message.The local variable m
is declared as a pointer to the struct
mMessage
(note the spelling, in order to distinguish it from themessage
field in an EventRecord). The GetMsg()
call, one of
the ten A/ROSE primitives, is in this case an A/ROSE Prep service.GetMsg()
returns a pointer to this message structure, which has
already been partially initialized: mId
is a statistically unique
identification number for the particular message, and mFrom
has
already been filled in with the sender's task ID (the number returned
by OpenQueue()
, or by the utility GetTID())
.
We need to identify the addressee in the mTo
field and we need to
specify mCode = ICC_GETCARDS
(this constant is defined in the
include file managers.h) in order to request information about
Macintosh Coprocessor Platform cards in the machine. On receiving
a message with this mCode
, the ICCM expects in mDataPtr
a
pointer to 64 bytes (according to mDataSize)
, and fills the array
cards[0..15] of tid_type for each slot with a value
- <0, if there is no Macintosh Coprocessor Platform card at all, or no ICCM
- =0, if there is an ICCM but no Namer Manager
- >0, if there is an ICCM and a Name Manager; the value is the Name Manager's TID
The rest is easy: For each Name Manager TID, a repeated call toLookup_Task()
returns successively all identifiers of tasks that
registered correctly with the Name Manager. The variable index
is initially set to zero and then passed by address; it is an internal
value that must be passed back to A/ROSE unchanged on subsequent
calls to Lookup_Task()
. This call is an example of an A/ROSE
utility call, which hides the underlying mechanism of sending a
message with a specific mCode
and mDataPtr
to a manager, and
getting the result back through a Receive()
call.
Sending a message now to the Name Manager in the current slot
with mCode = NM_LOOKUP_NAME
and with mDataPtr
pointing to an appropriate buffer, brings back the object name and
type name of the task, which is finally displayed.
BUILDING A DOWNLOAD FILE
To download code to a NuBus card, you have to build a code
resource. I will reproduce and discuss the required code (file
osmain.c) in a simplified form here.
main ()
{
struct ST_PB stpb, *pb; // Start parameter block.
// Init OS with cMaxMsg messages and cStackOS stack.
osinit (cMaxMsg, cOSStack);
pb = &stpb;
StartNameServer(pb); // The Name Manager.
StartICCManager(pb); // You guess it!
StartmyTask(pb); // And our sample task.
// Start all other required managers and tasks.
// Start operating system.
osstart (TICK_MIN_MAJ, TICKS_PS);
// Should never get here!
} // Main().
void StartmyTask(struct ST_PB *pb)
{
pb->CodeSegment = 0;
pb->DataSegment = 0;
pb->StartParmSegment = 0;
pb->InitRegs.A_Registers [5] = GetgCommon() -> gInitA5;
pb->ParentTID = GetTID();
pb->stack = 4096;
pb->heap = 0;
pb->priority = 10;
pb->InitRegs.PC = myTask; // Entry point of myTask.
if (StartTask (pb) == 0) // If the task does not get started,
illegal (); // go debugging.
}
The routines StartNameServer(pb)
andStartICCManager(pb)
are quite similar (except for slight
variations in some parameters and the priority level) toStartmyTask(pb)
.
So this is the code that will be downloaded to the card. The calls osinit()
and osstart()
are only meaningful in this context of
an initial load of the card. The first call takes two parameters whose
default values are cMaxMsg = 500
(maximum number of available
message buffers) and cOSStack = 4096
(size of OS stack). In
many cases, the cMaxMsg
value in particular can be safely
diminished, which allows optimization of memory usage on the card.
The second call, osstart (TICK_MIN_MAJ, TICKS_PS)
launches A/ROSE, with default values for the number of time-slicing
ticks per second, and for a subdivision of major ticks into minor ticks.
In between these two calls, all other required tasks need to be
initialized by means of their start parameter block; the required
minimum consists of the Name Manager task (the linker finds its
code under the name name_server
in the library OS.o), and the
InterCard Communication Manager task (again, its code is in OS.o).
In our example, we added our own myTask
, whose source code file
is compiled separately (compare this with the routine TaskProcessing()
in the TaskSample program):
static char my_object_name [] = "myTaskName";
static char my_type_name [] = "myTaskType";
void myTask()
// All it does at this point is to register with the Name Manager
// (in order to be recognized by possible clients looking for it)
// and then just send back the messages it receives.
{
mMessage *m;
if (!Register_Task (my_object_name, my_type_name,
Machine_Visible))
illegal (); // Go debugging: something mysterious happened.
while(1) // Forever !
{
m = Receive(OS_MATCH_ALL, OS_MATCH_ALL, OS_MATCH_ALL,
OS_NO_TIMEOUT);
if (m) {
if (m->mStatus != 0) {
// What happened? A real program would investigate but
// we'll just get rid of it here
FreeMsg(m);
}
else {
switch (m->mCode) {
case DUMMYCODE:
// Is there something to do with this message?
break;
// Handle here all the message codes you specified in your
// design.
default:
m->mCode |= 0x8000; // Unrecognized message code;
m->mStatus = OS_UNKNOWN_MESSAGE;
// defined in "managers.h."
break;
} // Switch.
// Send message back.
SwapTID(m); // Swap mFrom and mTo fields.
m->mCode++; // Response is one greater, by convention.
Send (m);
} // Message status was OK.
} // There was a message.
} // While.
} // End myTask().
Finally, we need to put everything together. The following MPW
shell commands do the trick. Adopting the convention on the
A/ROSE distribution disks, we'll use the filename Start for what we
will download. I recommend defining the MPW shell variables AROSE
, AROSEBin
,
and AROSEIncl
in a UserStartup file, which
holds the corresponding folder pathnames.
C osmain.c -i "{AROSEIncl}"
C myTask.c -i "{AROSEIncl}"
Link -t 'DMRP' -c 'RWM ' ð
-o start ð
osmain.c.o ð
myTask.c.o ð
"{AROSEBin}"OS.o ð
"{AROSEBin}"osglue.o
Finally, we need to download the file to the available Macintosh
Coprocessor Platform cards in your machine, by means of the
Download tool:
"{AROSE}Downloader:Download" start
The tool should reply with
Segment of size 00000040 is downloaded
Segment of size 00000054 is downloaded
Segment of size 00004D44 is downloaded
Now it's time to come back to our tool ShowTasks:
showtasks
slot = $0 :
$00000003: name "echo manager", type "echo manager"
slot = $D :
$0D000001: name "name manager", type "name manager"
$0D000003: name "myTaskName", type "myTaskType"
. . . and to go ahead and send messages to myTask on a Macintosh
Coprocessor Platform card. To try this, launch ClientAppli again, but
this time without the TaskSample application running. ClientAppli
now finds the "server" named
myTaskName
in its slot, and
messages sent to it are returned as expected.
A MANDELBROT SETS EXERCISE
The downloaded file in the Macintosh Coprocessor Platform card
works, but it gets boring fast: our server task is quite lazy, and doesn't
do anything besides echoing our messages. For a more interesting
programming exercise, open the MCPMB folder. The program you'll
find there computes Mandelbrot (MB) sets in parallel processing,
involving as many Macintosh Coprocessor Platform cards as you can
put into your machine.
For each line to be computed, a message is sent to an MBTask,
carrying along the required parameters and a pointer to where the
line fits into the bitmap. The MBTask allocates a local buffer each
time (for pedagogical reasons, I didn't optimize the design too much)
and sends the computed data back over the NuBus by means of a
NetCopy() call. Have a look at the source code on the CD, play with
it, and let me know what you did to improve on the error handling
and some other flaws in it.
THAT'S ALL, FOLKS . . .
This article has given you an idea of how to find your way around the
Macintosh Coprocessor Platform and A/ROSE. You now know
something about the origins, architecture, and implementation of this
generic hardware and software foundation, and have seen some
samples of A/ROSE programming. If you want to go on from here,
the APDAlog ® contains everything you need to know to order the
complete Macintosh Coprocessor Platform Developer's Kit, or the
A/ROSE Software Kit, or just the Macintosh Coprocessor Platform
Developer's Guide.
TOKENTALK AND A/ROSE
by Anumele Raja
TokenTalk® is a typical application that runs under A/ROSE on an intelligent NuBus card. The following is a
brief description of the TokenTalk hardware and how TokenTalk uses A/ROSE.
THE HARDWARE AND SOFTWARE
The TokenTalk NB card is the intelligent NuBus card that implements the Token Ring interface. The card
consists of a 68000 processor and a Token Ring interface chip set made by Texas Instruments. The card's
foundation is the Macintosh Coprocessor Platform. Besides TokenTalk, the card can also run MacAPPCTM,
MacDFT®, and MacSMB file transfer programs.
The Token Ring interface chip set is controlled by a program called Logical Link Control (LLC) that also
implements the Token Ring protocol. LLC runs as a task under A/ROSE.
TokenTalk itself is an A/ROSE task that serves as the interface between programs running on the Macintosh
and the LLC task. This task can be replaced by another task to implement other protocols like SNA.
THE DOWNLOAD PROCESS
When the user selects the TokenTalk device on the Network control panel, a resource file called TokenTalk
Prep is loaded into the Macintosh and executed. TokenTalk Prep first finds a TokenTalk card, downloads
A/ROSE if it is not already running on the card, and downloads the LLC task onto the card. The TokenTalk
part of the AppleTalk device driver downloads the card part of the TokenTalk task by using TokenTalk Prep
utilities, and starts the task.
Sound complicated? Let's take the operation sequence at a slower pace.
On the Macintosh side of TokenTalk, the operation sequence is as follows:
- The user selects TokenTalk on the Network
control panel.
- The TokenTalk Prep file is loaded and started.
- TokenTalk Prep makes sure that A/ROSE Prep is running on the Macintosh, and searches for a TokenTalk card
by calling the
NewFindcard()
routine. If a TokenTalk card is found, TokenTalk Prep checks to see if
A/ROSE is already running on that card by looking for a Name Manager. If A/ROSE is not running,
TokenTalk Prep downloads onto the card a version of A/ROSE that includes the Name Manager, the
InterCard Communication Manager, the Remote System Manager, and the Echo Manager. It then does aLookup_Task()
to find the LLC task. This task controls the Token Ring interface chip set and handles
interrupts. If the LLC task is not found, it downloads the LLC task using theDynamicDownload()
call.
- The Network CDev then talks to the TokenTalk driver to activate TokenTalk. The TokenTalk driver, which
resides on the Macintosh, is the interface between AppleTalk and the TokenTalk card. It operates by sending
control commands and data to the TokenTalk task on the card and receiving status information and data.
- The TokenTalk driver downloads the TokenTalk task to the NuBus card.
On the TokenTalk card side of TokenTalk, the operation sequence is as follows:
- When the LLC task is started up, it registers itself with the object name LLC and the type name TokenTalk
NB by calling the
Register_Task()
routine. It then calls the Receive()
primitive and waits for
messages. In the current implementation, the Receive()
is issued with a timeout parameter. The LLC
task runs under run-to-block mode.
- When the TokenTalk task is started up, it registers itself with the object name Token Talk 1 and the type
name TokenTalk NB by calling the
Register_Task()
routine. The TokenTalk task searches for the LLC
task by doing a Lookup_Task()
. It then waits for messages from the Macintosh to start an operation.
When requests are received from the Macintosh, the TokenTalk task sends commands to the LLC task to
carry out the various operations. The TokenTalk task does not control any hardware by itself.
- Data is transferred from the Token Ring interface to the memory and vice versa by a direct memory
access (DMA) mechanism built into the interface chip set. An interrupt is generated by the DMA device at
the completion of a data transfer.
Both the LLC task and the TokenTalk task run with a priority of 30 and allocate a stack of 2048 bytes. No
heap space is allocated by these tasks. TokenTalk Prep uses the start parameter block to pass information to
the LLC task. This information specifies the TokenTalk address for the node.
UTILITIES THAT MAKE IT EASIER TO DEVELOP A/ROSE PROGRAMS ON THE MACINTOSH
by Anumele Raja
The following Macintosh utilities, included on the A/ROSE distribution disks, facilitate development of
A/ROSE programs on the Macintosh:
Print Manager (nprm) is a Macintosh application that enables users to display information from a task
running on a Macintosh Coprocessor Platform card. It registers itself with A/ROSE by the object name Print
Manager and the type name Print Manager. Strings to be printed are sent to the Print Manager by the printf
routine supplied with the A/ROSE release. The first time printf is called, it looks for the Print Manager and
finds its Task ID. Subsequently, it sends all print strings as messages to the Print Manager, which puts up a
window and displays the strings it receives. Users can display diagnostic messages using printf. Print
Manager features can also be implemented in a user's program.
Dumpcard is an MPW tool that dumps the status of A/ROSE tasks running on any NuBus card. Available
options display the memory blocks, the messages waiting to be received by a task, and the task control
blocks of all tasks running under A/ROSE. If the card stops for any reason, like a bus error, the user can get
a trace of the stack to find the calling sequence that caused the exception. In addition, the user can request
disassembly of instructions around the break point.
Download application (ndld) is a Macintosh application
used to download A/ROSE and/or A/ROSE
tasks onto a specified card or cards either statically or dynamically. The file selection is done through a
standard Get File dialog box.
Download is an MPW tool that downloads A/ROSE and/or A/ROSE tasks onto a specified card or cards
either statically or dynamically. It is useful when the user wishes to download code from a shell script.
NuBug is a debugging application used to debug A/ROSE programs running on a card. NuBug looks and
works like MacsBug. All MacsBug commands that are not specific to the Mac OS are supported by NuBug.
In addition, NuBug provides commands specific to A/ROSE, dealing with task status, task names, and such.
NuBug is a multiwindow application that brings up as many windows as there are NuBus cards capable of
running A/ROSE. Because NuBug is implemented in C++, it can be enhanced very easily.
Users can look forward to a new, we hope official, version of NuBug very soon. The current release of
NuBug has not been tested formally and is not supported by Apple. Still, programmers find it so
helpful that they don't seem to mind if they encounter
a few glitches.
JOSEPH MAURER, an 18-month Apple veteran, studied mathematics and theoretical physics at universities in Munich and
Nice. Since then he's led a varied but somewhat theoretical life, which has included being a ballet school piano player,
bicycle racer (champion of lower Bavaria!), math researcher, mountain climber, university professor, and Apple European
technical support and training guru (you can decide for yourself if that one's theoretical or not). All in all, Joseph is
basically a man of numbers: he has one wife, two Macintosh computers, three bicycles, and four children. He says he
wants more Macs and more racing bikes (the ones he has are "slowing down"), but refuses to comment on wanting more
wives and/or children. *
Thanks to Our Technical ReviewersNed Buratovich, Dave Comstock, Chris Deighan, Ken Siemers, Steve Wang.
A special thanks to Anumele D. Raja. *