EEE8087 4W Rev. 1.5
Real Time Embedded Systems
Worksheet 4. The Time-Slicing Structure
This week we start work on the central components of an elementary real-time operating system -
we will call it a 'runtime system' - that divides the processor's time between separate user tasks.
These tasks will need to communicate with the system, and will request its services by means of a
'software interrupt'.
Implementation of Software Interrupts
A software interrupt is known as a 'trap'. It causes the processor to respond in a very similar way as
it does to a hardware interrupt, and so allows system calls from the user program and interrupts
from hardware devices to enter the operating system in a consistent way.
There are 16 trap instructions available, numbered 0 to 15, and written
trap #0
...
trap #15
Each of the 16 trap instructions may have its own interrupt service routine (ISR). After pushing the
PC and SR, the processor then accesses a table in low memory, at address 80H. As for the
hardware interrupts, the table contains a 4-byte value corresponding to the address of the ISR for
each software interrupt. Vectors for the two types of interrupt will normally be combined into a single
block of code.
;interrupt vectors
org $64 ;origin 64H
hvec1 dc.l hisr1 ;address of hardware ISR 1
hvec2 dc.l hisr2 ; ... etc
org $80 ;origin 80H
svec0 dc.l sisr0 ;address of software ISR 0
svec1 dc.l sisr1 ; ... etc
Controlling Interrupts
There is, however, an important difference between hardware and software interrupts. Hardware
interrupts are in order of priority, with 7 being the highest priority and 1 the lowest. If two hardware
interrupts occur at the same time, then the one at the higher priority will be accepted and the other
one will be kept waiting until the first ISR has completed. If a hardware interrupt occurs shortly after
another one, but while the ISR for the first interrupt is still in execution, then the processor will again
compare the priorities of the two interrupts. If the new interrupt is of a higher priority, then it will
interrupt the lower priority ISR. If the new interrupt is at a lower priority than the currently executing
ISR, it will be kept waiting until that ISR completes.
Software interrupts do not behave in an analogous way. Since the processor can only execute one
instruction at a time, it would be impossible for two software interrupts to occur at the same time,
and unless a programmer includes a trap instruction within an ISR, there will also be no occasions
on which a trap takes place during the processing of another trap. There is therefore no point in
prioritising the software interrupts, and all 16 are at the same priority. There is, however, the
question of the relative priority of the hardware and software interrupts. What if a hardware interrupt
is raised at the same time as the processor is executing a software interrupt instruction? This is
EEE8087 4W Rev. 1.5
handled by assigning all the software interrupts to priority level 0. Processing of a software interrupt
is therefore interruptible by a hardware interrupt at any of the priority levels 1 to 7.
Within your system, however, regardless of the type of interrupt being processed, you will want to
prevent the acceptance of any other interrupt. Your system will therefore be completely
uninterruptible. Once entered, it will always run to completion and then return to the user task that
was running when the interrupt was raised. You will therefore need to disable interrupt acceptance,
the procedure for which is explained now.
Using the simulator, examine the 16-bit status register. Bits 8, 9 and 10 (labelled 'INT') hold a 3-bit
value that represents the interrupt priority mask. When an interrupt is accepted, the mask is set to
the priority level of that interrupt. A hardware interrupt will only be accepted if its priority is greater
than the current setting in the mask. Normally, the mask is set to 000 (decimal 0) thereby allowing
the acceptance of any hardware interrupt. However, it will remain at zero during its response to a
software interrupt, since that is the priority of these interrupts, and will thereby allow the hardware to
interrupt the software ISR. If you wish to prevent this, then the following instruction, placed at the
very start of a software ISR, sets the mask to binary 111 (decimal 7). Any hardware interrupts will
now be disabled, and held pending until the mask is returned to zero.
or #$0700,sr ;disable hardware interrupts
The status register will have been automatically saved on the stack at the start of the interrupt
servicing. On execution of the 'return from exception' instruction (RTE), it will be restored, and the
mask reset to the zero value that it held previously, thereby allowing the acceptance of any
hardware interrupt that might have been raised in the meantime and is currently pending.
If you want to enable hardware interrupts at any other time, the following instruction will set the
mask to zero.
and #$f8ff,sr ;enable hardware interrupts*
EEE8087 4W Rev. 1.5
Practical Work
Assessment question
Work in groups of three on this question. Your submission should include the following.
Your software, including the run-time system and the test programmes you used to
demonstrate it. Submit the source code, not the assembler output listing.
Documentation: a .PDF file is preferable, otherwise .DOC.
An individual 5 - 10 minute video presentation from each member of the group, explaining
parts the system and how they were tested. Each presentation should deal with specific
aspects, and the three presentations together should cover the entire system.
The names and student numbers of all three group members should be shown on the software
heading and on the front page of the documentation. On the video presentations the name of the
group member should be clearly announced or displayed as a caption.
There are therefore five items to be submitted: a single software file, the documentation, and three
presentations. These items should be placed into a single zipped file, and uploaded to a Canvas
submission point to be advised. The submission deadline is 2pm on Friday 19th January, 2024.
Each item will now be described in detail.
The run-time system
The work consists of writing a basic time-slicing system, along the lines of the one discussed in the
lecture. It should allow the execution of several concurrent user tasks, with support for task
scheduling and inter-task communication. An outline programme is provided on Canvas, but you will
write the service routines (including reset), the scheduler, and the user tasks for each application.
The user tasks are located in memory, above the system itself. Each task has its own area of
memory, with the programme code at the lowest address, data above it, and top-of-stack at the next
address above this task's memory area. For example, the following task occupies memory between
2000H and 2FFFH. It has its code at 2000H, data at 2C00H, and stack at the top of the data area.
Address
2000H Programme code
3000H Top-of-stack
The system runs in the foreground, and is entered following either a timer interrupt, a software
interrupt from one of the tasks requesting service, or another hardware interrupt.
The following system calls should be supported by means of software interrupts. They can either
each be allocated to a separate trap number, or (as in the demonstration system) they can all be
called on the same trap, with one of the registers used to hold a value identifying the requested
function. Some of the calls also require additional parameters in other registers.
EEE8087 4W Rev. 1.5
1. Create task
Function: A currently unused TCB is marked as in use and set up for a new task.
It is placed on the ready list. The requesting task remains on the ready list. Two
parameters indicate the start and end of the memory area occupied by the new task
and its data.
Parameters: The start address of the new task,
The address of its top-of-stack.
2. Delete task
Function: The requesting task is terminated, its TCB is removed from the list and
marked as unused. Any memory allocated to it is returned to the system.
Parameters: None.
3. Wait mutex
Function: If the mutex variable is one, it is set to zero and the requesting task is placed
back onto the ready list. If the mutex is zero, the task is placed onto the wait
list, and subsequently transferred back to the ready list when another task
executes a signal mutex.
Parameters: None.
4. Signal mutex
Function: If the mutex variable is zero, and a task is waiting on the mutex, then that task
is transferred to the ready list and the mutex remains at zero. If the mutex is zero and
no task is waiting, the mutex is set to one. In either case, the requesting task remains
on the ready list.
Parameters: None.
5. Initialise mutex
Function: The mutex is set to the value 0 or 1, as specified in the parameter.
Parameters: 0 or 1.
6. Wait time
Function: The requesting task is placed onto the wait list until the passage of the
number of timer interrupts specified in the parameter, when it is transferred
back to the ready list.
Parameters: Number of timer intervals to wait.
The following functions are more difficult. Good results may be obtained without implementing these
functions, but it would be expected that they will be included in the best submissions. Function 7
will require the use of an additional interrupt at level 2, and you will need to consider the implications
of this for the correct working of the system. Function 8 will require some thought about the internal
record-keeping relating to which areas of memory are in use. You will need to devise your own test
programmes for these functions.
7. Wait I/O:
Function: The requesting task is placed onto the wait list, until an interrupt signifies
completion of an I/O operation, at which time the task is transferred back to the ready
list.
Parameters: None.
8. Allocate memory
Function: For tasks that require a large amount of memory, it is more efficient to allocate it as
required when the task runs. A large area of memory is therefore kept free within the
system, and a block from it, of say 16 kbytes, is returned to the requesting task. If the
request is satisfied, the requesting task remains on the ready list. If there is
insufficient free memory available, then the requesting task is put on the wait list
until memory is returned when another task terminates.
Parameters: On return, the start address of the allocated memory is held in the parameter register.
EEE8087 4W Rev. 1.5
An additional function is executed automatically at start-up, or if the user presses the reset button.
System reset
Function: The system is initialised: all internal variables are reset, and each TCB is
marked as unused. A TCB for task T0 is then created, and T0 becomes
the running task.
The system assumes that a default user task, T0, is present. The system runs this task immediately
after a reset. It will need to be located at a predetermined address, which will be coded into the
reset function.
Your system should be robust, and deal with errors in an intelligent way. For example, what if the
user tries to create more tasks than there are available TCBs?
Test programmes
Test your system using the following programmes. These are modified versions of the questions in
last week's worksheet, this time running under your RTS.
1. Testing create task and wait time functions.
A stopwatch counts in seconds. It starts when button 0 is pressed and stops when the button is
pressed again. It is programmed as follows.
Timer interrupts are set to 100ms. Task 0 starts task 1, and both tasks run concurrently. A shared
variable running is held in a memory location, and is set by T1 and read by T0. T0 displays a 2-
digit value on the 7-segment display, initialised to zero. If running is set, T0 increments the
display, then waits for 10 time intervals, and then repeats. If running is not set, T0 does nothing.
T1 tests pushbutton 0. Each time the button is pressed, running changes state.
2. Testing initialise mutex, wait mutex, and signal mutex functions.
On a gaming device, two players each have a button that fires bullets. Internal counters a and b
record the number of bullets fired by each player, and a third counter c records the total number of
bullets fired. It would be difficult to programme this application on the simulator because it is not
possible to press two buttons at once. This would often happen in reality, and dealing with it
represents the most difficult technical challenge in the programming. However, we can focus on this
specific problem by programming two tasks, one for each player, and having each task fire bullets
continuously.
Timer interrupts are set to 100ms. Task 0 starts task1, and both tasks run concurrently. A shared
variable c is held in a memory location (not a register) and is initialised to zero. Task 0 has a
variable a, also in memory, and runs in a loop that continuously increments a and c. Task 1 has a
variable b in memory, and runs in a loop that continuously increments b and c.
To check that the system is working, one of the tasks calculates a + b - c, which should be zero,
and displays it on the 7-segment display.
3. A test of your own to check the delete task function. Also tests of the wait I/O and allocate
memory functions, if you have implemented them.
Your test programmes should be included at the end of the RTS code. Place all the tests together.
An individual one can then be selected by commenting out the others.
EEE8087 4W Rev. 1.5
Documentation
This consists of a user manual. It will explain your system to a user, and will therefore focus on what
it does and how to use it. It will include a brief overview of how the system works internally, but only
to an extent that is required for the programmer to use the system correctly. It should be structured
as follows.
1.
A general description, including, for example:
explanation of the principles of multitasking and time-slicing, and how they are implemented;
description of the memory layout of the user tasks as shown above;
explanation of the startup behaviour: a default task runs, which may then start other tasks.
2.
A description of each of the user functions and their parameters. Explain all aspects of the
behaviour of each function that are of interest to the user. For example, when a new task is created,
does it run immediately, or is it put at the end of the ready queue? What if two tasks are waiting for
the same time, and so become ready together? Give some emphasis to the behaviour of any of the
more advanced features that you may have implemented. For example, if you have implemented
function 7, you should explain how the different interrupt priorities have been handled internally, and
how this affects the user. Your descriptions should also state how long each function takes to
execute. This time should be quoted in terms of instructions, and may be within a specified range,
e.g., a particular function might execute in 30 - 50 instructions.
3.
A description of any other relevant aspects of system behaviour. For example, have you arranged
for prioritised scheduling? If so, how does it behave? If not, then are all tasks that are ready likely to
receive a similar amount of run time? Is there any possibility that a task may receive too little time,
or none at all? State the timer interrupt period, and how long (in terms of instructions) the system
takes to switch tasks. Using a reasonable average for the instruction execution time, calculate the
proportion of time that is spent in the RTS, and how much is available for running the user tasks.
With normal typeface and spacing (such as used here) it would be reasonable to expect a length of
no more than five or six pages. Submission in .PDF format is preferred, but .DOC(X) is also
acceptable.
Individual presentation
Each of the three group members should include an individual video presentation explaining
particular parts of the system, how they worked, and how they were tested. These will probably be
the functions with which that member was most closely involved in writing. Between the three
presentations, the entire system should be explained.
These presentations will complement the written material in your documentation. Instead of
describing how to use the system, they will explain in more depth how it works internally. They do
not need to be long and detailed, but should focus on the underlying working principles of the
software. It is unlikely to be necessary to talk about individual machine instructions, unless you are
discussing a relevant concept such as mutual exclusion. You should also explain how each part of
the system was tested, and how the test results verify its correctness.
Each presentation should last 5 - 10 minutes.
EEE8087 4W Rev. 1.5
The Demonstration System
A system was demonstrated during the lecture. It recognises a hardware interrupt at level 1 from the
timer. It also allows system calls by means of software interrupts, all of which have been allocated to
trap 0. These system calls are programmed by placing a value that identifies the requested function
into data register 0, and any other parameters as required by each of the individual functions in
registers D1 onwards. For example, system call 1 is used to create a new task. Suppose that this
new task is called T1, and that its top-of-stack is to be located at address 6000H. It would be
programmed as follows.
move.l #1,d0 ;set id in d0
move.l #t1,d1 ;set address of new task in d1
move.l #$6000,d2 ;set stack address in d2
trap #0 ;call system
The main data structure used in this system is a list of task control blocks (TCBs). Each TCB
represents the state of one of the tasks. It contains a copy of all that task's registers, together with
some items of control information including a flag that indicates whether the TCB is in use.
At any time, each of the tasks will be in one of three states: the currently running task, ready to run
when its turn comes up, or unable to run because it is waiting for the occurrence of some event,
which could be a signal operation on a mutex, the expiry of a time interval, or an I/O interrupt.
At initialisation, all the TCBs in the list are marked as unused. As each new task is started, one of
the unused TCBs is allocated for it and marked as used. These TCBs are organised into two linked
lists, in which each element contains a pointer to the next element. These lists are called 'ready' and
'waiting'. Two more data items consist of pointers to the first element in each list. The pointer
rdytcb holds the address of the first element in the list of ready TCBs. The first element in this list
is the task that actually is running. The linkage in this list is circular, that is, the last entry points back
to the first, so making it easy to access each TCB in rotation. The pointer wttcb holds the address
of the first element in the list of TCBs that are waiting. Each element will contain an indication of the
event the task is waiting for. There is no need to access elements of this list in rotation, so the last
element has its pointer set to zero.
N N N N N 0
The example above shows a list of 8 elements, each of which has a pointer, labelled N, to the next
element. Elements 4, 6, 0 and 1 are on the ready list, with element 4 being the TCB for the running
task. Elements 3 and 7 are on the waiting list, and elements 2 and 5 are unused.
0 1 2 3 4 5 6 7
rdytcb
wttcb
EEE8087 4W Rev. 1.5
The system is organised into the following sections. Some of these sections, shown in italics, are
provided for you in the outline code on Canvas. These can be used in your own systems:
unchanged, modified, or rewritten as you wish.
Data definitions and equates
org 0
Interrupt vectors
Executable code
System reset
First-level interrupt handler entry
FLIH Service routines
Scheduler
Dispatcher
Data storage
Default user task T0
Data definitions
Each TCB represents the state of one of the current tasks, and is defined as follows.
tcb org 0 ;tcb record
tcbd0 ds.l 1 ; D register save
tcbd1 ds.l 1
tcbd2 ds.l 1
tcbd3 ds.l 1
tcbd4 ds.l 1
tcbd5 ds.l 1
tcbd6 ds.l 1
tcbd7 ds.l 1
tcba0 ds.l 1 ; A register save
tcba1 ds.l 1
tcba2 ds.l 1
tcba3 ds.l 1
tcba4 ds.l 1
tcba5 ds.l 1
tcba6 ds.l 1
tcba7 ds.l 1
tcbsr ds.l 1 ; SR (status reg) save
tcbpc ds.l 1 ; PC save
tcbnext ds.l 1 ; link to next record
tcbused ds.l 1 ; record in use flag
ds.l 1 ; other fields as required
ds.l 1 ;
tcblen equ * ; length of tcb record in bytes
Data storage
Storage for a list of TCBs is defined as in the first line below. The constant ntcb represents the
number of TCBs in the list, and should be set up as an equate. Other variables are described
throughout these notes.
tcblst ds.b tcblen*ntcb ;tcb list (length x no of tcbs)
rdytcb ds.l 1 ;^ ready tcb list
wttcb ds.l 1 ;^ waiting tcb list
a0sav ds.l 1 ;A0 temporary save
d0sav ds.l 1 ;D0 temporary save
id ds.l 1 ;function id
EEE8087 4W Rev. 1.5
Interrupt vectors
The interrupt vectors are addresses of the code that will be executed as a result of an interrupt. The
following three addresses are defined.
Address res is the location of the routine to which the processor branches when the it responds to
a hardware reset. Address fltint is the location to which the processor branches following a timer
interrupt at level 1, and flsint following a software interrupt. The address stk is the value that is
loaded into the stack pointer following a hardware reset.
dc.l stk ; initial SP
dc.l res ; reset
org $64
dc.l fltint ; interrupt 1 (timer)
org $80
dc.l flsint ; trap 0 (system call)
Executable Code
First-level interrupt handler
The first-level interrupt handler (FLIH) contains the code that services an interrupt. For convenience
it is split into the common FLIH entry section that is executed immediately following an interrupt, and
the FLIH service routines that carry out the processing specific to each type of interrupt.
FLIH entry
Hardware interrupts at level 1 are directed by the interrupt vector to enter the FLIH at fltint, while
software interrupts arrive at flsint. The FLIH performs three main functions.
It takes the pointer to the TCB of the currently executing task, stored at rdytcb, and saves the
values of the registers, including the PC and SR, within that TCB.
It also sets a value within a storage location, known as id, that identifies the source of the interrupt.
If an interrupt has been raised by the hardware timer, then id is set to 0. For a software interrupt,
id is set to the value, from 1 onwards, of the system call function number. The id will subsequently
be used to select the corresponding service routine for processing this interrupt.
Programming the above two operations requires particular care, because saving the value of the
user's registers as they were at the time of the interrupt requires the use of certain registers itself.
Registers D0 and A0 are in use for this purpose. These registers are therefore saved in temporary
locations, before being transferred to their long-term holding locations within the TCB.
The other function performed by the FLIH is to disable interrupts, if this has not already happened. A
level-1 hardware interrupt from the timer will have set the interrupt priority mask to 1, thereby
preventing any further interrupts. A software interrupt will have left the mask at 0, which would allow
the timer device to interrupt the processing of the software interrupt. Therefore the first action taken
at the software interrupt entry point is to disable hardware interrupts by setting the mask to 7.
EEE8087 4W Rev. 1.5
fltint ;ENTRY FROM TIMER INTERRUPT
move.l d0,d0sav ;save D0
move.l #$0,d0 ;set id = 0
move.l d0,id
move.l d0sav,d0 ;restore D0
bra fl1
flsint ;ENTRY FROM TRAP (SOFTWARE INTERRUPT)
or #%0000011100000000,sr ;disable hardware interrupts
move.l d0,id ;store id
bra fl1
fl1 move.l a0,a0sav ;save working reg
move.l rdytcb,a0 ;A0 ^ 1st ready tcb (ie running tcb)
move.l d0,tcbd0(a0) ;store registers
move.l d1,tcbd1(a0)
move.l d2,tcbd2(a0)
move.l d3,tcbd3(a0)
move.l d4,tcbd4(a0)
move.l d5,tcbd5(a0)
move.l d6,tcbd6(a0)
move.l d7,tcbd7(a0)
move.l a0sav,d0
move.l d0,tcba0(a0)
move.l a1,tcba1(a0)
move.l a2,tcba2(a0)
move.l a3,tcba3(a0)
move.l a4,tcba4(a0)
move.l a5,tcba5(a0)
move.l a6,tcba6(a0)
move (sp),d0 ;pop and store SR
add.l #2,sp
move.l d0,tcbsr(a0)
move.l (sp),d0 ;pop and store PC
add.l #4,sp
move.l d0,tcbpc(a0)
move.l a7,tcba7(a0) ;store SP
;START OF SERVICE ROUTINES
FLIH service routines, including system reset
The service routines are arranged as a large switch statement, using id as the case variable. Each
routine carries out one of the functions defined in the specification.
Scheduler
The scheduler examines the ready list, to which rdytcb points to the first element. This is the TCB
of the task that was executing when the system was invoked, and which has just been interrupted.
By following the links, the scheduler can locate each TCB that is currently ready to run. It selects
EEE8087 4W Rev. 1.5
one of these tasks for running, and adjusts the value in rdytcb to point to the TCB for this task.
This TCB will be then used by the dispatcher to resume execution of the task.
The scheduler may make the decision as to which task will run next by doing nothing more than
following the link in the current TCB to the next one in the chain. This will result in each ready task
running in rotation, receiving an approximately equal amount of run time each. Alternatively, it would
be possible to assign a priority to each task as it is created, by adding another parameter to the
'create task' system call. Higher priority tasks would then receive a larger proportion of the available
run time.
Dispatcher
The dispatcher reverses the action taken by the FLIH. Using the newly set value in rdytcb, it
restores the registers of the selected task to the values that were stored when that task was
interrupted. Careful housekeeping is again necessary, as this operation itself requires the use of
registers D0 and A0. The dispatcher finishes by recreating the state of the stack as it was after the
task was interrupted. The processor then uses a 'return from exception' instruction, as though it
were returning from any normal interrupt, to transfer control back to the selected task.
; ;END OF SCHEDULER
move.l rdytcb,a0 ;A0 ^ new running tcb
move.l tcbd1(a0),d1 ;restore registers
move.l tcbd2(a0),d2
move.l tcbd3(a0),d3
move.l tcbd4(a0),d4
move.l tcbd5(a0),d5
move.l tcbd6(a0),d6
move.l tcbd7(a0),d7
move.l tcba1(a0),a1
move.l tcba2(a0),a2
move.l tcba3(a0),a3
move.l tcba4(a0),a4
move.l tcba5(a0),a5
move.l tcba6(a0),a6
move.l tcba7(a0),a7
sub.l #4,sp ;push PC
move.l tcbpc(a0),d0
move.l d0,(sp)
sub.l #2,sp
move.l tcbsr(a0),d0 ;push SR
move d0,(sp)
move.l tcbd0(a0),d0 ;restore remaining registers
move.l tcba0(a0),a0
rte ;return
EEE8087 4W Rev. 1.5
An example of a user programme running under this system is shown here. It consists of two
concurrent tasks. Task T0 calls the system to start task T1, then switches on the RH LED. Task T1
calls the system to wait for 3 timer intervals, then switches on the LH LED. From then on, the two
tasks run alternately. If the timer is set to interrupt at one-second intervals, the result is that the RH
LED lights immediately, then after 3 seconds the two LEDs start alternating.
;system call equates
sys equ 0 ; system call trap (trap 0)
syscr equ 1 ; create new task
sysdel equ 2 ; delete task
syswttm equ 6 ; wait on timer
org usrcode
led equ $e00010 ;led
sw equ $e00014 ;switch
t0: ;TASK 0
move.l #sysscr,d0 ;start task 1
move.l #t1,d1 ; address
move.l #$4000,d2 ; top of stack
trap #sys
;repeat
t00: move.l #$01,d1 ; set led 0
move.b d1,led
bra t00
t1: ;TASK 1
move.l #syswttm,d0 ;wait for 3 clocks
move.l #3,d1
trap #sys
;repeat
t10: move.l #$02,d0 ; set led 1
move.b d0,led
bra t10
END res