Multitask OS  (MOS)


MOS is a simple round-robin scheduling preemptive OS.  I will show how to use interrupt to write a concurrent program supported by a task-switcher.  A process consists of its data structure called Process Control Block (PCB).  PCB contains process's PC, process's SP (and other attributes). Each process has its own Stack. The main part of MOS is its scheduler.  The scheduler has a queue of the pointer of PCB.  

In the following example, I will show how two processes are created and then the operating system starts.
 
process1()
  ...
process2()
  ...

main()
  p = newp()   // create a new process
  p.PC = &process1
  p.SP = newStack()
  enqueue(p)   // put it in the process queue
  p = newp()
  p.PC = &process2
  p.SP = newStack()
  enqueue(p)
  boot()       // start the scheduler


We create two PCBs for process1() and process2() and put them in the process queue.  The scheduler is an Interrupt Service Routine.  It is invoked when an interrupt occurs.  The task of the scheduler is to save the current active process (its PC and SP including local states, all registers) and set up for the next process to run by restoring its PC, SP, registers.  How the actual execution of the real processor instruction occurs will be shown. It required the knowledge of low level instruction of a real processor.

// this is invoked when interrupt occurs
tswitch()
  p = current process
  save current context of p
  p = nextp()                //  get the next one in the process queue
  restore the context of p   // it uses p.SP
  PC = p.PC
  reti                       //   return from int, jump to the current p


That is all for a task-switcher.  When an interrupt occurs, the current process is interrupted. Its context is saved in its Stack.  The next process is selected and its context is restored. Then, finally, jump to that process which make it active until the next interrupt occurs. 

The rest is just the handling of data structure and operations on them that are required.

newp()
  allocate a block of memory for storing PC, SP and other attributes
  return a pointer to this block (called PCB)

newStack()
  allocate a block of memory as the stack for a process
  return a pointer to this block


Next is the operations on the process queue.  The process queue is a fixed size array storing pointers to PCBs.  A global variable "nump" stores the number of process in the queue.  The end of queue is denoted by 0.  The terminated process is denoted by 1. 

nextp()
  from the present queue index
  scan for the next process
  update the queue index
  return the pointer to PCB

terminate()
  terminate the current process
  nump--

enqueue(p)
  put p to the end of the queue
  nump++


The last mysterious thing is boot(). Actually it is rather simple. It works in a part just like the switcher.  It sets up PC and SP then jumps to start the first process in the queue.

boot()
  PC = currentp.PC
  SP = currentp.SP
  jump to PC        // this act needs low level instruction sequence


The remaining question is "What happen when all processes terminate?".  Some how we must check that there is at least one process in the queue (during nextp()). When "nump" is zero the operating system should shut down (the simulation stops).

Data structure

Process control block (PCB)

PCB contains the state of a process.  It has the following fields:

0  Program counter
1  stack pointer
2  frame pointer
3  return address reg
4  return value reg
5  state: active/not-active

Process queue and semaphore

We use singly linked list with header for both. The header has two fields: head, end.  This will make appending a new cell at the end of list easy.  For semaphore wait list, the operations are

append-list(L,a)    append a cell at the end of list
delete-head(L)       delete a cell from the head

We keep the deleted cell in "freelist" and reuse it when a new cell is required.  For process queue, we need a pointer to identify the current process. Also, we made the list a circular list in order to find the next process.  The members of this list are never deleted because deleting an arbitrary element required searching for the previous link.  In order to make a process being "out" of the list, we use a flag in the PCB instead denoting its state: active/not-active.  When we mark the state as not-active, it will be skip when considering the next process. The operations on this list are:

nextp()               find next process on the process list (from the current process)
enqueue(p)          put a new member p, at the end of process list
dequeue()           dequeue the current process by marking its PCB state not-active

To know when to terminate the program, we keep "nump" the number of process in the process list.  enqueue() increases this number.  dequeue() decreases this number.  when nump is zero the simulation will terminate.

Let us makes all the above code concrete by coding in S2 assembly language.

Extended Instructions

These are supporting function for implementing OS.  They are "trap r1 #n" forms.

trap r1 #15              disable interrupt, r1 contains interrupt number
trap r1 #16              enable interrupt, r1 contains interrupt number

These instruction allows manipulation the internal RetAds register (to save/restore return address from interrupt service routine) and save/restore process context.

xch r1                       exchange RetAds with r1      
pushm sp          push multiple register r0..r15 to stack
popm  sp          pop multiple register r0..r15 from stack

MOS in Assembly

Before we write assembly code we must layout the memory.  To assign location of all variables and data, because we need to know absolute addresses. 

Memory Map

0..900             code
1000               interrupt vector
1500..1999    global variables
2000              process queue
2100              process control block
2300              stack

Task Switcher

The main part of MOS in the task switcher. The implementation is S2 assembly language follows the pseudo code explained earlier.

:tswitch       
    ld r27 currentp
    xch r20           ;  get RetAds
    st r20 @0 r27     ;  p.PC = RetAds
    pushm sp          ;  save current context
    st sp @1 r27      ;  p.SP = sp
    jal link nextp
    jf r27 exit       ;  no process in the queue

    st r27 currentp   ;  update currentp
    ld sp @1 r27      ;  get p.SP
    popm sp           ;  restore context
    ld r20 @0 r27     ;  get p.PC
    xch r20           ;  RetAds = p.PC
    reti
:exit   
   
    trap r0 #stop


This section of code saves the current process context: PC, SP, Stack (register r0..r15). sp (r29) is a register that is persistent.  It lives across all process switches. 
    ld r27 currentp
    xch r20           ;  get RetAds
    st r20 @0 r27     ;  p.PC = RetAds
    pushm sp          ;  save current context
    st sp @1 r27      ;  p.SP = sp


Then we get the next process (trap nextp returns r27) and check whether there is any process in the queue.

    jal link nextp
    jf r27 exit       ;  no process in the queue

And restore the context of the next process.  r27 holds the pointer to the process (PCB).

    st r27 currentp   ;  update currentp
    ld sp @1 r27      ;  get p.SP
    popm sp           ;  restore context

Then switch to the next process using "swap r20" follows by "reti"

    ld r20 @0 r27     ;  get p.PC
    xch r20           ;  RetAds = p.PC
    reti

Critical Section

The task-switcher is not interruptible, so we must protect this section of the code.  Using Disable Interrupt (di) and Enable Interrupt (ei) to enclose the code to prevent interrupt during running of this code.  This is called "critical section".

:tswitch
     trap r0 #di
    ...  < critical section>
    ...
     trap r0 #ei
    reti

Here is the assembly code (mos2.txt) that implement  newp(), newStack(), enqueue(), terminate(), nextp() in S2 assembly code.  The way to boot MOS is slightly improved.  We made a dummy zeroth process which is not in the process queue and start the task switcher to switch to the task in the process queue by software interrupt, int #0.  

    jal link newp
    st sp @1 retval
    st retval currentp  ; zeroth p
    trap r0 #ei
    int #0              ; start by switch p

Example

I create two processes, both count 1 to 10.  The interrupt interval is set to 100. The output shows the distinction between process1 n  and process2 [n].  Please observe the frequency of interrupt and the concurrency of two processes.  Once the process1 is terminated, process2 continues to its end.   This application program is tacked at the end of the operating system code (mos2.txt). In pseudo code it looks like this:

main()
    set up int vec to tswitch()
    initialize  OS variables
    create process1() and put it in the process queue
    create process2() and put it in the process queue
    start OS by forcing tswitch() with interrupt


global   cnt, cnt2

process1()
    cnt = 0
    while( cnt < 10)
        cnt = cnt + 1
        print(cnt)

process2()
    cnt2 = 0
    while( cnt2 < 10)
        cnt2 = cnt2 + 1
        print(cnt2)

Here is the screen dump.

C:>\iot-rz\test>sim21 mos2.obj
load program, last address 200
>g
interrupt0
[1]interrupt0
1 2 3 4 5 6 7 interrupt0
[2][3][4][5][6][7][8]interrupt0
8 9 10 interrupt0
interrupt0
[9][10]interrupt0
stop, clock 577, execute 577 instructions
>

You can trace the execution of this program step-by-step and watch what happen when the interrupt occurs, how the task-switch works.

Please try it.  I set the interrupt interval to 100 (in s21.h).  If you change it (you have to recompile the simulator), you can observe the change in the number of interrupt. 

Enjoy!

last update  18 Feb 2017