Writing Your Own OS

Advance C, Operating System Development / By
Operating System is nothing but collection of programs for managing system resources like CPU, memory, storage device etc. Study of the Operating System is one of the vastest areas. This chapter does not deal with the details about Operating System. And in this chapter I would like to show you how OS can be written in Turbo C. However you may not be able to code your Operating System without depth knowledge of memory management, processor scheduling etc. So I strongly recommend you to go through a good Operating System book for indepth knowledge. According to me most of the people are not using Turbo C to write OS, because Turbo C is 16bit. Also people mainly hangout with Assembly language for a better and tight code.
EZOS_86
EZOS_86 is a simple multitasking kernel written in Turbo C for x86 machines in 1996-97. Operating Systems are usually protected and licensed according to GNU’s General Public License and so this EZOS_86! So if you modify or rewrite this source code, you must acknowledge the author Scott A. Christensen and you are expected to keep the name of the revised OS as EZOS_86, but you can change the version. Regarding OS and other software, violation of copyright is treated as high offense. So beware of the licenses!
Notes
The author added following note:
EZOS_86 is a simple multitasking kernel for the x86 family. It is written in 100% C source (it uses Turbo C extensions to access the registers). If you need a tight, fast, hand-coded, assembly kernel, forget this one! The main emphasis here is to keep it simple: no linked lists, no dynamic allocation, no complicated task scheduling, no assembly language, etc. Yes, this can be embedded! The scheduler is very rudimentary. It is preemptive, but with a strictly prioritized order.
There is no protection from starvation; if a higher priority task spins the CPU, the lower priority tasks will never execute. Programs for embedded applications are often event driven and properly written will work fine. On the other hand, it wouldn’t be that hard to change the scheduler to a round robin method if desired.
The scheduler always traverses the Task Control Block (TCB) array from the beginning (&tcb[0]). The first task encountered that is eligible to run is the one executed. At least one task MUST always be eligible to run; hence the “null” task, which is created as the lowest priority and NEVER, sleeps. The same task function can have multiple instances. For example you could call OsTaskCreate( ) three times and pass task0 as the function all three times. Of course you must specify a unique stack and tcb. The parameter passed to task0 can identify the particular instance of the function.
Reentrancy issues:
  • use the runtime library at your own risk (reason for direct video)
  • floating point is not reentrant; use semaphore protection or only do floating point in one task.
Semaphores:
  • clearing semaphore does not cause task switch; call OsSchedule( ) to yield. This can throttle throughput. One could have null task continuously scan TCBs for eligible task and yield.
  • OsSemClear( ) returns TRUE if higher priority task waiting on sem
  • multiple tasks can sleep on same semaphore
  • ok to clear semaphore from within interrupt routine
As written this code will run a demo on an IBM clones and even clean up upon exit returning nicely backs to DOS. It creates the file “out” to dump the stack contents. Interrupt routines use the current task’s stack. Be careful not to exceed your allocated stack space; very strange results can occur. Compile it with Turbo C with optimization off.
Wishlist:
  • simple file functions to read/write directly to IDE HD with FAT16
  • multitasking capable floating point support
  • some sort of built in debugging capability (TBUG.ZIP looks like a good start)
  • runtime calculation of cpu utilization
  • a _simplified_ malloc for embedded applications
Kernel Source Code
/*

* ezos_86.c

*

*

* All Rights Reserved

*

* This file is part of the EZOS_86 multitasking kernel.

*

*

* version description

* ————————————————————–

* 0.01.00 initial release

*

*/

#include <dos.h>

#include <stdio.h>

#include <conio.h>

#include <stdarg.h>

/*————————————————————–*/

#define TRUE (0 == 0)

#define FALSE (0 != 0)

#define RUNNING 0

#define RUN_ASAP 1

#define SLEEPING 2

#define PENDING 3

#define SUSPENDED 4

#define KILLED 5

#define ALL_KILLED -2

#define NOT_STARTED -1

#define TICK_VECT 8

#define MAX_TASKS 10

#define STACKSIZE 1024

#define PENDING_SEM_REQUEST 0

#define PENDING_SEM_WAIT 1

#define TSK_ERR_ -1000

#define TSK_ERR_TIMEOUT (TSK_ERR_ – 0)

#define OS_INFINITE_WAIT -1L

#define OS_IMMEDIATE_RETURN 0L

#define OsEnable() enable()

#define OsDisable() disable()

#define ATTR ((unsigned int) (((BLACK<<4)|WHITE)<<8))

#define schedule() \

{ \

int si; \

 

static PTCB_REC pTCBsi; \

static PTCB_REC pTCBsc; \

\

if(killedTasks == numTasks) \

{ \

_SP = mainSP; \

_SS = mainSS; \

mainSleep = FALSE; \

curTask = ALL_KILLED; \

} \

else \

{ \

for(si = 0, pTCBsi = tcb; si < numTasks; si++, pTCBsi++) \

{ \

if(pTCBsi->taskStatus == RUNNING) \

break; \

if(pTCBsi->taskStatus == RUN_ASAP) \

{ \

pTCBsc = &tcb[curTask]; \

if(pTCBsc->taskStatus == RUNNING) \

pTCBsc->taskStatus = RUN_ASAP; \

pTCBsc->taskSP = _SP; \

pTCBsc->taskSS = _SS; \

pTCBsi->taskStatus = RUNNING; \

_SP = pTCBsi->taskSP; \

_SS = pTCBsi->taskSS; \

curTask = si; \

break; \

} \

} \

} \

}

/*————————————————————–*/

typedef void (far cdecl *FUNCPTR)();

typedef struct

{

unsigned int r_bp;

unsigned int r_di;

unsigned int r_si;

unsigned int r_ds;

unsigned int r_es;

unsigned int r_dx;

unsigned int r_cx;

unsigned int r_bx;

unsigned int r_ax;

 

FUNCPTR taskStartAddr;

unsigned int r_flags;

FUNCPTR taskExitReturn;

void * pTaskParam;

} STACK_REC;

typedef struct

{

unsigned int taskStatus;

unsigned int taskSP;

unsigned int taskSS;

long ticks;

int semState;

int * pSem;

} TCB_REC, *PTCB_REC;

/*————————————————————–*/

void far interrupt OsTickIsr(void);

int far interrupt OsSchedule(void);

void far OsTaskKill(void);

void OsTaskCreate(PTCB_REC, FUNCPTR, void *,

unsigned char far *, int);

long OsTranslateMilsToTicks(long);

void OsInstall(void);

void OsRun(void);

void OsDeinstall(void);

void OsSleep(long);

void OsSleepTicks(long);

int OsSemClear(int *);

void OsSemSet(int *);

int OsSemWait(int *, long);

int OsSemSetWait(int *, long);

int OsSemRequest(int *, long);

int OsDisableStat(void);

void dumpStack(FILE *, unsigned char *, int);

void tprintf(const char *, …);

void tputs(const char *);

void sout(char *);

void incRow(void);

void far task0(void *);

void far task1(void *);

void far task2(void *);

void far taskNull(void *);

/*————————————————————–*/

void (far interrupt *oldTickIsr)(void);

int numTasks = 0;

int killedTasks = 0;

int curTask = NOT_STARTED;

int mainSleep = TRUE;

unsigned int mainSP;

unsigned int mainSS;

TCB_REC tcb[MAX_TASKS];

unsigned int _stklen = (STACKSIZE * MAX_TASKS) + 1024;

int itick = 0;

unsigned int (far *screen)[80];

int row = 0;

int col = 0;

int tickSem = 1;

int goSem = 1;

int screenSem = 0;

/*————————————————————–*/

/*————————————————————–*/

/*————————————————————–*/

void main()

{

unsigned char stack0[STACKSIZE];

unsigned char stack1[STACKSIZE];

unsigned char stack2[STACKSIZE];

unsigned char stackNull[STACKSIZE];

FILE * f;

clrscr();

puts(“\n\n EZOS_86 multitasking kernel”);

puts(” Copyright (C) 1996-97 Scott A. Christensen”);

delay(5000);

clrscr();

gotoxy(1, 24);

screen = MK_FP(0xB800, 0);

OsTaskCreate(&tcb[0], task0, (void *) 100, stack0, STACKSIZE);

OsTaskCreate(&tcb[1], task1, (void *) 101, stack1, STACKSIZE);

OsTaskCreate(&tcb[2], task2, (void *) 102, stack2, STACKSIZE);

OsTaskCreate(&tcb[3], taskNull, NULL, stackNull, STACKSIZE);

OsInstall();

OsRun();

OsDeinstall();

f = fopen(“out”, “wb”);

dumpStack(f, stack0, STACKSIZE);

dumpStack(f, stack1, STACKSIZE);

dumpStack(f, stack2, STACKSIZE);

dumpStack(f, stackNull, STACKSIZE);

fclose(f);

puts(“done, hit key to continue…”);

getch();

}

/*————————————————————–*/

void dumpStack(

FILE * f,

unsigned char * stack,

int size

)

{

int i;

char buf[80];

char string[80];

string[0] = 0;

for(i = 0; i < size; i++)

{

if(i % 16 == 0)

fprintf(f, “%04X:%04X “, FP_SEG(&stack[i]), FP_OFF(&stack[i]));

fprintf(f, “%02X “, stack[i]);

if(isalnum(stack[i]) || stack[i] == ‘ ‘)

{

buf[0] = stack[i];

buf[1] = 0;

strcat(string, buf);

}

else

strcat(string, “.”);

if(i % 16 == 15)

{

fprintf(f, ” %s\r\n”, string);

string[0] = 0;

}

}

fprintf(f, “\r\n”);

}

/*————————————————————–*/

void OsInstall()

{

oldTickIsr = getvect(TICK_VECT);

setvect(TICK_VECT, OsTickIsr);

}

/*————————————————————–*/

void OsRun()

{

while(mainSleep);

}

/*————————————————————–*/

void OsDeinstall()

{

setvect(TICK_VECT, oldTickIsr);

}

/*————————————————————–*/

void far interrupt OsTickIsr()

{

int i;

static PTCB_REC pTCBi;

switch(curTask)

{

case ALL_KILLED:

break;

case NOT_STARTED:

mainSP = _SP;

mainSS = _SS;

pTCBi = tcb;

pTCBi->taskStatus = RUNNING;

_SP = pTCBi->taskSP;

_SS = pTCBi->taskSS;

curTask = 0;

break;

default:

itick++;

for(i = 0, pTCBi = tcb; i < numTasks; i++, pTCBi++)

{

if((pTCBi->taskStatus == SLEEPING) ||

(pTCBi->taskStatus == PENDING))

if(pTCBi->ticks > 0L)

if(–(pTCBi->ticks) == 0L)

pTCBi->taskStatus = RUN_ASAP;

}

schedule();

break;

}

oldTickIsr();

}

/*————————————————————–*/

int far interrupt OsSchedule()

{

OsDisable();

schedule();

return _AX; /* dummy value */

}

/*————————————————————–*/

void far OsTaskKill()

{

OsDisable();

killedTasks++;

tcb[curTask].taskStatus = KILLED;

OsSchedule();

}

/*————————————————————–*/

void OsTaskCreate(

PTCB_REC pTCB,

FUNCPTR func,

void * pTaskParam,

unsigned char far * pStack,

int stackSize

)

{

STACK_REC far * pStackRec;

int i;

for(i = 0; i < stackSize; i++)

pStack[i] = 0xFF;

pStackRec = (STACK_REC far *) (pStack + stackSize –

sizeof(STACK_REC));

pStackRec->r_bp = 0;

pStackRec->r_di = 0;

pStackRec->r_si = 0;

pStackRec->r_ds = _DS;

pStackRec->r_es = _DS;

pStackRec->r_dx = 0;

pStackRec->r_cx = 0;

pStackRec->r_bx = 0;

pStackRec->r_ax = 0;

pStackRec->taskStartAddr = func;

pStackRec->r_flags = 0x0200;

pStackRec->taskExitReturn = OsTaskKill;

pStackRec->pTaskParam = pTaskParam;

pTCB->taskStatus = RUN_ASAP;

pTCB->taskSP = FP_OFF(pStackRec);

pTCB->taskSS = FP_SEG(pStackRec);

numTasks++;

}

/*————————————————————–*/

long OsTranslateMilsToTicks(

long mils

)

{

long x;

if(mils < 0L)

return -1L;

if(!mils)

return 0L;

x = ((mils * 91L) / 5000L) + 1L; /* 18.2 ticks per sec */

return x;

}

/*————————————————————–*/

void OsSleep(

 

long mils

)

{

long ticks;

ticks = OsTranslateMilsToTicks(mils);

OsSleepTicks(ticks);

}

/*————————————————————–*/

void OsSleepTicks(

long ticks

)

{

PTCB_REC pTCB;

if(ticks <= 0L)

return;

OsDisable();

pTCB = &tcb[curTask];

pTCB->taskStatus = SLEEPING;

pTCB->ticks = ticks;

OsSchedule();

}

/*————————————————————–*/

int OsSemClear(

int * pSem

)

{

int i;

STACK_REC far * pStackRec;

int processedRequest;

PTCB_REC pTCB;

int higherEligible;

int intsEnabled;

intsEnabled = OsDisableStat();

if(!*pSem)

{

if(intsEnabled)

OsEnable();

 

return FALSE;

}

*pSem = 0;

processedRequest = FALSE;

higherEligible = FALSE;

for(i = 0, pTCB = tcb; i < numTasks; i++, pTCB++)

{

if((pTCB->taskStatus == PENDING) && (pTCB->pSem == pSem))

{

switch(pTCB->semState)

{

case PENDING_SEM_REQUEST:

if(processedRequest)

break;

processedRequest = TRUE;

*pSem = 1;

/* !!! no break here !!! */

case PENDING_SEM_WAIT:

pStackRec = MK_FP(pTCB->taskSS, pTCB->taskSP);

pStackRec->r_ax = 0;

pTCB->taskStatus = RUN_ASAP;

if(i < curTask)

higherEligible = TRUE;

break;

}

}

}

if(intsEnabled)

OsEnable();

return higherEligible;

}

/*————————————————————–*/

void OsSemSet(

int * pSem

)

{

int intsEnabled;

intsEnabled = OsDisableStat();

*pSem = 1;

 

if(intsEnabled)

OsEnable();

}

/*————————————————————–*/

int OsSemWait(

int * pSem,

long mils

)

{

long ticks;

PTCB_REC pTCB;

OsDisable();

if(!*pSem)

{

OsEnable();

return 0;

}

ticks = OsTranslateMilsToTicks(mils);

if(!ticks)

{

OsEnable();

return TSK_ERR_TIMEOUT;

}

pTCB = &tcb[curTask];

pTCB->taskStatus = PENDING;

pTCB->semState = PENDING_SEM_WAIT;

pTCB->pSem = pSem;

pTCB->ticks = ticks;

_AX = TSK_ERR_TIMEOUT;

return OsSchedule();

}

/*————————————————————–*/

int OsSemSetWait(

int * pSem,

 

long mils

)

{

OsDisable();

OsSemSet(pSem);

return OsSemWait(pSem, mils);

}

/*————————————————————–*/

int OsSemRequest(

int * pSem,

long mils

)

{

long ticks;

PTCB_REC pTCB;

OsDisable();

if(!*pSem)

{

*pSem = 1;

OsEnable();

return 0;

}

ticks = OsTranslateMilsToTicks(mils);

if(!ticks)

{

OsEnable();

return TSK_ERR_TIMEOUT;

}

pTCB = &tcb[curTask];

pTCB->taskStatus = PENDING;

pTCB->semState = PENDING_SEM_REQUEST;

pTCB->pSem = pSem;

pTCB->ticks = ticks;

_AX = TSK_ERR_TIMEOUT;

 

return OsSchedule();

}

/*————————————————————–*/

int OsDisableStat()

{

unsigned int flags;

flags = _FLAGS;

OsDisable();

return flags & 0x0200;

}

/*————————————————————–*/

void tprintf(

const char * format,

…

)

{

va_list argPtr;

char buf[100];

struct time t;

va_start(argPtr, format);

vsprintf(buf + 18, format, argPtr);

va_end(argPtr);

OsSemRequest(&screenSem, OS_INFINITE_WAIT);

gettime(&t);

sprintf(buf, “-T%02d(%02d:%02d:%02d.%02d)”,

curTask, t.ti_hour, t.ti_min, t.ti_sec, t.ti_hund);

buf[17] = ‘ ‘;

sout(buf);

OsSemClear(&screenSem);

}

/*————————————————————–*/

void tputs(

 

const char * string

)

{

struct time t;

char buf[100];

OsSemRequest(&screenSem, OS_INFINITE_WAIT);

gettime(&t);

sprintf(buf, “-T%02d(%02d:%02d:%02d.%02d) %s\n”,

curTask, t.ti_hour, t.ti_min, t.ti_sec, t.ti_hund, string);

sout(buf);

OsSemClear(&screenSem);

}

/*————————————————————–*/

void sout(

char * p

)

{

while(*p)

{

switch(*p)

{

case ‘\r’:

col = 0;

break;

case ‘\n’:

col = 0;

incRow();

break;

case ‘\t’:

sout(” “);

break;

default:

screen[row][col] = ATTR | ((unsigned int) *p);

if(++col > 79)

{

col = 0;

 

incRow();

}

break;

}

p++;

}

}

/*————————————————————–*/

void incRow()

{

int r;

int c;

if(++row > 24)

{

for(r = 0; r < 24; r++)

for(c = 0; c < 80; c++)

screen[r][c] = screen[r + 1][c];

for(c = 0; c < 80; c++)

screen[24][c] = ATTR | ((unsigned int) ‘ ‘);

row = 24;

}

}

/*————————————————————–*/

void far task0(

void * pTaskParam

)

{

int val = (int) pTaskParam;

int i;

long j;

int rc;

OsSemWait(&goSem, OS_INFINITE_WAIT);

tprintf(“init val passed = %d\n”, val);

for(i = 0; i < 7; i++)

{

rc = OsSemWait(&tickSem, 300L);

switch(rc)

{

case 0:

 

tputs(“OsSemWait successful”);

OsSleep(150L);

break;

case TSK_ERR_TIMEOUT:

tputs(“OsSemWait failed, error = TSK_ERR_TIMEOUT”);

break;

default:

tprintf(“OsSemWait failed, error = %d\n”, rc);

break;

}

OsSleep(100L);

}

}

/*————————————————————–*/

void far task1(

void * pTaskParam

)

{

int val = (int) pTaskParam;

int i;

OsSemWait(&goSem, OS_INFINITE_WAIT);

tprintf(“init val passed = %d\n”, val);

for(i = 0; i < 3; i++)

{

OsSleep(500L);

tputs(“”);

}

tputs(“clearing tickSem”);

OsSemClear(&tickSem);

OsSleep(1000L);

tputs(“”);

}

/*————————————————————–*/

void far task2(

 

void * pTaskParam

)

{

int val = (int) pTaskParam;

int i;

int j;

tprintf(“init val passed = %d\n”, val);

OsSleep(2000L);

OsSemClear(&goSem);

for(i = 0; i < 3; i++)

{

OsSleepTicks(18L);

tputs(“”);

}

}

/*————————————————————–*/

void far taskNull(

void * pTaskParam

)

{

while(killedTasks != numTasks – 1);

}
Good Luck!
Because of the success of Linux, many people are hanging out with the creation of OS.
Writing an efficient and neat OS is considered to be tough task because you may need to know more OS fundamentals and hardware details. If you could be able to come out with a new OS, the World would really appreciate you! Good Luck!
Scroll to Top