How to write an operating system

Writing an operating system is something that can not only be interesting (if you're one of those people that get turned on by Int 13....) but it is also a great learning experience. Through creating your own operating system you will learn exactly what goes on behind the scenes, elevating you above the average programmer that just writes in Visual Basic.

In this tutorial you will be tought by examples, and by the end you should have created your own operating system.

Tools:

EasyOsNew version of EasyOSMost reliable version of Easyos.zipeasyos.zip 300kb
EasyOS is a very simple operating system, it contains all the tools needed to build an operating system. (Not written by me, although I did add bits to it and mess it up a bit)

A quick explanation of assembly: (see here for a good tutorial)

si and ah Think of si as something to put text into, and ah as something to put numbers into.
mov This (mov)es data. mov ah,0 would move 0 into ah. (The data on the right is moved into the left)
Int Think of these as functions. Different int's do different things when ah is different. Ahem. Eg. when ah = 0 and you call int 10 it prints si to the screen.
Stuff To put words and stuff in your program you can't just do mov si,'some words' (well, you can but you wont like the resutls) so instead you have to declare the words first. You do this by putting the name of what you want the words to be called by, then the data type (nearly always db) then the words themselves. Eg:
name db 'some words'
Jump To give sections of code a label, just type the label and add a : at the end, eg code: . You can then use jmp to jump to it, eg jmp codeIf To do an if in assembly you use cmp, eg cmp al,0 (if al=0). On the next line you then put je code, and if al=0 then the program jumps to the section of code called code. If you use jne code, then if al is not 0 the program will jump to code.
The stack The stack is where stuff is stored. push pushes stuff into it, pop pulls stuff out. The following example would put cx into dx:
push cx pop dx

Now you know everything there is to know about assembly, you can now understand most of the program that boot's EasyOs. Drives (hard drives and floppy's) are split into lots of bits, all 512 bytes long (enough to fit 512 letters in). These are calle sectors. The first sector is what the computer looks for when it boots. It is called the bootsector.
Open the folder src and then open boot.asm. Or if you are lazy, just look at the code below (its the same).

; This is a comment[ORG 0x7C00] ;This just tells the program where it is in the memory. Not important[BITS 16] ;Not important too.jmp start; Jump over BIOS parameter blockstart: ;The label for the start of the actual programpush cs ;Put cs onto the stackpop ds ;Take it out and put it into dsmov si,Print_loading     ;Print loading messagecall printstring ;Call is like jump, but it goes back. Like a function ;The complicated bit: Loads the next programmov ah,02h               ;When ah=, int13 reads a disk sectormov al,4  ;Al is how many sectors to readmov ch,0  ;The track to read frommov cl,2  ;Sector Idmov dh,0  ;Headmov dl,0  ;Drive (0 is floppy)mov bx,0x1000 ;Es and Bx put together are where to load the program too (see jmp 0x1000:0x00)mov es,bxmov bx,0x00int 13h;Int 13 is all functions for disksmov si,putdot           ;Print a ".".call printstringjmp 0x1000:0x00         ;Run Bootinit from stack.printstring:            ;Print string routine.mov ah,0eh;Mov ah into 0, so int 10 printsstringloop:;The following code loads each seperate charcter so it can be printedlodsbcmp al,00;If al =0, then the string has all been loadedje endstringint 10h;When int 10 is called, and ah=, it printsjmp stringloopendstring:ret;Ret returnsputdot        db '.',0Print_loading db 13,10,'Loading Easy OS v0.01a...',0times 425 db 0 ;wastes 425 bytes on purpose, so the sector is full (The program must be 512 bytes long)

You may have noticed that numbers, like int 10h, end in h. They don't have to, it just looks funky (there is a real reason, but it's boring).
Anyway, now copy the program called copyboot in the folder called utils to the folder with test.asm in. Open the dos prompt and type: (Make sure a blank floppy is inserted)
copyboot test.com 0
copyboot is the name of the program, test.com the name of the file to copy, and 0 is the sector.
In the program above all it does is print a string then load whats at sector 1. The program that easyos loads under the src folder called bootinit. If you assemble it with nasm, then copy it to sector 1 and restart, the bootsector will load it.
There isn't much more to be learnt from EasyOs, so run either setup.exe or make.bat to build the whole thing. The difference is setup.exe lets you setup a root password for EasyOs. If you just run make.bat the passwords is the default password: monty (named after my psychotic dog).
Now restart and be amazed. Wow. Pretty crappy, but it isn't that bad.

Fat 12

No, its not a fat people supprt group but a file system. If you try and access the floppy disk Windows will say it need s to be formatted. Formatted? You ask. Formatting is basically organising the sectors so you can give them names. Underneath Windows and Fat, you are still just accessing sectors. I won't go into Fat here, check out http://www.maverick.subnet.dk/ for some info. Anyway, the bootsector needs to have some information in it that when Windows reads it it tells it it is fat. The following BootSector was disgustingly ripped by me of NYAOS (I have no idea what i stands for). With your A* qualification in assembly language you can no doubt understand it.

; NYAOS Boot Sector (C) Copyright Sean Tash 1998; assemble with:; nasm -f bin -o bootsect.bin bootsect.asm            bits 16            org 0x7C00start:      jmp short begin            nopbsOEM       db "NYAOS1.0"               ; OEM StringbsSectSize  dw 512                      ; Bytes per sectorbsClustSize db 1                        ; Sectors per clusterbsRessect   dw 1                        ; # of reserved sectorsbsFatCnt    db 2                        ; # of fat copiesbsRootSize  dw 224                      ; size of root directorybsTotalSect dw 2880                     ; total # of sectors if < 32 megbsMedia     db 0xF0                     ; Media DescriptorbsFatSize   dw 9                        ; Size of each FATbsTrackSect dw 18                       ; Sectors per trackbsHeadCnt   dw 2                        ; number of read-write headsbsHidenSect dd 0                        ; number of hidden sectorsbsHugeSect  dd 0                        ; if bsTotalSect is 0 this value is                                        ; the number of sectorsbsBootDrv   db 0                        ; holds drive that the bs came frombsReserv    db 0                        ; not used for anythingbsBootSign  db 29h                      ; boot signature 29hbsVolID     dd 0                        ; Disk volume ID also used for temp                                        ; sector # / # sectors to loadbsVoLabel   db "NO NAME    "            ; Volume LabelbsFSType    db "FAT12   "               ; File System typebegin:      cli                         ; disable interrupts            mov [bsBootDrv],dl          ; save drive number            mov ax,0x9000               ; put stack at 0x98000            mov ss,ax            mov sp,0x8000            mov cx,[bsTrackSect]        ; update int 1E FDC param table            mov bx,0x0078            lds si,[ds:bx]            mov byte [si+4], cl            mov byte [si+9], 0x0F            sti                         ; enable interrupts            push ds            mov dl,[bsBootDrv]          ; reset controller            xor ax,ax            int 0x13            pop ds            jc bootfail2                ; display error message            jmp _l1bootfail2:  jmp bootfail_l1:            mov ax,0x0000            mov es,ax            mov ds,ax            mov si,MsgLoad              ; display load message            call putstr            ; find the root directory            xor ax,ax            mov al,[bsFatCnt]            mov bx,[bsFatSize]            mul bx            add ax,word [bsHidenSect]            adc ax,word [bsHidenSect+2]            add ax,word [bsRessect]     ; ax holds root directory location            mov word [BootSig],ax            call checkroot            xor ax,ax            add ax,word [start]            add ax,word [bsVolID]       ; sector number            add ax,word [BootSig]            sub ax,2                    ; correction for a mis-calc            mov cx,word [bsVolID+2]     ; number of sectors            mov bx,0x8000            mov es,bxnextsector: push ax                     ; save registers            push cx            push dx            push es            xor bx,bx                   ; set zero offset            call readsect               ; read a sector            mov si,MsgDot               ; display a dot            call putstr            pop es                      ; restore registers            pop dx            pop cx            pop ax            mov bx,es            add bx,20h                  ; increment address 512 bytes            mov es,bx            inc ax                      ; read next sector            loopnz nextsector            mov ax,0x8000               ; set segment registers and jump            mov es,ax            mov ds,ax            push ax            mov ax,0            push ax            retfcheckroot:            push ax                     ; save registers            push bx            push cx            push dx            push si            push di                        mov ax,0x8000               ; put root directory at 0x80000            mov es,ax            mov ax,32                   ; AX = ((32*RootSize)/512) + 2            mul word [bsRootSize]            div word [bsSectSize]            mov cx,ax                   ; cx holds # of sectors in root            mov word [start],ax            mov ax,word [BootSig]       ; get prev. saved loc. for root dirr1:         xor bx,bx            push cx                     ; save count            push ax                     ; save sector number            push es            push dx            call readsect            xor bx,bxl_1:        mov di,bx                   ; set address to check from            mov cx,11                   ; check 11 bytes            mov si,FileName             ; address of string to check with            repz cmpsb            je foundit            add bx,32                   ; check next entry            cmp bx,[bsSectSize]         ; end of sector?            je l_2            jmp l_1l_2:        pop dx                      ; restore registers            pop es            pop ax            pop cx            inc ax                      ; read next sector            loopnz r1            jmp bootfailfoundit:    pop dx                      ; get these off the stack            pop es            pop ax            pop cx            mov di,0x1A                 ; get clustor #            add di,bx            push bx                     ; save bx for finding # of sectors            mov ax,[es:di]            xor bx,bx                   ; calculate sector #            mov bl,[bsClustSize]            mul bx                      ; ax holds sector #            mov word [bsVolID],ax            pop bx                      ; get location of directory entry            mov di,0x1C            add di,bx            mov ax,[es:di]              ; put number of bytes in ax            xor dx,dx            mov bx,[bsClustSize]        ; # of bytes / 512            div bx            inc ax            mov word [bsVolID+2],ax     ; save number of sectors to load            pop di                      ; restore registers            pop si            pop dx            pop cx            pop bx            pop ax                        ret                         ; return to caller            putstr:     ; SI = address of string to display            lodsb            or al,al            jz short putstrd            mov ah,0x0E            mov bx,0x0007            int 0x10            jmp putstrputstrd:    retn                        ; return to callerbootfail:   ; display failure message            mov si,MsgBad               ; display error message            call putstr            xor ax,ax                   ; wait for keypress            int 0x16            int 0x19                    ; rebootreadsect:   ; ES:BX = Location ; AX = Sector            mov si,[bsTrackSect]            div si                      ; divide logical sect by track size            inc dl                      ; sector # begins at 1            mov [bsReserv],dl           ; sector to read            xor dx,dx                   ; logical track left in ax            div word [bsHeadCnt]        ; leaves head in dl, cyl in ax            mov dh, [bsBootDrv]         ;            xchg dl,dh                  ; head to dh, drive to dl            mov cx,ax                   ; cyl to cx            xchg cl,ch                  ; low 8 bits of cyl to ch, hi 2 bits            shl cl,6                    ; shifted to bits 6 and 7            or cl, byte [bsReserv]      ; or with sector number            mov al,1                    ; number of sectors            mov ah,2                    ; use read function of int 0x13            int 0x13                    ; read sector            jc bootfail                 ; display error message            ret                         ; return to callerpadding     times 45 db 0FileName    db "OSLOADERCOM"MsgBad      db "Disk Error...",13,10,0MsgDot      db ".",0MsgLoad     db "doors loading",0BootSig     db 0x55, 0xAA

Anyways, copy the above program, save it, build it with nasm, copy it with copyboot.
As you can guess above, it loads a program called 'OSLOADER.COM' off of the floppy. So, if you want a particularily funky os, build the following with nasm:

;Funky squares;;Assembles with NASM;Made by Frej  somewhere between april and may 2000;Bits reprogrammed to run within DeviatorOS;;This demo is just to show you how small graphical demos can get ;);Reprogrammed for DeviatorOS as demo program        org 0x0000        mov ax,cs        mov ds,ax        mov es,ax         ; fix segment regsstart:  mov bx,cs         ;put codesegment to bx        add bh,0x20       ;add 2000 to bx        mov ds,bx         ;and put it to ds        mov ax,0x13       ;set ax to videomode 13        int 10h           ;and do thatMain:   push ds           ;put buffer seg to stack        pop es            ;and put that into es        in ax,0x40        ;generate "random" number (timer)        shl ax,4          ;multiply random # with 16        mov di,ax         ;box offset (random)        mov al,255        ;color of the box        mov bx,50         ;height=50pl:     add di,270        ;di+270 (320-width(50))        mov cx,50         ;# bytes to copy to buffer        rep stosb         ;and do it        dec bx            ;decrement bx        jnz pl            ;jump if bx not zero        mov bh,0xFA       ;assume bl = 0 (-> bx = FA00)Smudge: mov al,[bx+1]     ;right color to al        mov cl,[bx-1]     ;left color to cl        add ax,cx         ;and add it to ax        mov cl,[bx-320]   ;upper color to cl        add ax,cx         ;and add it to ax        mov cl,[bx+320]   ;lower color to cl        add ax,cx         ;and add it to ax        shr ax,2          ;divide with 4        mov [bx],al       ;and but the avarage color to buffer        dec bx            ;decrement bx        jnz Smudge        ;jump if bx not zero        mov ax,0xA000     ;vga seg        mov es,ax         ;put it to es        mov ch,0xFA       ;# bytes to copy to vga        xor di,di         ;zero vga offset        xor si,si         ;zero buffer offset        rep movsb         ;and do that        in al,0x60        ;check for keys        dec al            ;was it esc?        jnz Main          ;nope, continue        mov ax,3          ;text mode        int 10h           ;get back into text mode        xor ah,ah         ;yes, return to OS        int 0x18          ;back to good old kernel;Note:- This was, as you can guess, ripped by from an os. So when it goes back to the 'good ol kernel' it just restarts.

Build it with nasm, but rather than faffing around copyboot, just name is as OSLOADER.COM and copy it to the floppy.
Now restart and enjoy the funkiness. Woah dood.

C

Time to escape assembly language. The boot sector has to be written in assembly, but nothing else does. Unfortunately you can't just go and write a cool shell with Visual C++. First of all, its has to be a .com program, not .exe.
.exe programs are just .com with a bit of extra info. at the start giving some info on what the program is. It's very easy to add .exe capabililty to an os, or you can download a program called exe2com.
The serious problem though is that you can create your own ints to make things easier. EasyOs does this (look in kernel.asm under the src folder) and Dos does this too (Dos makes int 21). By default, compilers build a program with these ints.

For linux lovers

The following code is for Gcc on linux and nasm for dos or linux. Dos/ Windows users can download Djgpp from http://www.delorie.com/djgpp/, which is like gcc. The line
gcc -c -O2 -Wall -g -o hello.o hello.c
Tells gcc to build a plain binary. I don't know who wrote the following, e-mail me if you do.

The smallest pmode + C program I could write is shown below,in three files. I compiled with GCC, not GPP, and got nowarnings.This code doesn't check for a 32-bit CPU or V86 mode. If youtry to run it inside a DOS box, Windows will kill it.load.asm is assembled to aout format instead of COFF, becauseDJGPP COFF doesn't let you mix 16- and 32-bit code.;/****************************************************************************;       file load.asm;****************************************************************************/[SECTION .text][BITS 16][GLOBAL start]start:  xor ebx,ebx             ; now in real mode (16-bit)mov bx,csshl ebx,4mov eax,ebx; EAX=EBX=CS<<4lea eax,[ebx]mov [gdt2 + 2],ax; set descriptor base address=EAXmov [gdt3 + 2],axshr eax,16mov [gdt2 + 4],almov [gdt3 + 4],almov [gdt2 + 7],ahmov [gdt3 + 7],ahlea eax,[ebx + gdt]; point gdt_ptr to the gdtmov [gdt_ptr + 2],eax; EAX=linear address of gdtpush dword 0; zero EFLAGS (interrupts off,popfd;  IOPL=0, NT bit=0)lgdt [gdt_ptr]mov eax,cr0or al,1mov cr0,eaxjmp SYS_CODE_SEL:do_pm[BITS 32]do_pm:  mov ax,SYS_DATA_SEL     ; now in 32-bit pmodemov ds,eax; EAX works, one byte smaller :)mov ss,eax        nopmov es,eaxmov fs,eaxmov gs,eaxxor eax,eax; zero top 16 bits of ESPmov ax,spmov esp,eax[EXTERN _main]call _main; call C codejmp $; freeze[SECTION .data]; null descriptorgdt:dw 0; limit 15:0dw 0; base 15:0db 0; base 23:16db 0; typedb 0; limit 19:16, flagsdb 0; base 31:24; linear data segment descriptorLINEAR_SELequ$-gdt        dw 0xFFFF               ; limit 0xFFFFF (1 meg? 4 gig?)dw 0; base for this one is always 0db 0db 0x92; present,ring 0,data,expand-up,writable        db 0xCF                 ; page-granular (4 gig limit), 32-bitdb 0; code segment descriptorSYS_CODE_SELequ$-gdtgdt2:dw 0xFFFFdw 0; (base gets set above)db 0db 0x9A; present,ring 0,code,non-conforming,readabledb 0xCFdb 0; data segment descriptorSYS_DATA_SELequ$-gdtgdt3:dw 0xFFFFdw 0; (base gets set above)db 0db 0x92; present,ring 0,data,expand-up,writabledb 0xCFdb 0gdt_end:gdt_ptr:dw gdt_end - gdt - 1; GDT limitdd gdt; linear, physical address of GDT;/****************************************************************************;       file hello.c;****************************************************************************/#include  /* movedata() */#defineLINEAR_SEL0x08#defineSYS_DATA_SEL0x18int main(void){const char Msg[] = "h e l l o ";movedata(SYS_DATA_SEL, (unsigned)Msg,LINEAR_SEL, 0xB8000,sizeof(Msg));return 0;};/****************************************************************************;file build.bat;*****************************************************************************/nasm -f aout -o load.o load.asmgcc -c -O2 -Wall -g -o hello.o hello.cld -o pm.com -oformat binary -Ttext=0x100 load.o hello.o /djgpp/lib/libc.a

That's it

E-mail [e-mail] with questions
Or find me and others at alt.os.development