How to convert ld -Ttext option to a linker script?

Question

I've found this tutorial on creating an operating system, and I'm trying to convert the linking part in the make file into a linker script.

Here is the tutorial: https://github.com/ghaiklor/ghaiklor-os-gcc

Here is the make file:

SOURCES = $(shell find cpu drivers include kernel libc -name '*.c')
HEADERS = $(shell find cpu drivers include kernel libc -name '*.h')
OBJ = ${SOURCES:.c=.o cpu/interrupt.o}

ASM = nasm 
CC = gcc  
LD = ld -m elf_i386 
CFLAGS = -g -ffreestanding -Wall -Wextra -fno-exceptions -m32 -std=c11 -fno-pie

ifeq ($(shell uname -s),Darwin)     
    CC = i386-elf-gcc   
    LD = i386-elf-ld 
endif

all: os-image.bin

run: all    
    qemu-system-i386 os-image.bin

clean:  
    rm -rf *.dis *.o *.elf  
    rm -rf *.bin os-image.bin boot/*.bin kernel/*.bin   
    rm -rf boot/*.o cpu/*.o drivers/*.o kernel/*.o libc/*.o

os-image.bin: boot/boot.bin kernel/kernel.bin   
    cat $^ > os-image.bin

boot/boot.bin: boot/boot.asm    
    ${ASM} $< -f bin -o $@

kernel/kernel.bin: boot/kernel_entry.o ${OBJ}   
     ${LD} -o $@ -Ttext 0x1000 $^ --oformat binary

# ${LD} -o $@ -Tlinker.ld
# ${LD} -o $@ -Ttext 0x1000 $^ --oformat binary

%.o: %.c ${HEADERS}     ${CC} ${CFLAGS} -c $< -o $@

%.o: %.asm  ${ASM} $< -f elf -o $@

%.bin: %.asm    ${ASM} $< -f bin -o $@

Here is my attempt at creating a linker script for the linking phase of this make file:

ENTRY(_start)
OUTPUT_FORMAT(binary)
INPUT(boot/kernel_entry.o cpu/idt.o cpu/interrupt.o cpu/isr.o cpu/ports.o cpu/timer.o drivers/keyboard.o drivers/screen.o libc/mem.o libc/string.o libc/common.o kernel/kernel.o)
OUTPUT(kernel/kernel.bin)
SECTIONS
{
    . = 0x1000;
    .text : { *(.text) }
    end = .; _end = .; __end = .;
}

There is no .data or .bss for the boot/kernel_entry.o and this is why I did not include them into the linker script. I know that -Ttext 0x1000 is where .text section is suppose to be loaded and thats why I set the counter to start at the address of 0x1000. When I run the system with the new linking command in the makefile ${LD} -o $@ -Tlinker.ld the system isn't working like normal, so I'm doing something wrong. I've tried adding simple .data and .bss sections and all other sorts of things but still can't manage to get the thing to work correctly with a linking script. Any help would be great.

Thanks.

Just to confirm, it *is* working when you build with the original Makefile? So it's not something wrong with your test setup, but actually a different binary. — Peter Cordes, Mar 10 '19 at 08:40
Run objdump -h on the one that works and the one that doesn’t and compare. (You can add that output to the question if you need help interpreting it.) — prl, Mar 10 '19 at 09:43
In what way doesn't the system work as normal? DO you get a disk error, does the shell not appear? — Michael Petch, Mar 10 '19 at 11:40
One concern I'd have is how your boot.asm reads the kernel sectors.It hard codes the value 15 as the number of sectors to read. If your generated kernel.bin is larger than 7680 (15*512) bytes you could run into problems with not everything loaded in memory. If you have two few sectors then QEMU may barf with a read error because with hard drive emulation it may not be keen on reading sectors that may not be physically present in the disk image. — Michael Petch, Mar 10 '19 at 11:46
You say `There is no .data or .bss for the boot/kernel_entry.o` .The problem is that kernel_entry.o may not have a .data or .bss section but the other object files may have them. But even then, despite not listing those sections they will still be processed and added after the sections your linker script is processing. I also don't recommend having an INPUT line in your linker script. I'd be inclined to remove it and pass all the objects through LD in the Makefile and use `${LD} -o $@ -Tlinker.ld $^` instead. — Michael Petch, Mar 10 '19 at 11:54
I'd also recommend not using `OUTPUT` in the linker script either and allow the linker command line to specify the output file. This all makes things more generic. — Michael Petch, Mar 10 '19 at 11:56
Thank you for all the suggestions everyone. @MichaelPetch how the system differs when I try and write the linker script vs when I don't is that nothing shows up. I found out last night that the clear_screen() function runs and then nothing else happens on the screen. It is just a blank screen with a cursor blinking in the top left. — AaronV77, Mar 10 '19 at 15:19
As for reading in all the sectors from disk, I've always been concerned with why the number 15 was used and if it was actually enough for the whole system but with little other examples out there I assumed that it was correct. The only other way to figure out if it is enough is to get the size of the files used and do the calculation on if it is or not, I guess. As for `There is no .data or .bss for the boot/kernel_entry.o` that was my greatest fear that I had to do it for the whole application, because I haven't been able to find any answers yet. — AaronV77, Mar 10 '19 at 15:23
Your Makefile suggests you have made changes to the tutorial and that you are trying to build a 32-bit kernel instead of a 64-bit one? — Michael Petch, Mar 10 '19 at 17:42
Yes sir. I just wanted to focus on Real and Protected because of the learning curve. Then once I learned everything sufficiently I would then go into long mode. All I did to change the tutorial was remove the long mode section of it, and get the system to actually build since the original version was not compiling on my Ubuntu VM. — AaronV77, Mar 10 '19 at 18:36
Let us [continue this discussion in chat](https://chat.stackoverflow.com/rooms/189778/discussion-between-aaronv77-and-michael-petch). — AaronV77, Mar 10 '19 at 20:41

Michael Petch · Accepted Answer · 2019-03-11T01:15:48.133

The tutorial you linked to was for a 64-bit code.Your Makefile and subsequent comments suggest you are trying to modify it to assemble/compile/run as a 32-bit kernel. I have placed a copy of the revised project discussed below on my wesbite. A compressed tarball can be downloaded from here.

The tutorial you have is rather dumb when it comes to loading the kernel into memory. It requires you to know how many sectors to read and hard code the value. Getting this wrong can cause unusual behaviour. Rather than hard coding the value you can get NASM to include kernel.bin inside boot.bin so that the bootloader can compute the number of sectors to read at assembly time. Not all emulators and real machines support multi-track reads so I'd modify the bootloader to read one sector at a time using LBA addressing. To learn more about CHS to LBA conversion calculations you can see my other Stackoveflow answer on the topic. Modify boot/boot.asm to be:

STAGE2_ABS_ADDR  equ 0x01000
STAGE2_RUN_SEG   equ 0x0000
STAGE2_RUN_OFS   equ STAGE2_ABS_ADDR
                                ; Run stage2 with segment of 0x0000 and offset of 0x1000

STAGE2_LOAD_SEG  equ STAGE2_ABS_ADDR>>4
                                ; Segment to start reading Stage2 into
                                ;     right after bootloader

STAGE2_LBA_START equ 1          ; Logical Block Address(LBA) Stage2 starts on
                                ;     LBA 1 = sector after boot sector
STAGE2_LBA_END   equ STAGE2_LBA_START + NUM_STAGE2_SECTORS
                                ; Logical Block Address(LBA) Stage2 ends at
DISK_RETRIES     equ 3          ; Number of times to retry on disk error

bits 16
ORG 0x7c00

; Include a BPB (1.44MB floppy with FAT12) to be more comaptible with USB floppy media
; %include "bpb.inc"

boot_start:
    xor ax, ax                  ; DS=SS=ES=0 for stage2 loading
    mov ds, ax
    mov ss, ax                  ; Stack at 0x0000:0x0000
                                ;     (grows down fromtopof1st 64KiB segment)
    mov sp, 0x0000
    cld                         ; Set string instructions to use forward movement

    ; Read Stage2 1 sector at a time until stage2 is completely loaded
load_stage2:
    mov [bootDevice], dl        ; Save boot drive
    mov bx, MSG_LOAD_KERNEL
    call print_string

    mov di, STAGE2_LOAD_SEG     ; DI = Current segment to read into
    mov si, STAGE2_LBA_START    ; SI = LBA that stage2 starts at
    jmp .chk_for_last_lba       ; Check to see if we are last sector in stage2

.read_sector_loop:
    mov bp, DISK_RETRIES        ; Set disk retry count

    call lba_to_chs             ; Convert current LBA to CHS
    mov es, di                  ; Set ES to current segment number to read into
    xor bx, bx                  ; Offset zero in segment

.retry:
    mov ax, 0x0201              ; Call function 0x02 of int 13h (read sectors)
                                ;     AL = 1 = Sectors to read
    int 0x13                    ; BIOS Disk interrupt call
    jc .disk_error              ; If CF set then disk error

.success:
    add di, 512>>4              ; Advance to next 512 byte segment (0x20*16=512)
    inc si                      ; Next LBA

.chk_for_last_lba:
    cmp si, STAGE2_LBA_END      ; Have we reached the last stage2 sector?
    jl .read_sector_loop        ;     If we haven't then read next sector

.stage2_loaded:
    call switch_to_pm

.disk_error:
    xor ah, ah                  ; Int13h/AH=0 is drive reset
    int 0x13
    dec bp                      ; Decrease retry count
    jge .retry                  ; If retry count not exceeded then try again

error_end:
    ; Unrecoverable error; print drive error; enter infinite loop
    mov bx, diskErrorMsg        ; Display disk error message
    call print_string
    cli
.error_loop:
    hlt
    jmp .error_loop

;    Function: lba_to_chs
; Description: Translate Logical block address to CHS (Cylinder, Head, Sector).
;              Works for all valid FAT12 compatible disk geometries.
;
;   Resources: http://www.ctyme.com/intr/rb-0607.htm
;              https://en.wikipedia.org/wiki/Logical_block_addressing#CHS_conversion
;              https://stackoverflow.com/q/45434899/3857942
;              Sector    = (LBA mod SPT) + 1
;              Head      = (LBA / SPT) mod HEADS
;              Cylinder  = (LBA / SPT) / HEADS
;
;      Inputs: SI = LBA
;     Outputs: DL = Boot Drive Number
;              DH = Head
;              CH = Cylinder (lower 8 bits of 10-bit cylinder)
;              CL = Sector/Cylinder
;                   Upper 2 bits of 10-bit Cylinders in upper 2 bits of CL
;                   Sector in lower 6 bits of CL
;
;       Notes: Output registers match expectation of Int 13h/AH=2 inputs
;
lba_to_chs:
    push ax                     ; Preserve AX
    mov ax, si                  ; Copy LBA to AX
    xor dx, dx                  ; Upper 16-bit of 32-bit value set to 0 for DIV
    div word [sectorsPerTrack]  ; 32-bit by 16-bit DIV : LBA / SPT
    mov cl, dl                  ; CL = S = LBA mod SPT
    inc cl                      ; CL = S = (LBA mod SPT) + 1
    xor dx, dx                  ; Upper 16-bit of 32-bit value set to 0 for DIV
    div word [numHeads]         ; 32-bit by 16-bit DIV : (LBA / SPT) / HEADS
    mov dh, dl                  ; DH = H = (LBA / SPT) mod HEADS
    mov dl, [bootDevice]        ; boot device, not necessary to set but convenient
    mov ch, al                  ; CH = C(lower 8 bits) = (LBA / SPT) / HEADS
    shl ah, 6                   ; Store upper 2 bits of 10-bit Cylinder into
    or  cl, ah                  ;     upper 2 bits of Sector (CL)
    pop ax                      ; Restore scratch registers
    ret

%include "boot/print/print_string.asm"
%include "boot/pm/switch_to_pm.asm"
%include "boot/pm/gdt.asm"

bits 32

begin_pm:
jmp 0x1000

; Uncomment these lines if not using a BPB (via bpb.inc)
%ifndef WITH_BPB
numHeads:        dw 2         ; 1.44MB Floppy has 2 heads & 18 sector per track
sectorsPerTrack: dw 18
%endif

bootDevice:      db 0x00
diskErrorMsg:    db "Unrecoverable disk error!", 0
MSG_PROT_MODE db "Landed in 32-bit Protected Mode", 0
MSG_LOAD_KERNEL db "Loading kernel into memory", 0



; Pad boot sector to 510 bytes and add 2 byte boot signature for 512 total bytes
TIMES 510-($-$$) db  0
dw 0xaa55

; Beginning of stage2. This is at 0x1000 and will allow your stage2 to be 32.5KiB
; before running into problems. DL will be set to the drive number originally
; passed to us by the BIOS.

NUM_STAGE2_SECTORS equ (stage2_end-stage2_start+511) / 512
                                ; Number of 512 byte sectors stage2 uses.

stage2_start:
    ; Insert stage2 binary here. It is done this way since we
    ; can determine the size(and number of sectors) to load since
    ;     Size = stage2_end-stage2_start
    incbin "kernel/kernel.bin"

; End of stage2. Make sure this label is LAST in this file!
stage2_end:

Your Makefile can use some cleaning up. I noticed you added interrupts.o to OBJ manually (since it is an ASM file). I'd recommend collecting all the kernel related .asm files and adding them to the OBJ list. I recommend changing it to this:

SOURCES  = $(shell find cpu drivers include kernel libc -name '*.c')
KERN_ASM = $(shell find cpu drivers include kernel libc -name '*.asm')
HEADERS  = $(shell find cpu drivers include kernel libc -name '*.h')
OBJ      = ${SOURCES:.c=.o} ${KERN_ASM:.asm=.o}

ASM = nasm
CC = gcc
LD = ld -m elf_i386
OBJCOPY = objcopy

CFLAGS = -g -ffreestanding -Wall -Wextra -fno-exceptions -m32 -std=c11 -fno-pic \
         -fno-asynchronous-unwind-tables

ifeq ($(shell uname -s),Darwin)
        CC = i386-elf-gcc
        LD = i386-elf-ld
        OBJCOPY = i386-elf-objcopy
endif

all: os-image.bin

run: all
        qemu-system-i386 os-image.bin

clean:
        rm -rf *.dis *.o *.elf
        rm -rf *.bin os-image.bin boot/*.bin kernel/*.bin
        rm -rf boot/*.o cpu/*.o drivers/*.o kernel/*.o libc/*.o

# Make a 1.44MiB disk image. Can work for HDA and FDA booting
os-image.bin: kernel/kernel.bin boot/boot.bin
        dd if=/dev/zero of=$@ bs=1024 count=1440
        dd if=$(word 2,$^) of=$@ conv=notrunc

boot/boot.bin: boot/boot.asm
        ${ASM} $< -f bin -o $@

kernel/kernel.bin: kernel/kernel.elf
        ${OBJCOPY} -O binary $^ $@

kernel/kernel.elf: ${OBJ}
        ${LD} -o $@ -Tlinker.ld $^

%.o: %.c ${HEADERS}
        ${CC} ${CFLAGS} -c $< -o $@

%.o: %.asm
        ${ASM} $< -f elf -o $@

%.bin: %.asm
        ${ASM} $< -f bin -o $@

This make file uses DD to create a 1.44MiB floppy image that can be used as a floppy or hard drive disk image. You will notice I have removed kernel_entry.asm from the explicit dependency list. For this new Makefile to work You must MOVE boot/kernel_entry.asm to kernel/kernel_entry.asm. Ensure you REMOVE boot/kernel_entry.asm.

Modify kernel/kernel_entry.asm to use section .text.entry and zero out the BSS. It can look like this:

global _start

bits 32
extern kernel_main
extern __bss_start
extern __bss_sizel

section .text.entry

_start:
  ; Zero out the BSS memory area a DWORD at a time
  ; since the memory isn't guaranteed to already be zero
  xor eax, eax
  mov ecx, __bss_sizel
  mov edi, __bss_start
  rep stosd

  ; Call C entry point of kernel
  call kernel_main
  jmp $

The linker script linker.ld that works with these changes is as follows:

OUTPUT_FORMAT(elf32-i386)

SECTIONS {
    . = 0x1000;

    .text : SUBALIGN(4)
    {
        *(.text.entry)       /* Ensure .text.entry appears first */
        *(.text*)
        *(.rodata*)
        *(.data)
    }

    .bss : SUBALIGN(4) {
        __bss_start = .;
        *(COMMON)            /* all COMMON sections from all files */
        *(.bss)              /* all BSS sections from all files */
    }
    . = ALIGN(4);
    __bss_end = .;
    __bss_sizeb = __bss_end - __bss_start;       /* BSS size in bytes */
    __bss_sizel = (__bss_end - __bss_start) / 4; /* BSS size in longs/DWORDs */

    /DISCARD/ : {            /* Remove Unneeded sections */
        *(.eh_frame);
        *(.comment);
    }

    end = .; _end = .; __end = .;
}

It handles all the normal segments you'd generally see in an ELF file for a basic OS. It also uses a special .entry.text section to ensure the code in kernel/kernel_entry.asm comes first.

How to convert ld -Ttext option to a linker script?

1 Answers1

Linked