New versions of LD cannot take ELF files as input to link - linux

Question

I have been working on a 32-bit operating system project written in C using GNU LD version 2.34 as the linker.
As part of the build process, I use the following command:
ld -m elf_i386 -nostdlib -T ld/loader.ld build/bootloader/loader.o build/bootloader/loader.elf -o loader_full.elf
When using GNU LD version 2.34, this command succeeds. However, when I use a version higher than this, I get the following error:
ld: cannot use executable file 'build/bootloader/loader.elf' as input to a link
If necessary, here is the full Makefile script:
CC_FLAGS = -g -m32 -ffreestanding -nostartfiles -nostdlib -fno-stack-protector
LD_FLAGS = -m elf_i386 -nostdlib
CC := gcc ${CC_FLAGS}
LD := ld ${LD_FLAGS}
BOOTLOADER_DRIVERS = kernel/drivers/disk/ata.c kernel/drivers/io/screen.c kernel/drivers/utils/mem.c kernel/drivers/utils/ports.c
raw: prep os-image.bin
convert_vmdk: os-image.vmdk
all: prep os-image.bin
ESFS_raw_write: ESFS_raw_write.c
gcc $^ -o $#
# run OS in QEMU
run:
qemu-system-i386 -drive format=raw,file=os-image.bin
# assemble boot sector
build/bootloader/boot_sect.bin: boot/boot_sect.asm
nasm $^ -f bin -o $#
# compile second stage bootloader
build/bootloader/loader_2.o build/drivers/*.o: boot/*.c ${BOOTLOADER_DRIVERS}
${CC} -c $^
mv loader.o build/bootloader/loader_2.o
mv *.o build/drivers/
# link object files into kernel loader
build/bootloader/loader.elf: build/bootloader/loader_2.o build/drivers/*.o
${LD} -T ld/loader.ld $^ -o $#
rm build/bootloader/loader_2.o
# assemble first stage bootloader
build/bootloader/loader.o: boot/loader.asm
nasm $^ -f elf -o $#
# link first and second stage bootloaders
build/bootloader/loader_full.elf: build/bootloader/loader.o build/bootloader/loader.elf
${LD} -T ld/loader.ld $^ -o $#
# make bootloader binary file
build/bootloader/loader.bin: build/bootloader/loader_full.elf
objcopy $^ -O binary $#
./scripts/pad_loader.sh
# clean up unnecessary files
# rm build/bootloader/*.o build/bootloader/*.elf build/drivers/*.o
build/kernel/interrupt.o: kernel/cpu/interrupt.asm
nasm $^ -f elf -o $#
# compile kernel & write to 10MB raw drive image
build/kernel/hdd.bin: kernel/drivers/*/*.c kernel/cpu/*.c kernel/libc/*.c build/kernel/interrupt.o kernel/*.c
${CC} $^ -o build/kernel/kernel.o -T ld/kernel.ld
./scripts/write_kernel_to_drive.sh
# concat 3 boot stages into os-image file
os-image.bin: build/bootloader/boot_sect.bin build/bootloader/loader.bin build/kernel/hdd.bin
cat $^ > $#
os-image.vmdk:
VBoxManage convertfromraw os-image.bin os-image.vmdk --format VMDK
VBoxManage internalcommands sethduuid /home/tim/Dev/OSDev/os-image.vmdk 6372c00a-a62e-4241-9a21-90fa4c22f019
# prepare directory structure for build process
prep:
mkdir -p build/bootloader
mkdir -p build/drivers
mkdir -p build/kernel
# clean up build files and os-image binary
clean:
-rm -rf build/
-rm *.bin *.vmdk
-rm ESFS_raw_write
The project is also on GitHub, so you can build it for yourself:
http://github.com/TimCve/OSDev.git

Answer 1

I have just come up with a similar problem.
It seems ld version GNU ld (GNU Binutils) 2.36.1 links the file into an executable format.
To fix it, add -r in your ld command to output a relocatable format file.
From the ld manual:
-r
--relocateable
Generate relocatable output--i.e., generate an output file that can in turn serve as input to ld. This is often called partial linking. As a side effect, in environments that support standard Unix magic numbers, this option also sets the output file's magic number to OMAGIC. If this option is not specified, an absolute file is produced. When linking C++ programs, this option will not resolve references to constructors; to do that, use -Ur. This option does the same thing as -i.

Answer 2

It's not that ld can't take ELF files as input. It's that it won't take executable files as input. The error is:
ld: cannot use executable file '...' as input to a link
See https://sourceware.org/bugzilla/show_bug.cgi?id=26223
Does it work to simply use chmod -x $# after building the .elf file?

Answer 3

In my (eclipse-based) multicore MCUXpresso project, I had to add this line to my post-build steps to modify a byte inside of the generated "*.o" file so the linker step which used that file wouldn't complain anymore. This happened after I updated to version 11.6 which includes the LD version mentioned previously. I originally found this on the forums for NXP MCUXpresso, but I assume it will work similarly for just about anyone:
dd if=/dev/zero of="${BuildArtifactFileName}.o" bs=1 seek=16 conv=notrunc count=2
NXP MCUXpresso fix