Intro

I was working on a microkernel (CS 452: Real Time Programming), and I needed to do my compilations with a custom compiler, assembler and linker, so I could compile for ARM. However this required me to use an older version of gcc without the --with-as= flag, thus I needed to get the assembler to run as a seperate command.

Enter a complication: I needed this to run tests on my local machine (Mac), as well as on those of my partners (Windows running cygwin), I also needed this to compile on a remote Linux server, which could do the necessary cross-compilation.

My resulting makefile had the following basic structure.

# Create the assembly file so that we can use a seperate assembler in the next step
%.s: %.c
    $(CC) -S $(CFLAGS) -o $@ $<

# Use out assembler to compile an object file
%.o: %.s
    $(AS) $(ASFLAGS) -o $@ $<

# Combine the object files using the linker
%.elf: %.o
    $(LD) $(LDFLAGS) -Map $*.map -o $@ $*.o $(INC) -lgcc
    chmod a+r $@

First we generate the assembly files, next we assemble the resulting files into object files. Finally we produce the elf file that will represent my Kernel.

This will run properly on the linux machine, however when I try to use the same makefile to generate code locally, it fails.

Solution

The error I get looks like this.

test_cbuf.s:5:Unknown pseudo-op: .cfi_startproc
test_cbuf.s:9:Unknown pseudo-op: .cfi_def_cfa_offset

So the assemble results from gcc can't be read by the assembler?

It seems the gcc and the as progams that Mac has by default use differing standards.

The correct assembler to use uses a command like the following:

clang -c -x assembler $@

Now to solve my situation above:

• I can give a different AS = "clang -c -x assembler" variable to my makefile

  • But then I either need to call it that way always (a pain) or add different instructions

• I can add a simple bash script that calls the code above, one that is earlier in my $PATH

I opted for the second solution, mostly since this seems to be a more common use-case for me.