.align 3 not working gnu as spits out errors

[ITchimp]

linux 0.01's boot.s has a assembler directive stating .align that is causing assembler to spit out error message... I can try to replace 3 with 4... the error go away... but based on the gas manual, .align 3 is a valid statement... was compiling in a vm ubuntu 64 bit machine...


thanks for the help...

[alexfru]

Are you trying to align to a multiple of 3 instad of a multiple of 8?

[iansjack]

When using elf format, the default for gas nowadays, the argument to .align must be a power of 2. The behaviour is different when using a.out format (which that version of Linux would have used) where the argument specifies the power of 2.

So .align 3 (a.out) is equivalent to .align 8 (elf).

[ITchimp]

Is there a way to choose zmagic/amagic over elf? What switch/flag is there? there doesn't seem to have any at first glance... just curious...

also by the same token, is .align 4 in a.out equivalent to .align 16 in elf?

[alexfru]

Is there a way to choose zmagic/amagic over elf? What switch/flag is there? there doesn't seem to have any at first glance... just curious...

Try something like this (you'll likely need the object file, aout.o, already in a.out format, not ELF):

Code: Select all

ld -e __start --oformat=a.out-i386 --omagic aout.o -o aout.out
ld -e __start --oformat=a.out-i386 --nmagic aout.o -o aout.out
ld -e __start --oformat=a.out-i386 --qmagic aout.o -o aout.out

also by the same token, is .align 4 in a.out equivalent to .align 16 in elf?

I'd expect so.

[zaval]

When using elf format, the default for gas nowadays, the argument to .align must be a power of 2. The behaviour is different when using a.out format (which that version of Linux would have used) where the argument specifies the power of 2.

So .align 3 (a.out) is equivalent to .align 8 (elf).

but don't forget, as always with this mess, - this is valid only for x86.

for example on both arm, mips and ppc gccs emitting elf, .align 3 means "align on 2^3". which is sane imho.

[iansjack]

For compatible behaviour across architectures use .balign.

Of course, Linux 0.01 only supported x86 processors.

[zaval]

For compatible behaviour across architectures use .balign.

Of course, Linux 0.01 only supported x86 processors.

I noted that just in case, for the OP, maybe he/she is going to develop not only on x86.

I prefer the meaning where N in .align N means power of two. so .balign is BS. .p2align is better. of course if one wants to mess with gcc specific directives (which .balign an .p2align are).

[zaval]

For compatible behaviour across architectures use .balign.

Of course, Linux 0.01 only supported x86 processors.

I noted that just in case, for the OP, maybe he/she is going to develop not only on x86.

I prefer the meaning where N in .align N means log2(alignment), not the alignment itself. so .balign is BS. .p2align is better. of course if one wants to mess with gcc specific directives (which .balign an .p2align are).