2001-11-28 22:39:43

by Justin T. Gibbs

Subject: Kernel's X86 ffs() doesn't work on constants.

If you attempt to call ffs(SOME_CONSTAT) in an x86 kernel under
Linux, you get messages like this:

{standard input}: Assembler messages:
{standard input}:14864: Error: suffix or operands invalid for `bsf'

I'm not enough of a GCC asm syntax guru to understand why the
compiler/assembler doesn't handle this, but it is hightly anoying.

"Why not just code in the constant bit offset?", you ask? If
the constant the bit offset is based on is ever changed, I must
recognize that the change occured and change the second constant.
For constants that are maintained outside of my code, I'd rather
code the dependency once and let the compiler ensure that the constants
are in sync.

--
Justin

2001-11-28 23:01:27

by H. Peter Anvin

Subject: Re: Kernel's X86 ffs() doesn't work on constants.

Followup to: <[email protected]>
By author: "Justin T. Gibbs" <[email protected]>
In newsgroup: linux.dev.kernel
>
> If you attempt to call ffs(SOME_CONSTAT) in an x86 kernel under
> Linux, you get messages like this:
>
> {standard input}: Assembler messages:
> {standard input}:14864: Error: suffix or operands invalid for `bsf'
>
> I'm not enough of a GCC asm syntax guru to understand why the
> compiler/assembler doesn't handle this, but it is hightly anoying.
>
> "Why not just code in the constant bit offset?", you ask? If
> the constant the bit offset is based on is ever changed, I must
> recognize that the change occured and change the second constant.
> For constants that are maintained outside of my code, I'd rather
> code the dependency once and let the compiler ensure that the constants
> are in sync.
>

Try changing the "g" in the definition of ffs() (asm/bitops.h) to
"rm"; the "g" constrains incorrectly allows immediate operands.

-hpa
--
<[email protected]> at work, <[email protected]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt <[email protected]>