Re: [PATCH] fix check warnings in drivers/macintosh

[Linus Torvalds]

On Sat, 7 Jun 2003, Paul Mackerras wrote:
> 
> This patch removes the warnings that the `check' program came up with
> in drivers/macintosh.  This involves adding __user in various places
> and fixing some non-ANSI function definitions for functions that take
> no arguments.

Thanks, but please avoid doing things like this:

> diff -urN linux-2.5/drivers/macintosh/macserial.c pmac-2.5/drivers/macintosh/macserial.c
> --- linux-2.5/drivers/macintosh/macserial.c	2003-06-07 08:57:42.000000000 +1000
> +++ pmac-2.5/drivers/macintosh/macserial.c	2003-06-07 14:35:34.000000000 +1000
> @@ -1496,7 +1496,7 @@
>  			if (c <= 0)
>  				break;
>  
> -			c -= copy_from_user(tmp_buf, buf, c);
> +			c -= copy_from_user(tmp_buf, (void __user *) buf, c);

The problem here is that "buf" has the wrong type, and you're hiding it 
with a cast.

That's pointless - yes, it avoids a warning from "check", but the thing 
is, that warning was _correct_, and you just hid the problem.

If "buf" really is a user pointer, then it should have been marked that 
way in the first place.

I understand why you did it the way you did: the tty functions are badly 
designed, and the same function is used for both user and kernel pointers.

And that's something that check does and _should_ warn about. The function 
definition is crap, and I'd rather have the warning there and hope some 
tty layer person comes along and makes the user/kernel case into separate 
functions (they don't share that much code anyway, since most of the 
function is

	if (from_user) {
		user case
	} else {
		kernel case
	}

anyway.

I hate these kinds of "flag" arguments anyway. If a function does two
different things, it should be two different functions. Yeah, I know, the
"gfp_mask" thing for the memory allocators are the same way, and yeah,
that was a design mistake too.

		Linus