Re: recent IDE regression

[Ben Dooks <ben-linux@fluff.org>]

On Fri, Jul 25, 2008 at 09:37:25AM -0700, Linus Torvalds wrote:
> 
> 
> On Fri, 25 Jul 2008, Ben Dooks wrote:
> > 
> > personally, i would much prefer to see the loop being less evil
> > like:
> > 
> > 	for (p = s; p < end; p += 2)
> > 		be16_to_cpus((u16 *)p);
> 
> Well, in this case, the code actually depends on 'p' being back at the 
> start of the buffer by the end of it all, so it would need some more 
> changes than that. 

I'm not adverse to the loop running backwards, it is just that we end
up evaluating macro with side-effects to the pointer itself, which is
"just asking for trouble". 

		for (p = end ; p != s; p -= 2)
			be16_to_cpus((u16 *)p);

This removes the side-effect of be16_to_cpus(), which means that the
change to p is moved back into the for statement; After all, the for
construct has three code blocks for a reason!

I just wonder what would happen if be16_to_cpus(x) evaluated 'x' more
than once... it would be difficult to find problems introduced if this
happened as it would fail to hang, just not do the 'right' thing... 

> But yes, I applied David's patch, but I _also_ suspect that we would be 
> better off without code that does horrid things like casts and assignments 
> inside the function arguments.
> 
> So it would be nice to re-code that loop to be more readable. But due to 
> the reliance of 'p' being 's' after the loop, the minimal patch would be 
> something like the appended.

As noted above, I don't have any problems with the looping going backwards,
just the problems with potential side-effects.
 
> Bartlomiej - take this or not, I'm not going to commit it - I haven't 
> tested it, nor do I even have any machines that would trigger it. So this 
> is more a "maybe something like this" than anything else.
> 
> 		Linus
> 
> ---
>  drivers/ide/ide-iops.c |    8 +++++---
>  1 files changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/ide/ide-iops.c b/drivers/ide/ide-iops.c
> index 8aae917..ae03151 100644
> --- a/drivers/ide/ide-iops.c
> +++ b/drivers/ide/ide-iops.c
> @@ -506,14 +506,16 @@ void ide_fix_driveid (struct hd_driveid *id)
>  
>  void ide_fixstring (u8 *s, const int bytecount, const int byteswap)
>  {
> -	u8 *p = s, *end = &s[bytecount & ~1]; /* bytecount must be even */
> +	u8 *p, *end = &s[bytecount & ~1]; /* bytecount must be even */
>  
>  	if (byteswap) {
>  		/* convert from big-endian to host byte order */
> -		for (p = end ; p != s;)
> -			be16_to_cpus((u16 *)(p -= 2));
> +		for (p = s ; p != end ; p += 2)
> +			be16_to_cpus((u16 *) p);
>  	}
> +
>  	/* strip leading blanks */
> +	p = s;
>  	while (s != end && *s == ' ')
>  		++s;
>  	/* compress internal blanks and strip trailing blanks */

-- 
Ben

Q:      What's a light-year?
A:      One-third less calories than a regular year.