Re: [GIT] Networking

[Hannes Frederic Sowa]

Hi,

On Wed, Oct 28, 2015, at 15:27, Rasmus Villemoes wrote:
> On Wed, Oct 28 2015, Hannes Frederic Sowa <hannes@xxxxxxxxxxxxxxxxxxx>
> wrote:
> 
> > Hi Linus,
> >
> > On Wed, Oct 28, 2015, at 10:39, Linus Torvalds wrote:
> >> Get rid of it. And I don't *ever* want to see that shit again.
> >
> > I don't want to give up on that this easily:
> >
> > In future I would like to see an interface like this. It is often hard
> > to do correct overflow/wrap-around tests and it would be great if there
> > are helper functions which could easily and without a lot of thinking be
> > used by people to remove those problems from the kernel.
> 
> I agree - proper overflow checking can be really hard. Quick, assuming a
> and b have the same unsigned integer type, is 'a+b<a' sufficient to
> check overflow? Of course not (hint: promotion rules). And as you say,
> it gets even more complicated for signed types.
> 
> A few months ago I tried posting a complete set of fallbacks for older
> compilers (https://lkml.org/lkml/2015/7/19/358), but nothing really
> happened. Now I know where Linus stands, so I guess I can just delete
> that branch.

I actually like your approach of being type agnostic a bit more (in
comparison to static inline functions), mostly because of one specific
reason:

The type agnostic __builtin_*_overflow function even do the correct
things if you deal with types smaller than int. Imagine e.g. you want to
add to unsigned chars a and b,

unsigned char a, b;
if (a + b < a)
  goto overflow;
else
  a += b;

The overflow condition will never trigger, as the comparisons will
always be done in the integer domain and a + b < a is never true. I
actually think that this is easy to overlook and the functions should
handle that. The macro version does this quite nicely.

Bye,
Hannes
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/