Re: fork pagesize patch

[Linus Torvalds <torvalds@osdl.org>]

On Tue, 16 Nov 2004, David Howells wrote:
> 
> > > Please don't do that. What you've done causes a divide-by-zero error to be
> > > emitted by the compiler if PAGE_SIZE > THREAD_SIZE. That's why I used the
> > > preprocessor in the first place.
> > 
> > What kind of broken compiler are _you_ using? Fix your compiler.
> 
> Sorry... I meant warning not error. It doesn't actually stop it building a
> working kernel, but gcc _does_ complain, and not unreasonably, I think.

I think it _is_ unreasonable. It's like doing

	if (a)
		x /= a;

and the compiler complaining that "a" can be zero. The above is perfectly
reasonable code and may well have a constant zero as part of inlining or
macro expansion. A compiler that does that is being silly. It's being
_especially_ silly since it evaluates to

	if (0)
		..

at compile time in this case. Yeah, yeah, sparse gets it wrong too, but 
only if the left-hand side is zero as well.

But yes, at least the compiler isn't totally buggy if it's just a warning. 
After all, you can warn about anything you like ;)

Anyway, to make it not warn, why not change it to

	max_threads = mempages / (8*THREAD_SIZE/PAGE_SIZE);

instead, and be done with it? If the thread size is _that_ small that we 
still divide by zero, color me impressed.

		Linus