Re: SLUB: Return ZERO_SIZE_PTR for kmalloc(0)

[Linus Torvalds <torvalds@linux-foundation.org>]

On Mon, 4 Jun 2007, Pekka Enberg wrote:
> 
> Ok, makes sense. I guess I might as well throw my suggestion in the
> mix. Lets create a new kmalloc cache for zero-length objects where
> object size is zero but there are regular red-zones on both sides.

Well, the red-zones won't catch readers, and more importantly, even for 
writers they are *really* inconvenient, because it will just tell you 
something bad happened, it won't tell you *where* it happened.

Since comparing the addresses of two zero-sized allocations is insane and 
not done _anyway_, it's just much better to return an invalid address.

The thing is, why *should* we care about comparing addresses? We'll give 
the right result (you got many perfectly separate allocations, they're 
just zero bytes apart, exactly like you asked for!). The fact that C++ has 
some semantics for it is not a good argument - C++ is a broken language, 
and it's not the language we use for the kernel anyway.

		Linus