Re: SLUB: Return ZERO_SIZE_PTR for kmalloc(0)

[Linus Torvalds]

On Mon, 4 Jun 2007, Alan Cox wrote:

> > The thing is, why *should* we care about comparing addresses? We'll give 
> 
> Because people use it to tell objects apart. All over the kernel we do
> things like
> 
> 	if (inode1 == inode2)

But that only makes sense if your objects have meaning, which is not 
possible with a zero-sized object.

Let's take an example: one of the few reasons to check for equality (or 
inequality) is for locking order.

So on UP, the lock goes away, and let's say that you (insanely) only have 
a spinlock in the structure in question, so you have a zero-sized 
structure that you want to test ordering on because you want to avoid 
ABBA deadlocks.

So you write your code as

   double_lock()
   {
	if (ptr1 == ptr2) {
		lock(ptr1->lock);
		return;
	}
	if (ptr1 < ptr2) {
		lock(ptr1->lock);
		lock(ptr2->lock);
		return;
	}
	lock(ptr2->lock);
	lock(ptr1->lock);
    }

and the interesting thing is that this actually *works* even for the case 
where the lock has gone away: even though ptr1/ptr2 are always equal.

In other words, the only _valid_ reasons to compare pointers like this end 
up degenerating into working cases even for a zero-sized pointer.

The exception is if you use the memory allocator as a "ID allocator", but 
quite frankly, if you use a size of zero, it's your own damn problem. 
Insane code is not an argument for insane behaviour.

If people can't be bothered to create a "random ID generator" themselves, 
they had damn well better use "kmalloc(1)" rather than "kmalloc(0)" to get 
a unique cookie. Asking the allocator to do something idiotic because some 
idiot thinks a memory allocator is a cookie allocator is just crazy.

I can understand that things like user-level libraries have to take crazy 
people into account, but the kernel internal libraries definitely do not.

(Right now we warn once for zero-sized allocations anyway, and all the 
cases we've found so far are either bugs that would have been found with 
ZERO_ALLOC_PTR or would have been perfectly fine with it, so I don't think 
anybody really _is_ that insane in the kernel)

		Linus
-
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/