Re: Q, slab, kmemleak_erase() and redzone?

[Pekka Enberg <penberg@cs.helsinki.fi>]

Catalin Marinas kirjoitti:
> On Tue, 2009-12-01 at 11:49 +0000, Pekka Enberg wrote:
>> hooanon05@yahoo.co.jp kirjoitti:
>>> Pekka Enberg:
>>>> We are setting an element in the per CPU array to NULL so the the
>>>> kmemleak code in ____cache_alloc() is safe. Red-zoning is done at the
>>>> _object_ which is not touched by kmemleak. Looking at the oops, it
>>>> does seem likely that you have a bug in your module (or in some other
>>>> part of the kernel).
>>> Thanks for reply.
>>> In ____cache_alloc(), the variable 'ac' is assigned before
>>> cache_alloc_refill() call, and it is used for the parameter of
>>> kmemleak_erase(). The value may be changed by cache_alloc_refill(),
>>> isn't it?
>> No. The pointer returned by cpu_cache_get() is not changed by
>> cache_alloc_refill(). The contents of the array might change, yes. That
>> said, we should check if objp is NULL before calling kmemleak_erase().
> 
> Possibly but I don't understand why that's needed. The kmemleak_erase()
> call just sets the ac->entry[ac->avail] to NULL. If ac->avail is 0, it
> doesn't cause any harm.

No, you are absolutely correct. Can you please send an updated patch to 
Catalin that adds a comment on top of the cpu_cache_get() call that 
explains why we need it there?

			Pekka