Re: [PATCH 07/15] kmemleak: Add memleak_alloc callback from alloc_large_system_hash

[Catalin Marinas <catalin.marinas@arm.com>]

On Fri, 2008-12-12 at 09:17 -0800, Dave Hansen wrote:
> On Fri, 2008-12-12 at 17:04 +0000, Catalin Marinas wrote:
> > It looks to me like alloc_large_system_hash() could also be called at
> > some later point and it may even invoke __vmalloc() if hashdist is set.
> > So I would prefer not to introduce another hook and additional if's to
> > know which one to call. BTW, I think the callback should actually be (to
> > avoid duplicating the vmalloc call, with proper comment):
> > 
> >         if (!hashdist)
> >                 memleak_alloc(table, size, 1, GFP_ATOMIC);
> 
> Does memleak_alloc() detect if it gets called twice on the same memory?

It does, and panics (disables itself). I think if this happens it is a
kmemleak bug and some hook is missing or added twice.

> Also, is alloc_large_system_hash() contained in the tests that you can
> compile for kmemleak?

No. This memleak_alloc() callback was mainly added to avoid plenty of
false reports from various parts of the kernel (especially the IPv4
stack). It wasn't really meant to track the allocated hash blocks.

-- 
Catalin