From: ext Catalin Marinas <catalin.marinas@xxxxxxx>
Subject: Re: [PATCH v2 0/3] kmemleak: Fix false positive with special scan
Date: Wed, 2 Jun 2010 16:12:29 +0200
> On Wed, 2010-06-02 at 12:34 +0100, Hiroshi DOYU wrote:
>> From: ext Catalin Marinas <catalin.marinas@xxxxxxx>
>> > Can we not add a new prio tree (or just use the existing one) for
>> > pointer aliases? The advantage is that you only have a single function
>> > to call, something like kmemleak_add_alias() and you do it at the point
>> > the value was converted.
>>
>> Actually I considered the above aliasing a little bit but I gave up
>> soon.
>>
>> I was afraid that this method might consume way more memory since this
>> just adds another member for "struct kmemleak_object", but adding a
>> single member for all objects. The number of kmemleak_object is
>> usually numerous.
>
> We could use a different tree with a "struct kmemleak_alias" structure
> which is much smaller. Something like below:
>
> struct kmemleak_alias {
> struct list_head alias_list;
> struct prio_tree_node tree_node;
> struct kmemleak_object *object;
> }
The above seems to be better than I thought. I'll give this a try.
> And an alias_list member would be added to kmemleak_object as well.
>
> Would the alias tree need to allow overlapping? Like different IOMMU
> mappings with the same address (but pointing to different physical
> memory).
Not for omap iommu.
omap iommu can have multiple instances, multiple devices can have each
own address spaces respectively. This doesn't affect this kmemleak
false positive.
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
