Re: [PATCH 4/4] mm/slub: free KFENCE objects in slab_free_hook()

[Vlastimil Babka]

On 12/11/23 23:11, Andrey Konovalov wrote:
> On Wed, Dec 6, 2023 at 3:45 PM Marco Elver <elver@xxxxxxxxxx> wrote:
>>
>> The is_kfence_address() implementation tolerates tagged addresses,
>> i.e. if it receives a tagged non-kfence address, it will never return
>> true.

So just to be sure, it can't happen that a genuine kfence address would then
become KASAN tagged and handed out, and thus when tested by
is_kfence_address() it would be a false negative?

>> The KASAN_HW_TAGS patches and KFENCE patches were in development
>> concurrently, and at the time there was some conflict resolution that
>> happened when both were merged. The
>> is_kfence_address(kasan_reset_tag(..)) initially came from [1] but was
>> squashed into 2b8305260fb.
>>
>> [1] https://lore.kernel.org/all/9dc196006921b191d25d10f6e611316db7da2efc.1611946152.git.andreyknvl@xxxxxxxxxx/
>>
>> Andrey, do you recall what issue you encountered that needed kasan_reset_tag()?
> 
> I don't remember at this point, but this could have been just a safety measure.
> 
> If is_kfence_address tolerates tagged addresses, we should be able to
> drop these kasan_reset_tag calls.

Will drop it once the above is confirmed. Thanks!