Re: [PATCH] slab: remove store_stackinfo()

[Vlastimil Babka]

On 4/16/2019 8:50 PM, Thomas Gleixner wrote:
> On Tue, 16 Apr 2019, Vlastimil Babka wrote:
> 
>> On 4/16/19 4:22 PM, Qian Cai wrote:
>>> store_stackinfo() does not seem used in actual SLAB debugging.
>>> Potentially, it could be added to check_poison_obj() to provide more
>>> information, but this seems like an overkill due to the declining
>>> popularity of the SLAB, so just remove it instead.
>>>
>>> Signed-off-by: Qian Cai <cai@xxxxxx>
>>
>> I've acked Thomas' version already which was narrower, but no objection
>> to remove more stuff on top of that. Linus (and I later in another
>> thread) already pointed out /proc/slab_allocators. It only takes a look
>> at add_caller() there to not regret removing that one.
> 
> The issue why I was looking at this was a krobot complaint about the kernel
> crashing in that stack store function with my stackguard series applied. It
> was broken before the stackguard pages already, it just went unnoticed.
> 
> As you explained, nobody is caring about DEBUG_SLAB + DEBUG_PAGEALLOC
> anyway, so I'm happy to not care about krobot tripping over it either.
> 
> So we have 3 options:
> 
>    1) I ignore it and merge the stack guard series w/o it
> 
>    2) I can carry the minimal fix or Qian's version in the stackguard
>       branch
> 
>    3) We ship that minimal fix to Linus right now and then everyone can
>       base their stuff on top independently.

I think #3 is overkill for something that was broken for who knows how long and
nobody noticed. I'd go with 2) and perhaps Qian's version as nobody AFAIK uses
the caller+cpu as well as the stack trace.

For Qian's version also:
Acked-by: Vlastimil Babka <vbabka@xxxxxxx>

> #3 is probably the right thing to do.
> 
> Thanks,
> 
> 	tglx
>