On 1/7/19 9:06 PM, Qian Cai wrote:
>
>
> On 1/7/19 5:43 AM, Catalin Marinas wrote:
>> On Thu, Jan 03, 2019 at 06:07:35PM +0100, Michal Hocko wrote:
>>>>> On Wed 02-01-19 13:06:19, Qian Cai wrote:
>>>>> [...]
>>>>>> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
>>>>>> index f9d9dc250428..9e1aa3b7df75 100644
>>>>>> --- a/mm/kmemleak.c
>>>>>> +++ b/mm/kmemleak.c
>>>>>> @@ -576,6 +576,16 @@ static struct kmemleak_object *create_object(unsigned long ptr, size_t size,
>>>>>> struct rb_node **link, *rb_parent;
>>>>>>
>>>>>> object = kmem_cache_alloc(object_cache, gfp_kmemleak_mask(gfp));
>>>>>> +#ifdef CONFIG_PREEMPT_COUNT
>>>>>> + if (!object) {
>>>>>> + /* last-ditch effort in a low-memory situation */
>>>>>> + if (irqs_disabled() || is_idle_task(current) || in_atomic())
>>>>>> + gfp = GFP_ATOMIC;
>>>>>> + else
>>>>>> + gfp = gfp_kmemleak_mask(gfp) | __GFP_DIRECT_RECLAIM;
>>>>>> + object = kmem_cache_alloc(object_cache, gfp);
>>>>>> + }
>>>>>> +#endif
>> [...]
>>> I will not object to this workaround but I strongly believe that
>>> kmemleak should rethink the metadata allocation strategy to be really
>>> robust.
>>
>> This would be nice indeed and it was discussed last year. I just haven't
>> got around to trying anything yet:
>>
>> https://marc.info/?l=linux-mm&m=152812489819532
>>
>
> It could be helpful to apply this 10-line patch first if has no fundamental
> issue, as it survives probably 50 times running LTP oom* workloads without a
> single kmemleak allocation failure.
>
> Of course, if someone is going to embed kmemleak metadata into slab objects
> itself soon, this workaround is not needed.
>
Well, it is really hard to tell even if someone get eventually redesign kmemleak
to embed the metadata into slab objects alone would survive LTP oom* workloads,
because it seems still use separate metadata for non-slab objects where kmemleak
allocation could fail like it right now and disable itself again.
