On 1/2/19 8:28 PM, Tetsuo Handa wrote:
> On 2019/01/03 3:19, Qian Cai wrote:
>> On 1/2/19 1:06 PM, Mel Gorman wrote:
>>
>>> While I recognise there is no test case available, how often does this
>>> trigger in syzbot as it would be nice to have some confirmation any
>>> patch is really fixing the problem.
>>
>> I think I did manage to trigger this every time running a mmap() workload
>> causing swapping and a low-memory situation [1].
>>
>> [1]
>> https://github.com/linux-test-project/ltp/blob/master/testcases/kernel/mem/oom/oom01.c
>
> wakeup_kswapd() is called because tlb_next_batch() is doing GFP_NOWAIT
> allocation. But since tlb_next_batch() can tolerate allocation failure,
> does below change in tlb_next_batch() help?
>
> #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
>
> - batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
> + batch = (void *)__get_free_pages(__GFP_NOWARN, 0);
No. In oom01 case, it is from,
do_anonymous_page
__alloc_zeroed_user_highpage
alloc_page_vma(GFP_HIGHUSER ...
GFP_HIGHUSER -> GFP_USER -> __GFP_RECLAIM -> ___GFP_KSWAPD_RECLAIM
Then, it has this new code in steal_suitable_fallback() via 1c30844d2df (mm:
reclaim small amounts of memory when an external fragmentation event occurs)
/*
* Boost watermarks to increase reclaim pressure to reduce
* the likelihood of future fallbacks. Wake kswapd now as
* the node may be balanced overall and kswapd will not
* wake naturally.
*/
boost_watermark(zone);
if (alloc_flags & ALLOC_KSWAPD)
wakeup_kswapd(zone, 0, 0, zone_idx(zone));
