Re: dm bufio: Reduce dm_bufio_lock contention

From: jing xia
Date: Tue Jul 31 2018 - 22:48:03 EST


We reproduced this issue again and found out the root cause.
dm_bufio_prefetch() with dm_bufio_lock enters the direct reclaim and
takes a long time to do the soft_limit_reclaim, because of the huge
number of memory excess of the memcg.
Then, all the task who do shrink_slab() wait for dm_bufio_lock.

Any suggestions for this?Thanks.

On Thu, Jun 14, 2018 at 3:31 PM, Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> On Thu 14-06-18 15:18:58, jing xia wrote:
> [...]
>> PID: 22920 TASK: ffffffc0120f1a00 CPU: 1 COMMAND: "kworker/u8:2"
>> #0 [ffffffc0282af3d0] __switch_to at ffffff8008085e48
>> #1 [ffffffc0282af3f0] __schedule at ffffff8008850cc8
>> #2 [ffffffc0282af450] schedule at ffffff8008850f4c
>> #3 [ffffffc0282af470] schedule_timeout at ffffff8008853a0c
>> #4 [ffffffc0282af520] schedule_timeout_uninterruptible at ffffff8008853aa8
>> #5 [ffffffc0282af530] wait_iff_congested at ffffff8008181b40
>
> This trace doesn't provide the full picture unfortunately. Waiting in
> the direct reclaim means that the underlying bdi is congested. The real
> question is why it doesn't flush IO in time.
>
>> #6 [ffffffc0282af5b0] shrink_inactive_list at ffffff8008177c80
>> #7 [ffffffc0282af680] shrink_lruvec at ffffff8008178510
>> #8 [ffffffc0282af790] mem_cgroup_shrink_node_zone at ffffff80081793bc
>> #9 [ffffffc0282af840] mem_cgroup_soft_limit_reclaim at ffffff80081b6040
>> #10 [ffffffc0282af8f0] do_try_to_free_pages at ffffff8008178b6c
>> #11 [ffffffc0282af990] try_to_free_pages at ffffff8008178f3c
>> #12 [ffffffc0282afa30] __perform_reclaim at ffffff8008169130
>> #13 [ffffffc0282afab0] __alloc_pages_nodemask at ffffff800816c9b8
>> #14 [ffffffc0282afbd0] __get_free_pages at ffffff800816cd6c
>> #15 [ffffffc0282afbe0] alloc_buffer at ffffff8008591a94
>> #16 [ffffffc0282afc20] __bufio_new at ffffff8008592e94
>> #17 [ffffffc0282afc70] dm_bufio_prefetch at ffffff8008593198
>> #18 [ffffffc0282afd20] verity_prefetch_io at ffffff8008598384
>> #19 [ffffffc0282afd70] process_one_work at ffffff80080b5b3c
>> #20 [ffffffc0282afdc0] worker_thread at ffffff80080b64fc
>> #21 [ffffffc0282afe20] kthread at ffffff80080bae34
>>
>> > Mikulas
>
> --
> Michal Hocko
> SUSE Labs