Re: [PATCH AUTOSEL 5.15 20/29] block: reduce kblockd_mod_delayed_work_on() CPU consumption

[Jens Axboe]

On 12/21/21 8:35 AM, Michael Kelley (LINUX) wrote:
> From: Sasha Levin <sashal@xxxxxxxxxx> Sent: Monday, December 20, 2021 5:58 PM
>>
>> From: Jens Axboe <axboe@xxxxxxxxx>
>>
>> [ Upstream commit cb2ac2912a9ca7d3d26291c511939a41361d2d83 ]
>>
>> Dexuan reports that he's seeing spikes of very heavy CPU utilization when
>> running 24 disks and using the 'none' scheduler. This happens off the
>> sched restart path, because SCSI requires the queue to be restarted async,
>> and hence we're hammering on mod_delayed_work_on() to ensure that the work
>> item gets run appropriately.
>>
>> Avoid hammering on the timer and just use queue_work_on() if no delay
>> has been specified.
>>
>> Reported-and-tested-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
>> Link: https://lore.kernel.org/linux-block/BYAPR21MB1270C598ED214C0490F47400BF719@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>> Reviewed-by: Ming Lei <ming.lei@xxxxxxxxxx>
>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
>> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
>> ---
>>  block/blk-core.c | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/block/blk-core.c b/block/blk-core.c
>> index c2d912d0c976c..a728434fcff87 100644
>> --- a/block/blk-core.c
>> +++ b/block/blk-core.c
>> @@ -1625,6 +1625,8 @@ EXPORT_SYMBOL(kblockd_schedule_work);
>>  int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
>>  				unsigned long delay)
>>  {
>> +	if (!delay)
>> +		return queue_work_on(cpu, kblockd_workqueue, &dwork->work);
>>  	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
>>  }
>>  EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
>> --
>> 2.34.1
> 
> Sasha -- there are reports of this patch causing performance problems.
> See
> https://lore.kernel.org/lkml/1639853092.524jxfaem2.none@localhost/. I
> would suggest *not* backporting it to any of the stable branches until
> the issues are fully sorted out.

Both this and the revert were backported. Which arguably doesn't make a
lot of sense, but at least it's consistent and won't cause any issues...

-- 
Jens Axboe