lkml.org 
[lkml]   [2021]   [Dec]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
SubjectRE: Random high CPU utilization in blk-mq with the none scheduler
Date
From: Jens Axboe <axboe@kernel.dk>
> Sent: Monday, December 13, 2021 10:43 AM
>
> I wonder if this will help, SCSI artifact making us hit the
> mod delayed work path all the time.

Yes, I think this is what's happening here.

> Might be a race in there, but should be fine for testing.
>
>
> diff --git a/block/blk-core.c b/block/blk-core.c
> index 4d8f5fe91588..cb2f4d604bad 100644
> --- a/block/blk-core.c
> +++ b/block/blk-core.c
> @@ -1625,7 +1625,9 @@ EXPORT_SYMBOL(kblockd_schedule_work);
> int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
> unsigned long delay)
> {
> - return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
> + if (!work_pending(&dwork->work))
> + return mod_delayed_work_on(cpu, kblockd_workqueue, dwork,
> delay);
> + return true;
> }
> EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
> --
> Jens Axboe

Thanks very much! This path works!

In the test, "v5.16-rc4 + reverting dc5fc361d891
("block: attempt direct issue of plug list") + this patch" shows a good I/O
performance that's about the same as v5.16-rc4.

I suppose "v5.16-rc4 + this patch" would show a slightly even better
performance.

It would be great if you can commit this small patch so that it can be
backported to old stable kernels.

Thanks,
Dexuan
\
 
 \ /
  Last update: 2021-12-14 01:44    [W:2.187 / U:0.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site