On 14/04/2021 12:12, Ming Lei wrote:

On Wed, Apr 14, 2021 at 04:12:22PM +0530, Kashyap Desai wrote:

quoted

quoted

Hi Ming,

quoted

It is reported inside RH that CPU utilization is increased ~20% when
running simple FIO test inside VM which disk is built on image stored
on XFS/megaraid_sas.

When I try to investigate by reproducing the issue via scsi_debug, I
found IO hang when running randread IO(8k, direct IO, libaio) on
scsi_debug disk created by the following command:

	modprobe scsi_debug host_max_queue=128

submit_queues=$NR_CPUS

quoted

virtual_gb=256

So I can recreate this hang for using mq-deadline IO sched for scsi debug,
in
that fio does not exit. I'm using v5.12-rc7.

I can also recreate this issue using mq-deadline. Using <none>, there is no
IO hang issue.
Also if I run script to change scheduler periodically (none, mq-deadline),
sysfs entry hangs.

Here is call trace-
Call Trace:
[ 1229.879862]  __schedule+0x29d/0x7a0
[ 1229.879871]  schedule+0x3c/0xa0
[ 1229.879875]  blk_mq_freeze_queue_wait+0x62/0x90
[ 1229.879880]  ? finish_wait+0x80/0x80
[ 1229.879884]  elevator_switch+0x12/0x40
[ 1229.879888]  elv_iosched_store+0x79/0x120
[ 1229.879892]  ? kernfs_fop_write_iter+0xc7/0x1b0
[ 1229.879897]  queue_attr_store+0x42/0x70
[ 1229.879901]  kernfs_fop_write_iter+0x11f/0x1b0
[ 1229.879905]  new_sync_write+0x11f/0x1b0
[ 1229.879912]  vfs_write+0x184/0x250
[ 1229.879915]  ksys_write+0x59/0xd0
[ 1229.879917]  do_syscall_64+0x33/0x40
[ 1229.879922]  entry_SYSCALL_64_after_hwframe+0x44/0xae


I tried both - 5.12.0-rc1 and 5.11.0-rc2+ and there is a same behavior.
Let me also check  megaraid_sas and see if anything generic or this is a
special case of scsi_debug.

As I mentioned, it could be one generic issue wrt. SCHED_RESTART.
shared tags might have to restart all hctx since all share same tags.

I tested on hisi_sas v2 hw (which now sets host_tagset), and can 
reproduce. Seems to be combination of mq-deadline and fio rw=randread 
settings required to reproduce from limited experiments.

Incidentally, about the mq-deadline vs none IO scheduler on the same 
host, I get this with 6x SAS SSD:

rw=read
		CPU util			IOPs
mq-deadline	usr=26.80%, sys=52.78%		650K
none		usr=22.99%, sys=74.10%		475K

rw=randread
		CPU util			IOPs
mq-deadline	usr=21.72%, sys=44.18%,		423K
none		usr=23.15%, sys=74.01%		450K

Thanks,
John