Re: [PATCH] blk-mq: Use request queue-wide tags for tagset-wide sbitmap

[Douglas Gilbert]

On 2021-05-10 8:52 p.m., Ming Lei wrote:
> On Mon, May 03, 2021 at 06:22:13PM +0800, John Garry wrote:
> > The tags used for an IO scheduler are currently per hctx.
> >
> > As such, when q->nr_hw_queues grows, so does the request queue total IO
> > scheduler tag depth.
> >
> > This may cause problems for SCSI MQ HBAs whose total driver depth is
> > fixed.
> >
> > Ming and Yanhui report higher CPU usage and lower throughput in scenarios
> > where the fixed total driver tag depth is appreciably lower than the total
> > scheduler tag depth:
> > https://lore.kernel.org/linux-block/440dfcfc-1a2c-bd98-1161-cec4d78c6dfc@xxxxxxxxxx/T/#mc0d6d4f95275a2743d1c8c3e4dc9ff6c9aa3a76b
>
> No difference any more wrt. fio running on scsi_debug with this patch in
> Yanhui's test machine:
>
> 	modprobe scsi_debug host_max_queue=128 submit_queues=32 virtual_gb=256 delay=1
> vs.
> 	modprobe scsi_debug max_queue=128 submit_queues=1 virtual_gb=256 delay=1
>
> Without this patch, the latter's result is 30% higher than the former's.
>
> note: scsi_debug's queue depth needs to be updated to 128 for avoiding io hang,
> which is another scsi issue.

"scsi_debug: Fix cmd_per_lun, set to max_queue" made it into lk 5.13.0-rc1 as
commit fc09acb7de31badb2ea9e85d21e071be1a5736e4 . Is this the issue you are
referring to, or is there a separate issue in the wider scsi stack?

BTW Martin's 5.14/scsi-queue is up and running with lk 5.13.0-rc1 .

Doug Gilbert