All of lore.kernel.org
 help / color / mirror / Atom feed
From: Tejun Heo <tj@kernel.org>
To: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: len.brown@intel.com, linux-pm@vger.kernel.org,
	gregkh@linuxfoundation.org, linux-nvdimm@lists.01.org,
	jiangshanlai@gmail.com, linux-kernel@vger.kernel.org,
	zwisler@kernel.org, pavel@ucw.cz, rafael@kernel.org,
	akpm@linux-foundation.org
Subject: Re: [RFC workqueue/driver-core PATCH 1/5] workqueue: Provide queue_work_near to queue work near a given NUMA node
Date: Mon, 1 Oct 2018 09:01:42 -0700	[thread overview]
Message-ID: <20181001160142.GE270328@devbig004.ftw2.facebook.com> (raw)
In-Reply-To: <ba72f007-84e2-6fe0-b128-d876dadef5f5@linux.intel.com>

Hello,

On Wed, Sep 26, 2018 at 03:19:21PM -0700, Alexander Duyck wrote:
> On 9/26/2018 3:09 PM, Tejun Heo wrote:
> I could just use queue_work_on probably, but is there any issue if I
> am passing CPU values that are not in the wq_unbound_cpumask? That

That should be fine.  If it can't find any available cpu, it'll fall
back to round-robin.  We probably can improve it so that it can
consider the numa distance when falling back.

> was mostly my concern. Also for an unbound queue do I need to worry
> about the hotplug lock? I wasn't sure if that was the case or not as

Issuers don't need to worry about them.

> I know it is called out as something to be concerned with using
> queue_work_on, but in __queue_work the value is just used to
> determine which node to grab a work queue from.

It might be better to leave queue_work_on() to be used for per-cpu
workqueues and introduce queue_work_near() as you suggseted.  I just
don't want it to duplicate the node selection code in it.  Would that
work?

> I forgot to address your question about the advantages. They are
> pretty significant. The test system I was working with was
> initializing 3TB of nvdimm memory per node. If the node is aligned
> it takes something like 24 seconds, whereas an unaligned core can
> take 36 seconds or more.

Oh yeah, sure, numa affinity matters quite a bit on memory heavy
workloads.  I was mistaken that you were adding adding numa affinity
to per-cpu workqueues.

Thanks.

-- 
tejun
_______________________________________________
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

WARNING: multiple messages have this Message-ID
From: Tejun Heo <tj@kernel.org>
To: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: linux-nvdimm@lists.01.org, gregkh@linuxfoundation.org,
	linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org,
	akpm@linux-foundation.org, len.brown@intel.com,
	dave.jiang@intel.com, rafael@kernel.org,
	vishal.l.verma@intel.com, jiangshanlai@gmail.com, pavel@ucw.cz,
	zwisler@kernel.org, dan.j.williams@intel.com
Subject: Re: [RFC workqueue/driver-core PATCH 1/5] workqueue: Provide queue_work_near to queue work near a given NUMA node
Date: Mon, 1 Oct 2018 09:01:42 -0700	[thread overview]
Message-ID: <20181001160142.GE270328@devbig004.ftw2.facebook.com> (raw)
In-Reply-To: <ba72f007-84e2-6fe0-b128-d876dadef5f5@linux.intel.com>

Hello,

On Wed, Sep 26, 2018 at 03:19:21PM -0700, Alexander Duyck wrote:
> On 9/26/2018 3:09 PM, Tejun Heo wrote:
> I could just use queue_work_on probably, but is there any issue if I
> am passing CPU values that are not in the wq_unbound_cpumask? That

That should be fine.  If it can't find any available cpu, it'll fall
back to round-robin.  We probably can improve it so that it can
consider the numa distance when falling back.

> was mostly my concern. Also for an unbound queue do I need to worry
> about the hotplug lock? I wasn't sure if that was the case or not as

Issuers don't need to worry about them.

> I know it is called out as something to be concerned with using
> queue_work_on, but in __queue_work the value is just used to
> determine which node to grab a work queue from.

It might be better to leave queue_work_on() to be used for per-cpu
workqueues and introduce queue_work_near() as you suggseted.  I just
don't want it to duplicate the node selection code in it.  Would that
work?

> I forgot to address your question about the advantages. They are
> pretty significant. The test system I was working with was
> initializing 3TB of nvdimm memory per node. If the node is aligned
> it takes something like 24 seconds, whereas an unaligned core can
> take 36 seconds or more.

Oh yeah, sure, numa affinity matters quite a bit on memory heavy
workloads.  I was mistaken that you were adding adding numa affinity
to per-cpu workqueues.

Thanks.

-- 
tejun

WARNING: multiple messages have this Message-ID
From: Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
To: Alexander Duyck
	<alexander.h.duyck-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
Cc: len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org,
	linux-pm-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	gregkh-hQyY1W1yCW8ekmWlsbkhG0B+6BGkLq7r@public.gmane.org,
	linux-nvdimm-hn68Rpc1hR1g9hUCZPvPmw@public.gmane.org,
	jiangshanlai-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org,
	linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	zwisler-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org,
	pavel-+ZI9xUNit7I@public.gmane.org,
	rafael-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org,
	akpm-de/tnXTf+JLsfHDXvbKv3WD2FQJk+8+b@public.gmane.org
Subject: Re: [RFC workqueue/driver-core PATCH 1/5] workqueue: Provide queue_work_near to queue work near a given NUMA node
Date: Mon, 1 Oct 2018 09:01:42 -0700	[thread overview]
Message-ID: <20181001160142.GE270328@devbig004.ftw2.facebook.com> (raw)
In-Reply-To: <ba72f007-84e2-6fe0-b128-d876dadef5f5-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>

Hello,

On Wed, Sep 26, 2018 at 03:19:21PM -0700, Alexander Duyck wrote:
> On 9/26/2018 3:09 PM, Tejun Heo wrote:
> I could just use queue_work_on probably, but is there any issue if I
> am passing CPU values that are not in the wq_unbound_cpumask? That

That should be fine.  If it can't find any available cpu, it'll fall
back to round-robin.  We probably can improve it so that it can
consider the numa distance when falling back.

> was mostly my concern. Also for an unbound queue do I need to worry
> about the hotplug lock? I wasn't sure if that was the case or not as

Issuers don't need to worry about them.

> I know it is called out as something to be concerned with using
> queue_work_on, but in __queue_work the value is just used to
> determine which node to grab a work queue from.

It might be better to leave queue_work_on() to be used for per-cpu
workqueues and introduce queue_work_near() as you suggseted.  I just
don't want it to duplicate the node selection code in it.  Would that
work?

> I forgot to address your question about the advantages. They are
> pretty significant. The test system I was working with was
> initializing 3TB of nvdimm memory per node. If the node is aligned
> it takes something like 24 seconds, whereas an unaligned core can
> take 36 seconds or more.

Oh yeah, sure, numa affinity matters quite a bit on memory heavy
workloads.  I was mistaken that you were adding adding numa affinity
to per-cpu workqueues.

Thanks.

-- 
tejun

  reply	other threads:[~2018-10-01 16:01 UTC|newest]

Thread overview: 69+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-09-26 21:51 [RFC workqueue/driver-core PATCH 0/5] Add NUMA aware async_schedule calls Alexander Duyck
2018-09-26 21:51 ` Alexander Duyck
2018-09-26 21:51 ` Alexander Duyck
2018-09-26 21:51 ` [RFC workqueue/driver-core PATCH 1/5] workqueue: Provide queue_work_near to queue work near a given NUMA node Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-26 21:53   ` Tejun Heo
2018-09-26 21:53     ` Tejun Heo
2018-09-26 21:53     ` Tejun Heo
2018-09-26 22:05     ` Alexander Duyck
2018-09-26 22:05       ` Alexander Duyck
2018-09-26 22:09       ` Tejun Heo
2018-09-26 22:09         ` Tejun Heo
2018-09-26 22:09         ` Tejun Heo
2018-09-26 22:19         ` Alexander Duyck
2018-09-26 22:19           ` Alexander Duyck
2018-10-01 16:01           ` Tejun Heo [this message]
2018-10-01 16:01             ` Tejun Heo
2018-10-01 16:01             ` Tejun Heo
2018-10-01 21:54             ` Alexander Duyck
2018-10-01 21:54               ` Alexander Duyck
2018-10-01 21:54               ` Alexander Duyck
2018-10-02 17:41               ` Tejun Heo
2018-10-02 17:41                 ` Tejun Heo
2018-10-02 17:41                 ` Tejun Heo
2018-10-02 18:23                 ` Alexander Duyck
2018-10-02 18:23                   ` Alexander Duyck
2018-10-02 18:23                   ` Alexander Duyck
2018-10-02 18:41                   ` Tejun Heo
2018-10-02 18:41                     ` Tejun Heo
2018-10-02 18:41                     ` Tejun Heo
2018-10-02 20:49                     ` Alexander Duyck
2018-10-02 20:49                       ` Alexander Duyck
2018-10-02 20:49                       ` Alexander Duyck
2018-09-26 21:51 ` [RFC workqueue/driver-core PATCH 2/5] async: Add support for queueing on specific " Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-27  0:31   ` Dan Williams
2018-09-27  0:31     ` Dan Williams
2018-09-27  0:31     ` Dan Williams
2018-09-27 15:16     ` Alexander Duyck
2018-09-27 15:16       ` Alexander Duyck
2018-09-27 15:16       ` Alexander Duyck
2018-09-27 19:48       ` Dan Williams
2018-09-27 19:48         ` Dan Williams
2018-09-27 20:03         ` Alexander Duyck
2018-09-27 20:03           ` Alexander Duyck
2018-09-26 21:51 ` [RFC workqueue/driver-core PATCH 3/5] driver core: Probe devices asynchronously instead of the driver Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-27  0:48   ` Dan Williams
2018-09-27  0:48     ` Dan Williams
2018-09-27  0:48     ` Dan Williams
2018-09-27 15:27     ` Alexander Duyck
2018-09-27 15:27       ` Alexander Duyck
2018-09-27 15:27       ` Alexander Duyck
2018-09-28  2:48       ` Dan Williams
2018-09-28  2:48         ` Dan Williams
2018-09-28  2:48         ` Dan Williams
2018-09-26 21:51 ` [RFC workqueue/driver-core PATCH 4/5] driver core: Use new async_schedule_dev command Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-26 21:51   ` Alexander Duyck
2018-09-28 17:42   ` Dan Williams
2018-09-28 17:42     ` Dan Williams
2018-09-28 17:42     ` Dan Williams
2018-09-26 21:52 ` [RFC workqueue/driver-core PATCH 5/5] nvdimm: Schedule device registration on node local to the device Alexander Duyck
2018-09-26 21:52   ` Alexander Duyck
2018-09-26 21:52   ` Alexander Duyck
2018-09-28 17:46   ` Dan Williams
2018-09-28 17:46     ` Dan Williams

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20181001160142.GE270328@devbig004.ftw2.facebook.com \
    --to=tj@kernel.org \
    --cc=akpm@linux-foundation.org \
    --cc=alexander.h.duyck@linux.intel.com \
    --cc=gregkh@linuxfoundation.org \
    --cc=jiangshanlai@gmail.com \
    --cc=len.brown@intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-nvdimm@lists.01.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=pavel@ucw.cz \
    --cc=rafael@kernel.org \
    --cc=zwisler@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.