From: David Hildenbrand <david@redhat.com>
To: Wei Yang <richard.weiyang@gmail.com>
Cc: mhocko@suse.com, osalvador@suse.de, akpm@linux-foundation.org,
linux-mm@kvack.org
Subject: Re: [PATCH v2] mm, hotplug: move init_currently_empty_zone() under zone_span_lock protection
Date: Thu, 22 Nov 2018 22:53:31 +0100 [thread overview]
Message-ID: <8a130cbe-8f1a-420f-9a82-f0905f4fc46d@redhat.com> (raw)
In-Reply-To: <20181122212822.ypedpcbhrxpa3tyv@master>
On 22.11.18 22:28, Wei Yang wrote:
> On Thu, Nov 22, 2018 at 04:26:40PM +0100, David Hildenbrand wrote:
>> On 22.11.18 11:12, Wei Yang wrote:
>>> During online_pages phase, pgdat->nr_zones will be updated in case this
>>> zone is empty.
>>>
>>> Currently the online_pages phase is protected by the global lock
>>> mem_hotplug_begin(), which ensures there is no contention during the
>>> update of nr_zones. But this global lock introduces scalability issues.
>>>
>>> This patch is a preparation for removing the global lock during
>>> online_pages phase. Also this patch changes the documentation of
>>> node_size_lock to include the protectioin of nr_zones.
>>
>> I looked into locking recently, and there is more to it.
>>
>> Please read:
>>
>> commit dee6da22efac451d361f5224a60be2796d847b51
>> Author: David Hildenbrand <david@redhat.com>
>> Date: Tue Oct 30 15:10:44 2018 -0700
>>
>> memory-hotplug.rst: add some details about locking internals
>>
>> Let's document the magic a bit, especially why device_hotplug_lock is
>> required when adding/removing memory and how it all play together with
>> requests to online/offline memory from user space.
>>
>> Short summary: Onlining/offlining of memory requires the device_hotplug_lock
>> as of now.
>>
>> mem_hotplug_begin() is just an internal optimization. (we don't want
>> everybody to take the device lock)
>>
>
> Hi, David
>
> Thanks for your comment.
My last sentence should have been "we don't want everybody to take the
device hotplug lock" :) That caused confusion.
>
> Hmm... I didn't catch your point.
>
> Related to memory hot-plug, there are (at least) three locks,
>
> * device_hotplug_lock (global)
> * device lock (device scope)
> * mem_hotplug_lock (global)
>
> But with two different hold sequence in two cases:
>
> * device_online()
>
> device_hotplug_lock
> device_lock
> mem_hotplug_lock
>
> * add_memory_resource()
>
> device_hotplug_lock
> mem_hotplug_lock
> device_lock
> ^
> |
> I don't find where this is hold in add_memory_resource().
> Would you mind giving me a hint?
>
> If my understanding is correct, what is your point?
>
The point I was trying to make:
Right now all onlining/offlining/adding/removing is protected by the
device_hotplug_lock (and that's a good thing, things are fragile enough
already).
mem_hotplug_lock() is used in addition for get_online_mems().
"This patch is a preparation for removing the global lock during
online_pages phase." - is more like "one global lock".
> I guess your point is : just remove mem_hotplug_lock is not enough to
> resolve the scalability issue?
Depends on which scalability issue :)
Getting rid of / removing the impact of mem_hotplug_lock is certainly a
very good idea. And improves scalability of all callers of
get_online_mems(). If that is the intention, very good :)
If the intention is to make onlining/offlining more scalable (e.g. in
parallel or such), then scalability is limited by device_hotplug_lock.
>
> Please correct me, if I am not. :-)
>
Guess I was just wondering which scalability issue we are trying to solve :)
--
Thanks,
David / dhildenb
next prev parent reply other threads:[~2018-11-22 21:53 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-11-20 1:48 [PATCH] mm, hotplug: protect nr_zones with pgdat_resize_lock() Wei Yang
2018-11-20 7:31 ` Michal Hocko
2018-11-20 7:58 ` osalvador
2018-11-20 8:48 ` Michal Hocko
2018-11-21 2:52 ` Wei Yang
2018-11-21 7:15 ` Michal Hocko
2018-11-22 1:52 ` Wei Yang
2018-11-22 8:39 ` Michal Hocko
2018-11-26 2:28 ` Wei Yang
2018-11-26 8:16 ` Michal Hocko
2018-11-26 9:06 ` Wei Yang
2018-11-26 10:03 ` Michal Hocko
2018-11-27 0:18 ` Wei Yang
2018-11-27 3:12 ` Wei Yang
2018-11-27 13:16 ` Michal Hocko
2018-11-27 23:56 ` Wei Yang
2018-11-21 8:24 ` osalvador
2018-11-21 2:44 ` Wei Yang
2018-11-21 7:14 ` Michal Hocko
2018-11-22 10:12 ` [PATCH v2] mm, hotplug: move init_currently_empty_zone() under zone_span_lock protection Wei Yang
2018-11-22 10:15 ` Wei Yang
2018-11-22 10:29 ` Michal Hocko
2018-11-22 14:27 ` Wei Yang
2018-11-22 10:37 ` osalvador
2018-11-22 14:28 ` Wei Yang
2018-11-22 15:26 ` David Hildenbrand
2018-11-22 21:28 ` Wei Yang
2018-11-22 21:53 ` David Hildenbrand [this message]
2018-11-22 23:53 ` Wei Yang
2018-11-23 8:42 ` Michal Hocko
2018-11-23 8:46 ` David Hildenbrand
2018-11-26 1:44 ` Wei Yang
2018-11-26 9:24 ` David Hildenbrand
2018-11-27 0:23 ` Wei Yang
2018-11-30 6:58 ` [PATCH v3] " Wei Yang
2018-11-30 9:30 ` David Hildenbrand
2018-12-01 0:27 ` Wei Yang
2018-12-03 10:09 ` David Hildenbrand
2018-12-03 20:37 ` Wei Yang
2018-12-03 20:50 ` [PATCH v4] " Wei Yang
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=8a130cbe-8f1a-420f-9a82-f0905f4fc46d@redhat.com \
--to=david@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@suse.com \
--cc=osalvador@suse.de \
--cc=richard.weiyang@gmail.com \
/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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).