linux-kernel - Re: [RFC v2 2/2] mm/memory_hotplug: Shrink spanned pages when offlining memory

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20180822075020.GA14550@techadventures.net>
Date:   Wed, 22 Aug 2018 09:50:20 +0200
From:   Oscar Salvador <osalvador@...hadventures.net>
To:     David Hildenbrand <david@...hat.com>
Cc:     akpm@...ux-foundation.org, mhocko@...e.com,
        dan.j.williams@...el.com, jglisse@...hat.com,
        jonathan.cameron@...wei.com, Pavel.Tatashin@...rosoft.com,
        yasu.isimatu@...il.com, logang@...tatee.com, dave.jiang@...el.com,
        linux-mm@...ck.org, linux-kernel@...r.kernel.org,
        Oscar Salvador <osalvador@...e.de>
Subject: Re: [RFC v2 2/2] mm/memory_hotplug: Shrink spanned pages when
 offlining memory

On Tue, Aug 21, 2018 at 03:17:10PM +0200, David Hildenbrand wrote:
> > add_device_memory is in charge of
> 
> I wouldn't use the terminology of onlining/offlining here. That applies
> rather to memory that is exposed to the rest of the system (e.g. buddy
> allocator, has underlying memory block devices). I guess it is rather a
> pure setup/teardown of that device memory.

Hi David,

I am not sure if you are referring to:

"
a) calling either arch_add_memory() or add_pages(), depending on whether
   we want a linear mapping
b) online the memory sections that correspond to the pfn range
c) calling move_pfn_range_to_zone() being zone ZONE_DEVICE to
   expand zone/pgdat spanned pages and initialize its pages
"

Well, that is partialy true.
I mean, in order to make this work, we need to offline/online the memory
sections, because shrink_pages will rely on that from now on.
Is what we do when online/offline pages, but since device memory
does not go through the "official" channels, we need to do it there
as well.

Sure I can use another terminology, but since that is what
offline/online_mem_sections do, I just came up with that.

> I would really like to see the mem_hotplug_begin/end also getting moved
> inside add_device_memory()/del_device_memory(). (just like for
> add/remove_memory)
> 
> I wonder if kasan_ stuff actually requires this lock, or if it could
> also be somehow moved inside add_device_memory/del_device_memory.

Yes, that was my first approach, but then I saw that the kasan stuff is being
handled whithin those locks, so I was not sure and I backed off leaving the
mem_hotplug_begin/end where they were.

Maybe Jerome can shed some light and, and we can just handle the kasan stuff
out of the locks.

> Maybe shorten that a bit
> 
> "HMM/devm memory does not have IORESOURCE_SYSTEM_RAM set. They use
>  devm_request_mem_region/devm_release_mem_region to add/release a
>  resource. Just back off here."

Uhm, fair enough.

> Any reason for these indirections?

I wanted to hide the internals in the memory_hotplug code.
I thought about removing them, but I finally left them.
If people think that we are better off without them, I can just
remove them.

> I guess for readability, this patch could be split up into several
> patches. E.g. factoring out of add_device_memory/del_device_memory,
> release_mem_region_adjustable change ...

Yes, really true.
But I wanted first to gather feedback mainly from HMM/devm people to see
if they saw an outright bug within the series because I am not so
familiar with that part of the code.

Feedback from Jerome/Dan will be appreciate as well to see if this is a good
direction.

But you are right, in the end, this will have to be slipt up into several
parts to ease the review.

Thanks for reviewing this David!
I will try to address your concerns.

Thanks 
-- 
Oscar Salvador
SUSE L3