linux-kernel - Re: [RFC PATCH 0/3] Allocate memmap from hotadded memory (per device)

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 for Android: free password hash cracker in your pocket

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

Date:   Thu, 29 Oct 2020 08:49:10 +0100
From:   David Hildenbrand <david@...hat.com>
To:     Mike Kravetz <mike.kravetz@...cle.com>,
        Oscar Salvador <osalvador@...e.de>
Cc:     mhocko@...nel.org, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, vbabka@...e.cz, pasha.tatashin@...een.com
Subject: Re: [RFC PATCH 0/3] Allocate memmap from hotadded memory (per device)

On 28.10.20 19:47, Mike Kravetz wrote:
> On 10/27/20 8:58 AM, Oscar Salvador wrote:
>> On Tue, Oct 27, 2020 at 04:44:33PM +0100, David Hildenbrand wrote:
>>> I'm planning on looking into patch #2/3 later this or next week (this week
>>> is open source summit / KVM Forum).
>>
>> Sure, aprecciated the time ;-)
>>
>>>
>>> One thing to look into right now is how to make this fly this with vmemmap
>>> optimizations for hugetlb pages.
>>>
>>> https://lkml.kernel.org/r/20201026145114.59424-1-songmuchun@bytedance.com
>>
>> I was about to have a look at that series eitherway, but good you mentioned.
>>
> 
> More eyes on that series would be appreciated.
> 
> That series will dynamically free and allocate memmap pages as hugetlb
> pages are allocated or freed.  I haven't looked through this series, but
> my first thought is that we would need to ensure those allocs/frees are
> directed to the device.  Not sure if there are interfaces for that.

Directing to the device might be part of the solution, but does not have 
to be. You really want to free the pages to the OS in the end, otherwise 
you lose the whole benefit of the vmemmap optimization.

You would want to actually free the pages (making sure whatever 
generic_online_page() does was done to these special vmemmap pages). But 
then, you cannot simply skip all X first pages of a memory block when 
offlining, you can only skip the once that are still vmemmap pages 
(e.g., marked via page type), and have to isolate/migrate off the 
no-longer vmemmap pages.

-- 
Thanks,

David / dhildenb