linux-kernel - Re: [PATCH 0/1] prctl: allow overriding system THP policy to always

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: <e971dcf1-9ea4-4a20-bbc2-7ffc1ff15d37@gmail.com>
Date: Thu, 8 May 2025 19:05:17 +0100
From: Usama Arif <usamaarif642@...il.com>
To: David Hildenbrand <david@...hat.com>,
 Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org
Cc: hannes@...xchg.org, shakeel.butt@...ux.dev, riel@...riel.com,
 ziy@...dia.com, baolin.wang@...ux.alibaba.com, lorenzo.stoakes@...cle.com,
 Liam.Howlett@...cle.com, npache@...hat.com, ryan.roberts@....com,
 linux-kernel@...r.kernel.org, kernel-team@...a.com
Subject: Re: [PATCH 0/1] prctl: allow overriding system THP policy to always



On 08/05/2025 18:39, David Hildenbrand wrote:
>>> We did something similar with PR_SET_MEMORY_MERGE.
>>>
>>
>> For this you mean the prctl command would do for_each_vma and set VM_HUGEPAGE to implement point 2.
> 
> Yes. The expectation would be that the prctl is either a) triggered early when there are not that many VMAs yet or b) inherited during fork+exec such that the target process immediately gets it enabled for all VMAs.

Yes that is the case for us as well, For us its mostly going to be a, i.e. exec_invoke called by systemd. But b will be good as well.
> 
>> For having point 1, I think we will still need extra mm->flags, i.e. MMF_VM_THP_MADVISE/DEFER/ALWAYS/NEVER.
> 
> Yes, and it's unclear what we really want here.

Let me make a v2 and we might be able to have a better discussion once its done. I will move it to RFC.
> 
>>
>> I think it would have the same affect as what this patch is trying to do? But would be just more
>> expensive in terms of both code changes and the cost of the actual call as you now have to walk
>> all vmas. On the other hand you wont need the below diff in from v1. I do feel the current approach
>> in the patch is simpler? But if your point 3 is better in terms of code maintainability, happy to make
>> it the change to it in v2.
> 
> Yes. Having a toggle that
> 
> a) Changes the default of new VMAs to be VM_HUGEPAGE
> 
> b) Modifies all existing VMAs that are not VM_NOHUGEPAGE to be VM_HUGEPAGE
> 
> c) Is inherited during fork+exec
> 
> Would not add any additional checks in our already-complicated THP allocation logic, it simply changes the default/value of VM_HUGEPAGE.
> 
> 
> At least to me, this sounds like better semantics. Would that make your use case happy?
> 

Yes, this would work for us. Let me try and make this work in the way you described and also make it work for
MADVISE and NEVER, will send that in v2.

Thanks for the feedback!!