[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <43595c4b-7f3f-f812-b4fe-66c900718e85@oracle.com>
Date: Wed, 24 Mar 2021 09:38:17 -0700
From: Mike Kravetz <mike.kravetz@...cle.com>
To: Michal Hocko <mhocko@...e.com>
Cc: linux-mm@...ck.org, linux-kernel@...r.kernel.org,
Shakeel Butt <shakeelb@...gle.com>,
Oscar Salvador <osalvador@...e.de>,
David Hildenbrand <david@...hat.com>,
Muchun Song <songmuchun@...edance.com>,
David Rientjes <rientjes@...gle.com>,
Miaohe Lin <linmiaohe@...wei.com>,
Peter Zijlstra <peterz@...radead.org>,
Matthew Wilcox <willy@...radead.org>,
HORIGUCHI NAOYA <naoya.horiguchi@....com>,
"Aneesh Kumar K . V" <aneesh.kumar@...ux.ibm.com>,
Waiman Long <longman@...hat.com>, Peter Xu <peterx@...hat.com>,
Mina Almasry <almasrymina@...gle.com>,
Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [RFC PATCH 5/8] hugetlb: change free_pool_huge_page to
remove_pool_huge_page
On 3/24/21 1:40 AM, Michal Hocko wrote:
> On Tue 23-03-21 18:03:07, Mike Kravetz wrote:
> [...]
>> Since you brought up cgroups ... what is your opinion on lock hold time
>> in hugetlb_cgroup_css_offline? We could potentially be calling
>> hugetlb_cgroup_move_parent for every hugetlb page while holding the lock
>> with interrupts disabled.
>
> I am not familiar with hugetlb cgroup code TBH. But from a quick look
> there is not much of heavy lifting there. If we find out that this is
> really visible we can do the lock dance with cond_resched and retry with
> the iteration again. Or is there any strong reason to process the list
> in a single go?
AFAICT, the primary reason for processing the list in a single go is
that the lock protects the list. If you drop the lock, the list can
change ...
I have come up with a (not so pretty) way of processing the list in
batches of pages. But, I dod not want to introduce that if there is no
need. Perhaps just take a wait and see approach for now.
I'll see if I can come up with some timing information to determine
if/when we may have an issue.
--
Mike Kravetz
Powered by blists - more mailing lists