Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Rei Yamamoto <yamamoto.rei@jp.fujitsu.com>
To:     akpm@linux-foundation.org
Cc:     aquini@redhat.com, ddutile@redhat.com, linmiaohe@huawei.com,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        mgorman@techsingularity.net, vvghjk1234@gmail.com,
        yamamoto.rei@jp.fujitsu.com
Subject: Re: [PATCH] mm, compaction: fast_find_migrateblock() should return pfn in the target zone
Date:   Mon, 16 May 2022 11:41:22 +0900
Message-Id: <20220516024122.71543-1-yamamoto.rei@jp.fujitsu.com>
In-Reply-To: <20220513140141.3b2accfc8fbae08ec4177781@linux-foundation.org>
References: <20220513140141.3b2accfc8fbae08ec4177781@linux-foundation.org>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Fri, 13 May 2022 14:01:41 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:
> On Fri, 13 May 2022 13:11:12 +0900 Rei Yamamoto <yamamoto.rei@jp.fujitsu.com> wrote:
>
>> On Thu, 12 May 2022 13:49:45 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:
>> > On Thu, 12 May 2022 13:27:33 +0900 Rei Yamamoto <yamamoto.rei@jp.fujitsu.com> wrote:
>> >
>> >> >> Thank you for your review.
>> >> >> The runtime effect is that compaction become unintended behavior.
>> >> >> For example, pages not in the target zone are added to cc->migratepages list in isolate_migratepages_block().
>> >> >> As a result, pages migrate between nodes unintentionally.
>> >> >
>> >> > Many thanks for clarifying. :) Is this worth a Fixes tag or even CC stable?
>> >> >
>> >> > Thanks!
>> >> 
>> >> Thank you for your reply.
>> >> 
>> >> If add a Fixes tag, I think the following commit:
>> >>   Fixes: 70b4459 ("mm, compaction: use free lists to quickly locate a migration source")
>> >> 
>> >> Andrew, how do you think about this? 
>> >
>> > Thanks, I added that and also a paragraph describing the effect of the bug:
>> >
>> > https://git.kernel.org/pub/scm/linux/kernel/git/akpm/25-new.git/tree/patches/mm-compaction-fast_find_migrateblock-should-return-pfn-in-the-target-zone.patch
>> >
>> > I assume this problem isn't sufficiently serious to require a -stable
>> > backport of the fix?
>> 
>> This would be a serious problem for older kernels without commit a984226, 
>> because it can corrupt the lru list by handling pages in list without holding proper lru_lock.
>
> Thanks, I added the above to the changelog.
>
> The patch applies OK to older kernels (I tried v5.10).  So I guess we
> put a cc:stable in this, so it gets backported?

Sounds great.
I think that's fine.

Thanks,
Rei