Linux-mm Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "David Hildenbrand (Arm)" <david@kernel.org>
To: Joanne Koong <joannelkoong@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Kiryl Shutsemau <kas@kernel.org>,
	Usama Arif <usama.arif@linux.dev>, Barry Song <baohua@kernel.org>,
	Lance Yang <lance.yang@linux.dev>,
	Lorenzo Stoakes <ljs@kernel.org>, Rik van Riel <riel@surriel.com>,
	"Liam R . Howlett" <liam@infradead.org>,
	Vlastimil Babka <vbabka@kernel.org>, Harry Yoo <harry@kernel.org>,
	Jann Horn <jannh@google.com>, Zi Yan <ziy@nvidia.com>,
	SJ Park <sj@kernel.org>, Johannes Weiner <hannes@cmpxchg.org>,
	linux-mm@kvack.org, Sashiko <sashiko-bot@kernel.org>,
	azpijr@gmail.com
Subject: Re: [PATCH v1] mm/rmap: fix migration entries being treated as partial unmaps
Date: Sun, 22 Nov 2026 17:11:48 +0100	[thread overview]
Message-ID: <4a017c79-db13-4fe7-8161-36400596fd6c@kernel.org> (raw)
In-Reply-To: <CAJnrk1YWf=rJc2uB-sCwxSVFa8KperRn30bJ3m_1Q-026P68BQ@mail.gmail.com>

On 11/14/26 00:51, Joanne Koong wrote:
> On Fri, Nov 13, 2026 at 11:23 AM David Hildenbrand (Arm)
> <david@kernel.org> wrote:
>>
>> On 11/13/26 00:28, Joanne Koong wrote:
>>> On Tue, Nov 10, 2026 at 3:49 AM David Hildenbrand (Arm)
>>> <david@kernel.org> wrote:
>>>
>>> As I understand it, there's no simple way to check if a folio is
>>> partially mapped or not under CONFIG_NO_PAGE_MAPCOUNT without doing a
>>> rmap walk.
>>
>> Yes. With CONFIG_PAGE_MAPCOUNT you can (still) read the information from the
>> folio, but the grand plan is to phase that out and remove it.
>>
>> I just keep getting distracted :)
>>
>>> Would it make sense to do something simpler here instead
>>> where we just check the folio's partially mapped flag and cache that
>>> before we replace the PTEs with migration entries, and then for those
>>> edge cases that the batching solution doesn't cover, just restore that
>>> flag back to its original state?
>>
>> As a last resort if we cannot get something else working (and we really care).
>> It is rather hacky.
>>
>>>
>>> I think this would add coverage for the migration and reclaim paths,
>>> instead of just the deferred split shrinker, but would miss the race
>>> where someone tries to do a partial unmap while the folio has the
>>> migration PTEs, but I think this is already a race that exists in the
>>> batching case. imo it seems acceptable though given that the race is
>>> pretty rare and doesn't have any adverse effects except delaying
>>> freeing the unmapped pages until the whole folio gets unmapped.
>>>
>>> Do you think something like this makes sense, or would you prefer
>>> having the check in the deferred split handler?
>>
>> The question is, do we really care once we have the PTE batching in place?
>>
>> We would still split for if we cannot batch (multiple VMAs, PTE differences that
>> we have to respect). Do we have any idea how much that matters in practice?
>>
>> (I raised once that, just because a folio is fully mapped doesn't really telly
>> you anything. It could be mapped into various processes partially, or scattered
>> in a single process.)
>>
> 
> Sounds good! I'll put this on hold until it's proven to actually be
> needed after Shivank's series.

Thinking again, just saying "we're unmapping for migration, so just don't mess
with partial mapped state" through a new rmap flag that we pass does sound
reasonable.

But ideally this flag would have an abstracted meaning (e.g., this is temporary
system-triggered unmap due to migration/split/swapout/whatever) and not some
thing concrete like "defer split".

Yes, we'll not handle when someone actually triggers partial unmapping at the
same time by doing MADV_DONTNEED etc. But I guess we shouldn't worry about that
for now.

Doing the folio PTE batching still makes sense.

-- 
Cheers,

David


  reply	other threads:[~2026-11-22 16:12 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-09 22:09 [PATCH v1] mm/rmap: fix migration entries being treated as partial unmaps Joanne Koong
2026-11-10 11:48 ` David Hildenbrand (Arm)
2026-11-10 20:21   ` Joanne Koong
2026-11-10 20:55     ` David Hildenbrand (Arm)
2026-11-10 21:05       ` Zi Yan
2026-11-10 23:26         ` Joanne Koong
2026-11-12 23:28   ` Joanne Koong
2026-11-13 19:23     ` David Hildenbrand (Arm)
2026-11-13 23:51       ` Joanne Koong
2026-11-22 16:11         ` David Hildenbrand (Arm) [this message]
2026-11-10 22:13 ` sashiko-bot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4a017c79-db13-4fe7-8161-36400596fd6c@kernel.org \
    --to=david@kernel.org \
    --cc=akpm@linux-foundation.org \
    --cc=azpijr@gmail.com \
    --cc=baohua@kernel.org \
    --cc=hannes@cmpxchg.org \
    --cc=harry@kernel.org \
    --cc=jannh@google.com \
    --cc=joannelkoong@gmail.com \
    --cc=kas@kernel.org \
    --cc=lance.yang@linux.dev \
    --cc=liam@infradead.org \
    --cc=linux-mm@kvack.org \
    --cc=ljs@kernel.org \
    --cc=riel@surriel.com \
    --cc=sashiko-bot@kernel.org \
    --cc=sj@kernel.org \
    --cc=usama.arif@linux.dev \
    --cc=vbabka@kernel.org \
    --cc=ziy@nvidia.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox