From: Roman Gushchin <roman.gushchin@linux.dev>
To: Vasily Averin <vvs@openvz.org>
Cc: Shakeel Butt <shakeelb@google.com>,
Johannes Weiner <hannes@cmpxchg.org>,
kernel@openvz.org, LKML <linux-kernel@vger.kernel.org>,
Matthew Wilcox <willy@infradead.org>,
Cgroups <cgroups@vger.kernel.org>
Subject: Re: [PATCH] XArray: handle XA_FLAGS_ACCOUNT in xas_split_alloc
Date: Fri, 27 May 2022 10:47:46 -0700 [thread overview]
Message-ID: <YpEOwrBe2UzJZr9E@carbon> (raw)
In-Reply-To: <d4e81087-d057-4edb-5df0-47d99a1c72d9@openvz.org>
On Fri, May 27, 2022 at 02:22:19PM +0300, Vasily Averin wrote:
> On 5/27/22 04:40, Shakeel Butt wrote:
> > On Thu, May 26, 2022 at 6:21 PM Matthew Wilcox <willy@infradead.org> wrote:
> >>
> >> On Wed, May 25, 2022 at 11:26:37AM +0300, Vasily Averin wrote:
> >>> Commit 7b785645e8f1 ("mm: fix page cache convergence regression")
> >>> added support of new XA_FLAGS_ACCOUNT flag into all Xarray allocation
> >>> functions. Later commit 8fc75643c5e1 ("XArray: add xas_split")
> >>> introduced xas_split_alloc() but missed about XA_FLAGS_ACCOUNT
> >>> processing.
> >>
> >> Thanks, Vasily.
> >>
> >> Johannes, Shakeel, is this right? I don't fully understand the accounting
> >> stuff.
> >>
> >
> > If called from __filemap_add_folio() then this is correct.
> >
> > However from split_huge_page_to_list(), we can not use the memcg from
> > current as that codepath is called from reclaim which can be triggered
> > by processes of other memcgs.
> Btw, Shakeel, Johannes,
> I would like to understand, when Xarray should use XA_FLAGS_ACCOUNT ?
>
> From my point of view, this should be useless:
> a) if Xarray stores some index (idr?) - his memory is quite small,
> and his accounting can be ignored.
> b) if Xarray stores some accounted - the size of the corresponding Xarray
> infrastructure is usually significantly smaller than the size of the stored object,
> sо his accounting can be skipped too.
> c) if Xarray stores some non-accounted objects - it makes no sense to account
> corresponding Xarray infrastructure. In case of necessary it makes much more sense
> to enable accounting for stored objects (and return to case b).
>
> Am I missed something important perhaps?
>
> I looked for the description of 7b785645e8f1, but o be honest I'm still not sure
> that I understand correctly why XA_FLAGS_ACCOUNT flag solved the described problem.
>
> Could you please explain this in more details?
>
> Was it because the non-accounted Xarray kept a reference to the stored object
> and thus prevents it from being reclaimed?
>
> If so, was it some special case, or should it affect all such cases,
> and my b) statement above is not correct?
It's all about shadow entries, which are small, so b) is not true for them.
There is a good description on how it works on top of mm/workingset.c
prev parent reply other threads:[~2022-05-27 17:48 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-05-25 8:26 [PATCH] XArray: handle XA_FLAGS_ACCOUNT in xas_split_alloc Vasily Averin
2022-05-27 1:21 ` Matthew Wilcox
2022-05-27 1:40 ` Shakeel Butt
2022-05-27 11:22 ` Vasily Averin
2022-05-27 17:47 ` Roman Gushchin [this message]
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=YpEOwrBe2UzJZr9E@carbon \
--to=roman.gushchin@linux.dev \
--cc=cgroups@vger.kernel.org \
--cc=hannes@cmpxchg.org \
--cc=kernel@openvz.org \
--cc=linux-kernel@vger.kernel.org \
--cc=shakeelb@google.com \
--cc=vvs@openvz.org \
--cc=willy@infradead.org \
/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;
as well as URLs for NNTP newsgroup(s).