Re: [PATCH] mm: Mark idle page tracking as BROKEN

From: SeongJae Park <sj38.park@gmail.com>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: SeongJae Park <sj38.park@gmail.com>,
	"Matthew Wilcox (Oracle)" <willy@infradead.org>,
	akpm@linuxfoundation.org, linux-mm@kvack.org,
	Heiko Carstens <hca@linux.ibm.com>,
	Rafael Aquini <aquini@redhat.com>,
	Vlastimil Babka <vbabka@suse.cz>, Yu Zhao <yuzhao@google.com>,
	Vladimir Davydov <vdavydov.dev@gmail.com>,
	kirill.shutemov@linux.intel.com, amit@kernel.org,
	sjpark@amazon.de
Subject: Re: [PATCH] mm: Mark idle page tracking as BROKEN
Date: Tue, 15 Jun 2021 06:40:28 +0000	[thread overview]
Message-ID: <20210615064028.18813-1-sjpark@amazon.de> (raw)
In-Reply-To: <20210614190420.af6bb3ca193541cdb606517d@linux-foundation.org>

From: SeongJae Park <sjpark@amazon.de>

On Mon, 14 Jun 2021 19:04:20 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:

> On Mon, 14 Jun 2021 13:49:26 +0000 SeongJae Park <sj38.park@gmail.com> wrote:
> 
> > From: SeongJae Park <sjpark@amazon.de>
> > 
> > Hello Matthew,
> > 
> > On Sat, 12 Jun 2021 01:07:14 +0100 "Matthew Wilcox (Oracle)" <willy@infradead.org> wrote:
> > 
> > > In discussion with other MM developers around how idle page tracking
> > > should be fixed for transparent huge pages, several expressed the opinion
> > > that it should be removed as it is inefficient at accomplishing the
> > > job that it is supposed to, and we have better mechanisms (eg uffd) for
> > > accomplishing the same goals these days.
> > 
> > I think the THP case[1] is an intended behavior[2].  Could you please share a
> > link to the discussion or a detailed summary if possible?
> > 
> > > 
> > > Mark the feature as BROKEN for now and we can remove it entirely in a
> > > few months if nobody complains.  It is not enabled by Android, ChromeOS,
> > > Debian, Fedora or SUSE.  Red Hat enabled it with RHEL-8.1 and UEK followed
> > > suit, but I have been unable to find why RHEL enabled it.
> > 
> > Amazon Linux is also using it[3], for DAMON[4].  In detail, DAMON doesn't use
> > Idle Page Tracking but PG_Idle in kernel space, to avoid interfering the
> > reclaim logic[5].  So, I'm ok with removing the Idle Page Tracking user space
> > interface, but gonna be opposed to removing PG_Idle.
> > 
> > Nevertheless, the interference is not a real problem to DAMON, as DAMON is
> > aimed to provide just a reasonable quality of the monitoring, rather than
> > strict correctness.  Hence, if people think the interference is also not a
> > problem for the reclaim logic (after all, it does nothing unless sysadmin
> > manually turns it on in runtime, and can be turned off at anytime), I would
> > simply update DAMON code to don't use PG_Idle, add warnings in the doc, and
> > wouldn't be opposed to this change.
> 
> Couldn't the DAMON patchset simply re-add PG_Idle?  Perhaps with a new
> name which is more appropriate to the DAMON usage?

Good point, that makes sense.  Thank you, Andrew.

Nevertheless, I still not fully understand why Idle Page Tracking is considered
broken, and therefore worrying if I will end up reintroducing PG_Idle with
DAMON patchset in the broken form.  It would still be great if a link to the
dicussion could be provided.

Thanks,
SeongJae Park