Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 20 Sep 2021 12:16:39 -0700
From:   Sultan Alsawaf <sultan@kerneltoast.com>
To:     Michal Hocko <mhocko@suse.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        David Rientjes <rientjes@google.com>,
        Mel Gorman <mgorman@suse.de>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm: Mark the OOM reaper thread as freezable
Message-ID: <YUjeF6YsHKljSFis@sultan-box.localdomain>
References: <20210918233920.9174-1-sultan@kerneltoast.com>
 <YUiBRdrkjIdB/rSN@dhcp22.suse.cz>
 <YUiu42krQjSTVPnc@sultan-box.localdomain>
 <YUjGIuQciY7HNj+Y@dhcp22.suse.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YUjGIuQciY7HNj+Y@dhcp22.suse.cz>
Precedence: bulk

On Mon, Sep 20, 2021 at 07:34:26PM +0200, Michal Hocko wrote:
> The intention and the scope of the patch should be in the changelog.
> Your Fixes tag suggests there is a problem to fixed.

I guess References would be more appropriate here? I'm not familiar with every
subsystem's way of doing things, so I just rolled with Fixes to leave a
breadcrumb trail to the original commit implicated in my change. What would you
suggest in a case like this for mm patches?

> My memory has faded but I suspect it was to make sure that the oom
> reaper is not blocking the system wide freezing. The operation mode of
> the thread is to wait for oom victims and then do the unmapping without
> any blocking. While it can be frozen during the operation I do not
> remember that causing any problems and the waiting is exactly the point
> when that is obviously safe - hence wait_event_freezable which I believe
> is the proper API to use.

This isn't clear to me. Kthreads come with PF_NOFREEZE set by default, so the
system-wide freezing will already ignore the reaper thread as-is, although it
will make that determination from inside freeze_task() and thus
freezing_slow_path(), which involves acquiring a lock. You could set
PF_FREEZER_SKIP to skip the slowpath evaluation entirely.

Furthermore, the use of wait_event_freezable() will make the reaper thread enter
try_to_freeze() every time it's woken up. This seems wasteful considering that
the reaper thread will never actually freeze.

Sultan