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;
MIME-Version: 1.0
References: <CGME20200803061805epcas2p20faeeff0b31b23d1bc4464972285b561@epcas2p2.samsung.com>
 <1596435031-41837-1-git-send-email-pullip.cho@samsung.com>
 <5f41af0f-4593-3441-12f4-5b0f7e6999ac@redhat.com> <CAOJsxLHTL__N3KE_HUhQrVR=rkri80=kU1e64-6nBgMKHZRi4g@mail.gmail.com>
 <c0e87116-2633-c408-455b-c00c86607491@redhat.com> <20200818085444.GA72202@KEI>
In-Reply-To: <20200818085444.GA72202@KEI>
From:   Pekka Enberg <penberg@gmail.com>
Date:   Wed, 19 Aug 2020 16:12:32 +0300
Message-ID: <CAOJsxLGP3MVhfehfs5pYsAUmc7CnHCmgBFN7E-1u8spfD3MymA@mail.gmail.com>
Subject: Re: [PATCH] mm: sort freelist by rank number
To:     Cho KyongHo <pullip.cho@samsung.com>
Cc:     David Hildenbrand <david@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>, hyesoo.yu@samsung.com,
        janghyuck.kim@samsung.com
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi KyongHo and David,

On 07.08.20 09:08, Pekka Enberg wrote:
> > > I think having more knowledge of DRAM controller details in the OS
> > > would be potentially beneficial for better page allocation policy, so
> > > maybe try come up with something more generic, even if the fallback to
> > > providing this information is a kernel command line option.

On Tue, Aug 18, 2020 at 12:02 PM Cho KyongHo <pullip.cho@samsung.com> wrote:
> I don't find if there is a way to deliver detailed DRAM information
> through ACPI, ATAG or something similar. But I didn't find.

Perhaps that's still the case today then.

In the context of this patch, what I am hoping is that we'd turn the
"dram_rank_granule" parameter -- even if it's still a manually
configured thing -- into an abstraction that we can later extend. IOW,
something similar to what the "energy model"
(kernel/power/energy_model.c) today is.

On Mon, Aug 10, 2020 at 09:32:18AM +0200, David Hildenbrand wrote:
> I guess one universal approach is by measuring access times ... not what
> we might be looking for :)

I don't think it's an unreasonable approach, but measurement accuracy
and speed will determine if it's actually practical. There are
examples of this kind of approach elsewhere too. For example, MCTOP
[1] which aims to provide topology-aware OS abstractions, uses cache
latency measurements to infer the topology.

1. https://timharris.uk/papers/2017-eurosys-mctop.pdf

Regards,

- Pekka