Date:   Tue, 23 Nov 2021 09:25:23 -0400
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     David Hildenbrand <david@redhat.com>
Cc:     Jens Axboe <axboe@kernel.dk>,
        Andrew Dona-Couch <andrew@donacou.ch>,
        Andrew Morton <akpm@linux-foundation.org>,
        Drew DeVault <sir@cmpwn.com>,
        Ammar Faizi <ammarfaizi2@gnuweeb.org>,
        linux-kernel@vger.kernel.org, linux-api@vger.kernel.org,
        io_uring Mailing List <io-uring@vger.kernel.org>,
        Pavel Begunkov <asml.silence@gmail.com>, linux-mm@kvack.org
Subject: Re: [PATCH] Increase default MLOCK_LIMIT to 8 MiB
Message-ID: <20211123132523.GA5112@ziepe.ca>
References: <20211115203530.62ff33fdae14927b48ef6e5f@linux-foundation.org>
 <CFQZSHV700KV.18Y62SACP8KOO@taiga>
 <20211116114727.601021d0763be1f1efe2a6f9@linux-foundation.org>
 <CFRGQ58D9IFX.PEH1JI9FGHV4@taiga>
 <20211116133750.0f625f73a1e4843daf13b8f7@linux-foundation.org>
 <b84bc345-d4ea-96de-0076-12ff245c5e29@redhat.com>
 <8f219a64-a39f-45f0-a7ad-708a33888a3b@www.fastmail.com>
 <333cb52b-5b02-648e-af7a-090e23261801@redhat.com>
 <ca96bb88-295c-ccad-ed2f-abc585cb4904@kernel.dk>
 <5f998bb7-7b5d-9253-2337-b1d9ea59c796@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <5f998bb7-7b5d-9253-2337-b1d9ea59c796@redhat.com>
Precedence: bulk

On Mon, Nov 22, 2021 at 09:08:47PM +0100, David Hildenbrand wrote:

> > You can't use mmu notifiers without impacting the fast path. This isn't
> > just about io_uring, there are other users of memlock right now (like
> > bpf) which just makes it even worse.
> 
> 1) Do we have a performance evaluation? Did someone try and come up with
> a conclusion how bad it would be?

It needs additional locking between page readers and the mmu notifier.

One of the virtio things does this thing and they used rcu on the page
readers and a synchronize rcu in a mmu notifier - which I think is
pretty bad.

> 2) Could be provide a mmu variant to ordinary users that's just good
> enough but maybe not as fast as what we have today? And limit
> FOLL_LONGTERM to special, privileged users?

rdma has never been privileged

Jason