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;
Content-Type: text/plain;
        charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 13.4 \(3608.120.23.2.4\))
Subject: Re: [RFC PATCH 11/13] fs/userfaultfd: complete write asynchronously
From:   Nadav Amit <nadav.amit@gmail.com>
In-Reply-To: <20201129004548.1619714-12-namit@vmware.com>
Date:   Tue, 1 Dec 2020 23:12:42 -0800
Cc:     Jens Axboe <axboe@kernel.dk>,
        Andrea Arcangeli <aarcange@redhat.com>,
        Peter Xu <peterx@redhat.com>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        io-uring@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org
Content-Transfer-Encoding: 7bit
Message-Id: <8405A413-239A-47E8-9D46-B6060EF86A68@gmail.com>
References: <20201129004548.1619714-1-namit@vmware.com>
 <20201129004548.1619714-12-namit@vmware.com>
To:     linux-fsdevel@vger.kernel.org
Precedence: bulk

> On Nov 28, 2020, at 4:45 PM, Nadav Amit <nadav.amit@gmail.com> wrote:
> 
> From: Nadav Amit <namit@vmware.com>
> 
> Userfaultfd writes can now be used for copy/zeroing. When using iouring
> with userfaultfd, performing the copying/zeroing on the faulting thread
> instead of the handler/iouring thread has several advantages:
> 
> (1) The data of the faulting thread will be available on the local
> caches, which would make subsequent memory accesses faster.
> 
> (2) find_vma() would be able to use the vma-cache, which cannot be done
> from a different process or io-uring kernel thread.
> 
> (3) The page is more likely to be allocated on the correct NUMA node.
> 
> To do so, userfaultfd work queue structs are extended to hold the
> information that is required for the faulting thread to copy/zero. The
> handler wakes one of the faulting threads to perform the copy/zero and
> that thread wakes the other threads after the zero/copy is done.

I noticed some bugs of mine in this patch, but more importantly I realized
that the there may be a more performant solution to do the copying on the
faulting thread - without async-writes.

Please do not review this patch and the next one (12/13).

Feedback for the rest of the series is of course welcomed.

Regards,
Nadav