MIME-Version: 1.0
References: <20211201193750.2097885-1-catalin.marinas@arm.com>
 <CAHc6FU7gXfZk7=Xj+RjxCqkmsrcAhenfbeoqa4AmHd5+vgja7g@mail.gmail.com>
 <CAHk-=wiQAQTGdMNLCKwgnt4EiAXf7Bm6p7NQx5-31S9-qPD8jg@mail.gmail.com> <CAHc6FU6r-CsMHkWzxEm237mV2vZ2O9g_D7BbCPeaA2qX0dpi0g@mail.gmail.com>
In-Reply-To: <CAHc6FU6r-CsMHkWzxEm237mV2vZ2O9g_D7BbCPeaA2qX0dpi0g@mail.gmail.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Fri, 3 Dec 2021 10:25:17 -0800
Message-ID: <CAHk-=wi12AStfPrXLjki_SLc5qqDwYX21bJLp10mynNQj7u8FA@mail.gmail.com>
Subject: Re: [PATCH v2 0/4] Avoid live-lock in fault-in+uaccess loops with
 sub-page faults
To:     Andreas Gruenbacher <agruenba@redhat.com>
Cc:     Catalin Marinas <catalin.marinas@arm.com>,
        Josef Bacik <josef@toxicpanda.com>,
        David Sterba <dsterba@suse.com>,
        Al Viro <viro@zeniv.linux.org.uk>,
        Andrew Morton <akpm@linux-foundation.org>,
        Will Deacon <will@kernel.org>,
        Matthew Wilcox <willy@infradead.org>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        linux-btrfs <linux-btrfs@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Dec 3, 2021 at 10:12 AM Andreas Gruenbacher <agruenba@redhat.com> wrote:
>
> It happens when you mmap a file and write the mmapped region to
> another file, for example.

Do you actually have such loads? Nobody should use mmap() for
single-access file copy purposes. It's slower than just doing the copy
exactly due to page fault overhead.

In other words, you seem to be worrying about the performance of a
load that is _explicitly_ badly written. You should be fixing the
application, not making the kernel do stupid things.

Also, it's worth noting that that situation should be caught by the
page-in code, which will map multiple pages in one go
(do_fault_around() - for when the pages are cached), and do the
readahead logic (filemap_fault() - for when the pages aren't in the
page cache).

Both of which are a lot more important than the "synchronously fault
in pages one at a time".

                Linus