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: <20200428032745.133556-1-jannh@google.com> <20200428032745.133556-6-jannh@google.com>
 <CAHk-=wgBNSQhH1gyjo+Z2NFy4tOQnBQB4rra-jh+3XTpOjnThQ@mail.gmail.com>
In-Reply-To: <CAHk-=wgBNSQhH1gyjo+Z2NFy4tOQnBQB4rra-jh+3XTpOjnThQ@mail.gmail.com>
From:   Jann Horn <jannh@google.com>
Date:   Tue, 28 Apr 2020 08:10:51 +0200
Message-ID: <CAG48ez3kLgXQBOnSJJ+XuYpM__XnKc_AqJMoiPBhFbTdmoWxag@mail.gmail.com>
Subject: Re: [PATCH 5/5] mm/gup: Take mmap_sem in get_dump_page()
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Christoph Hellwig <hch@lst.de>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Linux-MM <linux-mm@kvack.org>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        "Eric W . Biederman" <ebiederm@xmission.com>,
        Oleg Nesterov <oleg@redhat.com>,
        Russell King <linux@armlinux.org.uk>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>,
        Mark Salter <msalter@redhat.com>,
        Aurelien Jacquiot <jacquiot.aurelien@gmail.com>,
        linux-c6x-dev@linux-c6x.org,
        Yoshinori Sato <ysato@users.sourceforge.jp>,
        Rich Felker <dalias@libc.org>,
        Linux-sh list <linux-sh@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Apr 28, 2020 at 5:50 AM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> On Mon, Apr 27, 2020 at 8:28 PM Jann Horn <jannh@google.com> wrote:
> >
> > Properly take the mmap_sem before calling into the GUP code from
> > get_dump_page(); and play nice, allowing __get_user_pages_locked() to drop
> > the mmap_sem if it has to sleep.
>
> This makes my skin crawl.
>
> The only reason for this all is that page cache flushing.
>
> My gut feeling is that it should be done by get_user_pages() anyway,
> since all the other users presumably want it to be coherent in the
> cache.
>
> And in fact, looking at __get_user_pages(), it already does that
>
>                 if (pages) {
>                         pages[i] = page;
>                         flush_anon_page(vma, page, start);
>                         flush_dcache_page(page);
>                         ctx.page_mask = 0;
>                 }
>
> and I think that the get_dump_page() logic is unnecessary to begin with.

Ah! And even though flush_cache_page() is broader than
flush_dcache_page(), that's actually unnecessary, right? Since the
kernel only wants to read from the page, and therefore e.g. the icache
is irrelevant?

Yay! :) I did think this was a bit gnarly, and it's nice to know that
this can be simplified.

(And now I'm going to avert my eyes from the GUP code before I start
thinking too hard about how much it sucks that FOLL_LONGTERM doesn't
drop the mmap_sem across the access and how much I dislike the whole
idea of FOLL_LONGTERM in general...)