Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 18 Apr 2023 15:21:09 -0700
From:   Andrew Morton <akpm@linux-foundation.org>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Vishal Moola <vishal.moola@gmail.com>,
        Peng Zhang <zhangpeng362@huawei.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, mike.kravetz@oracle.com,
        sidhartha.kumar@oracle.com, muchun.song@linux.dev,
        wangkefeng.wang@huawei.com, sunnanyong@huawei.com
Subject: Re: [PATCH v5 3/6] userfaultfd: convert copy_huge_page_from_user()
 to copy_folio_from_user()
Message-Id: <20230418152109.32a90ebecc905ae60e2d072f@linux-foundation.org>
In-Reply-To: <ZDTWoTTtS1JcuiI6@casper.infradead.org>
References: <20230331093937.945725-1-zhangpeng362@huawei.com>
        <20230331093937.945725-4-zhangpeng362@huawei.com>
        <CAOzc2pzr7VJRdsx1ud_ceBhbu2XP7Ay72jFETtN8eOt5yR7S=Q@mail.gmail.com>
        <ZDTWoTTtS1JcuiI6@casper.infradead.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Tue, 11 Apr 2023 04:40:17 +0100 Matthew Wilcox <willy@infradead.org> wrote:

> On Thu, Apr 06, 2023 at 07:28:44PM -0700, Vishal Moola wrote:
> > > -               flush_dcache_page(subpage);
> > > -
> > >                 cond_resched();
> > >         }
> > > +       flush_dcache_folio(dst_folio);
> > >         return ret_val;
> > >  }
> > 
> > Moving the flush_dcache_page() outside the loop to be
> > flush_dcache_folio() changes the behavior of the function.
> > 
> > Initially, if it fails to copy the entire page, the function breaks out
> > of the loop and returns the number of unwritten bytes without
> > flushing the page from the cache. Now if it fails, it will still flush
> > out the page it failed on, as well as any later pages it may not
> > have gotten to yet.
> 
> I'm not sure this is worth worrying about.  Failing to copy the entire
> folio is unlikely, and if we do, flushing the entire folio instead of just
> a few pages in it is harmless.  Plus I have patches which significantly
> optiise flush_dcache_folio() over flush_dcache_page() (for the majority
> of architectures) and so I think this change is actually beneficial in
> the long term.

Thanks, I'll send the series in for the next merge window as-is.  If
others remain unhappy with the flushing issue, please propose something
during the next -rc cycle.