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;
Message-ID: <e2580cfa-a529-934d-861a-091c4a9714d4@redhat.com>
Date:   Thu, 20 Jan 2022 16:26:22 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.4.0
Content-Language: en-US
To:     Matthew Wilcox <willy@infradead.org>
Cc:     "zhangliang (AG)" <zhangliang5@huawei.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linux-MM <linux-mm@kvack.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        wangzhigang17@huawei.com,
        Linus Torvalds <torvalds@linux-foundation.org>
References: <20220113140318.11117-1-zhangliang5@huawei.com>
 <YeA5oP/iaxtVPHb3@casper.infradead.org>
 <CAHk-=wjB0i-B=U-DhpAajQx3f6bp1X==neLOrg0jwq29mgz=3g@mail.gmail.com>
 <172ccfbb-7e24-db21-7d84-8c8d8c3805fd@redhat.com>
 <a93988da-80fb-dd32-4717-a6a0bae9e4ee@huawei.com>
 <dc415c4a-63aa-19b0-0fbc-795989970f6d@redhat.com>
 <fb02087a-b102-c91e-ab65-fb02cc8ee0a2@huawei.com>
 <9cd7eee2-91fd-ddb8-e47d-e8585e5baa05@redhat.com>
 <b6df4f7f-c080-ad6c-d1ad-098115f016f3@huawei.com>
 <747ff31c-6c9e-df6c-f14d-c43aa1c77b4a@redhat.com>
 <Yel0BXVyj8uvsWJX@casper.infradead.org>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat
Subject: Re: [PATCH] mm: reuse the unshared swapcache page in do_wp_page
In-Reply-To: <Yel0BXVyj8uvsWJX@casper.infradead.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 20.01.22 15:39, Matthew Wilcox wrote:
> On Thu, Jan 20, 2022 at 03:15:37PM +0100, David Hildenbrand wrote:
>> On 17.01.22 14:31, zhangliang (AG) wrote:
>>> Sure, I will do that :)
>>
>> I'm polishing up / testing the patches and might send something out for discussion shortly.
>> Just a note that on my branch was a version with a wrong condition that should have been fixed now.
>>
>> I am still thinking about PTE mapped THP. For these, we'll always
>> have page_count() > 1, essentially corresponding to the number of still-mapped sub-pages.
>>
>> So if we end up with a R/O mapped part of a THP, we'll always have to COW and cannot reuse ever,
>> although it's really just a single process mapping the THP via PTEs.
>>
>> One approach would be to scan the currently locked page table for entries mapping
>> this same page. If page_count() corresponds to that value, we know that only we are
>> mapping the THP and there are no additional references. That would be a special case
>> if we find an anon THP in do_wp_page(). Hm.
> 
> You're starting to optimise for some pretty weird cases at that point.

So your claim is that read-only, PTE mapped pages are weird? How do you
come to that conclusion?

If we adjust the THP reuse logic to split on additional references
(page_count() == 1) -- similarly as suggested by Linus to fix the CVE --
we're going to end up with exactly that more frequently.

> Anon THP is always going to start out aligned (and can be moved by
> mremap()).  Arguably it should be broken up if it's moved so it can be
> reformed into aligned THPs by khugepaged.

Can you elaborate, I'm missing the point where something gets moved. I
don't care about mremap() at all here.


1. You have a read-only, PTE mapped THP
2. Write fault on the THP
3. We PTE-map the THP because we run into a false positive in our COW
   logic to handle COW on PTE
4. Write fault on the PTE
5. We always have to COW each and every sub-page and can never reuse,
   because page_count() > 1

That's essentially what reuse_swap_page() tried to handle before.
Eventually optimizing for this is certainly the next step, but I'd like
to document which effect the removal of reuse_swap_page() will have to THP.

-- 
Thanks,

David / dhildenb