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;
Message-ID: <efc5f0c3-bfb7-e36e-fa5b-60f94b49b7d4@redhat.com>
Date:   Fri, 2 Dec 2022 18:18:58 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.5.0
Content-Language: en-US
To:     Andrew Morton <akpm@linux-foundation.org>
Cc:     Peter Xu <peterx@redhat.com>, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Nadav Amit <nadav.amit@gmail.com>,
        Andrea Arcangeli <aarcange@redhat.com>,
        Ives van Hoorne <ives@codesandbox.io>,
        Axel Rasmussen <axelrasmussen@google.com>,
        Alistair Popple <apopple@nvidia.com>, stable@vger.kernel.org
References: <20221114000447.1681003-1-peterx@redhat.com>
 <20221114000447.1681003-2-peterx@redhat.com>
 <5ddf1310-b49f-6e66-a22a-6de361602558@redhat.com>
 <20221130142425.6a7fdfa3e5954f3c305a77ee@linux-foundation.org>
 <Y4jIHureiOd8XjDX@x1n> <a215fe2f-ef9b-1a15-f1c2-2f0bb5d5f490@redhat.com>
 <20221201143058.80296541cc6802d1e5990033@linux-foundation.org>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat
Subject: Re: [PATCH v3 1/2] mm/migrate: Fix read-only page got writable when
 recover pte
In-Reply-To: <20221201143058.80296541cc6802d1e5990033@linux-foundation.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 01.12.22 23:30, Andrew Morton wrote:
> On Thu, 1 Dec 2022 16:42:52 +0100 David Hildenbrand <david@redhat.com> wrote:
> 
>> On 01.12.22 16:28, Peter Xu wrote:
>>>
>>> I didn't reply here because I have already replied with the question in
>>> previous version with a few attempts.  Quotting myself:
>>>
>>> https://lore.kernel.org/all/Y3KgYeMTdTM0FN5W@x1n/
>>>
>>>           The thing is recovering the pte into its original form is the
>>>           safest approach to me, so I think we need justification on why it's
>>>           always safe to set the write bit.
>>>
>>> I've also got another longer email trying to explain why I think it's the
>>> other way round to be justfied, rather than justifying removal of the write
>>> bit for a read migration entry, here:
>>>
>>
>> And I disagree for this patch that is supposed to fix this hunk:
>>
>>
>> @@ -243,11 +243,15 @@ static bool remove_migration_pte(struct page *page, struct vm_area_struct *vma,
>>                   entry = pte_to_swp_entry(*pvmw.pte);
>>                   if (is_write_migration_entry(entry))
>>                           pte = maybe_mkwrite(pte, vma);
>> +               else if (pte_swp_uffd_wp(*pvmw.pte))
>> +                       pte = pte_mkuffd_wp(pte);
>>    
>>                   if (unlikely(is_zone_device_page(new))) {
>>                           if (is_device_private_page(new)) {
>>                                   entry = make_device_private_entry(new, pte_write(pte));
>>                                   pte = swp_entry_to_pte(entry);
>> +                               if (pte_swp_uffd_wp(*pvmw.pte))
>> +                                       pte = pte_mkuffd_wp(pte);
>>                           }
>>                   }
> 
> David, I'm unclear on what you mean by the above.  Can you please
> expand?
> 
>>
>> There is really nothing to justify the other way around here.
>> If it's broken fix it independently and properly backport it independenty.
>>
>> But we don't know about any such broken case.
>>
>> I have no energy to spare to argue further ;)
> 
> This is a silent data loss bug, which is about as bad as it gets.
> Under obscure conditions, fortunately.  But please let's keep working
> it.  Let's aim for something minimal for backporting purposes.  We can
> revisit any cleanliness issues later.
> 
> David, do you feel that the proposed fix will at least address the bug
> without adverse side-effects?

Just to answer that question clearly: it will fix this bug, but it's 
likely that other similar bugs remain (suspecting NUMA hinting).

Adverse side effect will be that some PTEs that could we writable won't 
be writable. I assume it's not too bad in practice for this particular case.

I proposed an alternative fix and identified other possible broken 
cases. Again, I don't NAK this patch as is, it just logically doesn't 
make sense to me to handle this case differently to the other 
vma->vm_page_prot users. (more details in the other thread)

-- 
Thanks,

David / dhildenb