Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates
 216.228.117.160 as permitted sender) receiver=protection.outlook.com;
 client-ip=216.228.117.160; helo=mail.nvidia.com; pr=C
Message-ID: <d07c45bb-4eba-484c-bba1-b586136f8cdb@nvidia.com>
Date:   Mon, 30 Oct 2023 16:25:38 -0700
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v6 5/9] mm: thp: Extend THP to allocate anonymous large
 folios
Content-Language: en-US
To:     Ryan Roberts <ryan.roberts@arm.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Matthew Wilcox <willy@infradead.org>,
        "Yin Fengwei" <fengwei.yin@intel.com>,
        David Hildenbrand <david@redhat.com>,
        "Yu Zhao" <yuzhao@google.com>,
        Catalin Marinas <catalin.marinas@arm.com>,
        "Anshuman Khandual" <anshuman.khandual@arm.com>,
        Yang Shi <shy828301@gmail.com>,
        "Huang, Ying" <ying.huang@intel.com>, Zi Yan <ziy@nvidia.com>,
        Luis Chamberlain <mcgrof@kernel.org>,
        Itaru Kitayama <itaru.kitayama@gmail.com>,
        "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
        David Rientjes <rientjes@google.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Hugh Dickins <hughd@google.com>
CC:     <linux-mm@kvack.org>, <linux-kernel@vger.kernel.org>,
        <linux-arm-kernel@lists.infradead.org>
References: <20230929114421.3761121-1-ryan.roberts@arm.com>
 <20230929114421.3761121-6-ryan.roberts@arm.com>
 <8a72da61-b2ef-48ad-ae59-0bae7ac2ce10@nvidia.com>
 <5993c198-0d27-46c3-b757-3a02c2aacfc9@arm.com>
From:   John Hubbard <jhubbard@nvidia.com>
In-Reply-To: <5993c198-0d27-46c3-b757-3a02c2aacfc9@arm.com>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 8bit
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 30 Oct 2023 23:25:54.3960
 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: 4ad3013c-3984-4902-8b3b-08dbd99f902e
X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a;Ip=[216.228.117.160];Helo=[mail.nvidia.com]
X-MS-Exchange-CrossTenant-AuthSource: CY4PEPF0000E9CE.namprd03.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: BL3PR12MB6596
X-Spam-Status: No, score=-1.6 required=5.0 tests=BAYES_00,DKIMWL_WL_HIGH,
        DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FORGED_SPF_HELO,
        RCVD_IN_DNSWL_BLOCKED,RCVD_IN_MSPIKE_H2,SPF_HELO_PASS,SPF_NONE
        autolearn=no autolearn_force=no version=3.4.6
X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on
        lindbergh.monkeyblade.net
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
X-Greylist: Sender passed SPF test, not delayed by milter-greylist-4.6.4 (snail.vger.email [0.0.0.0]); Mon, 30 Oct 2023 16:26:03 -0700 (PDT)

On 10/30/23 04:43, Ryan Roberts wrote:
> On 28/10/2023 00:04, John Hubbard wrote:
>> On 9/29/23 04:44, Ryan Roberts wrote:
...
>>>    +static bool vmf_pte_range_changed(struct vm_fault *vmf, int nr_pages)
>>> +{
>>> +    int i;
>>> +
>>> +    if (nr_pages == 1)
>>> +        return vmf_pte_changed(vmf);
>>> +
>>> +    for (i = 0; i < nr_pages; i++) {
>>> +        if (!pte_none(ptep_get_lockless(vmf->pte + i)))
>>> +            return true;
>>
>> This seems like something different than the function name implies.
>> It's really confusing: for a single page case, return true if the
>> pte in the page tables has changed, yes that is very clear.
>>
>> But then for multiple page cases, which is really the main
>> focus here--for that, claim that the range has changed if any
>> pte is present (!pte_none). Can you please help me understand
>> what this means?
> 
> Yes I understand your confusion. Although I'm confident that the code is
> correct, its a bad name - I'll make the excuse that this has evolved through
> rebasing to cope with additions to UFFD. Perhaps something like
> vmf_is_large_folio_suitable() is a better name.
> 
> It used to be that we would only take the do_anonymous_page() path if the pte
> was none; i.e. this is the first time we are faulting on an address covered by
> an anon VMA and we need to allocate some memory. But more recently we also end
> up here if the pte is a uffd_wp marker. So for a single pte, instead of checking
> none, we can check if the pte has changed from our original check (where we
> determined it was a uffd_wp marker or none). But for multiple ptes, we don't
> have storage to store all the original ptes from the first check.
> 
> Fortunately, if uffd is in use for a vma, then we don't want to use a large
> folio anyway (this would break uffd semantics because we would no longer get a
> fault for every page). So we only care about the "same but not none" case for
> nr_pages=1.
> 
> Would changing the name to vmf_is_large_folio_suitable() help here?

Yes it would! And adding in a sentence or two from above about the uffd, as
a function-level comment might be just the right of demystification for
the code.

...
pte_offset_map() can only fail due to:
>>
>>      a) Wrong pmd type. These include:
>>          pmd_none
>>          pmd_bad
>>          pmd migration entry
>>          pmd_trans_huge
>>          pmd_devmap
>>
>>      b) __pte_map() failure
>>
>> For (a), why is it that -EAGAIN is used here? I see that that
>> will lead to a re-fault, I got that far, but am missing something
>> still.
>>
>> For (b), same question, actually. I'm not completely sure why
>> why a retry is going to fix a __pte_map() failure?
> 
> I'm not going to claim to understand all the details of this. But this is due to
> a change that Hugh introduced and we concluded at [1] that its always correct to
> return EAGAIN here to rerun the fault. In fact, with the current implementation
> pte_offset_map() should never fail for anon IIUC, but the view was that EAGAIN
> makes it safe for tomorrow, and because this would only fail due to a race,
> retrying is correct.
> 
> [1] https://lore.kernel.org/linux-mm/8bdfd8d8-5662-4615-86dc-d60259bd16d@google.com/
> 

OK, got it.

...
>> And finally: is it accurate to say that there are *no* special
>> page flags being set, for PTE-mapped THPs? I don't see any here,
>> but want to confirm.
> 
> The page flags are coming from 'gfp = vma_thp_gfp_mask(vma)', which pulls in the
> correct flags based on transparent_hugepage/defrag file.
> 

OK that all is pretty clear now, thanks for the answers!


thanks,
-- 
John Hubbard
NVIDIA