linux-kernel - Re: [PATCH 1/4] mm: migrate: use a folio in add_page_for

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <F2621E68-F36E-493C-8619-ADFE05050823@nvidia.com>
Date:   Thu, 03 Aug 2023 22:42:22 -0400
From:   Zi Yan <ziy@...dia.com>
To:     Kefeng Wang <wangkefeng.wang@...wei.com>
Cc:     Matthew Wilcox <willy@...radead.org>,
        Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, Huang Ying <ying.huang@...el.com>,
        David Hildenbrand <david@...hat.com>
Subject: Re: [PATCH 1/4] mm: migrate: use a folio in add_page_for_migration()

On 3 Aug 2023, at 21:45, Kefeng Wang wrote:

> On 2023/8/3 20:30, Matthew Wilcox wrote:
>> On Thu, Aug 03, 2023 at 03:13:21PM +0800, Kefeng Wang wrote:
>>>
>>>
>>> On 2023/8/2 20:21, Matthew Wilcox wrote:
>>>> On Wed, Aug 02, 2023 at 05:53:43PM +0800, Kefeng Wang wrote:
>>>>>    	err = -EACCES;
>>>>> -	if (page_mapcount(page) > 1 && !migrate_all)
>>>>> -		goto out_putpage;
>>>>> +	if (folio_estimated_sharers(folio) > 1 && !migrate_all)
>>>>> +		goto out_putfolio;
>>>>
>>>> I do not think this is the correct change.  Maybe leave this line
>>>> alone.
>>>
>>> Ok, I am aware of the discussion about this in other mail, will not
>>> change it(also the next two patch about this function), or wait the
>>> new work of David.
>>>>
>>>>> -	if (PageHuge(page)) {
>>>>> -		if (PageHead(page)) {
>>>>> -			isolated = isolate_hugetlb(page_folio(page), pagelist);
>>>>> +	if (folio_test_hugetlb(folio)) {
>>>>> +		if (folio_test_large(folio)) {
>>>>
>>>> This makes no sense when you read it.  All hugetlb folios are large,
>>>> by definition.  Think about what this code used to do, and what it
>>>> should be changed to.
>>>
>>> hugetlb folio is self large folio, will drop redundant check
>>
>> No, that's not the difference.  Keep thinking about it.  This is not
>> a mechanical translation!
>
>
>   if (PageHuge(page))  // page must be a hugetlb page
> 	if (PageHead(page)) // page must be a head page, not tail
>              isolate_hugetlb() // isolate the hugetlb page if head
>
> After using folio,
>
>   if (folio_test_hugetlb(folio)) // only check folio is hugetlb or not
>
> I don't check the page is head or not, since the follow_page could
> return a sub-page, so the check PageHead need be retained, right?

Right. It will prevent the kernel from trying to isolate the same hugetlb page
twice when two pages are in the same hugetlb folio. But looking at the
code, if you try to isolate an already-isolated hugetlb folio, isolate_hugetlb()
would return false, no error would show up. But it changes err value
from -EACCES to -EBUSY and user will see a different page status than before.

I wonder why we do not have follow_folio() and returns -ENOENT error pointer
when addr points to a non head page. It would make this patch more folio if
follow_folio() can be used in place of follow_page(). One caveat is that
user will see -ENOENT instead of -EACCES after this change.


--
Best Regards,
Yan, Zi

Download attachment "signature.asc" of type "application/pgp-signature" (855 bytes)