linux-kernel - Re: [PATCH 4/9] mm/compaction: simplify pfn iteration in isolate_freepages

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: <d9403099-7fa8-ba34-4260-21da36175432@linux.alibaba.com>
Date:   Sat, 19 Aug 2023 19:58:50 +0800
From:   Baolin Wang <baolin.wang@...ux.alibaba.com>
To:     Kemeng Shi <shikemeng@...weicloud.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, akpm@...ux-foundation.org,
        mgorman@...hsingularity.net, david@...hat.com
Subject: Re: [PATCH 4/9] mm/compaction: simplify pfn iteration in
 isolate_freepages_range



On 8/15/2023 6:37 PM, Kemeng Shi wrote:
> 
> 
> on 8/15/2023 6:07 PM, Baolin Wang wrote:
>>
>>
>> On 8/15/2023 5:32 PM, Kemeng Shi wrote:
>>>
>>>
>>> on 8/15/2023 4:38 PM, Baolin Wang wrote:
>>>>
>>>>
>>>> On 8/5/2023 7:07 PM, Kemeng Shi wrote:
>>>>> We call isolate_freepages_block in strict mode, continuous pages in
>>>>> pageblock will be isolated if isolate_freepages_block successed.
>>>>> Then pfn + isolated will point to start of next pageblock to scan
>>>>> no matter how many pageblocks are isolated in isolate_freepages_block.
>>>>> Use pfn + isolated as start of next pageblock to scan to simplify the
>>>>> iteration.
>>>>
>>>> IIUC, the isolate_freepages_block() can isolate high-order free pages, which means the pfn + isolated can be larger than the block_end_pfn. So in your patch, the 'block_start_pfn' and 'block_end_pfn' can be in different pageblocks, that will break pageblock_pfn_to_page().
>>>>
>>> In for update statement, we always update block_start_pfn to pfn and
>>
>> I mean, you changed to:
>> 1) pfn += isolated;
>> 2) block_start_pfn = pfn;
>> 3) block_end_pfn = pfn + pageblock_nr_pages;
>>
>> But in 1) pfn + isolated can go outside of the currnet pageblock if isolating a high-order page, for example, located in the middle of the next pageblock. So that the block_start_pfn can point to the middle of the next pageblock, not the start position. Meanwhile after 3), the block_end_pfn can point another pageblock. Or I missed something else?
>>
> Ah, I miss to explain this in changelog.
> In case we could we have buddy page with order higher than pageblock:
> 1. page in buddy page is aligned with it's order
> 2. order of page is higher than pageblock order
> Then page is aligned with pageblock order. So pfn of page and isolated pages
> count are both aligned pageblock order. So pfn + isolated is pageblock order
> aligned.

That's not what I mean. pfn + isolated is not always pageblock-aligned, 
since the isolate_freepages_block() can isolated high-order free pages 
(for example: order-1, order-2 ...).

Suppose the pageblock size is 2M, when isolating a pageblock (suppose 
the pfn range is 0 - 511 to make the arithmetic easy) by 
isolate_freepages_block(), and suppose pfn 0 to pfn 510 are all order-0 
page, but pfn 511 is order-1 page, so you will isolate 513 pages from 
this pageblock, which will make 'pfn + isolated' not pageblock aligned.

>>> update block_end_pfn to pfn + pageblock_nr_pages. So they should point
>>> to the same pageblock. I guess you missed the change to update of
>>> block_end_pfn. :)
>>>>>
>>>>> Signed-off-by: Kemeng Shi <shikemeng@...weicloud.com>
>>>>> ---
>>>>>     mm/compaction.c | 14 ++------------
>>>>>     1 file changed, 2 insertions(+), 12 deletions(-)
>>>>>
>>>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>>>> index 684f6e6cd8bc..8d7d38073d30 100644
>>>>> --- a/mm/compaction.c
>>>>> +++ b/mm/compaction.c
>>>>> @@ -733,21 +733,11 @@ isolate_freepages_range(struct compact_control *cc,
>>>>>         block_end_pfn = pageblock_end_pfn(pfn);
>>>>>           for (; pfn < end_pfn; pfn += isolated,
>>>>> -                block_start_pfn = block_end_pfn,
>>>>> -                block_end_pfn += pageblock_nr_pages) {
>>>>> +                block_start_pfn = pfn,
>>>>> +                block_end_pfn = pfn + pageblock_nr_pages) {
>>>>>             /* Protect pfn from changing by isolate_freepages_block */
>>>>>             unsigned long isolate_start_pfn = pfn;
>>>>>     -        /*
>>>>> -         * pfn could pass the block_end_pfn if isolated freepage
>>>>> -         * is more than pageblock order. In this case, we adjust
>>>>> -         * scanning range to right one.
>>>>> -         */
>>>>> -        if (pfn >= block_end_pfn) {
>>>>> -            block_start_pfn = pageblock_start_pfn(pfn);
>>>>> -            block_end_pfn = pageblock_end_pfn(pfn);
>>>>> -        }
>>>>> -
>>>>>             block_end_pfn = min(block_end_pfn, end_pfn);
>>>>>               if (!pageblock_pfn_to_page(block_start_pfn,
>>>>
>>