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;
Subject: Re: [PATCH 1/1] mm: compaction: avoid fast_isolate_around() to set
 pageblock_skip on reserved pages
To:     Mel Gorman <mgorman@suse.de>
Cc:     Andrea Arcangeli <aarcange@redhat.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Andrew Morton <akpm@linux-foundation.org>, linux-mm@kvack.org,
        Qian Cai <cai@lca.pw>, Michal Hocko <mhocko@kernel.org>,
        linux-kernel@vger.kernel.org, Mike Rapoport <rppt@linux.ibm.com>,
        Baoquan He <bhe@redhat.com>
References: <X73s8fxDKPRD6wET@redhat.com>
 <35F8AADA-6CAA-4BD6-A4CF-6F29B3F402A4@redhat.com>
 <20201125103933.GM3306@suse.de>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat GmbH
Message-ID: <5f01bde6-fe31-9b0e-f288-06b82598a8b3@redhat.com>
Date:   Wed, 25 Nov 2020 12:04:15 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.4.0
MIME-Version: 1.0
In-Reply-To: <20201125103933.GM3306@suse.de>
Content-Type: text/plain; charset=iso-8859-15
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 25.11.20 11:39, Mel Gorman wrote:
> On Wed, Nov 25, 2020 at 07:45:30AM +0100, David Hildenbrand wrote:
>>> Something must have changed more recently than v5.1 that caused the
>>> zoneid of reserved pages to be wrong, a possible candidate for the
>>> real would be this change below:
>>>
>>> +               __init_single_page(pfn_to_page(pfn), pfn, 0, 0);
>>>
>>
>> Before that change, the memmap of memory holes were only zeroed out. So the zones/nid was 0, however, pages were not reserved and had a refcount of zero - resulting in other issues.
>>
>> Most pfn walkers shouldn???t mess with reserved pages and simply skip them. That would be the right fix here.
>>
> 
> Ordinarily yes, pfn walkers should not care about reserved pages but it's
> still surprising that the node/zone linkages would be wrong for memory
> holes. If they are in the middle of a zone, it means that a hole with
> valid struct pages could be mistaken for overlapping nodes (if the hole
> was in node 1 for example) or overlapping zones which is just broken.

I agree within zones - but AFAIU, the issue is reserved memory between
zones, right?

Assume your end of memory falls within a section - what would be the
right node/zone for such a memory hole at the end of the section? With
memory hotplug after such a hole, we can easily have multiple
nodes/zones spanning such a hole, unknown before hotplug.

IMHO, marking memory holes properly (as discussed) would be the cleanest
approach. For now, we use node/zone 0 + PageReserved - because memory
hotunplug (zone shrinking etc.) doesn't really care about ZONE_DMA.

> 
>>>
>>> Whenever pfn_valid is true, it's better that the zoneid/nid is correct
>>> all times, otherwise if the second stage fails we end up in a bug with
>>> weird side effects.
>>
>> Memory holes with a valid memmap might not have a zone/nid. For now, skipping reserved pages should be good enough, no?
>>
> 
> It would partially paper over the issue that setting the pageblock type
> based on a reserved page. I agree that compaction should not be returning
> pfns that are outside of the zone range because that is buggy in itself
> but valid struct pages should have valid information. I don't think we
> want to paper over that with unnecessary PageReserved checks.

Agreed as long as we can handle that issue using range checks.

-- 
Thanks,

David / dhildenb