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] mm/page_isolation: don't putback unisolated page
To:     Miaohe Lin <linmiaohe@huawei.com>, akpm@linux-foundation.org
Cc:     vbabka@suse.cz, iamjoonsoo.kim@lge.com, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
References: <20210904091839.20270-1-linmiaohe@huawei.com>
 <3b36529f-ab97-ddfe-0407-66f0cd1fd38d@redhat.com>
 <2d06db75-5c26-8fe2-6883-ac99056a9894@redhat.com>
 <b0a2947b-360a-40c2-03e4-f0f67845f4c3@huawei.com>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat
Message-ID: <c60dc5e2-6f19-3be8-56be-555033cc9ca4@redhat.com>
Date:   Mon, 6 Sep 2021 14:49:33 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.11.0
MIME-Version: 1.0
In-Reply-To: <b0a2947b-360a-40c2-03e4-f0f67845f4c3@huawei.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 06.09.21 14:45, Miaohe Lin wrote:
> On 2021/9/6 20:11, David Hildenbrand wrote:
>> On 06.09.21 14:02, David Hildenbrand wrote:
>>> On 04.09.21 11:18, Miaohe Lin wrote:
>>>> If __isolate_free_page() failed, due to zone watermark check, the page is
>>>> still on the free list. But this page will be put back to free list again
>>>> via __putback_isolated_page() now. This may trigger page->flags checks in
>>>> __free_one_page() if PageReported is set. Or we will corrupt the free list
>>>> because list_add() will be called for pages already on another list.
>>>>
>>>> Fixes: 3c605096d315 ("mm/page_alloc: restrict max order of merging on isolated pageblock")
>>>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>>>> ---
>>>>     mm/page_isolation.c | 6 ++----
>>>>     1 file changed, 2 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
>>>> index 9bb562d5d194..7d70d772525c 100644
>>>> --- a/mm/page_isolation.c
>>>> +++ b/mm/page_isolation.c
>>>> @@ -93,10 +93,8 @@ static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
>>>>                 buddy_pfn = __find_buddy_pfn(pfn, order);
>>>>                 buddy = page + (buddy_pfn - pfn);
>>>>     -            if (!is_migrate_isolate_page(buddy)) {
>>>> -                __isolate_free_page(page, order);
>>>> -                isolated_page = true;
>>>> -            }
>>>> +            if (!is_migrate_isolate_page(buddy))
>>>> +                isolated_page = !!__isolate_free_page(page, order);
>>>>             }
>>>>         }
>>>>    
>>>
>>> Thanks!
>>>
>>> Reviewed-by: David Hildenbrand <david@redhat.com>
>>>
>>
>> To make the confusion perfect (sorry) :D I tripple-checked:
>>
>> In unset_migratetype_isolate() we check that is_migrate_isolate_page(page) holds, otherwise we return.
>>
>> We call __isolate_free_page() only for such pages.
>>
>> __isolate_free_page() won't perform watermark checks on is_migrate_isolate().
>>
>> Consequently, __isolate_free_page() should never fail when called from unset_migratetype_isolate()
>>
>> If that's correct then we  could instead maybe add a VM_BUG_ON() and a comment why this can't fail.
>>
>>
>> Makes sense or am I missing something?
> 
> I think you're right. __isolate_free_page() should never fail when called from unset_migratetype_isolate()
> as explained by you. But it might be too fragile to reply on the failure conditions of __isolate_free_page().
> If that changes, VM_BUG_ON() here might trigger unexpectedly. Or am I just over-worried as failure conditions
> of __isolate_free_page() can hardly change?

Maybe

isolated_page = !!__isolate_free_page(page, order);
/*
  * Isolating a free page in an isolated pageblock is expected to always
  * work as watermarks don't apply here.
  */
VM_BUG_ON(isolated_page);


VM_BUG_ON() allows us to detect any issues when testing. Combined with 
the comment it tells everybody messing with __isolate_free_page() what 
we expect in this function.

In production system, we would handle it gracefully.

-- 
Thanks,

David / dhildenb