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] hugetlb: fix uninitialized subpool pointer
From:   Mike Kravetz <mike.kravetz@oracle.com>
To:     Oscar Salvador <osalvador@suse.de>
Cc:     linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        Gerald Schaefer <gerald.schaefer@linux.ibm.com>,
        Michal Hocko <mhocko@suse.com>,
        Muchun Song <songmuchun@bytedance.com>,
        Heiko Carstens <hca@linux.ibm.com>,
        Sven Schnelle <svens@linux.ibm.com>,
        Andrew Morton <akpm@linux-foundation.org>
References: <20210223215544.313871-1-mike.kravetz@oracle.com>
 <20210223224540.GB2740@localhost.localdomain>
 <53527e9d-d09b-7287-9f79-ebdbf4e9bc7a@oracle.com>
 <a7f063ea6b5eae4a4fcf038268e3a604@suse.de>
 <ef62ac45-9ec9-2582-3e58-7efc0609221f@oracle.com>
Message-ID: <3590f7a0-854d-04eb-6d73-7b3ef2ac49b5@oracle.com>
Date:   Tue, 23 Feb 2021 16:27:27 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.1.1
MIME-Version: 1.0
In-Reply-To: <ef62ac45-9ec9-2582-3e58-7efc0609221f@oracle.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2/23/21 3:21 PM, Mike Kravetz wrote:
> On 2/23/21 2:58 PM, Oscar Salvador wrote:
>> On 2021-02-23 23:55, Mike Kravetz wrote:
>>> Yes, that is the more common case where the once active hugetlb page
>>> will be simply added to the free list via enqueue_huge_page().  This
>>> path does not go through prep_new_huge_page.
>>
>> Right, I see.
>>
>> Thanks
> 
> You got me thinking ...
> When we dynamically allocate gigantic pages via alloc_contig_pages, we
> will not use the buddy allocator.  Therefore, the usual 'page prepping'
> will not take place.  Specifically, I could not find anything in that
> path which clears page->private of the head page.
> Am I missing that somewhere?  If not, then we need to clear that as well
> in prep_compound_gigantic_page.  Or, just clear it in prep_new_huge_page
> to handle any change in assumptions about the buddy allocator.
> 
> This is not something introduced with the recent field shuffling, it
> looks like something that existed for some time.

nm, we do end up calling the same page prepping code (post_alloc_hook)
from alloc_contig_range->isolate_freepages_range.

Just to make sure, I purpously dirtied page->private of every page as it
was being freed.  Gigantic page allocation was just fine, and I even ran
ltp mm tests with this dirtying in place.
-- 
Mike Kravetz