Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.36 as permitted sender) client-ip=23.128.96.36;
Message-ID: <3b7221768bebd9cac5ab8004dca901c4f2faf3cb.camel@surriel.com>
Subject: Re: [PATCH 2/3] hugetlbfs: close race between MADV_DONTNEED and
 page fault
From:   Rik van Riel <riel@surriel.com>
To:     Mike Kravetz <mike.kravetz@oracle.com>
Cc:     linux-kernel@vger.kernel.org, kernel-team@meta.com,
        linux-mm@kvack.org, akpm@linux-foundation.org,
        muchun.song@linux.dev, leit@meta.com, willy@infradead.org
Date:   Mon, 25 Sep 2023 20:46:52 -0400
In-Reply-To: <20230925222539.GC11309@monkey>
References: <20230925203030.703439-1-riel@surriel.com>
         <20230925203030.703439-3-riel@surriel.com> <20230925222539.GC11309@monkey>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Sender: riel@surriel.com
Precedence: bulk

On Mon, 2023-09-25 at 15:25 -0700, Mike Kravetz wrote:
> On 09/25/23 16:28, riel@surriel.com wrote:
> > 
> > -void __unmap_hugepage_range_final(struct mmu_gather *tlb,
> > -                         struct vm_area_struct *vma, unsigned long
> > start,
> > -                         unsigned long end, struct page *ref_page,
> > -                         zap_flags_t zap_flags)
> > +void __hugetlb_zap_begin(struct vm_area_struct *vma,
> > +                        unsigned long *start, unsigned long *end)
> >  {
> > +       adjust_range_if_pmd_sharing_possible(vma, start, end);
> >         hugetlb_vma_lock_write(vma);
> >         i_mmap_lock_write(vma->vm_file->f_mapping);
> > +}
> 
> __unmap_hugepage_range_final() was called from unmap_single_vma.
> unmap_single_vma has two callers, zap_page_range_single and
> unmap_vmas.
> 
> When the locking was moved into hugetlb_zap_begin, it was only added
> to the
> zap_page_range_single call path.  Calls from unmap_vmas are missing
> the
> locking.

Oh, that's a fun one.

I suppose the locking of the f_mapping lock, and calling
adjust_range_if_pmd_sharing_possible matters for the call
from unmap_vmas, while the call tho hugetlb_vma_lock_write
really doesn't matter, since unmap_vmas is called with the
mmap_sem held for write, which already excludes page faults.

I'll add the call there for v4.

Good catch.

-- 
All Rights Reversed.