Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 23 May 2023 19:50:16 -0700 (PDT)
From:   Hugh Dickins <hughd@google.com>
To:     Qi Zheng <qi.zheng@linux.dev>
cc:     Alistair Popple <apopple@nvidia.com>,
        Hugh Dickins <hughd@google.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Mike Kravetz <mike.kravetz@oracle.com>,
        Mike Rapoport <rppt@kernel.org>,
        "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
        Matthew Wilcox <willy@infradead.org>,
        David Hildenbrand <david@redhat.com>,
        Suren Baghdasaryan <surenb@google.com>,
        Qi Zheng <zhengqi.arch@bytedance.com>,
        Yang Shi <shy828301@gmail.com>,
        Mel Gorman <mgorman@techsingularity.net>,
        Peter Xu <peterx@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Will Deacon <will@kernel.org>, Yu Zhao <yuzhao@google.com>,
        Ralph Campbell <rcampbell@nvidia.com>,
        Ira Weiny <ira.weiny@intel.com>,
        Steven Price <steven.price@arm.com>,
        SeongJae Park <sj@kernel.org>,
        Naoya Horiguchi <naoya.horiguchi@nec.com>,
        Christophe Leroy <christophe.leroy@csgroup.eu>,
        Zack Rusin <zackr@vmware.com>, Jason Gunthorpe <jgg@ziepe.ca>,
        Axel Rasmussen <axelrasmussen@google.com>,
        Anshuman Khandual <anshuman.khandual@arm.com>,
        Pasha Tatashin <pasha.tatashin@soleen.com>,
        Miaohe Lin <linmiaohe@huawei.com>,
        Minchan Kim <minchan@kernel.org>,
        Christoph Hellwig <hch@infradead.org>,
        Song Liu <song@kernel.org>,
        Thomas Hellstrom <thomas.hellstrom@linux.intel.com>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [PATCH 13/31] mm/hmm: retry if pte_offset_map() fails
In-Reply-To: <1d5d7f30-1a6a-5ea1-165f-65d34e7658e9@linux.dev>
Message-ID: <89461cb2-ec2c-719c-8a50-1f76e8048ef@google.com>
References: <68a97fbe-5c1e-7ac6-72c-7b9c6290b370@google.com> <2edc4657-b6ff-3d6e-2342-6b60bfccc5b@google.com> <0e3c5937-2f04-25d9-98eb-01e4c2b04150@linux.dev> <87353n9354.fsf@nvidia.com> <1d5d7f30-1a6a-5ea1-165f-65d34e7658e9@linux.dev>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

On Tue, 23 May 2023, Qi Zheng wrote:
> On 2023/5/23 10:39, Alistair Popple wrote:
> > Qi Zheng <qi.zheng@linux.dev> writes:
> >> On 2023/5/22 13:05, Hugh Dickins wrote:
> >>> hmm_vma_walk_pmd() is called through mm_walk, but already has a goto
> >>> again loop of its own, so take part in that if pte_offset_map() fails.
> >>> Signed-off-by: Hugh Dickins <hughd@google.com>
> >>> ---
> >>>    mm/hmm.c | 2 ++
> >>>    1 file changed, 2 insertions(+)
> >>> diff --git a/mm/hmm.c b/mm/hmm.c
> >>> index e23043345615..b1a9159d7c92 100644
> >>> --- a/mm/hmm.c
> >>> +++ b/mm/hmm.c
> >>> @@ -381,6 +381,8 @@ static int hmm_vma_walk_pmd(pmd_t *pmdp,
> >>>     }
> >>>      	ptep = pte_offset_map(pmdp, addr);
> >>> +	if (!ptep)
> >>> +		goto again;
> >>>     for (; addr < end; addr += PAGE_SIZE, ptep++, hmm_pfns++) {
> >>>      int r;
> >>>    
> >>
> >> I haven't read the entire patch set yet, but taking a note here.
> >> The hmm_vma_handle_pte() will unmap pte and then call
> >> migration_entry_wait() to remap pte, so this may fail, we need to
> >> handle this case like below:
> > 
> > I don't see a problem here. Sure, hmm_vma_handle_pte() might return
> > -EBUSY but that will get returned up to hmm_range_fault() which will
> > retry the whole thing again and presumably fail when looking at the PMD.
> 
> Yeah. There is no problem with this and the modification to
> migration_entry_wait() can be simplified. My previous thought was that
> we can finish the retry logic in hmm_vma_walk_pmd() without handing it
> over to the caller. :)

Okay, Alistair has resolved this one, thanks, I agree; but what is
"the modification to migration_entry_wait()" that you refer to there?

I don't think there's any need to make it a bool, it's normal for there
to be races on entry to migration_entry_wait(), and we're used to just
returning to caller (and back up to userspace) when it does not wait.

Hugh