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;
MIME-Version: 1.0
References: <D90B73BA-22EC-407E-838F-2BA646C60DE0@lca.pw> <CAM4kBBKkOGOjT5fddQ_vongNx_cXmv0tCQzD9ZtrGgkuPTKfTg@mail.gmail.com>
In-Reply-To: <CAM4kBBKkOGOjT5fddQ_vongNx_cXmv0tCQzD9ZtrGgkuPTKfTg@mail.gmail.com>
From:   Qian Cai <cai@lca.pw>
Date:   Mon, 18 May 2020 23:50:05 -0400
Message-ID: <CAG=TAF65nmW2m9-5z5xvLmfRAD0LZAxgARJOEAqvzA3GdN-iJA@mail.gmail.com>
Subject: Re: zswap z3fold + memory offline = infinite loop
To:     Vitaly Wool <vitaly.wool@konsulko.com>
Cc:     Michal Hocko <mhocko@suse.com>,
        David Hildenbrand <david@redhat.com>,
        Linux-MM <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Minchan Kim <minchan@kernel.org>,
        Seth Jennings <sjenning@redhat.com>,
        Dan Streetman <ddstreet@ieee.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, May 13, 2020 at 4:28 AM Vitaly Wool <vitaly.wool@konsulko.com> wrote:
>
>
>
> On Wed, May 13, 2020, 2:36 AM Qian Cai <cai@lca.pw> wrote:
>>
>> Put zswap z3fold pages into the memory and then offline those memory would trigger an infinite loop here in
>>
>> __offline_pages() --> do_migrate_range() because there is no error handling,
>>
>>                         if (pfn) {
>>                                 /*
>>                                  * TODO: fatal migration failures should bail
>>                                  * out
>>                                  */
>>                                 do_migrate_range(pfn, end_pfn);
>>
>> There, isolate_movable_page() will always return -EBUSY  because,
>>
>>         if (!mapping->a_ops->isolate_page(page, mode))
>>                 goto out_no_isolated;
>>
>> i.e., z3fold_page_isolate() will always return false because,
>>
>> zhdr->mapped_count == 2
>
>
> So who mapped these pages? The whole zswap operation presumes that objects are mapped for a short while to run some I/O and so, most of the time zhdr->mapped_count would be 0.

I have no clue why those pages have been mapped for so long, but it is
trivial to reproduce using the above reproducer. Also, zbud has no
such issue. Alternatively, if you could send me some debug patches to
narrow it down, I'll be happy to run for you.

>
> Removing that check in ->isolate() is not a big deal, but ->migratepage() shall not allow actual migration anyway if there are mapped objects.

Is that worse than an endless loop here?