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: <20200623094258.6705-1-richard.weiyang@linux.alibaba.com>
 <20200623151828.GA31426@dhcp22.suse.cz> <20200624061340.GA11552@L-31X9LVDL-1304.local>
In-Reply-To: <20200624061340.GA11552@L-31X9LVDL-1304.local>
From:   Dan Williams <dan.j.williams@intel.com>
Date:   Wed, 24 Jun 2020 09:10:09 -0700
Message-ID: <CAPcyv4gEUq0yyz00AbQQSyo_acj1wUCZPse5vEff_Bq7GmeksQ@mail.gmail.com>
Subject: Re: [PATCH] mm/spase: never partially remove memmap for early section
To:     Wei Yang <richard.weiyang@linux.alibaba.com>
Cc:     Michal Hocko <mhocko@kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Oscar Salvador <osalvador@suse.de>,
        Linux MM <linux-mm@kvack.org>, Baoquan He <bhe@redhat.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        David Hildenbrand <david@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Jun 23, 2020 at 11:14 PM Wei Yang
<richard.weiyang@linux.alibaba.com> wrote:
>
> On Tue, Jun 23, 2020 at 05:18:28PM +0200, Michal Hocko wrote:
> >On Tue 23-06-20 17:42:58, Wei Yang wrote:
> >> For early sections, we assumes its memmap will never be partially
> >> removed. But current behavior breaks this.
> >>
> >> Let's correct it.
> >>
> >> Fixes: ba72b4c8cf60 ("mm/sparsemem: support sub-section hotplug")
> >> Signed-off-by: Wei Yang <richard.weiyang@linux.alibaba.com>
> >
> >Can a user trigger this or is this a theoretical bug?
>
> Let me rewrite the changelog a little. Look forward any comments.
>
>    For early sections, its memmap is handled specially even sub-section is
>    enabled. The memmap could only be populated as a whole.
>
>    Quoted from the comment of section_activate():
>
>        * The early init code does not consider partially populated
>        * initial sections, it simply assumes that memory will never be
>        * referenced.  If we hot-add memory into such a section then we
>        * do not need to populate the memmap and can simply reuse what
>        * is already there.
>
>    While current section_deactivate() breaks this rule. When hot-remove a
>    sub-section, section_deactivate() would depopulate its memmap. The
>    consequence is if we hot-add this subsection again, its memmap never get
>    proper populated.

Ok, forgive the latency as re-fetched this logic into my mental cache.
So what I was remembering was the initial state of the code that
special cased early sections, and that still seems to be the case in
pfn_valid(). IIRC early_sections / bootmem are blocked from being
removed entirely. Partial / subsection removals are ok.