Re: memory offline infinite loop after soft offline

[David Hildenbrand]

On 17.10.19 12:01, Michal Hocko wrote:
> On Thu 17-10-19 09:34:10, Naoya Horiguchi wrote:
> > On Mon, Oct 14, 2019 at 10:39:14AM +0200, Michal Hocko wrote:
> [...]
> > > diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> > > index 89c19c0feadb..5fb3fee16fde 100644
> > > --- a/mm/page_isolation.c
> > > +++ b/mm/page_isolation.c
> > > @@ -274,7 +274,7 @@ __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
> > >   			 * simple way to verify that as VM_BUG_ON(), though.
> > >   			 */
> > >   			pfn += 1 << page_order(page);
> > > -		else if (skip_hwpoisoned_pages && PageHWPoison(page))
> > > +		else if (skip_hwpoisoned_pages && PageHWPoison(compound_head(page)))
> > >   			/* A HWPoisoned page cannot be also PageBuddy */
> > >   			pfn++;
> > >   		else
> >
> > This fix looks good to me. The original code only addresses hwpoisoned 4kB-page,
> > we seem to have this issue since the following commit,
>
> Thanks a lot for double checking Naoya!
>   
> >    commit b023f46813cde6e3b8a8c24f432ff9c1fd8e9a64
> >    Author: Wen Congyang <wency@xxxxxxxxxxxxxx>
> >    Date:   Tue Dec 11 16:00:45 2012 -0800
> >    
> >        memory-hotplug: skip HWPoisoned page when offlining pages
> >
> > and extension of LTP coverage finally discovered this.
>
> Qian, could you give the patch some testing?
> ---
>
>  From 441a9515dcdb29bb0ca39ff995632907d959032f Mon Sep 17 00:00:00 2001
> From: Michal Hocko <mhocko@xxxxxxxx>
> Date: Thu, 17 Oct 2019 11:49:15 +0200
> Subject: [PATCH] hugetlb, memory_hotplug: fix HWPoisoned tail pages properly
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> Qian Cai has noticed that hwpoisoned hugetlb pages prevent memory
> offlining from making a forward progress. He has nailed down the issue
> to be __test_page_isolated_in_pageblock always returning EBUSY because
> of soft offlined page:
> [ÂÂ101.665160][ T8885] pfn = 77501, end_pfn = 78000
> [ÂÂ101.665245][ T8885] page:c00c000001dd4040 refcount:0 mapcount:0
> mapping:0000000000000000 index:0x0
> [ÂÂ101.665329][ T8885] flags: 0x3fffc000000000()
> [ÂÂ101.665391][ T8885] raw: 003fffc000000000 0000000000000000 ffffffff01dd0500
> 0000000000000000
> [ÂÂ101.665498][ T8885] raw: 0000000000000000 0000000000000000 00000000ffffffff
> 0000000000000000
> [ÂÂ101.665588][ T8885] page dumped because: soft_offline
> [ÂÂ101.665639][ T8885] page_owner tracks the page as freed
> [ÂÂ101.665697][ T8885] page last allocated via order 5, migratetype Movable,
> gfp_mask
> 0x346cca(GFP_HIGHUSER_MOVABLE|__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_COMP|__GFP_
> THISNODE)
> [ÂÂ101.665924][ T8885]ÂÂprep_new_page+0x3c0/0x440
> [ÂÂ101.665962][ T8885]ÂÂget_page_from_freelist+0x2568/0x2bb0
> [ÂÂ101.666059][ T8885]ÂÂ__alloc_pages_nodemask+0x1b4/0x670
> [ÂÂ101.666115][ T8885]ÂÂalloc_fresh_huge_page+0x244/0x6e0
> [ÂÂ101.666183][ T8885]ÂÂalloc_migrate_huge_page+0x30/0x70
> [ÂÂ101.666254][ T8885]ÂÂalloc_new_node_page+0xc4/0x380
> [ÂÂ101.666325][ T8885]ÂÂmigrate_pages+0x3b4/0x19e0
> [ÂÂ101.666375][ T8885]ÂÂdo_move_pages_to_node.isra.29.part.30+0x44/0xa0
> [ÂÂ101.666464][ T8885]ÂÂkernel_move_pages+0x498/0xfc0
> [ÂÂ101.666520][ T8885]ÂÂsys_move_pages+0x28/0x40
> [ÂÂ101.666643][ T8885]ÂÂsystem_call+0x5c/0x68
> [ÂÂ101.666665][ T8885] page last free stack trace:
> [ÂÂ101.666704][ T8885]ÂÂ__free_pages_ok+0xa4c/0xd40
> [ÂÂ101.666773][ T8885]ÂÂupdate_and_free_page+0x2dc/0x5b0
> [ÂÂ101.666821][ T8885]ÂÂfree_huge_page+0x2dc/0x740
> [ÂÂ101.666875][ T8885]ÂÂ__put_compound_page+0x64/0xc0
> [ÂÂ101.666926][ T8885]ÂÂputback_active_hugepage+0x228/0x390
> [ÂÂ101.666990][ T8885]ÂÂmigrate_pages+0xa78/0x19e0
> [ÂÂ101.667048][ T8885]ÂÂsoft_offline_page+0x314/0x1050
> [ÂÂ101.667117][ T8885]ÂÂsys_madvise+0x1068/0x1080
> [ÂÂ101.667185][ T8885]ÂÂsystem_call+0x5c/0x68
>
> The reason is that __test_page_isolated_in_pageblock doesn't recognize
> hugetlb tail pages as the HWPoison bit is not transferred from the head
> page. Pfn walker then doesn't recognize those pages and so EBUSY is
> returned up the call chain.
>
> The proper fix would be to handle HWPoison throughout the huge page but
> considering there is a WIP to rework that code considerably let's go
> with a simple and easily backportable workaround and simply check the
> the head of a compound page for the HWPoison flag.
>
> Reported-and-analyzed-by: Qian Cai <cai@xxxxxx>
> Fixes: b023f46813cd ("memory-hotplug: skip HWPoisoned page when offlining pages")
> Cc: stable
> Signed-off-by: Michal Hocko <mhocko@xxxxxxxx>
> ---
>   mm/page_isolation.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> index 89c19c0feadb..5fb3fee16fde 100644
> --- a/mm/page_isolation.c
> +++ b/mm/page_isolation.c
> @@ -274,7 +274,7 @@ __test_page_isolated_in_pageblock(unsigned long pfn, unsigned long end_pfn,
>   			 * simple way to verify that as VM_BUG_ON(), though.
>   			 */
>   			pfn += 1 << page_order(page);
> -		else if (skip_hwpoisoned_pages && PageHWPoison(page))
> +		else if (skip_hwpoisoned_pages && PageHWPoison(compound_head(page)))
>   			/* A HWPoisoned page cannot be also PageBuddy */
>   			pfn++;
>   		else

With the extended description, this makes sense to me now :)

Acked-by: David Hildenbrand <david@xxxxxxxxxx>

--

Thanks,

David / dhildenb