Re: memory offline infinite loop after soft offline

From: Qian Cai
Date: Thu Oct 17 2019 - 14:07:18 EST

Next message: Alan Maguire: "[PATCH v3 linux-kselftest-test 0/6] kunit: support building core/tests as modules"
Previous message: Sasha Levin: "Re: [PATCH v3] tpm/tpm_ftpm_tee: add shutdown call back"
In reply to: David Hildenbrand: "Re: memory offline infinite loop after soft offline"
Next in thread: Michal Hocko: "Re: memory offline infinite loop after soft offline"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Thu, 2019-10-17 at 12:01 +0200, 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?

Unfortunately, this does not solve the problem.ÂIt looks to me that in
soft_offline_huge_page(), set_hwpoison_free_buddy_page() will only set
PG_hwpoison for buddy pages, so the even the compound_head() has no PG_hwpoison
set.

if (PageBuddy(page_head) && page_order(page_head) >= order) {
if (!TestSetPageHWPoison(page))
hwpoisoned = true;

The below is the dump_page() of the compound_head().

[ÂÂ113.796632][ T8907] page:c00c000800458040 refcount:0 mapcount:0
mapping:0000000000000000 index:0x0
[ÂÂ113.796716][ T8907] flags: 0x83fffc000000000()
[ÂÂ113.796764][ T8907] raw: 083fffc000000000 0000000000000000 ffffffff00450500
0000000000000000
[ÂÂ113.796870][ T8907] raw: 0000000000000000 0000000000000000 00000000ffffffff
0000000000000000
[ÂÂ113.796959][ T8907] page dumped because: soft offline compound_head()
[ÂÂ113.797037][ T8907] page_owner tracks the page as freed
[ÂÂ113.797086][ T8907] page last allocated via order 5, migratetype Movable,
gfp_mask
0x346cca(GFP_HIGHUSER_MOVABLE|__GFP_NOWARN|__GFP_RETRY_MAYFAIL|__GFP_COMP|__GFP_
THISNODE)
[ÂÂ113.797262][ T8907]ÂÂprep_new_page+0x3c0/0x440
[ÂÂ113.797316][ T8907]ÂÂget_page_from_freelist+0x2568/0x2bb0
[ÂÂ113.797395][ T8907]ÂÂ__alloc_pages_nodemask+0x1b4/0x670
[ÂÂ113.797443][ T8907]ÂÂalloc_fresh_huge_page+0xb8/0x300
[ÂÂ113.797493][ T8907]ÂÂalloc_migrate_huge_page+0x30/0x70
[ÂÂ113.797550][ T8907]ÂÂalloc_new_node_page+0xc4/0x380
[ÂÂ113.797609][ T8907]ÂÂmigrate_pages+0x3b4/0x19e0
[ÂÂ113.797649][ T8907]ÂÂdo_move_pages_to_node.isra.29.part.30+0x44/0xa0
[ÂÂ113.797713][ T8907]ÂÂkernel_move_pages+0x498/0xfc0
[ÂÂ113.797784][ T8907]ÂÂsys_move_pages+0x28/0x40
[ÂÂ113.797843][ T8907]ÂÂsystem_call+0x5c/0x68
[ÂÂ113.797895][ T8907] page last free stack trace:
[ÂÂ113.797947][ T8907]ÂÂ__free_pages_ok+0xa4c/0xd40
[ÂÂ113.797991][ T8907]ÂÂupdate_and_free_page+0x2dc/0x5b0
[ÂÂ113.798059][ T8907]ÂÂfree_huge_page+0x2dc/0x740
[ÂÂ113.798103][ T8907]ÂÂ__put_compound_page+0x64/0xc0
[ÂÂ113.798171][ T8907]ÂÂputback_active_hugepage+0x228/0x390
[ÂÂ113.798219][ T8907]ÂÂmigrate_pages+0xa78/0x19e0
[ÂÂ113.798273][ T8907]ÂÂsoft_offline_page+0x314/0x1050
[ÂÂ113.798319][ T8907]ÂÂsys_madvise+0x1068/0x1080
[ÂÂ113.798381][ T8907]ÂÂsystem_call+0x5c/0x68

> ---
>
> 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
> --
> 2.20.1
>

Next message: Alan Maguire: "[PATCH v3 linux-kselftest-test 0/6] kunit: support building core/tests as modules"
Previous message: Sasha Levin: "Re: [PATCH v3] tpm/tpm_ftpm_tee: add shutdown call back"
In reply to: David Hildenbrand: "Re: memory offline infinite loop after soft offline"
Next in thread: Michal Hocko: "Re: memory offline infinite loop after soft offline"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]