[Resend Patch 0/2] mm/memory_hotplug: fix hot remove bug

[Resend Patch 0/2] mm/memory_hotplug: fix hot remove bug

[Yasuaki Ishimatsu]

Here are two patches for memory hotplug:

Yasuaki Ishimatsu (2):
   mm/sparse: use page_private() to get page->private value
   mm/memory_hotplug: set magic number to page->freelsit instead
     of page->lru.next

  arch/x86/mm/init_64.c | 2 +-
  mm/memory_hotplug.c   | 5 +++--
  mm/sparse.c           | 4 ++--
  3 files changed, 6 insertions(+), 5 deletions(-)

[Resend PATCH 1/2] mm/sparse: use page_private() to get page->private value

[Yasuaki Ishimatsu]

free_map_bootmem() uses page->private directly to set
removing_section_nr argument. But to get page->private
value, page_private() has been prepared.

So free_map_bootmem() should use page_private() instead of
page->private.

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
---
  mm/sparse.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/sparse.c b/mm/sparse.c
index 1e168bf..dc30a70 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -667,7 +667,7 @@ static void free_map_bootmem(struct page *memmap)
  		BUG_ON(magic == NODE_INFO);

  		maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
-		removing_section_nr = page->private;
+		removing_section_nr = page_private(page);

  		/*
  		 * When this function is called, the removing section is
-- 
1.8.3.1

[Resend PATCH 2/2] mm/memory_hotplug: set magic number to page->freelsit instead of page->lru.next

[Yasuaki Ishimatsu]

To identify that pages of page table are allocated from bootmem
allocator, magic number sets to page->lru.next. But page->lru
list is initialized in reserve_bootmem_region(). So when calling
free_pagetable(), the function cannot find the magic number of
pages. And free_pagetable() frees the pages by free_reserved_page()
not put_page_bootmem().

But if the pages are allocated from bootmem allocator and used as
page table, the pages have private flag. So before freeing the
pages, we should clear the private flag by put_page_bootmem().

Before applying the commit 7bfec6f47bb0 ("mm, page_alloc: check
multiple page fields with a single branch"), we could find the
following visible issue:

  BUG: Bad page state in process kworker/u1024:1
  page:ffffea103cfd8040 count:0 mapcount:0 mappi
  flags: 0x6fffff80000800(private)
  page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
  bad because of flags: 0x800(private)
  <snip>
  Call Trace:
  [...] dump_stack+0x63/0x87
  [...] bad_page+0x114/0x130
  [...] free_pages_prepare+0x299/0x2d0
  [...] free_hot_cold_page+0x31/0x150
  [...] __free_pages+0x25/0x30
  [...] free_pagetable+0x6f/0xb4
  [...] remove_pagetable+0x379/0x7ff
  [...] vmemmap_free+0x10/0x20
  [...] sparse_remove_one_section+0x149/0x180
  [...] __remove_pages+0x2e9/0x4f0
  [...] arch_remove_memory+0x63/0xc0
  [...] remove_memory+0x8c/0xc0
  [...] acpi_memory_device_remove+0x79/0xa5
  [...] acpi_bus_trim+0x5a/0x8d
  [...] acpi_bus_trim+0x38/0x8d
  [...] acpi_device_hotplug+0x1b7/0x418
  [...] acpi_hotplug_work_fn+0x1e/0x29
  [...] process_one_work+0x152/0x400
  [...] worker_thread+0x125/0x4b0
  [...] ? __schedule+0x345/0x960
  [...] ? rescuer_thread+0x380/0x380
  [...] kthread+0xd8/0xf0
  [...] ret_from_fork+0x22/0x40
  [...] ? kthread_park+0x60/0x60

And the issue still silently occurs.

Until freeing the pages of page table allocated from bootmem allocator,
the page->freelist is never used. So the patch sets magic number to
page->freelist instead of page->lru.next.

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
---
  arch/x86/mm/init_64.c | 2 +-
  mm/memory_hotplug.c   | 5 +++--
  mm/sparse.c           | 2 +-
  3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 963895f..b35e525 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -679,7 +679,7 @@ static void __meminit free_pagetable(struct page *page, int order)
  	if (PageReserved(page)) {
  		__ClearPageReserved(page);

-		magic = (unsigned long)page->lru.next;
+		magic = (unsigned long)page->freelist;
  		if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
  			while (nr_pages--)
  				put_page_bootmem(page++);
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index e43142c1..9847e4a 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -179,7 +179,7 @@ static void release_memory_resource(struct resource *res)
  void get_page_bootmem(unsigned long info,  struct page *page,
  		      unsigned long type)
  {
-	page->lru.next = (struct list_head *) type;
+	page->freelist = (void *) type;
  	SetPagePrivate(page);
  	set_page_private(page, info);
  	page_ref_inc(page);
@@ -189,11 +189,12 @@ void put_page_bootmem(struct page *page)
  {
  	unsigned long type;

-	type = (unsigned long) page->lru.next;
+	type = (unsigned long) page->freelist;
  	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
  	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);

  	if (page_ref_dec_return(page) == 1) {
+		page->freelist = NULL;
  		ClearPagePrivate(page);
  		set_page_private(page, 0);
  		INIT_LIST_HEAD(&page->lru);
diff --git a/mm/sparse.c b/mm/sparse.c
index dc30a70..db6bf3c 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -662,7 +662,7 @@ static void free_map_bootmem(struct page *memmap)
  		>> PAGE_SHIFT;

  	for (i = 0; i < nr_pages; i++, page++) {
-		magic = (unsigned long) page->lru.next;
+		magic = (unsigned long) page->freelist;

  		BUG_ON(magic == NODE_INFO);

-- 
1.8.3.1

Re: [Resend PATCH 2/2] mm/memory_hotplug: set magic number to page->freelsit instead of page->lru.next

[Yasuaki Ishimatsu]

Hi Andrew,

Please apply the following patch into your tree because patch
("mm/memory_hotplug: set magic number to page->freelsit instead of page->lru.next")
is not applied correctly.

---
From: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Date: Fri, 3 Feb 2017 15:18:03 -0500
Subject: [PATCH] Remove unnecessary code from get_page_bootmem()

The following patch is not applied correctly.
http://lkml.kernel.org/r/2c29bd9f-5b67-02d0-18a3-8828e78bbb6f@gmail.com

So the following unnecessary code still remains.

get_page_bootmem()
{
...
        page->lru.next = (struct list_head *)type;
...

The patch removeｓ this code from get_page_bootmem()

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>

---
  mm/memory_hotplug.c | 1 -
  1 file changed, 1 deletion(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 19b460a..50b586c 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -179,7 +179,6 @@ static void release_memory_resource(struct resource *res)
  void get_page_bootmem(unsigned long info,  struct page *page,
  		      unsigned long type)
  {
-	page->lru.next = (struct list_head *)type;
  	page->freelist = (void *)type;
  	SetPagePrivate(page);
  	set_page_private(page, info);
-- 
1.8.3.1

Re: [Resend PATCH 2/2] mm/memory_hotplug: set magic number to page->freelsit instead of page->lru.next

[Yasuaki Ishimatsu]

TAB was replaced to white spaces. So please apply this one.

---
From: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Date: Fri, 3 Feb 2017 15:18:03 -0500
Subject: [PATCH] mm/memory_hotplug: Remove unnecessary code from get_page_bootmem()

The following patch is not applied correctly.
http://lkml.kernel.org/r/2c29bd9f-5b67-02d0-18a3-8828e78bbb6f@gmail.com

So the following unnecessary code still remains.

void get_page_bootmem()
{
...
        page->lru.next = (struct list_head *)type;
...

The patch removes this code from get_page_bootmem()
---
  mm/memory_hotplug.c | 1 -
  1 file changed, 1 deletion(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 7409f25..d67787d 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -179,7 +179,6 @@ static void release_memory_resource(struct resource *res)
  void get_page_bootmem(unsigned long info,  struct page *page,
  		      unsigned long type)
  {
-	page->lru.next = (struct list_head *)type;
  	page->freelist = (void *)type;
  	SetPagePrivate(page);
  	set_page_private(page, info);
-- 
1.8.3.1