Skip to contentSkip to conversation

A few fixup patches for memory failure

PATCH Superseded Miaohe Lin 5 participants 20 messages Mar 12, 2022 – Mar 16, 2022

Subsystems (5)

v2 of 4 Superseded 2 of 3 applied Latest is v4 →

lore
Share & tools
lore

Miaohe Lin

author maintainer
#0 lore raw

v2 cover A few fixup patches for memory failure

Hi everyone,
This series contains a few patches to fix the race with changing page compound page, make non-LRU movable pages unhandlable and so on. More details can be found in the respective changelogs. Thanks!

---
v1->v2:
drop "mm/memory-failure.c: fix wrong user reference report" make non-LRU movable pages unhandlable
fix confusing commit log and introduce MF_MSG_DIFFERENT_PAGE_SIZE Many thanks Naoya, Mike and Yang Shi for review!
---

Miaohe Lin (3):
mm/memory-failure.c: fix race with changing page compound again mm/memory-failure.c: avoid calling invalidate_inode_page() with
unexpected pages
mm/memory-failure.c: make non-LRU movable pages unhandlable

 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 34 ++++++++++++++++++++++++++--------
 3 files changed, 28 insertions(+), 8 deletions(-)
Signature · 1 line
2.23.0
author maintainer
#1 lore raw

PATCH v2 1/3 mm/memory-failure.c: fix race with changing page compound again

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
 	MF_MSG_BUDDY,
 	MF_MSG_DAX,
 	MF_MSG_UNSPLIT_THP,
+	MF_MSG_DIFFERENT_PAGE_SIZE,
 	MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
 	EM ( MF_MSG_BUDDY, "free buddy page" )				\
 	EM ( MF_MSG_DAX, "dax page" )					\
 	EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )			\
+	EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )	\
 	EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +12 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
 	[MF_MSG_BUDDY]			= "free buddy page",
 	[MF_MSG_DAX]			= "dax page",
 	[MF_MSG_UNSPLIT_THP]		= "unsplit thp",
+	[MF_MSG_DIFFERENT_PAGE_SIZE]	= "different page size",
 	[MF_MSG_UNKNOWN]		= "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
 	}
 
 	lock_page(head);
+
+	/**
+	 * The page could have changed compound pages due to race window.
+	 * If this happens just bail out.
+	 */
+	if (!PageHuge(p) || compound_head(p) != head) {
+		action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+		res = -EBUSY;
+		goto out;
+	}
+
 	page_flags = head->flags;
 
 	if (hwpoison_filter(p)) {
Signature · 1 line
2.23.0
#2 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 59 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
 	MF_MSG_BUDDY,
 	MF_MSG_DAX,
 	MF_MSG_UNSPLIT_THP,
+	MF_MSG_DIFFERENT_PAGE_SIZE,
 	MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
 	EM ( MF_MSG_BUDDY, "free buddy page" )				\
 	EM ( MF_MSG_DAX, "dax page" )					\
 	EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )			\
+	EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )	\
 	EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
 	[MF_MSG_BUDDY]			= "free buddy page",
 	[MF_MSG_DAX]			= "dax page",
 	[MF_MSG_UNSPLIT_THP]		= "unsplit thp",
+	[MF_MSG_DIFFERENT_PAGE_SIZE]	= "different page size",
 	[MF_MSG_UNKNOWN]		= "unknown page",
 };
 

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1539

Quoted 5 lines
mm/memory-failure.c +2 -0
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
 	}
 
 	lock_page(head);
+
+	/**

This comment section starting with '/**' is considered as kernel-doc comment, maybe that's not expected because it just describes an implementation detail. So normal comment section with '/*' would be better.

Otherwise, looks good to me.

Quoted 14 lines
mm/memory-failure.c +9 -0
+	 * The page could have changed compound pages due to race window.
+	 * If this happens just bail out.
+	 */
+	if (!PageHuge(p) || compound_head(p) != head) {
+		action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+		res = -EBUSY;
+		goto out;
+	}
+
 	page_flags = head->flags;
 
 	if (hwpoison_filter(p)) {
Signature · 1 line
2.23.0
author maintainer
#3 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 62 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 61 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
 	MF_MSG_BUDDY,
 	MF_MSG_DAX,
 	MF_MSG_UNSPLIT_THP,
+	MF_MSG_DIFFERENT_PAGE_SIZE,
 	MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
 	EM ( MF_MSG_BUDDY, "free buddy page" )				\
 	EM ( MF_MSG_DAX, "dax page" )					\
 	EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )			\
+	EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )	\
 	EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
 	[MF_MSG_BUDDY]			= "free buddy page",
 	[MF_MSG_DAX]			= "dax page",
 	[MF_MSG_UNSPLIT_THP]		= "unsplit thp",
+	[MF_MSG_DIFFERENT_PAGE_SIZE]	= "different page size",
 	[MF_MSG_UNKNOWN]		= "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
 	}
 
Quoted 8 lines

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1539

Quoted 3 lines
mm/memory-failure.c +2 -0
 	lock_page(head);
+
+	/**

This comment section starting with '/**' is considered as kernel-doc comment, maybe that's not expected because it just describes an implementation detail. So normal comment section with '/*' would be better.

I see. Will change to use "/*".

Quoted 3 lines

Otherwise, looks good to me.

Many thanks for review and comment.

Quoted 15 lines
Quoted 15 lines
mm/memory-failure.c +9 -0
+	 * The page could have changed compound pages due to race window.
+	 * If this happens just bail out.
+	 */
+	if (!PageHuge(p) || compound_head(p) != head) {
+		action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+		res = -EBUSY;
+		goto out;
+	}
+
 	page_flags = head->flags;
 
 	if (hwpoison_filter(p)) {
Signature · 1 line
2.23.0
#4 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

In reply to Miaohe Lin
Quoted 60 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
 	MF_MSG_BUDDY,
 	MF_MSG_DAX,
 	MF_MSG_UNSPLIT_THP,
+	MF_MSG_DIFFERENT_PAGE_SIZE,
 	MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
 	EM ( MF_MSG_BUDDY, "free buddy page" )				\
 	EM ( MF_MSG_DAX, "dax page" )					\
 	EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )			\
+	EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )	\
 	EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
 	[MF_MSG_BUDDY]			= "free buddy page",
 	[MF_MSG_DAX]			= "dax page",
 	[MF_MSG_UNSPLIT_THP]		= "unsplit thp",
+	[MF_MSG_DIFFERENT_PAGE_SIZE]	= "different page size",
 	[MF_MSG_UNKNOWN]		= "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
 	}

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1545

Quoted 10 lines
mm/memory-failure.c +8 -0
 
 	lock_page(head);
+
+	/**
+	 * The page could have changed compound pages due to race window.
+	 * If this happens just bail out.
+	 */
+	if (!PageHuge(p) || compound_head(p) != head) {
+		action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+		res = -EBUSY;

We have discussed this race in other versions of the patch. When we encounter the race, we have likely marked poison on the wrong page. Correct?

Instead of printing a "different page size", would it be better to perhaps:
- Print a message that wrong page may be marked for poison?
- Clear the poison flag in the "head page" previously set?

Signature · 1 line
Mike Kravetz
Quoted 6 lines
mm/memory-failure.c +3 -0
+		goto out;
+	}
+
 	page_flags = head->flags;
 
 	if (hwpoison_filter(p)) {
author maintainer
#5 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 62 lines · Mike Kravetz
Quoted 61 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
 	MF_MSG_BUDDY,
 	MF_MSG_DAX,
 	MF_MSG_UNSPLIT_THP,
+	MF_MSG_DIFFERENT_PAGE_SIZE,
 	MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
 	EM ( MF_MSG_BUDDY, "free buddy page" )				\
 	EM ( MF_MSG_DAX, "dax page" )					\
 	EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )			\
+	EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )	\
 	EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
 	[MF_MSG_BUDDY]			= "free buddy page",
 	[MF_MSG_DAX]			= "dax page",
 	[MF_MSG_UNSPLIT_THP]		= "unsplit thp",
+	[MF_MSG_DIFFERENT_PAGE_SIZE]	= "different page size",
 	[MF_MSG_UNKNOWN]		= "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
 	}
 
Quoted 13 lines

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1545

Quoted 9 lines
mm/memory-failure.c +8 -0
 	lock_page(head);
+
+	/**
+	 * The page could have changed compound pages due to race window.
+	 * If this happens just bail out.
+	 */
+	if (!PageHuge(p) || compound_head(p) != head) {
+		action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+		res = -EBUSY;

We have discussed this race in other versions of the patch. When we encounter the race, we have likely marked poison on the wrong page. Correct?

Many thanks for comment.
I assume that Naoya's patch "mm/hwpoison: set PageHWPoison after taking page lock in memory_failure_hugetlb()" would set the PageHWPoison after the above check. So I think the below operation is not needed as PageHWPoison is not set yet. Does this makes sense for you?

Thanks.

Quoted 4 lines

Instead of printing a "different page size", would it be better to perhaps:
- Print a message that wrong page may be marked for poison?
- Clear the poison flag in the "head page" previously set?

#6 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 78 lines · Miaohe Lin
Quoted 62 lines · Mike Kravetz
Quoted 61 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
     MF_MSG_BUDDY,
     MF_MSG_DAX,
     MF_MSG_UNSPLIT_THP,
+    MF_MSG_DIFFERENT_PAGE_SIZE,
     MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
     EM ( MF_MSG_BUDDY, "free buddy page" )                          \
     EM ( MF_MSG_DAX, "dax page" )                                   \
     EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )                        \
+    EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )        \
     EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
     [MF_MSG_BUDDY]                  = "free buddy page",
     [MF_MSG_DAX]                    = "dax page",
     [MF_MSG_UNSPLIT_THP]            = "unsplit thp",
+    [MF_MSG_DIFFERENT_PAGE_SIZE]    = "different page size",
     [MF_MSG_UNKNOWN]                = "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
     }
 
Quoted 13 lines

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1545

Quoted 9 lines
mm/memory-failure.c +8 -0
     lock_page(head);
+
+    /**
+     * The page could have changed compound pages due to race window.
+     * If this happens just bail out.
+     */
+    if (!PageHuge(p) || compound_head(p) != head) {
+            action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+            res = -EBUSY;

We have discussed this race in other versions of the patch. When we encounter the race, we have likely marked poison on the wrong page. Correct?

Quoted 5 lines

Many thanks for comment.
I assume that Naoya's patch "mm/hwpoison: set PageHWPoison after taking page lock in memory_failure_hugetlb()" would set the PageHWPoison after the above check. So I think the below operation is not needed as PageHWPoison is not set yet. Does this makes sense for you?

I'm wondering if it might be better and helpful for review to squash this patch with Naoya's patch together? It seems we always missed the other part when reviewing the patches.

Quoted 8 lines

Thanks.

Quoted 4 lines

Instead of printing a "different page size", would it be better to perhaps:
- Print a message that wrong page may be marked for poison?
- Clear the poison flag in the "head page" previously set?

author maintainer
#7 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 85 lines · Yang Shi
Quoted 78 lines · Miaohe Lin
Quoted 62 lines · Mike Kravetz
Quoted 61 lines · Miaohe Lin

There is a race window where we got the compound_head, the hugetlb page could be freed to buddy, or even changed to another compound page just before we try to get hwpoison page. Think about the below race window:
CPU 1 CPU 2
memory_failure_hugetlb
struct page *head = compound_head(p);
hugetlb page might be freed to buddy, or even changed to another compound page.

get_hwpoison_page -- page is not what we want now...

If this race happens, just bail out. Also MF_MSG_DIFFERENT_PAGE_SIZE is introduced to record this event.

---
 include/linux/mm.h      |  1 +
 include/ras/ras_event.h |  1 +
 mm/memory-failure.c     | 12 ++++++++++++
 3 files changed, 14 insertions(+)
include/linux/mm.h +1 -0
index c9bada4096ac..ef98cff2b253 100644
@@ -3253,6 +3253,7 @@ enum mf_action_page_type {
     MF_MSG_BUDDY,
     MF_MSG_DAX,
     MF_MSG_UNSPLIT_THP,
+    MF_MSG_DIFFERENT_PAGE_SIZE,
     MF_MSG_UNKNOWN,
 };
 
include/ras/ras_event.h +1 -0
index d0337a41141c..1e694fd239b9 100644
@@ -374,6 +374,7 @@ TRACE_EVENT(aer_event,
     EM ( MF_MSG_BUDDY, "free buddy page" )                          \
     EM ( MF_MSG_DAX, "dax page" )                                   \
     EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" )                        \
+    EM ( MF_MSG_DIFFERENT_PAGE_SIZE, "different page size" )        \
     EMe ( MF_MSG_UNKNOWN, "unknown page" )
 
 /*
mm/memory-failure.c +1 -0
index 5444a8ef4867..dabecd87ad3f 100644
@@ -733,6 +733,7 @@ static const char * const action_page_types[] = {
     [MF_MSG_BUDDY]                  = "free buddy page",
     [MF_MSG_DAX]                    = "dax page",
     [MF_MSG_UNSPLIT_THP]            = "unsplit thp",
+    [MF_MSG_DIFFERENT_PAGE_SIZE]    = "different page size",
     [MF_MSG_UNKNOWN]                = "unknown page",
 };
 
@@ -1534,6 +1535,17 @@ static int memory_failure_hugetlb(unsigned long pfn, int flags)
     }
 
Quoted 13 lines

mm/memory-failure.c static int memory_failure_hugetlb(unsigned long pfn, int flags) · line 1545

Quoted 9 lines
mm/memory-failure.c +8 -0
     lock_page(head);
+
+    /**
+     * The page could have changed compound pages due to race window.
+     * If this happens just bail out.
+     */
+    if (!PageHuge(p) || compound_head(p) != head) {
+            action_result(pfn, MF_MSG_DIFFERENT_PAGE_SIZE, MF_IGNORED);
+            res = -EBUSY;

We have discussed this race in other versions of the patch. When we encounter the race, we have likely marked poison on the wrong page. Correct?

Quoted 5 lines

Many thanks for comment.
I assume that Naoya's patch "mm/hwpoison: set PageHWPoison after taking page lock in memory_failure_hugetlb()" would set the PageHWPoison after the above check. So I think the below operation is not needed as PageHWPoison is not set yet. Does this makes sense for you?

Quoted 4 lines

I'm wondering if it might be better and helpful for review to squash this patch with Naoya's patch together? It seems we always missed the other part when reviewing the patches.

Sounds like a good idea. This would make the reviewer's life easier. I'm fine if this patch is squashed into Naoya's patch altogether. But we might have to consult the opinion from Naoya.

Thanks.

Quoted 10 lines
Quoted 8 lines

Thanks.

Quoted 4 lines

Instead of printing a "different page size", would it be better to perhaps:
- Print a message that wrong page may be marked for poison?
- Clear the poison flag in the "head page" previously set?

.

#8 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 2 lines · Miaohe Lin
Quoted 1 line · Yang Shi

On Tue, Mar 15, 2022 at 7:19 AM Miaohe Lin <[email protected]> wrote:

...

Quoted 8 lines
Quoted 8 lines
Quoted 7 lines

Many thanks for comment.
I assume that Naoya's patch "mm/hwpoison: set PageHWPoison after taking page lock in memory_failure_hugetlb()" would set the PageHWPoison after the above check. So I think the below operation is not needed as PageHWPoison is not set yet. Does this makes sense for you?

Quoted 8 lines
Quoted 4 lines

I'm wondering if it might be better and helpful for review to squash this patch with Naoya's patch together? It seems we always missed the other part when reviewing the patches.

Sounds like a good idea. This would make the reviewer's life easier. I'm fine if this patch is squashed into Naoya's patch altogether. But we might have to consult the opinion from Naoya.

I'm fine with the squashing, so I'll send v4.

Thanks,
Naoya Horiguchi

author maintainer
#9 lore raw

Re: [PATCH v2 1/3] mm/memory-failure.c: fix race with changing page compound again

Quoted 17 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 2 lines · Miaohe Lin
Quoted 1 line · Yang Shi

On Tue, Mar 15, 2022 at 7:19 AM Miaohe Lin <[email protected]> wrote:

...

Quoted 13 lines
Quoted 12 lines
Quoted 7 lines

Many thanks for comment.
I assume that Naoya's patch "mm/hwpoison: set PageHWPoison after taking page lock in memory_failure_hugetlb()" would set the PageHWPoison after the above check. So I think the below operation is not needed as PageHWPoison is not set yet. Does this makes sense for you?

I'm wondering if it might be better and helpful for review to squash this patch with Naoya's patch together? It seems we always missed the other part when reviewing the patches.

Quoted 5 lines
Quoted 3 lines

Sounds like a good idea. This would make the reviewer's life easier. I'm fine if this patch is squashed into Naoya's patch altogether. But we might have to consult the opinion from Naoya.

I'm fine with the squashing, so I'll send v4.

Many thanks for doing this.

So I'll send v3 later to fix the "stale commit id" in the commit log of the [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages.

Thanks.

Quoted 4 lines

Thanks,
Naoya Horiguchi

author maintainer
#10 lore raw

PATCH v2 2/3 mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

In reply to Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page() into its one caller"), invalidate_inode_page() can invalidate the pages in the swap cache because the check of page->mapping != mapping is removed. But invalidate_inode_page() is not expected to deal with the pages in swap cache. Also non-lru movable page can reach here too. They're not page cache pages. Skip these pages by checking PageSwapCache and PageLRU.

---
 mm/memory-failure.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memory-failure.c +1 -1
index dabecd87ad3f..2ff7dd2078c4 100644
@@ -2190,7 +2190,7 @@ static int __soft_offline_page(struct page *page)
 		return 0;
 	}
 
-	if (!PageHuge(page))
+	if (!PageHuge(page) && PageLRU(page) && !PageSwapCache(page))
 		/*
 		 * Try to invalidate first. This should work for
 		 * non dirty unmapped page cache pages.
Signature · 1 line
2.23.0
#11 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

Thanks,
Naoya Horiguchi

Quoted 26 lines

into its one caller"), invalidate_inode_page() can invalidate the pages in the swap cache because the check of page->mapping != mapping is removed. But invalidate_inode_page() is not expected to deal with the pages in swap cache. Also non-lru movable page can reach here too. They're not page cache pages. Skip these pages by checking PageSwapCache and PageLRU.

---
 mm/memory-failure.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memory-failure.c +1 -1
index dabecd87ad3f..2ff7dd2078c4 100644
@@ -2190,7 +2190,7 @@ static int __soft_offline_page(struct page *page)
 		return 0;
 	}
 
-	if (!PageHuge(page))
+	if (!PageHuge(page) && PageLRU(page) && !PageSwapCache(page))
 		/*
 		 * Try to invalidate first. This should work for
 		 * non dirty unmapped page cache pages.
Signature · 1 line
2.23.0
author maintainer Reviewed-by John Hubbard Reviewed-by Christoph Hellwig
#12 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 6 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

This commit is in the mmotm but not in mainline yet:

commit 042c4f32323beb28146c658202d3e69899e4f245

Author: Matthew Wilcox (Oracle) <[email protected]>
Date:   Sat Feb 12 15:27:42 2022 -0500

mm/truncate: Inline invalidate_complete_page() into its one caller

invalidate_inode_page() is the only caller of invalidate_complete_page() and inlining it reveals that the first check is unnecessary (because we hold the page locked, and we just retrieved the mapping from the page). Actually, it does make a difference, in that tail pages no longer fail at this check, so it's now possible to remove a tail page from a mapping.

Am I "not" supposed to use this commit id as it's not "stable" now?

Will update this part in next version. Many thanks.

Quoted 29 lines

Thanks,
Naoya Horiguchi

Quoted 26 lines

into its one caller"), invalidate_inode_page() can invalidate the pages in the swap cache because the check of page->mapping != mapping is removed. But invalidate_inode_page() is not expected to deal with the pages in swap cache. Also non-lru movable page can reach here too. They're not page cache pages. Skip these pages by checking PageSwapCache and PageLRU.

---
 mm/memory-failure.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/memory-failure.c +1 -1
index dabecd87ad3f..2ff7dd2078c4 100644
@@ -2190,7 +2190,7 @@ static int __soft_offline_page(struct page *page)
 		return 0;
 	}
 
-	if (!PageHuge(page))
+	if (!PageHuge(page) && PageLRU(page) && !PageSwapCache(page))
 		/*
 		 * Try to invalidate first. This should work for
 		 * non dirty unmapped page cache pages.
Signature · 1 line
2.23.0
#13 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 22 lines · Miaohe Lin
Quoted 6 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

This commit is in the mmotm but not in mainline yet:

commit 042c4f32323beb28146c658202d3e69899e4f245

Author: Matthew Wilcox (Oracle) <[email protected]>
Date:   Sat Feb 12 15:27:42 2022 -0500

mm/truncate: Inline invalidate_complete_page() into its one caller

invalidate_inode_page() is the only caller of invalidate_complete_page() and inlining it reveals that the first check is unnecessary (because we hold the page locked, and we just retrieved the mapping from the page). Actually, it does make a difference, in that tail pages no longer fail at this check, so it's now possible to remove a tail page from a mapping.

Quoted 5 lines

Am I "not" supposed to use this commit id as it's not "stable" now?

No, it's not stable yet. In whatever way you get the above commit (I guess you get it from https://github.com/hnaz/linux-mm), all acked mm-related patches are sent to Linus by Andrew *by email*, so the eventual commit IDs should be determined when they are applied to mainline.

Thanks,
Naoya Horiguchi

author maintainer
#14 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 27 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 26 lines · Miaohe Lin
Quoted 6 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

This commit is in the mmotm but not in mainline yet:

commit 042c4f32323beb28146c658202d3e69899e4f245

Author: Matthew Wilcox (Oracle) <[email protected]>
Date:   Sat Feb 12 15:27:42 2022 -0500

mm/truncate: Inline invalidate_complete_page() into its one caller

invalidate_inode_page() is the only caller of invalidate_complete_page() and inlining it reveals that the first check is unnecessary (because we hold the page locked, and we just retrieved the mapping from the page). Actually, it does make a difference, in that tail pages no longer fail at this check, so it's now possible to remove a tail page from a mapping.

Quoted 7 lines
Quoted 1 line

Am I "not" supposed to use this commit id as it's not "stable" now?

No, it's not stable yet. In whatever way you get the above commit (I guess you get it from https://github.com/hnaz/linux-mm), all acked mm-related patches are sent to Linus by Andrew *by email*, so the eventual commit IDs should be determined when they are applied to mainline.

Many thanks for your explanation. (I get this commit id from linux-next tree:
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git)
So I should remember always to get the commit id from mainline.

Thanks again. :)

Quoted 3 lines

Thanks,
Naoya Horiguchi

maintainer
#15 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 36 lines · On Mon, 14 Mar 2022 10:59:40 +0800 Miaohe Lin <[email protected]> wrote:
Quoted 27 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 26 lines · Miaohe Lin
Quoted 6 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

This commit is in the mmotm but not in mainline yet:

commit 042c4f32323beb28146c658202d3e69899e4f245

Author: Matthew Wilcox (Oracle) <[email protected]>
Date:   Sat Feb 12 15:27:42 2022 -0500

mm/truncate: Inline invalidate_complete_page() into its one caller

invalidate_inode_page() is the only caller of invalidate_complete_page() and inlining it reveals that the first check is unnecessary (because we hold the page locked, and we just retrieved the mapping from the page). Actually, it does make a difference, in that tail pages no longer fail at this check, so it's now possible to remove a tail page from a mapping.

Quoted 7 lines
Quoted 1 line

Am I "not" supposed to use this commit id as it's not "stable" now?

No, it's not stable yet. In whatever way you get the above commit (I guess you get it from https://github.com/hnaz/linux-mm), all acked mm-related patches are sent to Linus by Andrew *by email*, so the eventual commit IDs should be determined when they are applied to mainline.

Quoted 3 lines

Many thanks for your explanation. (I get this commit id from linux-next tree:
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git)
So I should remember always to get the commit id from mainline.

It's likely that this commit ID will be the same once Matthew's patch goes into mainline.

But this is why we include the patch title ("mm/truncate: Inline ...") when identifying commits. Sometimes stuff happens...

author maintainer
#16 lore raw

Re: [PATCH v2 2/3] mm/memory-failure.c: avoid calling invalidate_inode_page() with unexpected pages

Quoted 42 lines · Andrew Morton
Quoted 36 lines · On Mon, 14 Mar 2022 10:59:40 +0800 Miaohe Lin <[email protected]> wrote:
Quoted 27 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 26 lines · Miaohe Lin
Quoted 6 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 1 line · Miaohe Lin

Since commit 042c4f32323b ("mm/truncate: Inline invalidate_complete_page()

This commit ID does not exist in mainline (or in the latest mmotm?), so you can't use it in patch description. Could you update this part?

This commit is in the mmotm but not in mainline yet:

commit 042c4f32323beb28146c658202d3e69899e4f245

Author: Matthew Wilcox (Oracle) <[email protected]>
Date:   Sat Feb 12 15:27:42 2022 -0500

mm/truncate: Inline invalidate_complete_page() into its one caller

invalidate_inode_page() is the only caller of invalidate_complete_page() and inlining it reveals that the first check is unnecessary (because we hold the page locked, and we just retrieved the mapping from the page). Actually, it does make a difference, in that tail pages no longer fail at this check, so it's now possible to remove a tail page from a mapping.

Quoted 7 lines
Quoted 1 line

Am I "not" supposed to use this commit id as it's not "stable" now?

No, it's not stable yet. In whatever way you get the above commit (I guess you get it from https://github.com/hnaz/linux-mm), all acked mm-related patches are sent to Linus by Andrew *by email*, so the eventual commit IDs should be determined when they are applied to mainline.

Quoted 3 lines

Many thanks for your explanation. (I get this commit id from linux-next tree:
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git)
So I should remember always to get the commit id from mainline.

Quoted 5 lines

It's likely that this commit ID will be the same once Matthew's patch goes into mainline.

But this is why we include the patch title ("mm/truncate: Inline ...") when identifying commits. Sometimes stuff happens...

I remember I used the stale commit id once in my past patch. I made this mistake again. Sorry about it. :(

Quoted 2 lines

.

author maintainer
#17 lore raw

PATCH v2 3/3 mm/memory-failure.c: make non-LRU movable pages unhandlable

In reply to Miaohe Lin

We can not really handle non-LRU movable pages in memory failure. Typically they are balloon, zsmalloc, etc. Assuming we run into a base (4K) non-LRU movable page, we could reach as far as identify_page_state(), it should not fall into any category except me_unknown. For the non-LRU compound movable pages, they could be taken for transhuge pages but it's unexpected to split non-LRU movable pages using split_huge_page_to_list in memory_failure. So we could just simply make non-LRU movable pages unhandlable to avoid these possible nasty cases.

---
 mm/memory-failure.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)
mm/memory-failure.c +13 -7
index 2ff7dd2078c4..ba621c6823ed 100644
@@ -1177,12 +1177,18 @@ void ClearPageHWPoisonTakenOff(struct page *page)
  * does not return true for hugetlb or device memory pages, so it's assumed
  * to be called only in the context where we never have such pages.
  */
-static inline bool HWPoisonHandlable(struct page *page)
+static inline bool HWPoisonHandlable(struct page *page, unsigned long flags)
 {
-	return PageLRU(page) || __PageMovable(page) || is_free_buddy_page(page);
+	bool movable = false;
+
+	/* Soft offline could mirgate non-LRU movable pages */
+	if ((flags & MF_SOFT_OFFLINE) && __PageMovable(page))
+		movable = true;
+
+	return movable || PageLRU(page) || is_free_buddy_page(page);
 }
 
-static int __get_hwpoison_page(struct page *page)
+static int __get_hwpoison_page(struct page *page, unsigned long flags)
 {
 	struct page *head = compound_head(page);
 	int ret = 0;
@@ -1197,7 +1203,7 @@ static int __get_hwpoison_page(struct page *page)
 	 * for any unsupported type of page in order to reduce the risk of
 	 * unexpected races caused by taking a page refcount.
 	 */
-	if (!HWPoisonHandlable(head))
+	if (!HWPoisonHandlable(head, flags))
 		return -EBUSY;
 
 	if (get_page_unless_zero(head)) {
@@ -1222,7 +1228,7 @@ static int get_any_page(struct page *p, unsigned long flags)
 
 try_again:
 	if (!count_increased) {
-		ret = __get_hwpoison_page(p);
+		ret = __get_hwpoison_page(p, flags);
 		if (!ret) {
 			if (page_count(p)) {
 				/* We raced with an allocation, retry. */
@@ -1250,7 +1256,7 @@ static int get_any_page(struct page *p, unsigned long flags)
 		}
 	}
 
-	if (PageHuge(p) || HWPoisonHandlable(p)) {
+	if (PageHuge(p) || HWPoisonHandlable(p, flags)) {
 		ret = 1;
 	} else {
 		/*
@@ -2308,7 +2314,7 @@ int soft_offline_page(unsigned long pfn, int flags)
 
 retry:
 	get_online_mems();
-	ret = get_hwpoison_page(page, flags);
+	ret = get_hwpoison_page(page, flags | MF_SOFT_OFFLINE);
 	put_online_mems();
 
 	if (ret > 0) {
Signature · 1 line
2.23.0
#18 lore raw

Re: [PATCH v2 3/3] mm/memory-failure.c: make non-LRU movable pages unhandlable

Quoted 11 lines · Miaohe Lin

We can not really handle non-LRU movable pages in memory failure. Typically they are balloon, zsmalloc, etc. Assuming we run into a base (4K) non-LRU movable page, we could reach as far as identify_page_state(), it should not fall into any category except me_unknown. For the non-LRU compound movable pages, they could be taken for transhuge pages but it's unexpected to split non-LRU movable pages using split_huge_page_to_list in memory_failure. So we could just simply make non-LRU movable pages unhandlable to avoid these possible nasty cases.

Looks good to me.

Reviewed-by on v2 3/3
#19 lore raw

Re: [PATCH v2 3/3] mm/memory-failure.c: make non-LRU movable pages unhandlable

In reply to Miaohe Lin
Quoted 10 lines · Miaohe Lin

We can not really handle non-LRU movable pages in memory failure. Typically they are balloon, zsmalloc, etc. Assuming we run into a base (4K) non-LRU movable page, we could reach as far as identify_page_state(), it should not fall into any category except me_unknown. For the non-LRU compound movable pages, they could be taken for transhuge pages but it's unexpected to split non-LRU movable pages using split_huge_page_to_list in memory_failure. So we could just simply make non-LRU movable pages unhandlable to avoid these possible nasty cases.

Quoted 2 lines
Quoted 69 lines
---
 mm/memory-failure.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)
mm/memory-failure.c +13 -8
index 2ff7dd2078c4..ba621c6823ed 100644
@@ -1177,12 +1177,18 @@ void ClearPageHWPoisonTakenOff(struct page *page)
  * does not return true for hugetlb or device memory pages, so it's assumed
  * to be called only in the context where we never have such pages.
  */
-static inline bool HWPoisonHandlable(struct page *page)
+static inline bool HWPoisonHandlable(struct page *page, unsigned long flags)
 {
-       return PageLRU(page) || __PageMovable(page) || is_free_buddy_page(page);
+       bool movable = false;
+
+       /* Soft offline could mirgate non-LRU movable pages */
+       if ((flags & MF_SOFT_OFFLINE) && __PageMovable(page))
+               movable = true;
+
+       return movable || PageLRU(page) || is_free_buddy_page(page);
 }
 
-static int __get_hwpoison_page(struct page *page)
+static int __get_hwpoison_page(struct page *page, unsigned long flags)
 {
        struct page *head = compound_head(page);
        int ret = 0;
@@ -1197,7 +1203,7 @@ static int __get_hwpoison_page(struct page *page)
         * for any unsupported type of page in order to reduce the risk of
         * unexpected races caused by taking a page refcount.
         */
-       if (!HWPoisonHandlable(head))
+       if (!HWPoisonHandlable(head, flags))
                return -EBUSY;
 
        if (get_page_unless_zero(head)) {
@@ -1222,7 +1228,7 @@ static int get_any_page(struct page *p, unsigned long flags)
 
 try_again:
        if (!count_increased) {
-               ret = __get_hwpoison_page(p);
+               ret = __get_hwpoison_page(p, flags);
                if (!ret) {
                        if (page_count(p)) {
                                /* We raced with an allocation, retry. */
@@ -1250,7 +1256,7 @@ static int get_any_page(struct page *p, unsigned long flags)
                }
        }
 
-       if (PageHuge(p) || HWPoisonHandlable(p)) {
+       if (PageHuge(p) || HWPoisonHandlable(p, flags)) {
                ret = 1;
        } else {
                /*
@@ -2308,7 +2314,7 @@ int soft_offline_page(unsigned long pfn, int flags)
 
 retry:
        get_online_mems();
-       ret = get_hwpoison_page(page, flags);
+       ret = get_hwpoison_page(page, flags | MF_SOFT_OFFLINE);
        put_online_mems();
 
        if (ret > 0) {
--

2.23.0