Skip to contentSkip to conversation

mm,hwpoison: set PG_hwpoison for busy hugetlb pages

PATCH In review Naoya Horiguchi 4 participants 10 messages May 11, 2022 – Jun 6, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm,hwpoison: set PG_hwpoison for busy hugetlb pages

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +4 -4
index 6a28d020a4da..e3269b991016 100644
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}
 
 	if (TestSetPageHWPoison(head)) {
@@ -1556,7 +1557,6 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
 	struct page *p = pfn_to_page(pfn);
 	struct page *head;
 	unsigned long page_flags;
-	bool retry = true;
 
 	*hugetlb = 1;
 retry:
@@ -1572,8 +1572,8 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
 		}
 		return res;
 	} else if (res == -EBUSY) {
-		if (retry) {
-			retry = false;
+		if (!(flags & MF_NO_RETRY)) {
+			flags |= MF_NO_RETRY;
 			goto retry;
 		}
 		action_result(pfn, MF_MSG_UNKNOWN, MF_IGNORED);
Signature · 1 line
2.25.1
#1 lore raw
Quoted 35 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

Signature · 1 line
Mike Kravetz
Quoted 21 lines
mm/memory-failure.c +2 -3
 	if (TestSetPageHWPoison(head)) {
@@ -1556,7 +1557,6 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
 	struct page *p = pfn_to_page(pfn);
 	struct page *head;
 	unsigned long page_flags;
-	bool retry = true;
 
 	*hugetlb = 1;
 retry:
@@ -1572,8 +1572,8 @@ static int try_memory_failure_hugetlb(unsigned long pfn, int flags, int *hugetlb
 		}
 		return res;
 	} else if (res == -EBUSY) {
-		if (retry) {
-			retry = false;
+		if (!(flags & MF_NO_RETRY)) {
+			flags |= MF_NO_RETRY;
 			goto retry;
 		}
 		action_result(pfn, MF_MSG_UNKNOWN, MF_IGNORED);
maintainer
#2 lore raw
Quoted 11 lines · Mike Kravetz
Quoted 10 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

I'm sorry but where is hugetlb_set_page_hwpoison() defined and used ? I can't find it.

Quoted 39 lines
Quoted 25 lines
Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

Quoted 7 lines

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

It seems HWPoison hugetlb page will still be accessed before this patch. Can we do a get_page_unless_zero first here to ensure that hugetlb page migration should fail due to this extra page reference and thus not access the page content? If hugetlb page is already freezed, corrupted memory will still be consumed though. :(

Thanks!

Quoted 1 line
#3 lore raw
Quoted 59 lines · Miaohe Lin
Quoted 11 lines · Mike Kravetz
Quoted 10 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

I'm sorry but where is hugetlb_set_page_hwpoison() defined and used ? I can't find it.

Quoted 44 lines
Quoted 25 lines
Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Quoted 7 lines
Quoted 2 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

It seems HWPoison hugetlb page will still be accessed before this patch. Can we do a get_page_unless_zero first here to ensure that hugetlb page migration should fail due to this extra page reference and thus not access the page content? If hugetlb page is already freezed, corrupted memory will still be consumed though. :(

Right. This potential issue was not introduced with this patch. Also, I am not sure but it might be an issue with non-hugetlb pages as well.

As mentioned, memory error handling is a best effort. Since errors can happen at any time, we can not handle all cases. Or, you could spend the rest of your life trying. :)

The question is, should we worry about errors that happen when a page is isolated for migration?

Signature · 1 line
Mike Kravetz
#4 lore raw
Quoted 71 lines · Mike Kravetz
Quoted 59 lines · Miaohe Lin
Quoted 11 lines · Mike Kravetz
Quoted 10 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

I'm sorry but where is hugetlb_set_page_hwpoison() defined and used ? I can't find it.

Quoted 44 lines
Quoted 25 lines
Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Quoted 7 lines
Quoted 2 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

It seems HWPoison hugetlb page will still be accessed before this patch. Can we do a get_page_unless_zero first here to ensure that hugetlb page migration should fail due to this extra page reference and thus not access the page content? If hugetlb page is already freezed, corrupted memory will still be consumed though. :(

Right. This potential issue was not introduced with this patch. Also, I am not sure but it might be an issue with non-hugetlb pages as well.

Quoted 6 lines

As mentioned, memory error handling is a best effort. Since errors can happen at any time, we can not handle all cases. Or, you could spend the rest of your life trying. :)

The question is, should we worry about errors that happen when a page is isolated for migration?

I think yes, but that's not to save current migration event, but to save us from future memory errors caused by the broken page.

Thanks,
Naoya Horiguchi

#5 lore raw
In reply to Miaohe Lin
Quoted 11 lines · Miaohe Lin
Quoted 10 lines · Mike Kravetz
Quoted 9 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

Quoted 3 lines
Quoted 1 line
Quoted 1 line

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

I'm sorry but where is hugetlb_set_page_hwpoison() defined and used ? I can't find it.

Sorry, this depends on the unmerged patch
https://lore.kernel.org/linux-mm/[email protected]/
, so should come after that. I'll do both into a single patchset next.

...

Quoted 24 lines
Quoted 24 lines

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 12 lines
mm/memory-failure.c +2 -1
index 6a28d020a4da..e3269b991016 100644
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Quoted 7 lines
Quoted 2 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

It seems HWPoison hugetlb page will still be accessed before this patch. Can we do a get_page_unless_zero first here to ensure that hugetlb page migration should fail due to this extra page reference and thus not access the page content? If hugetlb page is already freezed, corrupted memory will still be consumed though. :(

Right, I have no idea about this ...

Thanks,
Naoya Horiguchi

#6 lore raw
In reply to Mike Kravetz
Quoted 46 lines · Mike Kravetz
Quoted 35 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

Quoted 3 lines

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

Yes, and I also detected this issue by testing race between hugepage allocation and memory_failure().

Quoted 5 lines

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Yes, that could happen. This patch does not affect ongoing hugepage migration. But after the migration source hugepage is freed, the PG_hwpoison should work to prevent reusing.

Quoted 3 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

Right, so doing nothing for this case could be OK if doing something causes some issues or makes code too complicated. The motivation of this patch is that now I think memory_failure() should do something (at least setting PG_hwpoison) unless the page is already hwpoisoned or rejected by hwpoison_filter(), because of the effect after free as mentioned above.

This is also expected in other case too. For example, slab is a unhandlable type of page, but we do set PG_hwpoison. This flag should not affect any of ongoing slab-related process, but that's OK because it becomes effective after the slab page is freed.

So this patch is intended to align to the behavior. Allowing hugepage migration to do something good using PG_hwpoison seems to me an unsolved separate issue.

Thanks,
Naoya Horiguchi

maintainer
#7 lore raw
Quoted 74 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 46 lines · Mike Kravetz
Quoted 35 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

Quoted 3 lines

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

Yes, and I also detected this issue by testing race between hugepage allocation and memory_failure().

Quoted 5 lines

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Yes, that could happen. This patch does not affect ongoing hugepage migration. But after the migration source hugepage is freed, the PG_hwpoison should work to prevent reusing.

Quoted 3 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

Right, so doing nothing for this case could be OK if doing something causes some issues or makes code too complicated. The motivation of this patch is that now I think memory_failure() should do something (at least setting PG_hwpoison) unless the page is already hwpoisoned or rejected by hwpoison_filter(), because of the effect after free as mentioned above.

Quoted 8 lines

This is also expected in other case too. For example, slab is a unhandlable type of page, but we do set PG_hwpoison. This flag should not affect any of ongoing slab-related process, but that's OK because it becomes effective after the slab page is freed.

So this patch is intended to align to the behavior. Allowing hugepage migration to do something good using PG_hwpoison seems to me an unsolved separate issue.

I tend to agree with Naoya. And could we try to do it better? IMHO, we could do a get_page_unless_zero here to ensure that hugetlb page migration will fail due to this extra page reference and thus preventing the page content from being accessed. Does this work? Or am I miss something?

Thanks!

Quoted 4 lines

Thanks,
Naoya Horiguchi

#8 lore raw
Quoted 80 lines · Miaohe Lin
Quoted 79 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 46 lines · Mike Kravetz
Quoted 35 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

Quoted 3 lines

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

Yes, and I also detected this issue by testing race between hugepage allocation and memory_failure().

Quoted 5 lines

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Yes, that could happen. This patch does not affect ongoing hugepage migration. But after the migration source hugepage is freed, the PG_hwpoison should work to prevent reusing.

Quoted 3 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

Right, so doing nothing for this case could be OK if doing something causes some issues or makes code too complicated. The motivation of this patch is that now I think memory_failure() should do something (at least setting PG_hwpoison) unless the page is already hwpoisoned or rejected by hwpoison_filter(), because of the effect after free as mentioned above.

This is also expected in other case too. For example, slab is a unhandlable type of page, but we do set PG_hwpoison. This flag should not affect any of ongoing slab-related process, but that's OK because it becomes effective after the slab page is freed.

Quoted 8 lines
Quoted 3 lines

So this patch is intended to align to the behavior. Allowing hugepage migration to do something good using PG_hwpoison seems to me an unsolved separate issue.

I tend to agree with Naoya. And could we try to do it better? IMHO, we could do a get_page_unless_zero here to ensure that hugetlb page migration will fail due to this extra page reference and thus preventing the page content from being accessed. Does this work? Or am I miss something?

Sorry for my missing to answering the question,

Taking page refcount to prevent page migration could work. One concern is how we can distinguish hugepages under migration and those under allocation from buddy. Maybe this was also mentioned in discussion over
https://github.com/torvalds/linux/commit/405ce051236cc65b30bbfe490b28ce60ae6aed85
, there's a small window where an allocating compound page is refcounted and hugepage (having deconstructor COMPOUND_PAGE_DTOR), and not protected by hugetlb_lock, so simply get_page_unless_zero() might not work (might break allocation code).
If we have more reliable indicator to ensure that a hugepage is under migration, that would be helpful.

Thanks,
Naoya Horiguchi

maintainer
#9 lore raw
Quoted 85 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 84 lines · Miaohe Lin
Quoted 82 lines · HORIGUCHI NAOYA(堀口 直也)
Quoted 46 lines · Mike Kravetz
Quoted 35 lines · Naoya Horiguchi

From: Naoya Horiguchi <[email protected]>

If memory_failure() fails to grab page refcount on a hugetlb page because it's busy, it returns without setting PG_hwpoison on it. This not only loses a chance of error containment, but breaks the rule that action_result() should be called only when memory_failure() do any of handling work (even if that's just setting PG_hwpoison). This inconsistency could harm code maintainability.

So set PG_hwpoison and call hugetlb_set_page_hwpoison() for such a case.

Fixes 405ce051236c ("mm/hwpoison: fix race between hugetlb free/demotion and memory_failure_hugetlb()")
---
 include/linux/mm.h  | 1 +
 mm/memory-failure.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)
include/linux/mm.h +1 -0
index d446e834a3e5..04de0c3e4f9f 100644
@@ -3187,6 +3187,7 @@ enum mf_flags {
 	MF_MUST_KILL = 1 << 2,
 	MF_SOFT_OFFLINE = 1 << 3,
 	MF_UNPOISON = 1 << 4,
+	MF_NO_RETRY = 1 << 5,
 };
 extern int memory_failure(unsigned long pfn, int flags);
 extern void memory_failure_queue(unsigned long pfn, int flags);
mm/memory-failure.c +0 -0
index 6a28d020a4da..e3269b991016 100644

mm/memory-failure.c int __get_huge_page_for_hwpoison(unsigned long pfn, int flags) · line 1531

Quoted 7 lines
mm/memory-failure.c +2 -1
@@ -1526,7 +1526,8 @@ int __get_huge_page_for_hwpoison(unsigned long pfn, int flags)
 			count_increased = true;
 	} else {
 		ret = -EBUSY;
-		goto out;
+		if (!(flags & MF_NO_RETRY))
+			goto out;
 	}

Hi Naoya,

Quoted 3 lines

We are in the else block because !HPageFreed() and !HPageMigratable(). IIUC, this likely means the page is isolated. One common reason for isolation is migration. So, the page could be isolated and on a list for migration.

Yes, and I also detected this issue by testing race between hugepage allocation and memory_failure().

Quoted 5 lines

I took a quick look at the hugetlb migration code and did not see any checks for PageHWPoison after a hugetlb page is isolated. I could have missed something? If there are no checks, we will read the PageHWPoison page in kernel mode while copying to the migration target.

Yes, that could happen. This patch does not affect ongoing hugepage migration. But after the migration source hugepage is freed, the PG_hwpoison should work to prevent reusing.

Quoted 3 lines

Is this an issue? Is is something we need to be concerned with? Memory errors can happen at any time, and gracefully handling them is best effort.

Right, so doing nothing for this case could be OK if doing something causes some issues or makes code too complicated. The motivation of this patch is that now I think memory_failure() should do something (at least setting PG_hwpoison) unless the page is already hwpoisoned or rejected by hwpoison_filter(), because of the effect after free as mentioned above.

This is also expected in other case too. For example, slab is a unhandlable type of page, but we do set PG_hwpoison. This flag should not affect any of ongoing slab-related process, but that's OK because it becomes effective after the slab page is freed.

So this patch is intended to align to the behavior. Allowing hugepage migration to do something good using PG_hwpoison seems to me an unsolved separate issue.

Quoted 6 lines
Quoted 4 lines

I tend to agree with Naoya. And could we try to do it better? IMHO, we could do a get_page_unless_zero here to ensure that hugetlb page migration will fail due to this extra page reference and thus preventing the page content from being accessed. Does this work? Or am I miss something?

Sorry for my missing to answering the question,

Never mind. I almost forget it too. ;)

Quoted 11 lines

Taking page refcount to prevent page migration could work. One concern is how we can distinguish hugepages under migration and those under allocation from buddy. Maybe this was also mentioned in discussion over
https://github.com/torvalds/linux/commit/405ce051236cc65b30bbfe490b28ce60ae6aed85
, there's a small window where an allocating compound page is refcounted and hugepage (having deconstructor COMPOUND_PAGE_DTOR), and not protected by hugetlb_lock, so simply get_page_unless_zero() might not work (might break allocation code).
If we have more reliable indicator to ensure that a hugepage is under migration, that would be helpful.

Yes, I agree with you. If we have more reliable indicator to ensure that a hugepage is under migration, we could try to do this then. :)

Quoted 3 lines

Thanks,
Naoya Horiguchi

Many thanks for reply!

Quoted 1 line