[0/7] migrate_pages(): fix several bugs in error path

Commit Message

Huang, Ying June 24, 2022, 2:53 a.m. UTC

From: "Huang, Ying" <[e-mail]>

During review the code of migrate_pages() and build a test program for
it.  Several bugs in error path are identified and fixed in this
series.

Most patches are tested via

- Apply error-inject.patch in Linux kernel
- Compile test-migrate.c (with -lnuma)
- Test with test-migrate.sh

error-inject.patch, test-migrate.c, and test-migrate.sh are as below.
It turns out that error injection is an important tool to fix bugs in
error path.

Best Regards,
Huang, Ying

------------------------- error-inject.patch -------------------------
From 295ea21204f3f025a041fe39c68a2eaec8313c68 Mon Sep 17 00:00:00 2001
From: Huang Ying <[e-mail]>
Date: Tue, 21 Jun 2022 11:08:30 +0800
Subject: [PATCH] migrate_pages: error inject

---
 mm/migrate.c | 58 +++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 55 insertions(+), 3 deletions(-)

Comments

Baolin Wang June 24, 2022, 10:11 a.m. UTC | #1

在 6/24/2022 10:53 AM, Huang Ying 写道:
> From: "Huang, Ying" <[e-mail]>
> 
> During review the code of migrate_pages() and build a test program for
> it.  Several bugs in error path are identified and fixed in this
> series.
> 
> Most patches are tested via
> 
> - Apply error-inject.patch in Linux kernel
> - Compile test-migrate.c (with -lnuma)
> - Test with test-migrate.sh
> 
> error-inject.patch, test-migrate.c, and test-migrate.sh are as below.
> It turns out that error injection is an important tool to fix bugs in
> error path.
> 

Hi Ying,

The subject prefix shows there are 7 patches in your patch set, however 
I only received 5 patches in my mail box, and I checked the link [1], it 
also shows there are 5 patches. Seems you missed 2 patches?

https://lore.kernel.org/all/[e-mail]/

Huang, Ying June 27, 2022, 2:30 a.m. UTC | #2

Baolin Wang <[e-mail]> writes:

> ( 6/24/2022 10:53 AM, Huang Ying S:
>> From: "Huang, Ying" <[e-mail]>
>> During review the code of migrate_pages() and build a test program
>> for
>> it.  Several bugs in error path are identified and fixed in this
>> series.
>> Most patches are tested via
>> - Apply error-inject.patch in Linux kernel
>> - Compile test-migrate.c (with -lnuma)
>> - Test with test-migrate.sh
>> error-inject.patch, test-migrate.c, and test-migrate.sh are as
>> below.
>> It turns out that error injection is an important tool to fix bugs in
>> error path.
>> 
>
> Hi Ying,
>
> The subject prefix shows there are 7 patches in your patch set,
> however I only received 5 patches in my mail box, and I checked the
> link [1], it also shows there are 5 patches. Seems you missed 2
> patches?
>
> https://lore.kernel.org/all/[e-mail]/

Hi, Baolin,

Thanks a lot for your information!  It turns out to be a format error in
my [6/7] that prevent git-send-email to work properly.  I have fixed it
and send out [6/7] and [7/7] in this thread.

Best Regards,
Huang, Ying

Patch

diff --git a/mm/migrate.c b/mm/migrate.c
index 399904015d23..87d47064ec6c 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -337,6 +337,42 @@  void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd)
 }
 #endif
 
+#define EI_MP_ENOSYS		0x0001
+#define EI_MP_THP_ENOMEM	0x0002
+#define EI_MP_NP_ENOMEM		0x0004
+#define EI_MP_EAGAIN		0x0008
+#define EI_MP_EOTHER		0x0010
+#define EI_MP_NOSPLIT		0x0020
+#define EI_MP_SPLIT_FAIL	0x0040
+#define EI_MP_EAGAIN_PERM	0x0080
+#define EI_MP_EBUSY		0x0100
+
+static unsigned int ei_migrate_pages;
+
+module_param(ei_migrate_pages, uint, 0644);
+
+static bool ei_thp_migration_supported(void)
+{
+	if (ei_migrate_pages & EI_MP_ENOSYS)
+		return false;
+	else
+		return thp_migration_supported();
+}
+
+static int ei_trylock_page(struct page *page)
+{
+	if (ei_migrate_pages & EI_MP_EAGAIN)
+		return 0;
+	return trylock_page(page);
+}
+
+static int ei_split_huge_page_to_list(struct page *page, struct list_head *list)
+{
+	if (ei_migrate_pages & EI_MP_SPLIT_FAIL)
+		return -EBUSY;
+	return split_huge_page_to_list(page, list);
+}
+
 static int expected_page_refs(struct address_space *mapping, struct page *page)
 {
 	int expected_count = 1;
@@ -368,6 +404,9 @@  int folio_migrate_mapping(struct address_space *mapping,
 		if (folio_ref_count(folio) != expected_count)
 			return -EAGAIN;
 
+		if (ei_migrate_pages & EI_MP_EAGAIN_PERM)
+			return -EAGAIN;
+
 		/* No turning back from here */
 		newfolio->index = folio->index;
 		newfolio->mapping = folio->mapping;
@@ -929,7 +968,7 @@  static int __unmap_and_move(struct page *page, struct page *newpage,
 	struct anon_vma *anon_vma = NULL;
 	bool is_lru = !__PageMovable(page);
 
-	if (!trylock_page(page)) {
+	if (!ei_trylock_page(page)) {
 		if (!force || mode == MIGRATE_ASYNC)
 			goto out;
 
@@ -952,6 +991,11 @@  static int __unmap_and_move(struct page *page, struct page *newpage,
 		lock_page(page);
 	}
 
+	if (ei_migrate_pages & EI_MP_EBUSY) {
+		rc = -EBUSY;
+		goto out_unlock;
+	}
+
 	if (PageWriteback(page)) {
 		/*
 		 * Only in the case of a full synchronous migration is it
@@ -1086,7 +1130,7 @@  static int unmap_and_move(new_page_t get_new_page,
 	int rc = MIGRATEPAGE_SUCCESS;
 	struct page *newpage = NULL;
 
-	if (!thp_migration_supported() && PageTransHuge(page))
+	if (!ei_thp_migration_supported() && PageTransHuge(page))
 		return -ENOSYS;
 
 	if (page_count(page) == 1) {
@@ -1102,6 +1146,11 @@  static int unmap_and_move(new_page_t get_new_page,
 		goto out;
 	}
 
+	if ((ei_migrate_pages & EI_MP_THP_ENOMEM) && PageTransHuge(page))
+		return -ENOMEM;
+	if ((ei_migrate_pages & EI_MP_NP_ENOMEM) && !PageTransHuge(page))
+		return -ENOMEM;
+
 	newpage = get_new_page(page, private);
 	if (!newpage)
 		return -ENOMEM;
@@ -1305,7 +1354,7 @@  static inline int try_split_thp(struct page *page, struct list_head *split_pages
 	int rc;
 
 	lock_page(page);
-	rc = split_huge_page_to_list(page, split_pages);
+	rc = ei_split_huge_page_to_list(page, split_pages);
 	unlock_page(page);
 
 	return rc;
@@ -1358,6 +1407,9 @@  int migrate_pages(struct list_head *from, new_page_t get_new_page,
 	bool nosplit = (reason == MR_NUMA_MISPLACED);
 	bool no_subpage_counting = false;
 
+	if (ei_migrate_pages & EI_MP_NOSPLIT)
+		nosplit = true;
+
 	trace_mm_migrate_pages_start(mode, reason);
 
 thp_subpage_migration: