[SRU][P:linux][PATCH 1/1 v2] mm/page_alloc: fix deadlock on cpu_hotplug_lock in __accept_page()

[Thibault Ferrante]

From: "Kirill A. Shutemov" <kirill.shutemov at linux.intel.com>

BugLink: https://bugs.launchpad.net/bugs/2109543

When the last page in the zone is accepted, __accept_page() calls
static_branch_dec().  This function takes cpu_hotplug_lock, which can lead
to a deadlock if the allocation occurs during CPU bringup path as
_cpu_up() also takes the lock.

To prevent this deadlock, defer static_branch_dec() to a workqueue.

Call static_branch_dec() only when the workqueue is not yet initialized.
Workqueues are initialized before CPU bring up, so this will not conflict
with the first scenario.

Link: https://lkml.kernel.org/r/20250329171030.3942298-1-kirill.shutemov@linux.intel.com
Fixes: 55ad43e8ba0f ("mm: add a helper to accept page")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov at linux.intel.com>
Reported-by: Srikanth Aithal <sraithal at amd.com>
Tested-by: Srikanth Aithal <sraithal at amd.com>
Cc: Dave Hansen <dave.hansen at intel.com>
Cc: Ashish Kalra <ashish.kalra at amd.com>
Cc: David Hildenbrand <david at redhat.com>
Cc: "Edgecombe, Rick P" <rick.p.edgecombe at intel.com>
Cc: Mel Gorman <mgorman at techsingularity.net>
Cc: "Mike Rapoport (IBM)" <rppt at kernel.org>
Cc: Thomas Lendacky <thomas.lendacky at amd.com>
Cc: Vlastimil Babka <vbabka at suse.cz>
Signed-off-by: Andrew Morton <akpm at linux-foundation.org>
(cherry picked from commit 4067196a52278156d18d8d6fa7f43970611b1b49)
Signed-off-by: Thibault Ferrante <thibault.ferrante at canonical.com>
---
 include/linux/mmzone.h |  3 +++
 mm/internal.h          |  1 +
 mm/mm_init.c           |  1 +
 mm/page_alloc.c        | 28 ++++++++++++++++++++++++++--
 4 files changed, 31 insertions(+), 2 deletions(-)

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 9540b41894da..9027f751b619 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -964,6 +964,9 @@ struct zone {
 #ifdef CONFIG_UNACCEPTED_MEMORY
 	/* Pages to be accepted. All pages on the list are MAX_PAGE_ORDER */
 	struct list_head	unaccepted_pages;
+
+	/* To be called once the last page in the zone is accepted */
+	struct work_struct	unaccepted_cleanup;
 #endif
 
 	/* zone flags, see below */
diff --git a/mm/internal.h b/mm/internal.h
index 20b3535935a3..2742c601fe10 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -1517,6 +1517,7 @@ unsigned long move_page_tables(struct vm_area_struct *vma,
 
 #ifdef CONFIG_UNACCEPTED_MEMORY
 void accept_page(struct page *page);
+void unaccepted_cleanup_work(struct work_struct *work);
 #else /* CONFIG_UNACCEPTED_MEMORY */
 static inline void accept_page(struct page *page)
 {
diff --git a/mm/mm_init.c b/mm/mm_init.c
index 2630cc30147e..d5a51f65dc4d 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -1404,6 +1404,7 @@ static void __meminit zone_init_free_lists(struct zone *zone)
 
 #ifdef CONFIG_UNACCEPTED_MEMORY
 	INIT_LIST_HEAD(&zone->unaccepted_pages);
+	INIT_WORK(&zone->unaccepted_cleanup, unaccepted_cleanup_work);
 #endif
 }
 
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 542d25f77be8..50ffabed5dd9 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6921,6 +6921,11 @@ static DEFINE_STATIC_KEY_FALSE(zones_with_unaccepted_pages);
 
 static bool lazy_accept = true;
 
+void unaccepted_cleanup_work(struct work_struct *work)
+{
+	static_branch_dec(&zones_with_unaccepted_pages);
+}
+
 static int __init accept_memory_parse(char *p)
 {
 	if (!strcmp(p, "lazy")) {
@@ -6959,8 +6964,27 @@ static void __accept_page(struct zone *zone, unsigned long *flags,
 
 	__free_pages_ok(page, MAX_PAGE_ORDER, FPI_TO_TAIL);
 
-	if (last)
-		static_branch_dec(&zones_with_unaccepted_pages);
+	if (last) {
+		/*
+		 * There are two corner cases:
+		 *
+		 * - If allocation occurs during the CPU bring up,
+		 *   static_branch_dec() cannot be used directly as
+		 *   it causes a deadlock on cpu_hotplug_lock.
+		 *
+		 *   Instead, use schedule_work() to prevent deadlock.
+		 *
+		 * - If allocation occurs before workqueues are initialized,
+		 *   static_branch_dec() should be called directly.
+		 *
+		 *   Workqueues are initialized before CPU bring up, so this
+		 *   will not conflict with the first scenario.
+		 */
+		if (system_wq)
+			schedule_work(&zone->unaccepted_cleanup);
+		else
+			unaccepted_cleanup_work(&zone->unaccepted_cleanup);
+	}
 }
 
 void accept_page(struct page *page)
-- 
2.48.1