mm/page_alloc: fix deadlock on cpu_hotplug_lock in __accept_page()

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/[email protected]
Fixes: 55ad43e ("mm: add a helper to accept page")
Signed-off-by: Kirill A. Shutemov <[email protected]>
Reported-by: Srikanth Aithal <[email protected]>
Tested-by: Srikanth Aithal <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Ashish Kalra <[email protected]>
Cc: David Hildenbrand <[email protected]>
Cc: "Edgecombe, Rick P" <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: "Mike Rapoport (IBM)" <[email protected]>
Cc: Thomas Lendacky <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>

Author: kiryl

include/linux/mmzone.h

@@ -967,6 +967,9 @@ 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 */

mm/internal.h

@@ -1595,6 +1595,7 @@ unsigned long move_page_tables(struct pagetable_move_control *pmc);
 
 #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)
 {

mm/mm_init.c

@@ -1441,6 +1441,7 @@ 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
 }
 

mm/page_alloc.c

@@ -7191,6 +7191,11 @@ 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")) {
@@ -7229,8 +7234,27 @@ 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)