Re: Kernel falls apart under light memory pressure (i.e. linkingvmlinux)

[KAMEZAWA Hiroyuki]

On Fri, 20 May 2011 13:20:15 +0900
Minchan Kim <minchan.kim@xxxxxxxxx> wrote:

> So I want to resolve your problem asap.
> We don't have see report about that. Could you do git-bisect?
> FYI, Recently, big change of mm is compaction,transparent huge pages.
> Kame, could you point out thing related to memcg if you have a mind?
> 

I don't doubt memcg at this stage because it never modify page->flags.
Consdering the case, PageActive() is set against off-LRU pages after
clear_active_flags() clears it.

Hmm, I think I don't understand the lock system fully but...how do you
think this ?

==

At splitting a hugepage, the routine marks all pmd as "splitting".

But assume a racy case where 2 threads run into spit at the
same time, one thread wins compound_lock() and do split, another
thread should not touch splitted pages.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
Index: mmotm-May11/mm/huge_memory.c
===================================================================
--- mmotm-May11.orig/mm/huge_memory.c
+++ mmotm-May11/mm/huge_memory.c
@@ -1150,7 +1150,7 @@ static int __split_huge_page_splitting(s
 	return ret;
 }
 
-static void __split_huge_page_refcount(struct page *page)
+static bool __split_huge_page_refcount(struct page *page)
 {
 	int i;
 	unsigned long head_index = page->index;
@@ -1161,6 +1161,11 @@ static void __split_huge_page_refcount(s
 	spin_lock_irq(&zone->lru_lock);
 	compound_lock(page);
 
+	if (!PageCompound(page)) {
+		compound_unlock(page);
+		spin_unlock_irq(&zone->lru_lock);
+		return false;
+	}
 	for (i = 1; i < HPAGE_PMD_NR; i++) {
 		struct page *page_tail = page + i;
 
@@ -1258,6 +1263,7 @@ static void __split_huge_page_refcount(s
 	 * to be pinned by the caller.
 	 */
 	BUG_ON(page_count(page) <= 0);
+	return true;
 }
 
 static int __split_huge_page_map(struct page *page,
@@ -1367,7 +1373,8 @@ static void __split_huge_page(struct pag
 		       mapcount, page_mapcount(page));
 	BUG_ON(mapcount != page_mapcount(page));
 
-	__split_huge_page_refcount(page);
+	if (!__split_huge_page_refcount(page))
+		return;
 
 	mapcount2 = 0;
 	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/