Thread (5 messages) 5 messages, 4 authors, 2012-06-11

Re: [WARNING] lockdep and kmemcheck_alloc_shadow

From: David Rientjes <rientjes@google.com>
Date: 2012-05-20 21:10:13
Also in: lkml
Subsystem: memory management, slab allocator, the rest · Maintainers: Andrew Morton, Vlastimil Babka, Harry Yoo, Linus Torvalds

On Sun, 20 May 2012, Ben Hutchings wrote:
Seems to be a bug in SLUB:
quoted
static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
{
[...]
quoted
	if (flags & __GFP_WAIT)
		local_irq_disable();

	if (!page)
		return NULL;

	if (kmemcheck_enabled
		&& !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
		int pages = 1 << oo_order(oo);

		kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
[...]

So kmemcheck_alloc_shadow() runs with IRQs off, but uses the same flags
as the original allocation.

Looks like this was introduced by:

commit 7e0528dadc9f8b04e4de0dba48a075100c2afe75
Author: Christoph Lameter [off-list ref]
Date:   Wed Jun 1 12:25:44 2011 -0500

    slub: Push irq disable into allocate_slab()

which went into 3.1.
Agreed, slab handles this correctly and it looks like slub ends up 
disabling irqs too early.

Does this fix it?  If so, we'll need to annotate it for stable as Ben 
noted.


mm, slub: ensure irqs are not disabled for kmemcheck allocation

kmemcheck_alloc_shadow() cannot be called with irqs disabled, which is
possible with __GFP_WAIT.  Ensure we always have them enabled at
allocation time like slab does.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Suggested-by: Ben Hutchings <redacted>
Signed-off-by: David Rientjes <rientjes@google.com>
---
diff --git a/mm/slub.c b/mm/slub.c
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1314,13 +1314,7 @@ static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
 			stat(s, ORDER_FALLBACK);
 	}
 
-	if (flags & __GFP_WAIT)
-		local_irq_disable();
-
-	if (!page)
-		return NULL;
-
-	if (kmemcheck_enabled
+	if (page && kmemcheck_enabled
 		&& !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
 		int pages = 1 << oo_order(oo);
 
@@ -1336,6 +1330,11 @@ static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
 			kmemcheck_mark_unallocated_pages(page, pages);
 	}
 
+	if (flags & __GFP_WAIT)
+		local_irq_disable();
+	if (!page)
+		return NULL;
+
 	page->objects = oo_objects(oo);
 	mod_zone_page_state(page_zone(page),
 		(s->flags & SLAB_RECLAIM_ACCOUNT) ?
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help