Re: [PATCH mmotm] memcg: avoid using buggy kmap at swap_cgroup

[Hugh Dickins <hugh@veritas.com>]

On Wed, 19 Nov 2008, KAMEZAWA Hiroyuki wrote:
> Okay, how about this direction ?
>  1. at first, remove kmap_atomic from page_cgroup.c and use GFP_KERNEL
>     to allocate buffer.

Yes, that's sensible for now.

>  2. later, add kmap_atomic + HighMem buffer support in explicit style.
>     maybe KM_BOUNCE_READ...can be used.....

It's hardly appropriate (there's no bouncing here), and you could only
use it if you disable interrupts.  Oh, you do disable interrupts:
why's that?

> 
> patch for BUGFIX is attached.
> (Sorry, I have to use Web-Mail and can't make it inlined)

swap_cgroup's kmap logic conflicts shmem's kmap logic.
avoid to use HIGHMEM for now and revisit this later.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
---

 mm/page_cgroup.c |    8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

Index: temp/mm/page_cgroup.c
===================================================================
--- temp.orig/mm/page_cgroup.c
+++ temp/mm/page_cgroup.c
@@ -306,7 +306,7 @@ static int swap_cgroup_prepare(int type)
 	ctrl = &swap_cgroup_ctrl[type];
 
 	for (idx = 0; idx < ctrl->length; idx++) {
-		page = alloc_page(GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 		if (!page)
 			goto not_enough_page;
 		ctrl->map[idx] = page;
@@ -347,11 +347,10 @@ struct mem_cgroup *swap_cgroup_record(sw
 
 	mappage = ctrl->map[idx];
 	spin_lock_irqsave(&ctrl->lock, flags);
-	sc = kmap_atomic(mappage, KM_USER0);
+	sc = page_address(mappage);
 	sc += pos;
 	old = sc->val;
 	sc->val = mem;
-	kunmap_atomic((void *)sc, KM_USER0);
 	spin_unlock_irqrestore(&ctrl->lock, flags);
 	return old;
 }
@@ -382,10 +381,9 @@ struct mem_cgroup *lookup_swap_cgroup(sw
 	mappage = ctrl->map[idx];
 
 	spin_lock_irqsave(&ctrl->lock, flags);
-	sc = kmap_atomic(mappage, KM_USER0);
+	sc = page_address(mappage);
 	sc += pos;
 	ret = sc->val;
-	kunmap_atomic((void *)sc, KM_USER0);
 	spin_unlock_irqrestore(&ctrl->lock, flags);
 	return ret;
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>