Re: [PATCH] nommu page refcount bug fixing

[Luke Yang]

On 3/30/06, Nick Piggin <nickpiggin@xxxxxxxxxxxx> wrote:
> Luke Yang wrote:
>
> Yep, sorry this slipped into the kernel. It's my fault for not giving
> Andrew a fix for it.
>
> As you might know, page refcounting in nommu was already broken, so
> I'm working on a proper solution to fix it.
>
> In the meantime though, this is a step backwards and reintroduces
> NOMMU special-casing in page refcounting. As a temporary fix, what I
> think should happen is simply for all slab allocations to ask for
> __GFP_COMP pages.
>
> Could you check that fixes your problem?
  It works.  What's your plan to modify nommu mm? I would like to
help. And I am also interested in implementing the "non-power-of-2"
allocator in 2.6.

  New patch:
Signed-off-by: Luke Yang <luke.adi@xxxxxxxxx>

diff --git a/mm/slab.c b/mm/slab.c
index 4cbf8bb..f93d3d5 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -1454,7 +1454,7 @@ static void *kmem_getpages(struct kmem_c
        int i;

        flags |= cachep->gfpflags;
-       page = alloc_pages_node(nodeid, flags, cachep->gfporder);
+       page = alloc_pages_node(nodeid, (flags | __GFP_COMP), cachep->gfporder);
        if (!page)
                return NULL;
        addr = page_address(page);

--
Best regards,
Luke Yang (Kernel for Blackfin maintainer)
luke.adi@xxxxxxxxx
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
diff --git a/mm/slab.c b/mm/slab.c
index 4cbf8bb..f93d3d5 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -1454,7 +1454,7 @@ static void *kmem_getpages(struct kmem_c
 	int i;
 
 	flags |= cachep->gfpflags;
-	page = alloc_pages_node(nodeid, flags, cachep->gfporder);
+	page = alloc_pages_node(nodeid, (flags | __GFP_COMP), cachep->gfporder);
 	if (!page)
 		return NULL;
 	addr = page_address(page);