On Mon, 6 Jun 2011 14:45:19 -0700
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> Hopefully he can test this one for us as well, thanks.
>
A patch with better description (of mine) is here.
Anyway, I felt I needed a fix for ARM special case.
==
fix-init-page_cgroup-for-sparsemem-taking-care-of-broken-page-flags.patch
Even with SPARSEMEM, there are some magical memmap.
If a Node is not aligned to SECTION, memmap of pfn which is out of
Node's range is not initialized. And page->flags contains 0.
If Node(0) doesn't exist, NODE_DATA(pfn_to_nid(pfn)) causes error.
In another case, for example, ARM frees memmap which is never be used
even under SPARSEMEM. In that case, page->flags will contain broken
value.
This patch does a strict check on nid which is obtained by
pfn_to_page() and use proper NID for page_cgroup allocation.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
---
mm/page_cgroup.c | 36 +++++++++++++++++++++++++++++++++++-
1 file changed, 35 insertions(+), 1 deletion(-)
Index: linux-3.0-rc1/mm/page_cgroup.c
===================================================================
--- linux-3.0-rc1.orig/mm/page_cgroup.c
+++ linux-3.0-rc1/mm/page_cgroup.c
@@ -168,6 +168,7 @@ static int __meminit init_section_page_c
struct mem_section *section;
unsigned long table_size;
unsigned long nr;
+ unsigned long tmp;
int nid, index;
nr = pfn_to_section_nr(pfn);
@@ -175,8 +176,41 @@ static int __meminit init_section_page_c
if (section->page_cgroup)
return 0;
+ /*
+ * check Node-ID. Because we get 'pfn' which is obtained by calculation,
+ * the pfn may "not exist" or "alreay freed". Even if pfn_valid() returns
+ * true, page->flags may contain broken value and pfn_to_nid() returns
+ * bad value.
+ * (See CONFIG_ARCH_HAS_HOLES_MEMORYMODEL and ARM's free_memmap())
+ * So, we need to do careful check, here.
+ */
+ for (tmp = pfn;
+ tmp < pfn + PAGES_PER_SECTION;
+ tmp += MAX_ORDER_NR_PAGES, nid = -1) {
+ struct page *page;
+
+ if (!pfn_valid(tmp))
+ continue;
+
+ page = pfn_to_page(tmp);
+ nid = page_to_nid(page);
- nid = page_to_nid(pfn_to_page(pfn));
+ /*
+ * If 'page' isn't initialized or freed, it may contains broken
+ * information.
+ */
+ if (!node_state(nid, N_NORMAL_MEMORY))
+ continue;
+ if (page_to_pfn(pfn_to_page(tmp)) != tmp)
+ continue;
+ /*
+ * The page seems valid and this 'nid' is safe to access,
+ * at least.
+ */
+ break;
+ }
+ if (nid == -1)
+ return 0;
table_size = sizeof(struct page_cgroup) * PAGES_PER_SECTION;
base = alloc_page_cgroup(table_size, nid);
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>