On Mon, Mar 16, 2015 at 11:29:45AM -0700, Laura Abbott wrote:
> Commit 3c605096d315 ("mm/page_alloc: restrict max order of merging on isolated pageblock")
> changed the logic of unset_migratetype_isolate to check the buddy allocator
> and explicitly call __free_pages to merge. The page that is being freed in
> this path never had prep_new_page called so set_page_refcounted is called
> explicitly but there is no call to kernel_map_pages. With the default
> kernel_map_pages this is mostly harmless but if kernel_map_pages does any
> manipulation of the page tables (unmapping or setting pages to read only) this
> may trigger a fault:
>
> alloc_contig_range test_pages_isolated(ceb00, ced00) failed
> Unable to handle kernel paging request at virtual address ffffffc0cec00000
> pgd = ffffffc045fc4000
> [ffffffc0cec00000] *pgd=0000000000000000
> Internal error: Oops: 9600004f [#1] PREEMPT SMP
> Modules linked in: exfatfs
> CPU: 1 PID: 23237 Comm: TimedEventQueue Not tainted 3.10.49-gc72ad36-dirty #1
> task: ffffffc03de52100 ti: ffffffc015388000 task.ti: ffffffc015388000
> PC is at memset+0xc8/0x1c0
> LR is at kernel_map_pages+0x1ec/0x244
>
> Fix this by calling kernel_map_pages to ensure the page is set in the
> page table properly
>
> Fixes: 3c605096d315 ("mm/page_alloc: restrict max order of merging on isolated pageblock")
> Cc: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxx>
> Cc: Rik van Riel <riel@xxxxxxxxxx>
> Cc: Yasuaki Ishimatsu <isimatu.yasuaki@xxxxxxxxxxxxxx>
> Cc: Zhang Yanfei <zhangyanfei@xxxxxxxxxxxxxx>
> Cc: Xishi Qiu <qiuxishi@xxxxxxxxxx>
> Cc: Vladimir Davydov <vdavydov@xxxxxxxxxxxxx>
> Cc: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> Cc: Gioh Kim <gioh.kim@xxxxxxx>
> Cc: Michal Nazarewicz <mina86@xxxxxxxxxx>
> Cc: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Cc: Vlastimil Babka <vbabka@xxxxxxx>
> Signed-off-by: Laura Abbott <lauraa@xxxxxxxxxxxxxx>
> ---
> Note this was found on a backport to 3.10 and the code to make kernel_map_pages
> change the page table state is currently out of tree. The original had stable,
> so this may need to go into stable as well.
I found that some implementation of kernel_map_pages() in mainline also require
this change. Some implementation doesn't check previous state of page table, but
some others check previous state of page table when calling kernel_map_pages().
Acked-by: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
Thanks.
> ---
> mm/page_isolation.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> index 72f5ac3..755a42c 100644
> --- a/mm/page_isolation.c
> +++ b/mm/page_isolation.c
> @@ -103,6 +103,7 @@ void unset_migratetype_isolate(struct page *page, unsigned migratetype)
>
> if (!is_migrate_isolate_page(buddy)) {
> __isolate_free_page(page, order);
> + kernel_map_pages(page, (1 << order), 1);
> set_page_refcounted(page);
> isolated_page = page;
> }
> --
> Qualcomm Innovation Center, Inc.
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
> This e-mail address will be inactive after March 20, 2015
> Please contact privately for follow up after that date.
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxx. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>