On Thu, 11 Mar 2010, Robin Holt wrote:
>
> ksm.c's write_protect_page implements a lockless means of verifying a
> page does not have any users of the page which are not accounted for via
> other kernel tracking means. It does this by removing the writable pte
> with TLB flushes, checking the page_count against the total known users,
> and then using set_pte_at_notify to make it a read-only entry.
>
> An unneeded mmu_notifier callout is made in the case where the known
> users does not match the page_count. In that event, we are inserting
> the identical pte and there is no need for the set_pte_at_notify, but
> rather the simpler set_pte_at suffices.
>
> To: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Robin Holt <holt@xxxxxxx>
> Acked-by: Izik Eidus <ieidus@xxxxxxxxxx>
> Acked-by: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> Cc: Hugh Dickins <hugh.dickins@xxxxxxxxxxxxx>
Acked-by: Hugh Dickins <hugh.dickins@xxxxxxxxxxxxx>
> Cc: Chris Wright <chrisw@xxxxxxxxxx>
> Cc: linux-mm@xxxxxxxxx
>
> ---
>
> mm/ksm.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> Index: ksm_remove_notify/mm/ksm.c
> ===================================================================
> --- ksm_remove_notify.orig/mm/ksm.c 2010-03-11 11:21:57.000000000 -0600
> +++ ksm_remove_notify/mm/ksm.c 2010-03-11 11:21:59.000000000 -0600
> @@ -751,7 +751,7 @@ static int write_protect_page(struct vm_
> * page
> */
> if (page_mapcount(page) + 1 + swapped != page_count(page)) {
> - set_pte_at_notify(mm, addr, ptep, entry);
> + set_pte_at(mm, addr, ptep, entry);
> goto out_unlock;
> }
> entry = pte_wrprotect(entry);
--
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/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>