Migration Cache 4/5 - use migration cache for lazy migration
Separate migration cache patch for migrate_pages_unmap_only().
That function is only present when the MPOL_MF_LAZY feature of
the "Migrate-on-Fault" patch series is present.
Call __add_to_swap() with the 'migration' argument set to 1
to select the migration cache--if it's configured. Add removal
from mig cache to error path if try_to_unmap() fails. This
call is a no-op if any of the mappings to the page were successfully
unmapped because the migration cache entry will have a non-zero
ref count. Calling migration_remove_reference() with a decrement
count of '0' will only remove the entry if it already has zero
refs--i.e., no ptes reference it.
Note that here is a case where the migration cache is visible
outside of the swap infrastructure.
Added a check for migration cache() entry to migration_remove_reference()
to handle the case where a page is already in the real swap cache, but
try_to_unmap() fails when called from migrate_pages_unmap_only(). We
don't want to pass a real swap entry further down into the mig cache.
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@xxxxxx>
mm/migrate.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
Index: linux-2.6.36-mmotm-101103-1217/mm/migrate.c
===================================================================
--- linux-2.6.36-mmotm-101103-1217.orig/mm/migrate.c
+++ linux-2.6.36-mmotm-101103-1217/mm/migrate.c
@@ -952,7 +952,17 @@ int migrate_pages_unmap_only(struct list
goto unlock_page;
if (PageAnon(page)) {
- if (!PageSwapCache(page) && !add_to_swap(page)) {
+ if (PageSwapCache(page)) {
+ /*
+ * add an extra ref while we hold the page
+ * lock so that do_swap_page() can't rip the
+ * entry out from under us.
+ */
+ if (migration_add_reference_page(page)) {
+ nr_failed++;
+ goto unlock_page;
+ }
+ } else if (!__add_to_swap(page, 1)) {
nr_failed++;
goto unlock_page;
}
@@ -965,6 +975,11 @@ int migrate_pages_unmap_only(struct list
ret = try_to_unmap(page,
TTU_MIGRATE_DEFERRED|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
+ /*
+ * Remove extra ref. This will free the migration cache
+ * entry if try_to_unmap() failed to replace any ptes.
+ */
+ migration_remove_reference_page(page);
if (ret != SWAP_SUCCESS || page_mapped(page)) {
nr_failed++;
goto unlock_page;
--
To unsubscribe from this list: send the line "unsubscribe linux-numa" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
