Re: Warning on memory offline (and possible in usual migration?)

[Christoph Lameter <clameter@sgi.com>]

I guess we need the following patch to handle !uptodate pages. Wish we had 
a better solution that would allow the skipping of pages with buffers 
under read I/O.


Subject: Page migration: Do not migrate page that is not uptodate

If we are migrating pages that are not mapped into a processes address
space then we may encounter !Uptodate pages. Page migration is now used
for offlining memory which scans unmapped pages.

If a page is not uptodate then read I/O may be in progress against it.
So do not migrate it. On the other hand if the page has buffers then
read I/O will lock a buffer. In that case we can migrate an !Uptodate
page but then migration will stall in buffer_migrate_page() until the
read is complete.

Signed-off-by: Christoph Lameter <clameter@sgi.com>

---
 mm/migrate.c |   17 +++++++++++++++++
 1 file changed, 17 insertions(+)

Index: linux-2.6/mm/migrate.c
===================================================================
--- linux-2.6.orig/mm/migrate.c	2008-05-02 13:47:45.113707645 -0700
+++ linux-2.6/mm/migrate.c	2008-05-02 14:08:32.203644985 -0700
@@ -652,6 +652,23 @@ static int unmap_and_move(new_page_t get
 			goto unlock;
 		wait_on_page_writeback(page);
 	}
+
+	/*
+	 * Page may be under read I/O if its not uptodate and has no buffers.
+	 * In that case the page contents are not stable and should not be
+	 * migrated. So we just pass on that page and return -EAGAIN.
+	 *
+	 * If a page has buffers then the locks taken on the buffers
+	 * will indicate that read I/O is in progress. Then PageUptodate does
+	 * not matter because buffer_migrate_page() will stall until I/O is
+	 * complete. It would be better if we could catch that here and delay
+	 * migrating the page because we could migrate a the other pages on the
+	 * migrate list instead of waiting for I/O to complete on this page
+	 * (like done for writes in progress).
+	 */
+	if (!PageUptodate(page) && !page_has_buffers(page))
+		goto unlock;
+
 	/*
 	 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
 	 * we cannot notice that anon_vma is freed while we migrates a page.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>