Re: [PATCH 30/45] vmscan: lumpy pageout

[Wu Fengguang <fengguang.wu@intel.com>]

On Wed, Oct 07, 2009 at 11:50:58PM +0800, Nikita Danilov wrote:
> 2009/10/7 Wu Fengguang <fengguang.wu@intel.com>:
> 
> [...]
> 
> > +                */
> > +               if (current_is_kswapd() &&
> > +                   bdi_cap_writeback_dirty(mapping->backing_dev_info) &&
> > +                   !mapping->a_ops->writepages) {
> > +                       wbc.range_start = (page->index + 1) << PAGE_CACHE_SHIFT;
> > +                       wbc.nr_to_write = LUMPY_PAGEOUT_PAGES - 1;
> > +                       generic_writepages(mapping, &wbc);
> > +                       iput(inode);
> > +               }
> > +
> > +               /*
> 
> One potential problem with this is that generic_writepages() waits on
> page locks and this can stall kswapd (always bad). This can be worked
> around by replacing lock_page() with trylock_page() conditionally on
> wbc->for_reclaim (or wbc->nonblocking?), but then, this almost look
> like a separate function would be better.

IMHO trylock_page() is not necessary. Locked pages are rare in normal
states. kswapd already do lock_page() for all pages it try to examine
state for reclaim. So it makes sense for lumpy pageout to follow the
(simple) convention.

> On a good side, it seems I was wrong and pageout calls iput() already:
> shrink_slab()->prune_icache()->iput().

Not totally wrong ;) iput() will be called if __GFP_FS is on. However
pageout may be called on either __GFP_FS or (__GFP_IO && PageSwapCache).

So I updated the patch to do lumpy pageout for __GFP_FS. In long term,
it would be good to remove AOP_WRITEPAGE_ACTIVATE and ->writepage()
totally, and to support shmem as well :)

Thanks,
Fengguang
---

vmscan: lumpy pageout

When pageout a dirty page, try to piggy back more consecutive dirty
pages (up to 512KB) to improve IO efficiency.

Only ext3/reiserfs which don't have its own aops->writepages are
supported in this initial version.

CC: Hugh Dickins <hugh.dickins@tiscali.co.uk>
CC: Dave Chinner <david@fromorbit.com>
CC: Nikita Danilov <danilov@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
---
 mm/page-writeback.c |   12 ++++++++++++
 mm/vmscan.c         |   23 ++++++++++++++++++++++-
 2 files changed, 34 insertions(+), 1 deletion(-)

--- linux.orig/mm/vmscan.c	2009-10-08 07:35:06.000000000 +0800
+++ linux/mm/vmscan.c	2009-10-08 10:20:51.000000000 +0800
@@ -344,11 +344,14 @@ typedef enum {
 	PAGE_CLEAN,
 } pageout_t;
 
+#define LUMPY_PAGEOUT_PAGES	(512 * 1024 / PAGE_CACHE_SIZE)
+
 /*
  * pageout is called by shrink_page_list() for each dirty page.
  * Calls ->writepage().
  */
 static pageout_t pageout(struct page *page, struct address_space *mapping,
+						struct scan_control *sc,
 						enum pageout_io sync_writeback)
 {
 	/*
@@ -398,6 +401,10 @@ static pageout_t pageout(struct page *pa
 			.nonblocking = 1,
 			.for_reclaim = 1,
 		};
+		struct inode *inode = NULL;
+
+		if (sc->gfp_mask & __GFP_FS)
+			inode = igrab(mapping->host);
 
 		SetPageReclaim(page);
 		res = mapping->a_ops->writepage(page, &wbc);
@@ -405,10 +412,24 @@ static pageout_t pageout(struct page *pa
 			handle_write_error(mapping, page, res);
 		if (res == AOP_WRITEPAGE_ACTIVATE) {
 			ClearPageReclaim(page);
+			iput(inode);
 			return PAGE_ACTIVATE;
 		}
 
 		/*
+		 * only write_cache_pages() supports for_reclaim for now
+		 * ignore shmem for now, thanks to Nikita.
+		 */
+		if (current_is_kswapd() &&
+		    bdi_cap_writeback_dirty(mapping->backing_dev_info) &&
+		    !mapping->a_ops->writepages) {
+			wbc.range_start = (page->index + 1) << PAGE_CACHE_SHIFT;
+			wbc.nr_to_write = LUMPY_PAGEOUT_PAGES - 1;
+			generic_writepages(mapping, &wbc);
+			iput(inode);
+		}
+
+		/*
 		 * Wait on writeback if requested to. This happens when
 		 * direct reclaiming a large contiguous area and the
 		 * first attempt to free a range of pages fails.
@@ -684,7 +705,7 @@ static unsigned long shrink_page_list(st
 				goto keep_locked;
 
 			/* Page is dirty, try to write it out here */
-			switch (pageout(page, mapping, sync_writeback)) {
+			switch (pageout(page, mapping, sc, sync_writeback)) {
 			case PAGE_KEEP:
 				goto keep_locked;
 			case PAGE_ACTIVATE:
--- linux.orig/mm/page-writeback.c	2009-10-08 07:38:57.000000000 +0800
+++ linux/mm/page-writeback.c	2009-10-08 10:06:33.000000000 +0800
@@ -805,6 +805,11 @@ int write_cache_pages(struct address_spa
 				break;
 			}
 
+			if (wbc->for_reclaim && done_index != page->index) {
+				done = 1;
+				break;
+			}
+
 			if (nr_to_write != wbc->nr_to_write &&
 			    done_index + WB_SEGMENT_DIST < page->index &&
 			    --wbc->nr_segments <= 0) {
@@ -846,6 +851,13 @@ continue_unlock:
 			if (!clear_page_dirty_for_io(page))
 				goto continue_unlock;
 
+			/*
+			 * active and unevictable pages will be checked at
+			 * rotate time
+			 */
+			if (wbc->for_reclaim)
+				SetPageReclaim(page);
+
 			ret = (*writepage)(page, wbc, data);
 			if (unlikely(ret)) {
 				if (ret == AOP_WRITEPAGE_ACTIVATE) {
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html