mm: compaction: Partially revert capture of suitable high-order page

Eric Wong reported on 3.7 and 3.8-rc2 that ppoll() got stuck when
waiting for POLLIN on a local TCP socket. It was easier to trigger if
there was disk IO and dirty pages at the same time and he bisected it to
commit 1fb3f8ca0e92 ("mm: compaction: capture a suitable high-order page
immediately when it is made available").

The intention of that patch was to improve high-order allocations under
memory pressure after changes made to reclaim in 3.6 drastically hurt
THP allocations but the approach was flawed. For Eric, the problem was
that page->pfmemalloc was not being cleared for captured pages leading
to a poor interaction with swap-over-NFS support causing the packets to
be dropped. However, I identified a few more problems with the patch
including the fact that it can increase contention on zone->lock in some
cases which could result in async direct compaction being aborted early.

In retrospect the capture patch took the wrong approach. What it should
have done is mark the pageblock being migrated as MIGRATE_ISOLATE if it
was allocating for THP and avoided races that way. While the patch was
showing to improve allocation success rates at the time, the benefit is
marginal given the relative complexity and it should be revisited from
scratch in the context of the other reclaim-related changes that have
taken place since the patch was first written and tested. This patch
partially reverts commit 1fb3f8ca "mm: compaction: capture a suitable
high-order page immediately when it is made available".

Reported-and-tested-by: Eric Wong <[e-mail]>
Tested-by: Eric Dumazet <[e-mail]>
Cc: [e-mail]
Signed-off-by: Mel Gorman <[e-mail]>
Signed-off-by: Linus Torvalds <[e-mail]>

include/linux/compaction.h

···
 extern int fragmentation_index(struct zone *zone, unsigned int order);
 extern unsigned long try_to_compact_pages(struct zonelist *zonelist,
 			int order, gfp_t gfp_mask, nodemask_t *mask,
-			bool sync, bool *contended, struct page **page);
+			bool sync, bool *contended);
 extern int compact_pgdat(pg_data_t *pgdat, int order);
 extern void reset_isolation_suitable(pg_data_t *pgdat);
 extern unsigned long compaction_suitable(struct zone *zone, int order);
···
 #else
 static inline unsigned long try_to_compact_pages(struct zonelist *zonelist,
 			int order, gfp_t gfp_mask, nodemask_t *nodemask,
-			bool sync, bool *contended, struct page **page)
+			bool sync, bool *contended)
 {
 	return COMPACT_CONTINUE;
 }

include/linux/mm.h

···
 
 void split_page(struct page *page, unsigned int order);
 int split_free_page(struct page *page);
-int capture_free_page(struct page *page, int alloc_order, int migratetype);
 
 /*
  * Compound pages have a destructor function.  Provide a

mm/compaction.c

···
 static int compact_finished(struct zone *zone,
 			    struct compact_control *cc)
 {
+	unsigned int order;
 	unsigned long watermark;
 
 	if (fatal_signal_pending(current))
···
 		return COMPACT_CONTINUE;
 
 	/* Direct compactor: Is a suitable page free? */
-	if (cc->page) {
-		/* Was a suitable page captured? */
-		if (*cc->page)
-			return COMPACT_PARTIAL;
-	} else {
-		unsigned int order;
-		for (order = cc->order; order < MAX_ORDER; order++) {
-			struct free_area *area = &zone->free_area[cc->order];
-			/* Job done if page is free of the right migratetype */
-			if (!list_empty(&area->free_list[cc->migratetype]))
-				return COMPACT_PARTIAL;
+	for (order = cc->order; order < MAX_ORDER; order++) {
+		struct free_area *area = &zone->free_area[order];
 
-			/* Job done if allocation would set block type */
-			if (cc->order >= pageblock_order && area->nr_free)
-				return COMPACT_PARTIAL;
-		}
+		/* Job done if page is free of the right migratetype */
+		if (!list_empty(&area->free_list[cc->migratetype]))
+			return COMPACT_PARTIAL;
+
+		/* Job done if allocation would set block type */
+		if (cc->order >= pageblock_order && area->nr_free)
+			return COMPACT_PARTIAL;
 	}
 
 	return COMPACT_CONTINUE;
···
 		return COMPACT_PARTIAL;
 
 	return COMPACT_CONTINUE;
-}
-
-static void compact_capture_page(struct compact_control *cc)
-{
-	unsigned long flags;
-	int mtype, mtype_low, mtype_high;
-
-	if (!cc->page || *cc->page)
-		return;
-
-	/*
-	 * For MIGRATE_MOVABLE allocations we capture a suitable page ASAP
-	 * regardless of the migratetype of the freelist is is captured from.
-	 * This is fine because the order for a high-order MIGRATE_MOVABLE
-	 * allocation is typically at least a pageblock size and overall
-	 * fragmentation is not impaired. Other allocation types must
-	 * capture pages from their own migratelist because otherwise they
-	 * could pollute other pageblocks like MIGRATE_MOVABLE with
-	 * difficult to move pages and making fragmentation worse overall.
-	 */
-	if (cc->migratetype == MIGRATE_MOVABLE) {
-		mtype_low = 0;
-		mtype_high = MIGRATE_PCPTYPES;
-	} else {
-		mtype_low = cc->migratetype;
-		mtype_high = cc->migratetype + 1;
-	}
-
-	/* Speculatively examine the free lists without zone lock */
-	for (mtype = mtype_low; mtype < mtype_high; mtype++) {
-		int order;
-		for (order = cc->order; order < MAX_ORDER; order++) {
-			struct page *page;
-			struct free_area *area;
-			area = &(cc->zone->free_area[order]);
-			if (list_empty(&area->free_list[mtype]))
-				continue;
-
-			/* Take the lock and attempt capture of the page */
-			if (!compact_trylock_irqsave(&cc->zone->lock, &flags, cc))
-				return;
-			if (!list_empty(&area->free_list[mtype])) {
-				page = list_entry(area->free_list[mtype].next,
-							struct page, lru);
-				if (capture_free_page(page, cc->order, mtype)) {
-					spin_unlock_irqrestore(&cc->zone->lock,
-									flags);
-					*cc->page = page;
-					return;
-				}
-			}
-			spin_unlock_irqrestore(&cc->zone->lock, flags);
-		}
-	}
 }
 
 static int compact_zone(struct zone *zone, struct compact_control *cc)
···
 				goto out;
 			}
 		}
-
-		/* Capture a page now if it is a suitable size */
-		compact_capture_page(cc);
 	}
 
 out:
···
 
 static unsigned long compact_zone_order(struct zone *zone,
 				 int order, gfp_t gfp_mask,
-				 bool sync, bool *contended,
-				 struct page **page)
+				 bool sync, bool *contended)
 {
 	unsigned long ret;
 	struct compact_control cc = {
···
 		.migratetype = allocflags_to_migratetype(gfp_mask),
 		.zone = zone,
 		.sync = sync,
-		.page = page,
 	};
 	INIT_LIST_HEAD(&cc.freepages);
 	INIT_LIST_HEAD(&cc.migratepages);
···
  */
 unsigned long try_to_compact_pages(struct zonelist *zonelist,
 			int order, gfp_t gfp_mask, nodemask_t *nodemask,
-			bool sync, bool *contended, struct page **page)
+			bool sync, bool *contended)
 {
 	enum zone_type high_zoneidx = gfp_zone(gfp_mask);
 	int may_enter_fs = gfp_mask & __GFP_FS;
···
 		int status;
 
 		status = compact_zone_order(zone, order, gfp_mask, sync,
-						contended, page);
+						contended);
 		rc = max(status, rc);
 
 		/* If a normal allocation would succeed, stop compacting */
···
 	struct compact_control cc = {
 		.order = order,
 		.sync = false,
-		.page = NULL,
 	};
 
 	return __compact_pgdat(pgdat, &cc);
···
 	struct compact_control cc = {
 		.order = -1,
 		.sync = true,
-		.page = NULL,
 	};
 
 	return __compact_pgdat(NODE_DATA(nid), &cc);

mm/internal.h

···
 	int migratetype;		/* MOVABLE, RECLAIMABLE etc */
 	struct zone *zone;
 	bool contended;			/* True if a lock was contended */
-	struct page **page;		/* Page captured of requested size */
 };
 
 unsigned long

mm/page_alloc.c

···
 		set_page_refcounted(page + i);
 }
 
-/*
- * Similar to the split_page family of functions except that the page
- * required at the given order and being isolated now to prevent races
- * with parallel allocators
- */
-int capture_free_page(struct page *page, int alloc_order, int migratetype)
+static int __isolate_free_page(struct page *page, unsigned int order)
 {
-	unsigned int order;
 	unsigned long watermark;
 	struct zone *zone;
 	int mt;
···
 	BUG_ON(!PageBuddy(page));
 
 	zone = page_zone(page);
-	order = page_order(page);
 	mt = get_pageblock_migratetype(page);
 
 	if (mt != MIGRATE_ISOLATE) {
···
 		if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
 			return 0;
 
-		__mod_zone_freepage_state(zone, -(1UL << alloc_order), mt);
+		__mod_zone_freepage_state(zone, -(1UL << order), mt);
 	}
 
 	/* Remove page from free list */
···
 	zone->free_area[order].nr_free--;
 	rmv_page_order(page);
 
-	if (alloc_order != order)
-		expand(zone, page, alloc_order, order,
-			&zone->free_area[order], migratetype);
-
-	/* Set the pageblock if the captured page is at least a pageblock */
+	/* Set the pageblock if the isolated page is at least a pageblock */
 	if (order >= pageblock_order - 1) {
 		struct page *endpage = page + (1 << order) - 1;
 		for (; page < endpage; page += pageblock_nr_pages) {
···
 		}
 	}
 
-	return 1UL << alloc_order;
+	return 1UL << order;
 }
 
 /*
···
 	unsigned int order;
 	int nr_pages;
 
-	BUG_ON(!PageBuddy(page));
 	order = page_order(page);
 
-	nr_pages = capture_free_page(page, order, 0);
+	nr_pages = __isolate_free_page(page, order);
 	if (!nr_pages)
 		return 0;
 
···
 	bool *contended_compaction, bool *deferred_compaction,
 	unsigned long *did_some_progress)
 {
-	struct page *page = NULL;
-
 	if (!order)
 		return NULL;
 
···
 	current->flags |= PF_MEMALLOC;
 	*did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
 						nodemask, sync_migration,
-						contended_compaction, &page);
+						contended_compaction);
 	current->flags &= ~PF_MEMALLOC;
 
-	/* If compaction captured a page, prep and use it */
-	if (page) {
-		prep_new_page(page, order, gfp_mask);
-		goto got_page;
-	}
-
 	if (*did_some_progress != COMPACT_SKIPPED) {
+		struct page *page;
+
 		/* Page migration frees to the PCP lists but we want merging */
 		drain_pages(get_cpu());
 		put_cpu();
···
 				alloc_flags & ~ALLOC_NO_WATERMARKS,
 				preferred_zone, migratetype);
 		if (page) {
-got_page:
 			preferred_zone->compact_blockskip_flush = false;
 			preferred_zone->compact_considered = 0;
 			preferred_zone->compact_defer_shift = 0;