73  * the "fragmentation score" of a node/zone.
159 static void defer_compaction(struct zone *zone, int order)  in defer_compaction()  argument
161 	zone->compact_considered = 0;  in defer_compaction()
162 	zone->compact_defer_shift++;  in defer_compaction()
164 	if (order < zone->compact_order_failed)  in defer_compaction()
165 		zone->compact_order_failed = order;  in defer_compaction()
167 	if (zone->compact_defer_shift > COMPACT_MAX_DEFER_SHIFT)  in defer_compaction()
168 		zone->compact_defer_shift = COMPACT_MAX_DEFER_SHIFT;  in defer_compaction()
170 	trace_mm_compaction_defer_compaction(zone, order);  in defer_compaction()
174 static bool compaction_deferred(struct zone *zone, int order)  in compaction_deferred()  argument
176 	unsigned long defer_limit = 1UL << zone->compact_defer_shift;  in compaction_deferred()
178 	if (order < zone->compact_order_failed)  in compaction_deferred()
182 	if (++zone->compact_considered >= defer_limit) {  in compaction_deferred()
183 		zone->compact_considered = defer_limit;  in compaction_deferred()
187 	trace_mm_compaction_deferred(zone, order);  in compaction_deferred()
197 void compaction_defer_reset(struct zone *zone, int order,  in compaction_defer_reset()  argument
201 		zone->compact_considered = 0;  in compaction_defer_reset()
202 		zone->compact_defer_shift = 0;  in compaction_defer_reset()
204 	if (order >= zone->compact_order_failed)  in compaction_defer_reset()
205 		zone->compact_order_failed = order + 1;  in compaction_defer_reset()
207 	trace_mm_compaction_defer_reset(zone, order);  in compaction_defer_reset()
211 static bool compaction_restarting(struct zone *zone, int order)  in compaction_restarting()  argument
213 	if (order < zone->compact_order_failed)  in compaction_restarting()
216 	return zone->compact_defer_shift == COMPACT_MAX_DEFER_SHIFT &&  in compaction_restarting()
217 		zone->compact_considered >= 1UL << zone->compact_defer_shift;  in compaction_restarting()
230 static void reset_cached_positions(struct zone *zone)  in reset_cached_positions()  argument
232 	zone->compact_cached_migrate_pfn[0] = zone->zone_start_pfn;  in reset_cached_positions()
233 	zone->compact_cached_migrate_pfn[1] = zone->zone_start_pfn;  in reset_cached_positions()
234 	zone->compact_cached_free_pfn =  in reset_cached_positions()
235 				pageblock_start_pfn(zone_end_pfn(zone) - 1);  in reset_cached_positions()
309 __reset_isolation_pfn(struct zone *zone, unsigned long pfn, bool check_source,  in __reset_isolation_pfn()  argument
319 	if (zone != page_zone(page))  in __reset_isolation_pfn()
339 	/* Ensure the start of the pageblock or zone is online and valid */  in __reset_isolation_pfn()
341 	block_pfn = max(block_pfn, zone->zone_start_pfn);  in __reset_isolation_pfn()
348 	/* Ensure the end of the pageblock or zone is online and valid */  in __reset_isolation_pfn()
350 	block_pfn = min(block_pfn, zone_end_pfn(zone) - 1);  in __reset_isolation_pfn()
382 static void __reset_isolation_suitable(struct zone *zone)  in __reset_isolation_suitable()  argument
384 	unsigned long migrate_pfn = zone->zone_start_pfn;  in __reset_isolation_suitable()
385 	unsigned long free_pfn = zone_end_pfn(zone) - 1;  in __reset_isolation_suitable()
392 	if (!zone->compact_blockskip_flush)  in __reset_isolation_suitable()
395 	zone->compact_blockskip_flush = false;  in __reset_isolation_suitable()
398 	 * Walk the zone and update pageblock skip information. Source looks  in __reset_isolation_suitable()
408 		if (__reset_isolation_pfn(zone, migrate_pfn, true, source_set) &&  in __reset_isolation_suitable()
412 			zone->compact_init_migrate_pfn = reset_migrate;  in __reset_isolation_suitable()
413 			zone->compact_cached_migrate_pfn[0] = reset_migrate;  in __reset_isolation_suitable()
414 			zone->compact_cached_migrate_pfn[1] = reset_migrate;  in __reset_isolation_suitable()
418 		if (__reset_isolation_pfn(zone, free_pfn, free_set, true) &&  in __reset_isolation_suitable()
422 			zone->compact_init_free_pfn = reset_free;  in __reset_isolation_suitable()
423 			zone->compact_cached_free_pfn = reset_free;  in __reset_isolation_suitable()
429 		zone->compact_cached_migrate_pfn[0] = migrate_pfn;  in __reset_isolation_suitable()
430 		zone->compact_cached_migrate_pfn[1] = migrate_pfn;  in __reset_isolation_suitable()
431 		zone->compact_cached_free_pfn = free_pfn;  in __reset_isolation_suitable()
440 		struct zone *zone = &pgdat->node_zones[zoneid];  in reset_isolation_suitable()  local
441 		if (!populated_zone(zone))  in reset_isolation_suitable()
444 		__reset_isolation_suitable(zone);  in reset_isolation_suitable()
469 	struct zone *zone = cc->zone;  in update_cached_migrate()  local
478 	if (pfn > zone->compact_cached_migrate_pfn[0])  in update_cached_migrate()
479 		zone->compact_cached_migrate_pfn[0] = pfn;  in update_cached_migrate()
481 	    pfn > zone->compact_cached_migrate_pfn[1])  in update_cached_migrate()
482 		zone->compact_cached_migrate_pfn[1] = pfn;  in update_cached_migrate()
492 	struct zone *zone = cc->zone;  in update_pageblock_skip()  local
499 	if (pfn < zone->compact_cached_free_pfn)  in update_pageblock_skip()
500 		zone->compact_cached_free_pfn = pfn;  in update_pageblock_skip()
619 		    && compact_unlock_should_abort(&cc->zone->lock, flags,  in isolate_freepages_block()
649 			locked = compact_lock_irqsave(&cc->zone->lock,  in isolate_freepages_block()
685 		spin_unlock_irqrestore(&cc->zone->lock, flags);  in isolate_freepages_block()
719  * Non-free pages, invalid PFNs, or zone boundaries within the
739 	if (block_start_pfn < cc->zone->zone_start_pfn)  in isolate_freepages_range()
740 		block_start_pfn = cc->zone->zone_start_pfn;  in isolate_freepages_range()
762 					block_end_pfn, cc->zone))  in isolate_freepages_range()
796 	pg_data_t *pgdat = cc->zone->zone_pgdat;  in too_many_isolated()
873 	pg_data_t *pgdat = cc->zone->zone_pgdat;  in isolate_migratepages_block()
974 				    low_pfn == cc->zone->zone_start_pfn)) {  in isolate_migratepages_block()
1038 		 * Skip if free. We read page order here without zone lock  in isolate_migratepages_block()
1365 	if (block_start_pfn < cc->zone->zone_start_pfn)  in isolate_migratepages_range()
1366 		block_start_pfn = cc->zone->zone_start_pfn;  in isolate_migratepages_range()
1376 					block_end_pfn, cc->zone))  in isolate_migratepages_range()
1423 		 * We are checking page_order without zone->lock taken. But  in suitable_migration_target()
1508 	start_pfn = max(pageblock_start_pfn(pfn), cc->zone->zone_start_pfn);  in fast_isolate_around()
1509 	end_pfn = min(pageblock_end_pfn(pfn), zone_end_pfn(cc->zone));  in fast_isolate_around()
1511 	page = pageblock_pfn_to_page(start_pfn, end_pfn, cc->zone);  in fast_isolate_around()
1559 	if (cc->free_pfn >= cc->zone->compact_init_free_pfn) {  in fast_isolate_freepages()
1584 		struct free_area *area = &cc->zone->free_area[order];  in fast_isolate_freepages()
1594 		spin_lock_irqsave(&cc->zone->lock, flags);  in fast_isolate_freepages()
1605 					      cc->zone->zone_start_pfn);  in fast_isolate_freepages()
1653 		spin_unlock_irqrestore(&cc->zone->lock, flags);  in fast_isolate_freepages()
1685 						    zone_end_pfn(cc->zone)),  in fast_isolate_freepages()
1686 						cc->zone);  in fast_isolate_freepages()
1696 	if (highest && highest >= cc->zone->compact_cached_free_pfn) {  in fast_isolate_freepages()
1698 		cc->zone->compact_cached_free_pfn = highest;  in fast_isolate_freepages()
1715 	struct zone *zone = cc->zone;  in isolate_freepages()  local
1730 	 * successfully isolated from, zone-cached value, or the end of the  in isolate_freepages()
1731 	 * zone when isolating for the first time. For looping we also need  in isolate_freepages()
1735 	 * zone which ends in the middle of a pageblock.  in isolate_freepages()
1742 						zone_end_pfn(zone));  in isolate_freepages()
1758 		 * This can iterate a massively long zone without finding any  in isolate_freepages()
1765 									zone);  in isolate_freepages()
1888  * freelist.  All pages on the freelist are from the same zone, so there is no
1981 	 * If the migrate_pfn is not at the start of a zone or the start  in fast_find_migrateblock()
1985 	if (pfn != cc->zone->zone_start_pfn && pfn != pageblock_start_pfn(pfn))  in fast_find_migrateblock()
2012 	if (cc->migrate_pfn != cc->zone->zone_start_pfn)  in fast_find_migrateblock()
2019 		struct free_area *area = &cc->zone->free_area[order];  in fast_find_migrateblock()
2027 		spin_lock_irqsave(&cc->zone->lock, flags);  in fast_find_migrateblock()
2053 				if (pfn < cc->zone->zone_start_pfn)  in fast_find_migrateblock()
2054 					pfn = cc->zone->zone_start_pfn;  in fast_find_migrateblock()
2060 		spin_unlock_irqrestore(&cc->zone->lock, flags);  in fast_find_migrateblock()
2093 	 * Start at where we last stopped, or beginning of the zone as  in isolate_migratepages()
2099 	if (block_start_pfn < cc->zone->zone_start_pfn)  in isolate_migratepages()
2100 		block_start_pfn = cc->zone->zone_start_pfn;  in isolate_migratepages()
2123 		 * This can potentially iterate a massively long zone with  in isolate_migratepages()
2131 						block_end_pfn, cc->zone);  in isolate_migratepages()
2149 		     low_pfn == cc->zone->zone_start_pfn) &&  in isolate_migratepages()
2200  * A zone's fragmentation score is the external fragmentation wrt to the
2203 static unsigned int fragmentation_score_zone(struct zone *zone)  in fragmentation_score_zone()  argument
2205 	return extfrag_for_order(zone, COMPACTION_HPAGE_ORDER);  in fragmentation_score_zone()
2209  * A weighted zone's fragmentation score is the external fragmentation
2210  * wrt to the COMPACTION_HPAGE_ORDER scaled by the zone's size. It
2218 static unsigned int fragmentation_score_zone_weighted(struct zone *zone)  in fragmentation_score_zone_weighted()  argument
2222 	score = zone->present_pages * fragmentation_score_zone(zone);  in fragmentation_score_zone_weighted()
2223 	return div64_ul(score, zone->zone_pgdat->node_present_pages + 1);  in fragmentation_score_zone_weighted()
2239 		struct zone *zone;  in fragmentation_score_node()  local
2241 		zone = &pgdat->node_zones[zoneid];  in fragmentation_score_node()
2242 		if (!populated_zone(zone))  in fragmentation_score_node()
2244 		score += fragmentation_score_zone_weighted(zone);  in fragmentation_score_node()
2279 		reset_cached_positions(cc->zone);  in __compact_finished()
2288 			cc->zone->compact_blockskip_flush = true;  in __compact_finished()
2300 		pgdat = cc->zone->zone_pgdat;  in __compact_finished()
2304 		score = fragmentation_score_zone(cc->zone);  in __compact_finished()
2333 		if (__zone_watermark_ok(cc->zone, cc->order,  in __compact_finished()
2334 					high_wmark_pages(cc->zone),  in __compact_finished()
2336 					zone_page_state(cc->zone,  in __compact_finished()
2346 		struct free_area *area = &cc->zone->free_area[order];  in __compact_finished()
2386 	trace_mm_compaction_finished(cc->zone, cc->order, ret);  in compact_finished()
2393 static bool __compaction_suitable(struct zone *zone, int order,  in __compaction_suitable()  argument
2415 		watermark += low_wmark_pages(zone) - min_wmark_pages(zone);  in __compaction_suitable()
2416 	return __zone_watermark_ok(zone, 0, watermark, highest_zoneidx,  in __compaction_suitable()
2421  * compaction_suitable: Is this suitable to run compaction on this zone now?
2423 bool compaction_suitable(struct zone *zone, int order, unsigned long watermark,  in compaction_suitable()  argument
2429 	suitable = __compaction_suitable(zone, order, watermark, highest_zoneidx,  in compaction_suitable()
2430 					 zone_page_state(zone, NR_FREE_PAGES));  in compaction_suitable()
2450 			int fragindex = fragmentation_index(zone, order);  in compaction_suitable()
2462 	trace_mm_compaction_suitable(zone, order, compact_result);  in compaction_suitable()
2471 	struct zone *zone;  in compaction_zonelist_suitable()  local
2475 	 * Make sure at least one zone would pass __compaction_suitable if we continue  in compaction_zonelist_suitable()
2478 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,  in compaction_zonelist_suitable()
2488 		available = zone_reclaimable_pages(zone) / order;  in compaction_zonelist_suitable()
2489 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);  in compaction_zonelist_suitable()
2490 		if (__compaction_suitable(zone, order, min_wmark_pages(zone),  in compaction_zonelist_suitable()
2506 compaction_suit_allocation_order(struct zone *zone, unsigned int order,  in compaction_suit_allocation_order()  argument
2514 		free_pages = zone_page_state(zone, NR_FREE_PAGES_BLOCKS);  in compaction_suit_allocation_order()
2516 		free_pages = zone_page_state(zone, NR_FREE_PAGES);  in compaction_suit_allocation_order()
2518 	watermark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);  in compaction_suit_allocation_order()
2519 	if (__zone_watermark_ok(zone, order, watermark, highest_zoneidx,  in compaction_suit_allocation_order()
2534 		if (!__zone_watermark_ok(zone, 0, watermark + compact_gap(order),  in compaction_suit_allocation_order()
2536 					 zone_page_state(zone, NR_FREE_PAGES)))  in compaction_suit_allocation_order()
2540 	if (!compaction_suitable(zone, order, watermark, highest_zoneidx))  in compaction_suit_allocation_order()
2550 	unsigned long start_pfn = cc->zone->zone_start_pfn;  in compact_zone()
2551 	unsigned long end_pfn = zone_end_pfn(cc->zone);  in compact_zone()
2559 	 * These counters track activities during zone compaction.  Initialize  in compact_zone()
2560 	 * them before compacting a new zone.  in compact_zone()
2573 		ret = compaction_suit_allocation_order(cc->zone, cc->order,  in compact_zone()
2586 	if (compaction_restarting(cc->zone, cc->order))  in compact_zone()
2587 		__reset_isolation_suitable(cc->zone);  in compact_zone()
2590 	 * Setup to move all movable pages to the end of the zone. Used cached  in compact_zone()
2592 	 * want to compact the whole zone), but check that it is initialised  in compact_zone()
2593 	 * by ensuring the values are within zone boundaries.  in compact_zone()
2600 		cc->migrate_pfn = cc->zone->compact_cached_migrate_pfn[sync];  in compact_zone()
2601 		cc->free_pfn = cc->zone->compact_cached_free_pfn;  in compact_zone()
2604 			cc->zone->compact_cached_free_pfn = cc->free_pfn;  in compact_zone()
2608 			cc->zone->compact_cached_migrate_pfn[0] = cc->migrate_pfn;  in compact_zone()
2609 			cc->zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn;  in compact_zone()
2612 		if (cc->migrate_pfn <= cc->zone->compact_init_migrate_pfn)  in compact_zone()
2627 		cc->zone->compact_cached_migrate_pfn[0] == cc->zone->compact_cached_migrate_pfn[1];  in compact_zone()
2661 				cc->zone->compact_cached_migrate_pfn[1] =  in compact_zone()
2662 					cc->zone->compact_cached_migrate_pfn[0];  in compact_zone()
2673 			last_migrated_pfn = max(cc->zone->zone_start_pfn,  in compact_zone()
2748 				lru_add_drain_cpu_zone(cc->zone);  in compact_zone()
2769 		 * already reset to zone end in compact_finished()  in compact_zone()
2771 		if (free_pfn > cc->zone->compact_cached_free_pfn)  in compact_zone()
2772 			cc->zone->compact_cached_free_pfn = free_pfn;  in compact_zone()
2785 static enum compact_result compact_zone_order(struct zone *zone, int order,  in compact_zone_order()  argument
2795 		.zone = zone,  in compact_zone_order()
2855 	struct zone *zone;  in try_to_compact_pages()  local
2863 	/* Compact each zone in the list */  in try_to_compact_pages()
2864 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,  in try_to_compact_pages()
2870 			!__cpuset_zone_allowed(zone, gfp_mask))  in try_to_compact_pages()
2874 					&& compaction_deferred(zone, order)) {  in try_to_compact_pages()
2879 		status = compact_zone_order(zone, order, gfp_mask, prio,  in try_to_compact_pages()
2886 			 * We think the allocation will succeed in this zone,  in try_to_compact_pages()
2889 			 * succeeds in this zone.  in try_to_compact_pages()
2891 			compaction_defer_reset(zone, order, false);  in try_to_compact_pages()
2899 			 * We think that allocation won't succeed in this zone  in try_to_compact_pages()
2903 			defer_compaction(zone, order);  in try_to_compact_pages()
2923  * For proactive compaction, compact till each zone's fragmentation score
2927  * contention on per-node or per-zone locks.
2932 	struct zone *zone;  in compact_node()  local
2943 		zone = &pgdat->node_zones[zoneid];  in compact_node()
2944 		if (!populated_zone(zone))  in compact_node()
2950 		cc.zone = zone;  in compact_node()
3068 	struct zone *zone;  in kcompactd_node_suitable()  local
3075 		zone = &pgdat->node_zones[zoneid];  in kcompactd_node_suitable()
3077 		if (!populated_zone(zone))  in kcompactd_node_suitable()
3080 		ret = compaction_suit_allocation_order(zone,  in kcompactd_node_suitable()
3098 	struct zone *zone;  in kcompactd_do_work()  local
3117 		zone = &pgdat->node_zones[zoneid];  in kcompactd_do_work()
3118 		if (!populated_zone(zone))  in kcompactd_do_work()
3121 		if (compaction_deferred(zone, cc.order))  in kcompactd_do_work()
3124 		ret = compaction_suit_allocation_order(zone,  in kcompactd_do_work()
3133 		cc.zone = zone;  in kcompactd_do_work()
3137 			compaction_defer_reset(zone, cc.order, false);  in kcompactd_do_work()
3141 			 * otherwise coalesce on the zone's free area for  in kcompactd_do_work()
3145 			drain_all_pages(zone);  in kcompactd_do_work()
3151 			defer_compaction(zone, cc.order);  in kcompactd_do_work()
3233 			 * as the condition of the zone changing substantionally  in kcompactd()