aboutsummaryrefslogtreecommitdiffstats
path: root/mm
diff options
context:
space:
mode:
Diffstat (limited to 'mm')
-rw-r--r--mm/Kconfig1
-rw-r--r--mm/Kconfig.debug1
-rw-r--r--mm/backing-dev.c186
-rw-r--r--mm/cma.c31
-rw-r--r--mm/cma.h1
-rw-r--r--mm/cma_debug.c2
-rw-r--r--mm/compaction.c89
-rw-r--r--mm/filemap.c66
-rw-r--r--mm/frame_vector.c5
-rw-r--r--mm/gup.c150
-rw-r--r--mm/huge_memory.c40
-rw-r--r--mm/hwpoison-inject.c3
-rw-r--r--mm/internal.h29
-rw-r--r--mm/kasan/kasan.c5
-rw-r--r--mm/kasan/kasan.h2
-rw-r--r--mm/kasan/report.c187
-rw-r--r--mm/khugepaged.c12
-rw-r--r--mm/ksm.c16
-rw-r--r--mm/madvise.c56
-rw-r--r--mm/memblock.c56
-rw-r--r--mm/memcontrol.c248
-rw-r--r--mm/memory-failure.c79
-rw-r--r--mm/memory.c2
-rw-r--r--mm/memory_hotplug.c6
-rw-r--r--mm/migrate.c12
-rw-r--r--mm/mlock.c6
-rw-r--r--mm/mmap.c2
-rw-r--r--mm/nommu.c8
-rw-r--r--mm/oom_kill.c2
-rw-r--r--mm/page-writeback.c29
-rw-r--r--mm/page_alloc.c280
-rw-r--r--mm/page_ext.c13
-rw-r--r--mm/page_idle.c4
-rw-r--r--mm/page_isolation.c11
-rw-r--r--mm/page_poison.c77
-rw-r--r--mm/percpu.c40
-rw-r--r--mm/rmap.c148
-rw-r--r--mm/rodata_test.c17
-rw-r--r--mm/slab.c7
-rw-r--r--mm/sparse.c5
-rw-r--r--mm/swap.c59
-rw-r--r--mm/swap_slots.c23
-rw-r--r--mm/swap_state.c14
-rw-r--r--mm/swapfile.c45
-rw-r--r--mm/truncate.c13
-rw-r--r--mm/usercopy.c19
-rw-r--r--mm/util.c57
-rw-r--r--mm/vmalloc.c35
-rw-r--r--mm/vmscan.c525
-rw-r--r--mm/vmstat.c75
-rw-r--r--mm/workingset.c6
51 files changed, 1583 insertions, 1222 deletions
diff --git a/mm/Kconfig b/mm/Kconfig
index 9b8fccb969dc..beb7a455915d 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -312,7 +312,6 @@ config NEED_BOUNCE_POOL
312config NR_QUICK 312config NR_QUICK
313 int 313 int
314 depends on QUICKLIST 314 depends on QUICKLIST
315 default "2" if AVR32
316 default "1" 315 default "1"
317 316
318config VIRT_TO_BUS 317config VIRT_TO_BUS
diff --git a/mm/Kconfig.debug b/mm/Kconfig.debug
index 79d0fd13b5b3..5b0adf1435de 100644
--- a/mm/Kconfig.debug
+++ b/mm/Kconfig.debug
@@ -42,7 +42,6 @@ config DEBUG_PAGEALLOC_ENABLE_DEFAULT
42 42
43config PAGE_POISONING 43config PAGE_POISONING
44 bool "Poison pages after freeing" 44 bool "Poison pages after freeing"
45 select PAGE_EXTENSION
46 select PAGE_POISONING_NO_SANITY if HIBERNATION 45 select PAGE_POISONING_NO_SANITY if HIBERNATION
47 ---help--- 46 ---help---
48 Fill the pages with poison patterns after free_pages() and verify 47 Fill the pages with poison patterns after free_pages() and verify
diff --git a/mm/backing-dev.c b/mm/backing-dev.c
index c6f2a37028c2..f028a9a472fd 100644
--- a/mm/backing-dev.c
+++ b/mm/backing-dev.c
@@ -12,8 +12,6 @@
12#include <linux/device.h> 12#include <linux/device.h>
13#include <trace/events/writeback.h> 13#include <trace/events/writeback.h>
14 14
15static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16
17struct backing_dev_info noop_backing_dev_info = { 15struct backing_dev_info noop_backing_dev_info = {
18 .name = "noop", 16 .name = "noop",
19 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, 17 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
@@ -242,6 +240,8 @@ static __init int bdi_class_init(void)
242} 240}
243postcore_initcall(bdi_class_init); 241postcore_initcall(bdi_class_init);
244 242
243static int bdi_init(struct backing_dev_info *bdi);
244
245static int __init default_bdi_init(void) 245static int __init default_bdi_init(void)
246{ 246{
247 int err; 247 int err;
@@ -294,6 +294,8 @@ static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
294 294
295 memset(wb, 0, sizeof(*wb)); 295 memset(wb, 0, sizeof(*wb));
296 296
297 if (wb != &bdi->wb)
298 bdi_get(bdi);
297 wb->bdi = bdi; 299 wb->bdi = bdi;
298 wb->last_old_flush = jiffies; 300 wb->last_old_flush = jiffies;
299 INIT_LIST_HEAD(&wb->b_dirty); 301 INIT_LIST_HEAD(&wb->b_dirty);
@@ -314,8 +316,10 @@ static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
314 wb->dirty_sleep = jiffies; 316 wb->dirty_sleep = jiffies;
315 317
316 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp); 318 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
317 if (!wb->congested) 319 if (!wb->congested) {
318 return -ENOMEM; 320 err = -ENOMEM;
321 goto out_put_bdi;
322 }
319 323
320 err = fprop_local_init_percpu(&wb->completions, gfp); 324 err = fprop_local_init_percpu(&wb->completions, gfp);
321 if (err) 325 if (err)
@@ -335,9 +339,14 @@ out_destroy_stat:
335 fprop_local_destroy_percpu(&wb->completions); 339 fprop_local_destroy_percpu(&wb->completions);
336out_put_cong: 340out_put_cong:
337 wb_congested_put(wb->congested); 341 wb_congested_put(wb->congested);
342out_put_bdi:
343 if (wb != &bdi->wb)
344 bdi_put(bdi);
338 return err; 345 return err;
339} 346}
340 347
348static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
349
341/* 350/*
342 * Remove bdi from the global list and shutdown any threads we have running 351 * Remove bdi from the global list and shutdown any threads we have running
343 */ 352 */
@@ -347,10 +356,18 @@ static void wb_shutdown(struct bdi_writeback *wb)
347 spin_lock_bh(&wb->work_lock); 356 spin_lock_bh(&wb->work_lock);
348 if (!test_and_clear_bit(WB_registered, &wb->state)) { 357 if (!test_and_clear_bit(WB_registered, &wb->state)) {
349 spin_unlock_bh(&wb->work_lock); 358 spin_unlock_bh(&wb->work_lock);
359 /*
360 * Wait for wb shutdown to finish if someone else is just
361 * running wb_shutdown(). Otherwise we could proceed to wb /
362 * bdi destruction before wb_shutdown() is finished.
363 */
364 wait_on_bit(&wb->state, WB_shutting_down, TASK_UNINTERRUPTIBLE);
350 return; 365 return;
351 } 366 }
367 set_bit(WB_shutting_down, &wb->state);
352 spin_unlock_bh(&wb->work_lock); 368 spin_unlock_bh(&wb->work_lock);
353 369
370 cgwb_remove_from_bdi_list(wb);
354 /* 371 /*
355 * Drain work list and shutdown the delayed_work. !WB_registered 372 * Drain work list and shutdown the delayed_work. !WB_registered
356 * tells wb_workfn() that @wb is dying and its work_list needs to 373 * tells wb_workfn() that @wb is dying and its work_list needs to
@@ -359,6 +376,12 @@ static void wb_shutdown(struct bdi_writeback *wb)
359 mod_delayed_work(bdi_wq, &wb->dwork, 0); 376 mod_delayed_work(bdi_wq, &wb->dwork, 0);
360 flush_delayed_work(&wb->dwork); 377 flush_delayed_work(&wb->dwork);
361 WARN_ON(!list_empty(&wb->work_list)); 378 WARN_ON(!list_empty(&wb->work_list));
379 /*
380 * Make sure bit gets cleared after shutdown is finished. Matches with
381 * the barrier provided by test_and_clear_bit() above.
382 */
383 smp_wmb();
384 clear_bit(WB_shutting_down, &wb->state);
362} 385}
363 386
364static void wb_exit(struct bdi_writeback *wb) 387static void wb_exit(struct bdi_writeback *wb)
@@ -372,6 +395,8 @@ static void wb_exit(struct bdi_writeback *wb)
372 395
373 fprop_local_destroy_percpu(&wb->completions); 396 fprop_local_destroy_percpu(&wb->completions);
374 wb_congested_put(wb->congested); 397 wb_congested_put(wb->congested);
398 if (wb != &wb->bdi->wb)
399 bdi_put(wb->bdi);
375} 400}
376 401
377#ifdef CONFIG_CGROUP_WRITEBACK 402#ifdef CONFIG_CGROUP_WRITEBACK
@@ -381,11 +406,9 @@ static void wb_exit(struct bdi_writeback *wb)
381/* 406/*
382 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree, 407 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
383 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU 408 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
384 * protected. cgwb_release_wait is used to wait for the completion of cgwb 409 * protected.
385 * releases from bdi destruction path.
386 */ 410 */
387static DEFINE_SPINLOCK(cgwb_lock); 411static DEFINE_SPINLOCK(cgwb_lock);
388static DECLARE_WAIT_QUEUE_HEAD(cgwb_release_wait);
389 412
390/** 413/**
391 * wb_congested_get_create - get or create a wb_congested 414 * wb_congested_get_create - get or create a wb_congested
@@ -438,7 +461,7 @@ retry:
438 return NULL; 461 return NULL;
439 462
440 atomic_set(&new_congested->refcnt, 0); 463 atomic_set(&new_congested->refcnt, 0);
441 new_congested->bdi = bdi; 464 new_congested->__bdi = bdi;
442 new_congested->blkcg_id = blkcg_id; 465 new_congested->blkcg_id = blkcg_id;
443 goto retry; 466 goto retry;
444 467
@@ -466,10 +489,10 @@ void wb_congested_put(struct bdi_writeback_congested *congested)
466 } 489 }
467 490
468 /* bdi might already have been destroyed leaving @congested unlinked */ 491 /* bdi might already have been destroyed leaving @congested unlinked */
469 if (congested->bdi) { 492 if (congested->__bdi) {
470 rb_erase(&congested->rb_node, 493 rb_erase(&congested->rb_node,
471 &congested->bdi->cgwb_congested_tree); 494 &congested->__bdi->cgwb_congested_tree);
472 congested->bdi = NULL; 495 congested->__bdi = NULL;
473 } 496 }
474 497
475 spin_unlock_irqrestore(&cgwb_lock, flags); 498 spin_unlock_irqrestore(&cgwb_lock, flags);
@@ -480,11 +503,6 @@ static void cgwb_release_workfn(struct work_struct *work)
480{ 503{
481 struct bdi_writeback *wb = container_of(work, struct bdi_writeback, 504 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
482 release_work); 505 release_work);
483 struct backing_dev_info *bdi = wb->bdi;
484
485 spin_lock_irq(&cgwb_lock);
486 list_del_rcu(&wb->bdi_node);
487 spin_unlock_irq(&cgwb_lock);
488 506
489 wb_shutdown(wb); 507 wb_shutdown(wb);
490 508
@@ -495,9 +513,6 @@ static void cgwb_release_workfn(struct work_struct *work)
495 percpu_ref_exit(&wb->refcnt); 513 percpu_ref_exit(&wb->refcnt);
496 wb_exit(wb); 514 wb_exit(wb);
497 kfree_rcu(wb, rcu); 515 kfree_rcu(wb, rcu);
498
499 if (atomic_dec_and_test(&bdi->usage_cnt))
500 wake_up_all(&cgwb_release_wait);
501} 516}
502 517
503static void cgwb_release(struct percpu_ref *refcnt) 518static void cgwb_release(struct percpu_ref *refcnt)
@@ -517,6 +532,13 @@ static void cgwb_kill(struct bdi_writeback *wb)
517 percpu_ref_kill(&wb->refcnt); 532 percpu_ref_kill(&wb->refcnt);
518} 533}
519 534
535static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
536{
537 spin_lock_irq(&cgwb_lock);
538 list_del_rcu(&wb->bdi_node);
539 spin_unlock_irq(&cgwb_lock);
540}
541
520static int cgwb_create(struct backing_dev_info *bdi, 542static int cgwb_create(struct backing_dev_info *bdi,
521 struct cgroup_subsys_state *memcg_css, gfp_t gfp) 543 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
522{ 544{
@@ -580,7 +602,6 @@ static int cgwb_create(struct backing_dev_info *bdi,
580 /* we might have raced another instance of this function */ 602 /* we might have raced another instance of this function */
581 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); 603 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
582 if (!ret) { 604 if (!ret) {
583 atomic_inc(&bdi->usage_cnt);
584 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); 605 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
585 list_add(&wb->memcg_node, memcg_cgwb_list); 606 list_add(&wb->memcg_node, memcg_cgwb_list);
586 list_add(&wb->blkcg_node, blkcg_cgwb_list); 607 list_add(&wb->blkcg_node, blkcg_cgwb_list);
@@ -670,7 +691,6 @@ static int cgwb_bdi_init(struct backing_dev_info *bdi)
670 691
671 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); 692 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
672 bdi->cgwb_congested_tree = RB_ROOT; 693 bdi->cgwb_congested_tree = RB_ROOT;
673 atomic_set(&bdi->usage_cnt, 1);
674 694
675 ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); 695 ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
676 if (!ret) { 696 if (!ret) {
@@ -680,29 +700,26 @@ static int cgwb_bdi_init(struct backing_dev_info *bdi)
680 return ret; 700 return ret;
681} 701}
682 702
683static void cgwb_bdi_destroy(struct backing_dev_info *bdi) 703static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
684{ 704{
685 struct radix_tree_iter iter; 705 struct radix_tree_iter iter;
686 void **slot; 706 void **slot;
707 struct bdi_writeback *wb;
687 708
688 WARN_ON(test_bit(WB_registered, &bdi->wb.state)); 709 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
689 710
690 spin_lock_irq(&cgwb_lock); 711 spin_lock_irq(&cgwb_lock);
691 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) 712 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
692 cgwb_kill(*slot); 713 cgwb_kill(*slot);
693 spin_unlock_irq(&cgwb_lock);
694 714
695 /* 715 while (!list_empty(&bdi->wb_list)) {
696 * All cgwb's must be shutdown and released before returning. Drain 716 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
697 * the usage counter to wait for all cgwb's ever created on @bdi. 717 bdi_node);
698 */ 718 spin_unlock_irq(&cgwb_lock);
699 atomic_dec(&bdi->usage_cnt); 719 wb_shutdown(wb);
700 wait_event(cgwb_release_wait, !atomic_read(&bdi->usage_cnt)); 720 spin_lock_irq(&cgwb_lock);
701 /* 721 }
702 * Grab back our reference so that we hold it when @bdi gets 722 spin_unlock_irq(&cgwb_lock);
703 * re-registered.
704 */
705 atomic_inc(&bdi->usage_cnt);
706} 723}
707 724
708/** 725/**
@@ -752,11 +769,18 @@ static void cgwb_bdi_exit(struct backing_dev_info *bdi)
752 rb_entry(rbn, struct bdi_writeback_congested, rb_node); 769 rb_entry(rbn, struct bdi_writeback_congested, rb_node);
753 770
754 rb_erase(rbn, &bdi->cgwb_congested_tree); 771 rb_erase(rbn, &bdi->cgwb_congested_tree);
755 congested->bdi = NULL; /* mark @congested unlinked */ 772 congested->__bdi = NULL; /* mark @congested unlinked */
756 } 773 }
757 spin_unlock_irq(&cgwb_lock); 774 spin_unlock_irq(&cgwb_lock);
758} 775}
759 776
777static void cgwb_bdi_register(struct backing_dev_info *bdi)
778{
779 spin_lock_irq(&cgwb_lock);
780 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
781 spin_unlock_irq(&cgwb_lock);
782}
783
760#else /* CONFIG_CGROUP_WRITEBACK */ 784#else /* CONFIG_CGROUP_WRITEBACK */
761 785
762static int cgwb_bdi_init(struct backing_dev_info *bdi) 786static int cgwb_bdi_init(struct backing_dev_info *bdi)
@@ -777,16 +801,26 @@ static int cgwb_bdi_init(struct backing_dev_info *bdi)
777 return 0; 801 return 0;
778} 802}
779 803
780static void cgwb_bdi_destroy(struct backing_dev_info *bdi) { } 804static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
781 805
782static void cgwb_bdi_exit(struct backing_dev_info *bdi) 806static void cgwb_bdi_exit(struct backing_dev_info *bdi)
783{ 807{
784 wb_congested_put(bdi->wb_congested); 808 wb_congested_put(bdi->wb_congested);
785} 809}
786 810
811static void cgwb_bdi_register(struct backing_dev_info *bdi)
812{
813 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
814}
815
816static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
817{
818 list_del_rcu(&wb->bdi_node);
819}
820
787#endif /* CONFIG_CGROUP_WRITEBACK */ 821#endif /* CONFIG_CGROUP_WRITEBACK */
788 822
789int bdi_init(struct backing_dev_info *bdi) 823static int bdi_init(struct backing_dev_info *bdi)
790{ 824{
791 int ret; 825 int ret;
792 826
@@ -802,11 +836,8 @@ int bdi_init(struct backing_dev_info *bdi)
802 836
803 ret = cgwb_bdi_init(bdi); 837 ret = cgwb_bdi_init(bdi);
804 838
805 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
806
807 return ret; 839 return ret;
808} 840}
809EXPORT_SYMBOL(bdi_init);
810 841
811struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id) 842struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
812{ 843{
@@ -823,22 +854,20 @@ struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
823 } 854 }
824 return bdi; 855 return bdi;
825} 856}
857EXPORT_SYMBOL(bdi_alloc_node);
826 858
827int bdi_register(struct backing_dev_info *bdi, struct device *parent, 859int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
828 const char *fmt, ...)
829{ 860{
830 va_list args;
831 struct device *dev; 861 struct device *dev;
832 862
833 if (bdi->dev) /* The driver needs to use separate queues per device */ 863 if (bdi->dev) /* The driver needs to use separate queues per device */
834 return 0; 864 return 0;
835 865
836 va_start(args, fmt); 866 dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args);
837 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
838 va_end(args);
839 if (IS_ERR(dev)) 867 if (IS_ERR(dev))
840 return PTR_ERR(dev); 868 return PTR_ERR(dev);
841 869
870 cgwb_bdi_register(bdi);
842 bdi->dev = dev; 871 bdi->dev = dev;
843 872
844 bdi_debug_register(bdi, dev_name(dev)); 873 bdi_debug_register(bdi, dev_name(dev));
@@ -851,20 +880,25 @@ int bdi_register(struct backing_dev_info *bdi, struct device *parent,
851 trace_writeback_bdi_register(bdi); 880 trace_writeback_bdi_register(bdi);
852 return 0; 881 return 0;
853} 882}
854EXPORT_SYMBOL(bdi_register); 883EXPORT_SYMBOL(bdi_register_va);
855 884
856int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) 885int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
857{ 886{
858 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); 887 va_list args;
888 int ret;
889
890 va_start(args, fmt);
891 ret = bdi_register_va(bdi, fmt, args);
892 va_end(args);
893 return ret;
859} 894}
860EXPORT_SYMBOL(bdi_register_dev); 895EXPORT_SYMBOL(bdi_register);
861 896
862int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner) 897int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
863{ 898{
864 int rc; 899 int rc;
865 900
866 rc = bdi_register(bdi, NULL, "%u:%u", MAJOR(owner->devt), 901 rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt));
867 MINOR(owner->devt));
868 if (rc) 902 if (rc)
869 return rc; 903 return rc;
870 /* Leaking owner reference... */ 904 /* Leaking owner reference... */
@@ -892,7 +926,7 @@ void bdi_unregister(struct backing_dev_info *bdi)
892 /* make sure nobody finds us on the bdi_list anymore */ 926 /* make sure nobody finds us on the bdi_list anymore */
893 bdi_remove_from_list(bdi); 927 bdi_remove_from_list(bdi);
894 wb_shutdown(&bdi->wb); 928 wb_shutdown(&bdi->wb);
895 cgwb_bdi_destroy(bdi); 929 cgwb_bdi_unregister(bdi);
896 930
897 if (bdi->dev) { 931 if (bdi->dev) {
898 bdi_debug_unregister(bdi); 932 bdi_debug_unregister(bdi);
@@ -906,19 +940,16 @@ void bdi_unregister(struct backing_dev_info *bdi)
906 } 940 }
907} 941}
908 942
909static void bdi_exit(struct backing_dev_info *bdi)
910{
911 WARN_ON_ONCE(bdi->dev);
912 wb_exit(&bdi->wb);
913 cgwb_bdi_exit(bdi);
914}
915
916static void release_bdi(struct kref *ref) 943static void release_bdi(struct kref *ref)
917{ 944{
918 struct backing_dev_info *bdi = 945 struct backing_dev_info *bdi =
919 container_of(ref, struct backing_dev_info, refcnt); 946 container_of(ref, struct backing_dev_info, refcnt);
920 947
921 bdi_exit(bdi); 948 if (test_bit(WB_registered, &bdi->wb.state))
949 bdi_unregister(bdi);
950 WARN_ON_ONCE(bdi->dev);
951 wb_exit(&bdi->wb);
952 cgwb_bdi_exit(bdi);
922 kfree(bdi); 953 kfree(bdi);
923} 954}
924 955
@@ -926,38 +957,7 @@ void bdi_put(struct backing_dev_info *bdi)
926{ 957{
927 kref_put(&bdi->refcnt, release_bdi); 958 kref_put(&bdi->refcnt, release_bdi);
928} 959}
929 960EXPORT_SYMBOL(bdi_put);
930void bdi_destroy(struct backing_dev_info *bdi)
931{
932 bdi_unregister(bdi);
933 bdi_exit(bdi);
934}
935EXPORT_SYMBOL(bdi_destroy);
936
937/*
938 * For use from filesystems to quickly init and register a bdi associated
939 * with dirty writeback
940 */
941int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
942{
943 int err;
944
945 bdi->name = name;
946 bdi->capabilities = 0;
947 err = bdi_init(bdi);
948 if (err)
949 return err;
950
951 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
952 atomic_long_inc_return(&bdi_seq));
953 if (err) {
954 bdi_destroy(bdi);
955 return err;
956 }
957
958 return 0;
959}
960EXPORT_SYMBOL(bdi_setup_and_register);
961 961
962static wait_queue_head_t congestion_wqh[2] = { 962static wait_queue_head_t congestion_wqh[2] = {
963 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), 963 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
diff --git a/mm/cma.c b/mm/cma.c
index a6033e344430..978b4a1441ef 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -53,6 +53,11 @@ unsigned long cma_get_size(const struct cma *cma)
53 return cma->count << PAGE_SHIFT; 53 return cma->count << PAGE_SHIFT;
54} 54}
55 55
56const char *cma_get_name(const struct cma *cma)
57{
58 return cma->name ? cma->name : "(undefined)";
59}
60
56static unsigned long cma_bitmap_aligned_mask(const struct cma *cma, 61static unsigned long cma_bitmap_aligned_mask(const struct cma *cma,
57 int align_order) 62 int align_order)
58{ 63{
@@ -168,6 +173,7 @@ core_initcall(cma_init_reserved_areas);
168 */ 173 */
169int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size, 174int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
170 unsigned int order_per_bit, 175 unsigned int order_per_bit,
176 const char *name,
171 struct cma **res_cma) 177 struct cma **res_cma)
172{ 178{
173 struct cma *cma; 179 struct cma *cma;
@@ -198,6 +204,13 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
198 * subsystems (like slab allocator) are available. 204 * subsystems (like slab allocator) are available.
199 */ 205 */
200 cma = &cma_areas[cma_area_count]; 206 cma = &cma_areas[cma_area_count];
207 if (name) {
208 cma->name = name;
209 } else {
210 cma->name = kasprintf(GFP_KERNEL, "cma%d\n", cma_area_count);
211 if (!cma->name)
212 return -ENOMEM;
213 }
201 cma->base_pfn = PFN_DOWN(base); 214 cma->base_pfn = PFN_DOWN(base);
202 cma->count = size >> PAGE_SHIFT; 215 cma->count = size >> PAGE_SHIFT;
203 cma->order_per_bit = order_per_bit; 216 cma->order_per_bit = order_per_bit;
@@ -229,7 +242,7 @@ int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
229int __init cma_declare_contiguous(phys_addr_t base, 242int __init cma_declare_contiguous(phys_addr_t base,
230 phys_addr_t size, phys_addr_t limit, 243 phys_addr_t size, phys_addr_t limit,
231 phys_addr_t alignment, unsigned int order_per_bit, 244 phys_addr_t alignment, unsigned int order_per_bit,
232 bool fixed, struct cma **res_cma) 245 bool fixed, const char *name, struct cma **res_cma)
233{ 246{
234 phys_addr_t memblock_end = memblock_end_of_DRAM(); 247 phys_addr_t memblock_end = memblock_end_of_DRAM();
235 phys_addr_t highmem_start; 248 phys_addr_t highmem_start;
@@ -335,7 +348,7 @@ int __init cma_declare_contiguous(phys_addr_t base,
335 base = addr; 348 base = addr;
336 } 349 }
337 350
338 ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma); 351 ret = cma_init_reserved_mem(base, size, order_per_bit, name, res_cma);
339 if (ret) 352 if (ret)
340 goto err; 353 goto err;
341 354
@@ -491,3 +504,17 @@ bool cma_release(struct cma *cma, const struct page *pages, unsigned int count)
491 504
492 return true; 505 return true;
493} 506}
507
508int cma_for_each_area(int (*it)(struct cma *cma, void *data), void *data)
509{
510 int i;
511
512 for (i = 0; i < cma_area_count; i++) {
513 int ret = it(&cma_areas[i], data);
514
515 if (ret)
516 return ret;
517 }
518
519 return 0;
520}
diff --git a/mm/cma.h b/mm/cma.h
index 17c75a4246c8..49861286279d 100644
--- a/mm/cma.h
+++ b/mm/cma.h
@@ -11,6 +11,7 @@ struct cma {
11 struct hlist_head mem_head; 11 struct hlist_head mem_head;
12 spinlock_t mem_head_lock; 12 spinlock_t mem_head_lock;
13#endif 13#endif
14 const char *name;
14}; 15};
15 16
16extern struct cma cma_areas[MAX_CMA_AREAS]; 17extern struct cma cma_areas[MAX_CMA_AREAS];
diff --git a/mm/cma_debug.c b/mm/cma_debug.c
index ffc0c3d0ae64..595b757bef72 100644
--- a/mm/cma_debug.c
+++ b/mm/cma_debug.c
@@ -167,7 +167,7 @@ static void cma_debugfs_add_one(struct cma *cma, int idx)
167 char name[16]; 167 char name[16];
168 int u32s; 168 int u32s;
169 169
170 sprintf(name, "cma-%d", idx); 170 sprintf(name, "cma-%s", cma->name);
171 171
172 tmp = debugfs_create_dir(name, cma_debugfs_root); 172 tmp = debugfs_create_dir(name, cma_debugfs_root);
173 173
diff --git a/mm/compaction.c b/mm/compaction.c
index 81e1eaa2a2cf..613c59e928cb 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -89,11 +89,6 @@ static void map_pages(struct list_head *list)
89 list_splice(&tmp_list, list); 89 list_splice(&tmp_list, list);
90} 90}
91 91
92static inline bool migrate_async_suitable(int migratetype)
93{
94 return is_migrate_cma(migratetype) || migratetype == MIGRATE_MOVABLE;
95}
96
97#ifdef CONFIG_COMPACTION 92#ifdef CONFIG_COMPACTION
98 93
99int PageMovable(struct page *page) 94int PageMovable(struct page *page)
@@ -988,13 +983,26 @@ isolate_migratepages_range(struct compact_control *cc, unsigned long start_pfn,
988#endif /* CONFIG_COMPACTION || CONFIG_CMA */ 983#endif /* CONFIG_COMPACTION || CONFIG_CMA */
989#ifdef CONFIG_COMPACTION 984#ifdef CONFIG_COMPACTION
990 985
991/* Returns true if the page is within a block suitable for migration to */ 986static bool suitable_migration_source(struct compact_control *cc,
992static bool suitable_migration_target(struct compact_control *cc,
993 struct page *page) 987 struct page *page)
994{ 988{
995 if (cc->ignore_block_suitable) 989 int block_mt;
990
991 if ((cc->mode != MIGRATE_ASYNC) || !cc->direct_compaction)
996 return true; 992 return true;
997 993
994 block_mt = get_pageblock_migratetype(page);
995
996 if (cc->migratetype == MIGRATE_MOVABLE)
997 return is_migrate_movable(block_mt);
998 else
999 return block_mt == cc->migratetype;
1000}
1001
1002/* Returns true if the page is within a block suitable for migration to */
1003static bool suitable_migration_target(struct compact_control *cc,
1004 struct page *page)
1005{
998 /* If the page is a large free page, then disallow migration */ 1006 /* If the page is a large free page, then disallow migration */
999 if (PageBuddy(page)) { 1007 if (PageBuddy(page)) {
1000 /* 1008 /*
@@ -1006,8 +1014,11 @@ static bool suitable_migration_target(struct compact_control *cc,
1006 return false; 1014 return false;
1007 } 1015 }
1008 1016
1017 if (cc->ignore_block_suitable)
1018 return true;
1019
1009 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */ 1020 /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
1010 if (migrate_async_suitable(get_pageblock_migratetype(page))) 1021 if (is_migrate_movable(get_pageblock_migratetype(page)))
1011 return true; 1022 return true;
1012 1023
1013 /* Otherwise skip the block */ 1024 /* Otherwise skip the block */
@@ -1242,8 +1253,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
1242 * Async compaction is optimistic to see if the minimum amount 1253 * Async compaction is optimistic to see if the minimum amount
1243 * of work satisfies the allocation. 1254 * of work satisfies the allocation.
1244 */ 1255 */
1245 if (cc->mode == MIGRATE_ASYNC && 1256 if (!suitable_migration_source(cc, page))
1246 !migrate_async_suitable(get_pageblock_migratetype(page)))
1247 continue; 1257 continue;
1248 1258
1249 /* Perform the isolation */ 1259 /* Perform the isolation */
@@ -1276,11 +1286,11 @@ static inline bool is_via_compact_memory(int order)
1276 return order == -1; 1286 return order == -1;
1277} 1287}
1278 1288
1279static enum compact_result __compact_finished(struct zone *zone, struct compact_control *cc, 1289static enum compact_result __compact_finished(struct zone *zone,
1280 const int migratetype) 1290 struct compact_control *cc)
1281{ 1291{
1282 unsigned int order; 1292 unsigned int order;
1283 unsigned long watermark; 1293 const int migratetype = cc->migratetype;
1284 1294
1285 if (cc->contended || fatal_signal_pending(current)) 1295 if (cc->contended || fatal_signal_pending(current))
1286 return COMPACT_CONTENDED; 1296 return COMPACT_CONTENDED;
@@ -1308,12 +1318,16 @@ static enum compact_result __compact_finished(struct zone *zone, struct compact_
1308 if (is_via_compact_memory(cc->order)) 1318 if (is_via_compact_memory(cc->order))
1309 return COMPACT_CONTINUE; 1319 return COMPACT_CONTINUE;
1310 1320
1311 /* Compaction run is not finished if the watermark is not met */ 1321 if (cc->finishing_block) {
1312 watermark = zone->watermark[cc->alloc_flags & ALLOC_WMARK_MASK]; 1322 /*
1313 1323 * We have finished the pageblock, but better check again that
1314 if (!zone_watermark_ok(zone, cc->order, watermark, cc->classzone_idx, 1324 * we really succeeded.
1315 cc->alloc_flags)) 1325 */
1316 return COMPACT_CONTINUE; 1326 if (IS_ALIGNED(cc->migrate_pfn, pageblock_nr_pages))
1327 cc->finishing_block = false;
1328 else
1329 return COMPACT_CONTINUE;
1330 }
1317 1331
1318 /* Direct compactor: Is a suitable page free? */ 1332 /* Direct compactor: Is a suitable page free? */
1319 for (order = cc->order; order < MAX_ORDER; order++) { 1333 for (order = cc->order; order < MAX_ORDER; order++) {
@@ -1335,20 +1349,40 @@ static enum compact_result __compact_finished(struct zone *zone, struct compact_
1335 * other migratetype buddy lists. 1349 * other migratetype buddy lists.
1336 */ 1350 */
1337 if (find_suitable_fallback(area, order, migratetype, 1351 if (find_suitable_fallback(area, order, migratetype,
1338 true, &can_steal) != -1) 1352 true, &can_steal) != -1) {
1339 return COMPACT_SUCCESS; 1353
1354 /* movable pages are OK in any pageblock */
1355 if (migratetype == MIGRATE_MOVABLE)
1356 return COMPACT_SUCCESS;
1357
1358 /*
1359 * We are stealing for a non-movable allocation. Make
1360 * sure we finish compacting the current pageblock
1361 * first so it is as free as possible and we won't
1362 * have to steal another one soon. This only applies
1363 * to sync compaction, as async compaction operates
1364 * on pageblocks of the same migratetype.
1365 */
1366 if (cc->mode == MIGRATE_ASYNC ||
1367 IS_ALIGNED(cc->migrate_pfn,
1368 pageblock_nr_pages)) {
1369 return COMPACT_SUCCESS;
1370 }
1371
1372 cc->finishing_block = true;
1373 return COMPACT_CONTINUE;
1374 }
1340 } 1375 }
1341 1376
1342 return COMPACT_NO_SUITABLE_PAGE; 1377 return COMPACT_NO_SUITABLE_PAGE;
1343} 1378}
1344 1379
1345static enum compact_result compact_finished(struct zone *zone, 1380static enum compact_result compact_finished(struct zone *zone,
1346 struct compact_control *cc, 1381 struct compact_control *cc)
1347 const int migratetype)
1348{ 1382{
1349 int ret; 1383 int ret;
1350 1384
1351 ret = __compact_finished(zone, cc, migratetype); 1385 ret = __compact_finished(zone, cc);
1352 trace_mm_compaction_finished(zone, cc->order, ret); 1386 trace_mm_compaction_finished(zone, cc->order, ret);
1353 if (ret == COMPACT_NO_SUITABLE_PAGE) 1387 if (ret == COMPACT_NO_SUITABLE_PAGE)
1354 ret = COMPACT_CONTINUE; 1388 ret = COMPACT_CONTINUE;
@@ -1481,9 +1515,9 @@ static enum compact_result compact_zone(struct zone *zone, struct compact_contro
1481 enum compact_result ret; 1515 enum compact_result ret;
1482 unsigned long start_pfn = zone->zone_start_pfn; 1516 unsigned long start_pfn = zone->zone_start_pfn;
1483 unsigned long end_pfn = zone_end_pfn(zone); 1517 unsigned long end_pfn = zone_end_pfn(zone);
1484 const int migratetype = gfpflags_to_migratetype(cc->gfp_mask);
1485 const bool sync = cc->mode != MIGRATE_ASYNC; 1518 const bool sync = cc->mode != MIGRATE_ASYNC;
1486 1519
1520 cc->migratetype = gfpflags_to_migratetype(cc->gfp_mask);
1487 ret = compaction_suitable(zone, cc->order, cc->alloc_flags, 1521 ret = compaction_suitable(zone, cc->order, cc->alloc_flags,
1488 cc->classzone_idx); 1522 cc->classzone_idx);
1489 /* Compaction is likely to fail */ 1523 /* Compaction is likely to fail */
@@ -1533,8 +1567,7 @@ static enum compact_result compact_zone(struct zone *zone, struct compact_contro
1533 1567
1534 migrate_prep_local(); 1568 migrate_prep_local();
1535 1569
1536 while ((ret = compact_finished(zone, cc, migratetype)) == 1570 while ((ret = compact_finished(zone, cc)) == COMPACT_CONTINUE) {
1537 COMPACT_CONTINUE) {
1538 int err; 1571 int err;
1539 1572
1540 switch (isolate_migratepages(zone, cc)) { 1573 switch (isolate_migratepages(zone, cc)) {
diff --git a/mm/filemap.c b/mm/filemap.c
index d6e67be1802e..6f1be573a5e6 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -519,7 +519,7 @@ EXPORT_SYMBOL(filemap_write_and_wait);
519 * 519 *
520 * Write out and wait upon file offsets lstart->lend, inclusive. 520 * Write out and wait upon file offsets lstart->lend, inclusive.
521 * 521 *
522 * Note that `lend' is inclusive (describes the last byte to be written) so 522 * Note that @lend is inclusive (describes the last byte to be written) so
523 * that this function can be used to write to the very end-of-file (end = -1). 523 * that this function can be used to write to the very end-of-file (end = -1).
524 */ 524 */
525int filemap_write_and_wait_range(struct address_space *mapping, 525int filemap_write_and_wait_range(struct address_space *mapping,
@@ -1277,12 +1277,14 @@ EXPORT_SYMBOL(find_lock_entry);
1277 * 1277 *
1278 * PCG flags modify how the page is returned. 1278 * PCG flags modify how the page is returned.
1279 * 1279 *
1280 * FGP_ACCESSED: the page will be marked accessed 1280 * @fgp_flags can be:
1281 * FGP_LOCK: Page is return locked 1281 *
1282 * FGP_CREAT: If page is not present then a new page is allocated using 1282 * - FGP_ACCESSED: the page will be marked accessed
1283 * @gfp_mask and added to the page cache and the VM's LRU 1283 * - FGP_LOCK: Page is return locked
1284 * list. The page is returned locked and with an increased 1284 * - FGP_CREAT: If page is not present then a new page is allocated using
1285 * refcount. Otherwise, %NULL is returned. 1285 * @gfp_mask and added to the page cache and the VM's LRU
1286 * list. The page is returned locked and with an increased
1287 * refcount. Otherwise, NULL is returned.
1286 * 1288 *
1287 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even 1289 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1288 * if the GFP flags specified for FGP_CREAT are atomic. 1290 * if the GFP flags specified for FGP_CREAT are atomic.
@@ -2202,12 +2204,12 @@ int filemap_fault(struct vm_fault *vmf)
2202 struct file_ra_state *ra = &file->f_ra; 2204 struct file_ra_state *ra = &file->f_ra;
2203 struct inode *inode = mapping->host; 2205 struct inode *inode = mapping->host;
2204 pgoff_t offset = vmf->pgoff; 2206 pgoff_t offset = vmf->pgoff;
2207 pgoff_t max_off;
2205 struct page *page; 2208 struct page *page;
2206 loff_t size;
2207 int ret = 0; 2209 int ret = 0;
2208 2210
2209 size = round_up(i_size_read(inode), PAGE_SIZE); 2211 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2210 if (offset >= size >> PAGE_SHIFT) 2212 if (unlikely(offset >= max_off))
2211 return VM_FAULT_SIGBUS; 2213 return VM_FAULT_SIGBUS;
2212 2214
2213 /* 2215 /*
@@ -2256,8 +2258,8 @@ retry_find:
2256 * Found the page and have a reference on it. 2258 * Found the page and have a reference on it.
2257 * We must recheck i_size under page lock. 2259 * We must recheck i_size under page lock.
2258 */ 2260 */
2259 size = round_up(i_size_read(inode), PAGE_SIZE); 2261 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2260 if (unlikely(offset >= size >> PAGE_SHIFT)) { 2262 if (unlikely(offset >= max_off)) {
2261 unlock_page(page); 2263 unlock_page(page);
2262 put_page(page); 2264 put_page(page);
2263 return VM_FAULT_SIGBUS; 2265 return VM_FAULT_SIGBUS;
@@ -2323,7 +2325,7 @@ void filemap_map_pages(struct vm_fault *vmf,
2323 struct file *file = vmf->vma->vm_file; 2325 struct file *file = vmf->vma->vm_file;
2324 struct address_space *mapping = file->f_mapping; 2326 struct address_space *mapping = file->f_mapping;
2325 pgoff_t last_pgoff = start_pgoff; 2327 pgoff_t last_pgoff = start_pgoff;
2326 loff_t size; 2328 unsigned long max_idx;
2327 struct page *head, *page; 2329 struct page *head, *page;
2328 2330
2329 rcu_read_lock(); 2331 rcu_read_lock();
@@ -2369,8 +2371,8 @@ repeat:
2369 if (page->mapping != mapping || !PageUptodate(page)) 2371 if (page->mapping != mapping || !PageUptodate(page))
2370 goto unlock; 2372 goto unlock;
2371 2373
2372 size = round_up(i_size_read(mapping->host), PAGE_SIZE); 2374 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2373 if (page->index >= size >> PAGE_SHIFT) 2375 if (page->index >= max_idx)
2374 goto unlock; 2376 goto unlock;
2375 2377
2376 if (file->f_ra.mmap_miss > 0) 2378 if (file->f_ra.mmap_miss > 0)
@@ -2718,18 +2720,16 @@ generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2718 * about to write. We do this *before* the write so that we can return 2720 * about to write. We do this *before* the write so that we can return
2719 * without clobbering -EIOCBQUEUED from ->direct_IO(). 2721 * without clobbering -EIOCBQUEUED from ->direct_IO().
2720 */ 2722 */
2721 if (mapping->nrpages) { 2723 written = invalidate_inode_pages2_range(mapping,
2722 written = invalidate_inode_pages2_range(mapping,
2723 pos >> PAGE_SHIFT, end); 2724 pos >> PAGE_SHIFT, end);
2724 /* 2725 /*
2725 * If a page can not be invalidated, return 0 to fall back 2726 * If a page can not be invalidated, return 0 to fall back
2726 * to buffered write. 2727 * to buffered write.
2727 */ 2728 */
2728 if (written) { 2729 if (written) {
2729 if (written == -EBUSY) 2730 if (written == -EBUSY)
2730 return 0; 2731 return 0;
2731 goto out; 2732 goto out;
2732 }
2733 } 2733 }
2734 2734
2735 written = mapping->a_ops->direct_IO(iocb, from); 2735 written = mapping->a_ops->direct_IO(iocb, from);
@@ -2742,10 +2742,8 @@ generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2742 * so we don't support it 100%. If this invalidation 2742 * so we don't support it 100%. If this invalidation
2743 * fails, tough, the write still worked... 2743 * fails, tough, the write still worked...
2744 */ 2744 */
2745 if (mapping->nrpages) { 2745 invalidate_inode_pages2_range(mapping,
2746 invalidate_inode_pages2_range(mapping, 2746 pos >> PAGE_SHIFT, end);
2747 pos >> PAGE_SHIFT, end);
2748 }
2749 2747
2750 if (written > 0) { 2748 if (written > 0) {
2751 pos += written; 2749 pos += written;
@@ -2793,12 +2791,6 @@ ssize_t generic_perform_write(struct file *file,
2793 ssize_t written = 0; 2791 ssize_t written = 0;
2794 unsigned int flags = 0; 2792 unsigned int flags = 0;
2795 2793
2796 /*
2797 * Copies from kernel address space cannot fail (NFSD is a big user).
2798 */
2799 if (!iter_is_iovec(i))
2800 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2801
2802 do { 2794 do {
2803 struct page *page; 2795 struct page *page;
2804 unsigned long offset; /* Offset into pagecache page */ 2796 unsigned long offset; /* Offset into pagecache page */
@@ -3000,7 +2992,7 @@ EXPORT_SYMBOL(generic_file_write_iter);
3000 * @gfp_mask: memory allocation flags (and I/O mode) 2992 * @gfp_mask: memory allocation flags (and I/O mode)
3001 * 2993 *
3002 * The address_space is to try to release any data against the page 2994 * The address_space is to try to release any data against the page
3003 * (presumably at page->private). If the release was successful, return `1'. 2995 * (presumably at page->private). If the release was successful, return '1'.
3004 * Otherwise return zero. 2996 * Otherwise return zero.
3005 * 2997 *
3006 * This may also be called if PG_fscache is set on a page, indicating that the 2998 * This may also be called if PG_fscache is set on a page, indicating that the
diff --git a/mm/frame_vector.c b/mm/frame_vector.c
index db77dcb38afd..72ebec18629c 100644
--- a/mm/frame_vector.c
+++ b/mm/frame_vector.c
@@ -200,10 +200,7 @@ struct frame_vector *frame_vector_create(unsigned int nr_frames)
200 * Avoid higher order allocations, use vmalloc instead. It should 200 * Avoid higher order allocations, use vmalloc instead. It should
201 * be rare anyway. 201 * be rare anyway.
202 */ 202 */
203 if (size <= PAGE_SIZE) 203 vec = kvmalloc(size, GFP_KERNEL);
204 vec = kmalloc(size, GFP_KERNEL);
205 else
206 vec = vmalloc(size);
207 if (!vec) 204 if (!vec)
208 return NULL; 205 return NULL;
209 vec->nr_allocated = nr_frames; 206 vec->nr_allocated = nr_frames;
diff --git a/mm/gup.c b/mm/gup.c
index 04aa405350dc..d9e6fddcc51f 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -1189,34 +1189,57 @@ struct page *get_dump_page(unsigned long addr)
1189 */ 1189 */
1190#ifdef CONFIG_HAVE_GENERIC_RCU_GUP 1190#ifdef CONFIG_HAVE_GENERIC_RCU_GUP
1191 1191
1192#ifndef gup_get_pte
1193/*
1194 * We assume that the PTE can be read atomically. If this is not the case for
1195 * your architecture, please provide the helper.
1196 */
1197static inline pte_t gup_get_pte(pte_t *ptep)
1198{
1199 return READ_ONCE(*ptep);
1200}
1201#endif
1202
1203static void undo_dev_pagemap(int *nr, int nr_start, struct page **pages)
1204{
1205 while ((*nr) - nr_start) {
1206 struct page *page = pages[--(*nr)];
1207
1208 ClearPageReferenced(page);
1209 put_page(page);
1210 }
1211}
1212
1192#ifdef __HAVE_ARCH_PTE_SPECIAL 1213#ifdef __HAVE_ARCH_PTE_SPECIAL
1193static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, 1214static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1194 int write, struct page **pages, int *nr) 1215 int write, struct page **pages, int *nr)
1195{ 1216{
1217 struct dev_pagemap *pgmap = NULL;
1218 int nr_start = *nr, ret = 0;
1196 pte_t *ptep, *ptem; 1219 pte_t *ptep, *ptem;
1197 int ret = 0;
1198 1220
1199 ptem = ptep = pte_offset_map(&pmd, addr); 1221 ptem = ptep = pte_offset_map(&pmd, addr);
1200 do { 1222 do {
1201 /* 1223 pte_t pte = gup_get_pte(ptep);
1202 * In the line below we are assuming that the pte can be read
1203 * atomically. If this is not the case for your architecture,
1204 * please wrap this in a helper function!
1205 *
1206 * for an example see gup_get_pte in arch/x86/mm/gup.c
1207 */
1208 pte_t pte = READ_ONCE(*ptep);
1209 struct page *head, *page; 1224 struct page *head, *page;
1210 1225
1211 /* 1226 /*
1212 * Similar to the PMD case below, NUMA hinting must take slow 1227 * Similar to the PMD case below, NUMA hinting must take slow
1213 * path using the pte_protnone check. 1228 * path using the pte_protnone check.
1214 */ 1229 */
1215 if (!pte_present(pte) || pte_special(pte) || 1230 if (pte_protnone(pte))
1216 pte_protnone(pte) || (write && !pte_write(pte)))
1217 goto pte_unmap; 1231 goto pte_unmap;
1218 1232
1219 if (!arch_pte_access_permitted(pte, write)) 1233 if (!pte_access_permitted(pte, write))
1234 goto pte_unmap;
1235
1236 if (pte_devmap(pte)) {
1237 pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
1238 if (unlikely(!pgmap)) {
1239 undo_dev_pagemap(nr, nr_start, pages);
1240 goto pte_unmap;
1241 }
1242 } else if (pte_special(pte))
1220 goto pte_unmap; 1243 goto pte_unmap;
1221 1244
1222 VM_BUG_ON(!pfn_valid(pte_pfn(pte))); 1245 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
@@ -1232,6 +1255,9 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1232 } 1255 }
1233 1256
1234 VM_BUG_ON_PAGE(compound_head(page) != head, page); 1257 VM_BUG_ON_PAGE(compound_head(page) != head, page);
1258
1259 put_dev_pagemap(pgmap);
1260 SetPageReferenced(page);
1235 pages[*nr] = page; 1261 pages[*nr] = page;
1236 (*nr)++; 1262 (*nr)++;
1237 1263
@@ -1261,15 +1287,76 @@ static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1261} 1287}
1262#endif /* __HAVE_ARCH_PTE_SPECIAL */ 1288#endif /* __HAVE_ARCH_PTE_SPECIAL */
1263 1289
1290#ifdef __HAVE_ARCH_PTE_DEVMAP
1291static int __gup_device_huge(unsigned long pfn, unsigned long addr,
1292 unsigned long end, struct page **pages, int *nr)
1293{
1294 int nr_start = *nr;
1295 struct dev_pagemap *pgmap = NULL;
1296
1297 do {
1298 struct page *page = pfn_to_page(pfn);
1299
1300 pgmap = get_dev_pagemap(pfn, pgmap);
1301 if (unlikely(!pgmap)) {
1302 undo_dev_pagemap(nr, nr_start, pages);
1303 return 0;
1304 }
1305 SetPageReferenced(page);
1306 pages[*nr] = page;
1307 get_page(page);
1308 put_dev_pagemap(pgmap);
1309 (*nr)++;
1310 pfn++;
1311 } while (addr += PAGE_SIZE, addr != end);
1312 return 1;
1313}
1314
1315static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
1316 unsigned long end, struct page **pages, int *nr)
1317{
1318 unsigned long fault_pfn;
1319
1320 fault_pfn = pmd_pfn(pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1321 return __gup_device_huge(fault_pfn, addr, end, pages, nr);
1322}
1323
1324static int __gup_device_huge_pud(pud_t pud, unsigned long addr,
1325 unsigned long end, struct page **pages, int *nr)
1326{
1327 unsigned long fault_pfn;
1328
1329 fault_pfn = pud_pfn(pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
1330 return __gup_device_huge(fault_pfn, addr, end, pages, nr);
1331}
1332#else
1333static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
1334 unsigned long end, struct page **pages, int *nr)
1335{
1336 BUILD_BUG();
1337 return 0;
1338}
1339
1340static int __gup_device_huge_pud(pud_t pud, unsigned long addr,
1341 unsigned long end, struct page **pages, int *nr)
1342{
1343 BUILD_BUG();
1344 return 0;
1345}
1346#endif
1347
1264static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, 1348static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
1265 unsigned long end, int write, struct page **pages, int *nr) 1349 unsigned long end, int write, struct page **pages, int *nr)
1266{ 1350{
1267 struct page *head, *page; 1351 struct page *head, *page;
1268 int refs; 1352 int refs;
1269 1353
1270 if (write && !pmd_write(orig)) 1354 if (!pmd_access_permitted(orig, write))
1271 return 0; 1355 return 0;
1272 1356
1357 if (pmd_devmap(orig))
1358 return __gup_device_huge_pmd(orig, addr, end, pages, nr);
1359
1273 refs = 0; 1360 refs = 0;
1274 head = pmd_page(orig); 1361 head = pmd_page(orig);
1275 page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT); 1362 page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
@@ -1293,6 +1380,7 @@ static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
1293 return 0; 1380 return 0;
1294 } 1381 }
1295 1382
1383 SetPageReferenced(head);
1296 return 1; 1384 return 1;
1297} 1385}
1298 1386
@@ -1302,9 +1390,12 @@ static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
1302 struct page *head, *page; 1390 struct page *head, *page;
1303 int refs; 1391 int refs;
1304 1392
1305 if (write && !pud_write(orig)) 1393 if (!pud_access_permitted(orig, write))
1306 return 0; 1394 return 0;
1307 1395
1396 if (pud_devmap(orig))
1397 return __gup_device_huge_pud(orig, addr, end, pages, nr);
1398
1308 refs = 0; 1399 refs = 0;
1309 head = pud_page(orig); 1400 head = pud_page(orig);
1310 page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT); 1401 page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
@@ -1328,6 +1419,7 @@ static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
1328 return 0; 1419 return 0;
1329 } 1420 }
1330 1421
1422 SetPageReferenced(head);
1331 return 1; 1423 return 1;
1332} 1424}
1333 1425
@@ -1338,9 +1430,10 @@ static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
1338 int refs; 1430 int refs;
1339 struct page *head, *page; 1431 struct page *head, *page;
1340 1432
1341 if (write && !pgd_write(orig)) 1433 if (!pgd_access_permitted(orig, write))
1342 return 0; 1434 return 0;
1343 1435
1436 BUILD_BUG_ON(pgd_devmap(orig));
1344 refs = 0; 1437 refs = 0;
1345 head = pgd_page(orig); 1438 head = pgd_page(orig);
1346 page = head + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT); 1439 page = head + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT);
@@ -1364,6 +1457,7 @@ static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
1364 return 0; 1457 return 0;
1365 } 1458 }
1366 1459
1460 SetPageReferenced(head);
1367 return 1; 1461 return 1;
1368} 1462}
1369 1463
@@ -1481,7 +1575,7 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
1481 end = start + len; 1575 end = start + len;
1482 1576
1483 if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ, 1577 if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
1484 start, len))) 1578 (void __user *)start, len)))
1485 return 0; 1579 return 0;
1486 1580
1487 /* 1581 /*
@@ -1520,6 +1614,21 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
1520 return nr; 1614 return nr;
1521} 1615}
1522 1616
1617#ifndef gup_fast_permitted
1618/*
1619 * Check if it's allowed to use __get_user_pages_fast() for the range, or
1620 * we need to fall back to the slow version:
1621 */
1622bool gup_fast_permitted(unsigned long start, int nr_pages, int write)
1623{
1624 unsigned long len, end;
1625
1626 len = (unsigned long) nr_pages << PAGE_SHIFT;
1627 end = start + len;
1628 return end >= start;
1629}
1630#endif
1631
1523/** 1632/**
1524 * get_user_pages_fast() - pin user pages in memory 1633 * get_user_pages_fast() - pin user pages in memory
1525 * @start: starting user address 1634 * @start: starting user address
@@ -1539,11 +1648,14 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
1539int get_user_pages_fast(unsigned long start, int nr_pages, int write, 1648int get_user_pages_fast(unsigned long start, int nr_pages, int write,
1540 struct page **pages) 1649 struct page **pages)
1541{ 1650{
1542 int nr, ret; 1651 int nr = 0, ret = 0;
1543 1652
1544 start &= PAGE_MASK; 1653 start &= PAGE_MASK;
1545 nr = __get_user_pages_fast(start, nr_pages, write, pages); 1654
1546 ret = nr; 1655 if (gup_fast_permitted(start, nr_pages, write)) {
1656 nr = __get_user_pages_fast(start, nr_pages, write, pages);
1657 ret = nr;
1658 }
1547 1659
1548 if (nr < nr_pages) { 1660 if (nr < nr_pages) {
1549 /* Try to get the remaining pages with get_user_pages */ 1661 /* Try to get the remaining pages with get_user_pages */
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index f3c4f9d22821..a84909cf20d3 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -715,7 +715,8 @@ int do_huge_pmd_anonymous_page(struct vm_fault *vmf)
715} 715}
716 716
717static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr, 717static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
718 pmd_t *pmd, pfn_t pfn, pgprot_t prot, bool write) 718 pmd_t *pmd, pfn_t pfn, pgprot_t prot, bool write,
719 pgtable_t pgtable)
719{ 720{
720 struct mm_struct *mm = vma->vm_mm; 721 struct mm_struct *mm = vma->vm_mm;
721 pmd_t entry; 722 pmd_t entry;
@@ -729,6 +730,12 @@ static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
729 entry = pmd_mkyoung(pmd_mkdirty(entry)); 730 entry = pmd_mkyoung(pmd_mkdirty(entry));
730 entry = maybe_pmd_mkwrite(entry, vma); 731 entry = maybe_pmd_mkwrite(entry, vma);
731 } 732 }
733
734 if (pgtable) {
735 pgtable_trans_huge_deposit(mm, pmd, pgtable);
736 atomic_long_inc(&mm->nr_ptes);
737 }
738
732 set_pmd_at(mm, addr, pmd, entry); 739 set_pmd_at(mm, addr, pmd, entry);
733 update_mmu_cache_pmd(vma, addr, pmd); 740 update_mmu_cache_pmd(vma, addr, pmd);
734 spin_unlock(ptl); 741 spin_unlock(ptl);
@@ -738,6 +745,7 @@ int vmf_insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
738 pmd_t *pmd, pfn_t pfn, bool write) 745 pmd_t *pmd, pfn_t pfn, bool write)
739{ 746{
740 pgprot_t pgprot = vma->vm_page_prot; 747 pgprot_t pgprot = vma->vm_page_prot;
748 pgtable_t pgtable = NULL;
741 /* 749 /*
742 * If we had pmd_special, we could avoid all these restrictions, 750 * If we had pmd_special, we could avoid all these restrictions,
743 * but we need to be consistent with PTEs and architectures that 751 * but we need to be consistent with PTEs and architectures that
@@ -752,9 +760,15 @@ int vmf_insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
752 if (addr < vma->vm_start || addr >= vma->vm_end) 760 if (addr < vma->vm_start || addr >= vma->vm_end)
753 return VM_FAULT_SIGBUS; 761 return VM_FAULT_SIGBUS;
754 762
763 if (arch_needs_pgtable_deposit()) {
764 pgtable = pte_alloc_one(vma->vm_mm, addr);
765 if (!pgtable)
766 return VM_FAULT_OOM;
767 }
768
755 track_pfn_insert(vma, &pgprot, pfn); 769 track_pfn_insert(vma, &pgprot, pfn);
756 770
757 insert_pfn_pmd(vma, addr, pmd, pfn, pgprot, write); 771 insert_pfn_pmd(vma, addr, pmd, pfn, pgprot, write, pgtable);
758 return VM_FAULT_NOPAGE; 772 return VM_FAULT_NOPAGE;
759} 773}
760EXPORT_SYMBOL_GPL(vmf_insert_pfn_pmd); 774EXPORT_SYMBOL_GPL(vmf_insert_pfn_pmd);
@@ -1564,9 +1578,6 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
1564 ClearPageDirty(page); 1578 ClearPageDirty(page);
1565 unlock_page(page); 1579 unlock_page(page);
1566 1580
1567 if (PageActive(page))
1568 deactivate_page(page);
1569
1570 if (pmd_young(orig_pmd) || pmd_dirty(orig_pmd)) { 1581 if (pmd_young(orig_pmd) || pmd_dirty(orig_pmd)) {
1571 pmdp_invalidate(vma, addr, pmd); 1582 pmdp_invalidate(vma, addr, pmd);
1572 orig_pmd = pmd_mkold(orig_pmd); 1583 orig_pmd = pmd_mkold(orig_pmd);
@@ -1575,6 +1586,8 @@ bool madvise_free_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
1575 set_pmd_at(mm, addr, pmd, orig_pmd); 1586 set_pmd_at(mm, addr, pmd, orig_pmd);
1576 tlb_remove_pmd_tlb_entry(tlb, pmd, addr); 1587 tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
1577 } 1588 }
1589
1590 mark_page_lazyfree(page);
1578 ret = true; 1591 ret = true;
1579out: 1592out:
1580 spin_unlock(ptl); 1593 spin_unlock(ptl);
@@ -1612,12 +1625,13 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
1612 tlb->fullmm); 1625 tlb->fullmm);
1613 tlb_remove_pmd_tlb_entry(tlb, pmd, addr); 1626 tlb_remove_pmd_tlb_entry(tlb, pmd, addr);
1614 if (vma_is_dax(vma)) { 1627 if (vma_is_dax(vma)) {
1628 if (arch_needs_pgtable_deposit())
1629 zap_deposited_table(tlb->mm, pmd);
1615 spin_unlock(ptl); 1630 spin_unlock(ptl);
1616 if (is_huge_zero_pmd(orig_pmd)) 1631 if (is_huge_zero_pmd(orig_pmd))
1617 tlb_remove_page_size(tlb, pmd_page(orig_pmd), HPAGE_PMD_SIZE); 1632 tlb_remove_page_size(tlb, pmd_page(orig_pmd), HPAGE_PMD_SIZE);
1618 } else if (is_huge_zero_pmd(orig_pmd)) { 1633 } else if (is_huge_zero_pmd(orig_pmd)) {
1619 pte_free(tlb->mm, pgtable_trans_huge_withdraw(tlb->mm, pmd)); 1634 zap_deposited_table(tlb->mm, pmd);
1620 atomic_long_dec(&tlb->mm->nr_ptes);
1621 spin_unlock(ptl); 1635 spin_unlock(ptl);
1622 tlb_remove_page_size(tlb, pmd_page(orig_pmd), HPAGE_PMD_SIZE); 1636 tlb_remove_page_size(tlb, pmd_page(orig_pmd), HPAGE_PMD_SIZE);
1623 } else { 1637 } else {
@@ -1626,10 +1640,7 @@ int zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
1626 VM_BUG_ON_PAGE(page_mapcount(page) < 0, page); 1640 VM_BUG_ON_PAGE(page_mapcount(page) < 0, page);
1627 VM_BUG_ON_PAGE(!PageHead(page), page); 1641 VM_BUG_ON_PAGE(!PageHead(page), page);
1628 if (PageAnon(page)) { 1642 if (PageAnon(page)) {
1629 pgtable_t pgtable; 1643 zap_deposited_table(tlb->mm, pmd);
1630 pgtable = pgtable_trans_huge_withdraw(tlb->mm, pmd);
1631 pte_free(tlb->mm, pgtable);
1632 atomic_long_dec(&tlb->mm->nr_ptes);
1633 add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR); 1644 add_mm_counter(tlb->mm, MM_ANONPAGES, -HPAGE_PMD_NR);
1634 } else { 1645 } else {
1635 if (arch_needs_pgtable_deposit()) 1646 if (arch_needs_pgtable_deposit())
@@ -2145,15 +2156,15 @@ static void freeze_page(struct page *page)
2145{ 2156{
2146 enum ttu_flags ttu_flags = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS | 2157 enum ttu_flags ttu_flags = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS |
2147 TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD; 2158 TTU_RMAP_LOCKED | TTU_SPLIT_HUGE_PMD;
2148 int ret; 2159 bool unmap_success;
2149 2160
2150 VM_BUG_ON_PAGE(!PageHead(page), page); 2161 VM_BUG_ON_PAGE(!PageHead(page), page);
2151 2162
2152 if (PageAnon(page)) 2163 if (PageAnon(page))
2153 ttu_flags |= TTU_MIGRATION; 2164 ttu_flags |= TTU_MIGRATION;
2154 2165
2155 ret = try_to_unmap(page, ttu_flags); 2166 unmap_success = try_to_unmap(page, ttu_flags);
2156 VM_BUG_ON_PAGE(ret, page); 2167 VM_BUG_ON_PAGE(!unmap_success, page);
2157} 2168}
2158 2169
2159static void unfreeze_page(struct page *page) 2170static void unfreeze_page(struct page *page)
@@ -2399,7 +2410,6 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
2399 2410
2400 VM_BUG_ON_PAGE(is_huge_zero_page(page), page); 2411 VM_BUG_ON_PAGE(is_huge_zero_page(page), page);
2401 VM_BUG_ON_PAGE(!PageLocked(page), page); 2412 VM_BUG_ON_PAGE(!PageLocked(page), page);
2402 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
2403 VM_BUG_ON_PAGE(!PageCompound(page), page); 2413 VM_BUG_ON_PAGE(!PageCompound(page), page);
2404 2414
2405 if (PageAnon(head)) { 2415 if (PageAnon(head)) {
diff --git a/mm/hwpoison-inject.c b/mm/hwpoison-inject.c
index 9d26fd9fefe4..356df057a2a8 100644
--- a/mm/hwpoison-inject.c
+++ b/mm/hwpoison-inject.c
@@ -34,8 +34,7 @@ static int hwpoison_inject(void *data, u64 val)
34 if (!hwpoison_filter_enable) 34 if (!hwpoison_filter_enable)
35 goto inject; 35 goto inject;
36 36
37 if (!PageLRU(hpage) && !PageHuge(p)) 37 shake_page(hpage, 0);
38 shake_page(hpage, 0);
39 /* 38 /*
40 * This implies unable to support non-LRU pages. 39 * This implies unable to support non-LRU pages.
41 */ 40 */
diff --git a/mm/internal.h b/mm/internal.h
index 266efaeaa370..0e4f558412fb 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -81,11 +81,16 @@ static inline void set_page_refcounted(struct page *page)
81extern unsigned long highest_memmap_pfn; 81extern unsigned long highest_memmap_pfn;
82 82
83/* 83/*
84 * Maximum number of reclaim retries without progress before the OOM
85 * killer is consider the only way forward.
86 */
87#define MAX_RECLAIM_RETRIES 16
88
89/*
84 * in mm/vmscan.c: 90 * in mm/vmscan.c:
85 */ 91 */
86extern int isolate_lru_page(struct page *page); 92extern int isolate_lru_page(struct page *page);
87extern void putback_lru_page(struct page *page); 93extern void putback_lru_page(struct page *page);
88extern bool pgdat_reclaimable(struct pglist_data *pgdat);
89 94
90/* 95/*
91 * in mm/rmap.c: 96 * in mm/rmap.c:
@@ -178,6 +183,7 @@ extern int user_min_free_kbytes;
178struct compact_control { 183struct compact_control {
179 struct list_head freepages; /* List of free pages to migrate to */ 184 struct list_head freepages; /* List of free pages to migrate to */
180 struct list_head migratepages; /* List of pages being migrated */ 185 struct list_head migratepages; /* List of pages being migrated */
186 struct zone *zone;
181 unsigned long nr_freepages; /* Number of isolated free pages */ 187 unsigned long nr_freepages; /* Number of isolated free pages */
182 unsigned long nr_migratepages; /* Number of pages to migrate */ 188 unsigned long nr_migratepages; /* Number of pages to migrate */
183 unsigned long total_migrate_scanned; 189 unsigned long total_migrate_scanned;
@@ -185,17 +191,18 @@ struct compact_control {
185 unsigned long free_pfn; /* isolate_freepages search base */ 191 unsigned long free_pfn; /* isolate_freepages search base */
186 unsigned long migrate_pfn; /* isolate_migratepages search base */ 192 unsigned long migrate_pfn; /* isolate_migratepages search base */
187 unsigned long last_migrated_pfn;/* Not yet flushed page being freed */ 193 unsigned long last_migrated_pfn;/* Not yet flushed page being freed */
194 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
195 int order; /* order a direct compactor needs */
196 int migratetype; /* migratetype of direct compactor */
197 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
198 const int classzone_idx; /* zone index of a direct compactor */
188 enum migrate_mode mode; /* Async or sync migration mode */ 199 enum migrate_mode mode; /* Async or sync migration mode */
189 bool ignore_skip_hint; /* Scan blocks even if marked skip */ 200 bool ignore_skip_hint; /* Scan blocks even if marked skip */
190 bool ignore_block_suitable; /* Scan blocks considered unsuitable */ 201 bool ignore_block_suitable; /* Scan blocks considered unsuitable */
191 bool direct_compaction; /* False from kcompactd or /proc/... */ 202 bool direct_compaction; /* False from kcompactd or /proc/... */
192 bool whole_zone; /* Whole zone should/has been scanned */ 203 bool whole_zone; /* Whole zone should/has been scanned */
193 int order; /* order a direct compactor needs */
194 const gfp_t gfp_mask; /* gfp mask of a direct compactor */
195 const unsigned int alloc_flags; /* alloc flags of a direct compactor */
196 const int classzone_idx; /* zone index of a direct compactor */
197 struct zone *zone;
198 bool contended; /* Signal lock or sched contention */ 204 bool contended; /* Signal lock or sched contention */
205 bool finishing_block; /* Finishing current pageblock */
199}; 206};
200 207
201unsigned long 208unsigned long
@@ -505,4 +512,14 @@ extern const struct trace_print_flags pageflag_names[];
505extern const struct trace_print_flags vmaflag_names[]; 512extern const struct trace_print_flags vmaflag_names[];
506extern const struct trace_print_flags gfpflag_names[]; 513extern const struct trace_print_flags gfpflag_names[];
507 514
515static inline bool is_migrate_highatomic(enum migratetype migratetype)
516{
517 return migratetype == MIGRATE_HIGHATOMIC;
518}
519
520static inline bool is_migrate_highatomic_page(struct page *page)
521{
522 return get_pageblock_migratetype(page) == MIGRATE_HIGHATOMIC;
523}
524
508#endif /* __MM_INTERNAL_H */ 525#endif /* __MM_INTERNAL_H */
diff --git a/mm/kasan/kasan.c b/mm/kasan/kasan.c
index 98b27195e38b..b10da59cf765 100644
--- a/mm/kasan/kasan.c
+++ b/mm/kasan/kasan.c
@@ -577,7 +577,8 @@ bool kasan_slab_free(struct kmem_cache *cache, void *object)
577 577
578 shadow_byte = READ_ONCE(*(s8 *)kasan_mem_to_shadow(object)); 578 shadow_byte = READ_ONCE(*(s8 *)kasan_mem_to_shadow(object));
579 if (shadow_byte < 0 || shadow_byte >= KASAN_SHADOW_SCALE_SIZE) { 579 if (shadow_byte < 0 || shadow_byte >= KASAN_SHADOW_SCALE_SIZE) {
580 kasan_report_double_free(cache, object, shadow_byte); 580 kasan_report_double_free(cache, object,
581 __builtin_return_address(1));
581 return true; 582 return true;
582 } 583 }
583 584
@@ -690,7 +691,7 @@ int kasan_module_alloc(void *addr, size_t size)
690 691
691 ret = __vmalloc_node_range(shadow_size, 1, shadow_start, 692 ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
692 shadow_start + shadow_size, 693 shadow_start + shadow_size,
693 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 694 GFP_KERNEL | __GFP_ZERO,
694 PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE, 695 PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
695 __builtin_return_address(0)); 696 __builtin_return_address(0));
696 697
diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
index dd2dea8eb077..1229298cce64 100644
--- a/mm/kasan/kasan.h
+++ b/mm/kasan/kasan.h
@@ -99,7 +99,7 @@ static inline const void *kasan_shadow_to_mem(const void *shadow_addr)
99void kasan_report(unsigned long addr, size_t size, 99void kasan_report(unsigned long addr, size_t size,
100 bool is_write, unsigned long ip); 100 bool is_write, unsigned long ip);
101void kasan_report_double_free(struct kmem_cache *cache, void *object, 101void kasan_report_double_free(struct kmem_cache *cache, void *object,
102 s8 shadow); 102 void *ip);
103 103
104#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB) 104#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB)
105void quarantine_put(struct kasan_free_meta *info, struct kmem_cache *cache); 105void quarantine_put(struct kasan_free_meta *info, struct kmem_cache *cache);
diff --git a/mm/kasan/report.c b/mm/kasan/report.c
index ab42a0803f16..beee0e980e2d 100644
--- a/mm/kasan/report.c
+++ b/mm/kasan/report.c
@@ -51,7 +51,13 @@ static const void *find_first_bad_addr(const void *addr, size_t size)
51 return first_bad_addr; 51 return first_bad_addr;
52} 52}
53 53
54static void print_error_description(struct kasan_access_info *info) 54static bool addr_has_shadow(struct kasan_access_info *info)
55{
56 return (info->access_addr >=
57 kasan_shadow_to_mem((void *)KASAN_SHADOW_START));
58}
59
60static const char *get_shadow_bug_type(struct kasan_access_info *info)
55{ 61{
56 const char *bug_type = "unknown-crash"; 62 const char *bug_type = "unknown-crash";
57 u8 *shadow_addr; 63 u8 *shadow_addr;
@@ -98,12 +104,39 @@ static void print_error_description(struct kasan_access_info *info)
98 break; 104 break;
99 } 105 }
100 106
101 pr_err("BUG: KASAN: %s in %pS at addr %p\n", 107 return bug_type;
102 bug_type, (void *)info->ip, 108}
103 info->access_addr); 109
104 pr_err("%s of size %zu by task %s/%d\n", 110const char *get_wild_bug_type(struct kasan_access_info *info)
105 info->is_write ? "Write" : "Read", 111{
106 info->access_size, current->comm, task_pid_nr(current)); 112 const char *bug_type = "unknown-crash";
113
114 if ((unsigned long)info->access_addr < PAGE_SIZE)
115 bug_type = "null-ptr-deref";
116 else if ((unsigned long)info->access_addr < TASK_SIZE)
117 bug_type = "user-memory-access";
118 else
119 bug_type = "wild-memory-access";
120
121 return bug_type;
122}
123
124static const char *get_bug_type(struct kasan_access_info *info)
125{
126 if (addr_has_shadow(info))
127 return get_shadow_bug_type(info);
128 return get_wild_bug_type(info);
129}
130
131static void print_error_description(struct kasan_access_info *info)
132{
133 const char *bug_type = get_bug_type(info);
134
135 pr_err("BUG: KASAN: %s in %pS\n",
136 bug_type, (void *)info->ip);
137 pr_err("%s of size %zu at addr %p by task %s/%d\n",
138 info->is_write ? "Write" : "Read", info->access_size,
139 info->access_addr, current->comm, task_pid_nr(current));
107} 140}
108 141
109static inline bool kernel_or_module_addr(const void *addr) 142static inline bool kernel_or_module_addr(const void *addr)
@@ -144,9 +177,9 @@ static void kasan_end_report(unsigned long *flags)
144 kasan_enable_current(); 177 kasan_enable_current();
145} 178}
146 179
147static void print_track(struct kasan_track *track) 180static void print_track(struct kasan_track *track, const char *prefix)
148{ 181{
149 pr_err("PID = %u\n", track->pid); 182 pr_err("%s by task %u:\n", prefix, track->pid);
150 if (track->stack) { 183 if (track->stack) {
151 struct stack_trace trace; 184 struct stack_trace trace;
152 185
@@ -157,59 +190,84 @@ static void print_track(struct kasan_track *track)
157 } 190 }
158} 191}
159 192
160static void kasan_object_err(struct kmem_cache *cache, void *object) 193static struct page *addr_to_page(const void *addr)
161{ 194{
162 struct kasan_alloc_meta *alloc_info = get_alloc_info(cache, object); 195 if ((addr >= (void *)PAGE_OFFSET) &&
196 (addr < high_memory))
197 return virt_to_head_page(addr);
198 return NULL;
199}
163 200
164 dump_stack(); 201static void describe_object_addr(struct kmem_cache *cache, void *object,
165 pr_err("Object at %p, in cache %s size: %d\n", object, cache->name, 202 const void *addr)
166 cache->object_size); 203{
204 unsigned long access_addr = (unsigned long)addr;
205 unsigned long object_addr = (unsigned long)object;
206 const char *rel_type;
207 int rel_bytes;
167 208
168 if (!(cache->flags & SLAB_KASAN)) 209 pr_err("The buggy address belongs to the object at %p\n"
210 " which belongs to the cache %s of size %d\n",
211 object, cache->name, cache->object_size);
212
213 if (!addr)
169 return; 214 return;
170 215
171 pr_err("Allocated:\n"); 216 if (access_addr < object_addr) {
172 print_track(&alloc_info->alloc_track); 217 rel_type = "to the left";
173 pr_err("Freed:\n"); 218 rel_bytes = object_addr - access_addr;
174 print_track(&alloc_info->free_track); 219 } else if (access_addr >= object_addr + cache->object_size) {
220 rel_type = "to the right";
221 rel_bytes = access_addr - (object_addr + cache->object_size);
222 } else {
223 rel_type = "inside";
224 rel_bytes = access_addr - object_addr;
225 }
226
227 pr_err("The buggy address is located %d bytes %s of\n"
228 " %d-byte region [%p, %p)\n",
229 rel_bytes, rel_type, cache->object_size, (void *)object_addr,
230 (void *)(object_addr + cache->object_size));
175} 231}
176 232
177void kasan_report_double_free(struct kmem_cache *cache, void *object, 233static void describe_object(struct kmem_cache *cache, void *object,
178 s8 shadow) 234 const void *addr)
179{ 235{
180 unsigned long flags; 236 struct kasan_alloc_meta *alloc_info = get_alloc_info(cache, object);
181 237
182 kasan_start_report(&flags); 238 if (cache->flags & SLAB_KASAN) {
183 pr_err("BUG: Double free or freeing an invalid pointer\n"); 239 print_track(&alloc_info->alloc_track, "Allocated");
184 pr_err("Unexpected shadow byte: 0x%hhX\n", shadow); 240 pr_err("\n");
185 kasan_object_err(cache, object); 241 print_track(&alloc_info->free_track, "Freed");
186 kasan_end_report(&flags); 242 pr_err("\n");
243 }
244
245 describe_object_addr(cache, object, addr);
187} 246}
188 247
189static void print_address_description(struct kasan_access_info *info) 248static void print_address_description(void *addr)
190{ 249{
191 const void *addr = info->access_addr; 250 struct page *page = addr_to_page(addr);
192 251
193 if ((addr >= (void *)PAGE_OFFSET) && 252 dump_stack();
194 (addr < high_memory)) { 253 pr_err("\n");
195 struct page *page = virt_to_head_page(addr); 254
196 255 if (page && PageSlab(page)) {
197 if (PageSlab(page)) { 256 struct kmem_cache *cache = page->slab_cache;
198 void *object; 257 void *object = nearest_obj(cache, page, addr);
199 struct kmem_cache *cache = page->slab_cache; 258
200 object = nearest_obj(cache, page, 259 describe_object(cache, object, addr);
201 (void *)info->access_addr);
202 kasan_object_err(cache, object);
203 return;
204 }
205 dump_page(page, "kasan: bad access detected");
206 } 260 }
207 261
208 if (kernel_or_module_addr(addr)) { 262 if (kernel_or_module_addr(addr) && !init_task_stack_addr(addr)) {
209 if (!init_task_stack_addr(addr)) 263 pr_err("The buggy address belongs to the variable:\n");
210 pr_err("Address belongs to variable %pS\n", addr); 264 pr_err(" %pS\n", addr);
265 }
266
267 if (page) {
268 pr_err("The buggy address belongs to the page:\n");
269 dump_page(page, "kasan: bad access detected");
211 } 270 }
212 dump_stack();
213} 271}
214 272
215static bool row_is_guilty(const void *row, const void *guilty) 273static bool row_is_guilty(const void *row, const void *guilty)
@@ -264,31 +322,34 @@ static void print_shadow_for_address(const void *addr)
264 } 322 }
265} 323}
266 324
325void kasan_report_double_free(struct kmem_cache *cache, void *object,
326 void *ip)
327{
328 unsigned long flags;
329
330 kasan_start_report(&flags);
331 pr_err("BUG: KASAN: double-free or invalid-free in %pS\n", ip);
332 pr_err("\n");
333 print_address_description(object);
334 pr_err("\n");
335 print_shadow_for_address(object);
336 kasan_end_report(&flags);
337}
338
267static void kasan_report_error(struct kasan_access_info *info) 339static void kasan_report_error(struct kasan_access_info *info)
268{ 340{
269 unsigned long flags; 341 unsigned long flags;
270 const char *bug_type;
271 342
272 kasan_start_report(&flags); 343 kasan_start_report(&flags);
273 344
274 if (info->access_addr < 345 print_error_description(info);
275 kasan_shadow_to_mem((void *)KASAN_SHADOW_START)) { 346 pr_err("\n");
276 if ((unsigned long)info->access_addr < PAGE_SIZE) 347
277 bug_type = "null-ptr-deref"; 348 if (!addr_has_shadow(info)) {
278 else if ((unsigned long)info->access_addr < TASK_SIZE)
279 bug_type = "user-memory-access";
280 else
281 bug_type = "wild-memory-access";
282 pr_err("BUG: KASAN: %s on address %p\n",
283 bug_type, info->access_addr);
284 pr_err("%s of size %zu by task %s/%d\n",
285 info->is_write ? "Write" : "Read",
286 info->access_size, current->comm,
287 task_pid_nr(current));
288 dump_stack(); 349 dump_stack();
289 } else { 350 } else {
290 print_error_description(info); 351 print_address_description((void *)info->access_addr);
291 print_address_description(info); 352 pr_err("\n");
292 print_shadow_for_address(info->first_bad_addr); 353 print_shadow_for_address(info->first_bad_addr);
293 } 354 }
294 355
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index ba40b7f673f4..7cb9c88bb4a3 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -483,8 +483,7 @@ void __khugepaged_exit(struct mm_struct *mm)
483 483
484static void release_pte_page(struct page *page) 484static void release_pte_page(struct page *page)
485{ 485{
486 /* 0 stands for page_is_file_cache(page) == false */ 486 dec_node_page_state(page, NR_ISOLATED_ANON + page_is_file_cache(page));
487 dec_node_page_state(page, NR_ISOLATED_ANON + 0);
488 unlock_page(page); 487 unlock_page(page);
489 putback_lru_page(page); 488 putback_lru_page(page);
490} 489}
@@ -532,7 +531,6 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
532 531
533 VM_BUG_ON_PAGE(PageCompound(page), page); 532 VM_BUG_ON_PAGE(PageCompound(page), page);
534 VM_BUG_ON_PAGE(!PageAnon(page), page); 533 VM_BUG_ON_PAGE(!PageAnon(page), page);
535 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
536 534
537 /* 535 /*
538 * We can do it before isolate_lru_page because the 536 * We can do it before isolate_lru_page because the
@@ -550,7 +548,7 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
550 * The page must only be referenced by the scanned process 548 * The page must only be referenced by the scanned process
551 * and page swap cache. 549 * and page swap cache.
552 */ 550 */
553 if (page_count(page) != 1 + !!PageSwapCache(page)) { 551 if (page_count(page) != 1 + PageSwapCache(page)) {
554 unlock_page(page); 552 unlock_page(page);
555 result = SCAN_PAGE_COUNT; 553 result = SCAN_PAGE_COUNT;
556 goto out; 554 goto out;
@@ -579,8 +577,8 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
579 result = SCAN_DEL_PAGE_LRU; 577 result = SCAN_DEL_PAGE_LRU;
580 goto out; 578 goto out;
581 } 579 }
582 /* 0 stands for page_is_file_cache(page) == false */ 580 inc_node_page_state(page,
583 inc_node_page_state(page, NR_ISOLATED_ANON + 0); 581 NR_ISOLATED_ANON + page_is_file_cache(page));
584 VM_BUG_ON_PAGE(!PageLocked(page), page); 582 VM_BUG_ON_PAGE(!PageLocked(page), page);
585 VM_BUG_ON_PAGE(PageLRU(page), page); 583 VM_BUG_ON_PAGE(PageLRU(page), page);
586 584
@@ -1183,7 +1181,7 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
1183 * The page must only be referenced by the scanned process 1181 * The page must only be referenced by the scanned process
1184 * and page swap cache. 1182 * and page swap cache.
1185 */ 1183 */
1186 if (page_count(page) != 1 + !!PageSwapCache(page)) { 1184 if (page_count(page) != 1 + PageSwapCache(page)) {
1187 result = SCAN_PAGE_COUNT; 1185 result = SCAN_PAGE_COUNT;
1188 goto out_unmap; 1186 goto out_unmap;
1189 } 1187 }
diff --git a/mm/ksm.c b/mm/ksm.c
index 19b4f2dea7a5..d9fc0e456128 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -1933,11 +1933,10 @@ struct page *ksm_might_need_to_copy(struct page *page,
1933 return new_page; 1933 return new_page;
1934} 1934}
1935 1935
1936int rmap_walk_ksm(struct page *page, struct rmap_walk_control *rwc) 1936void rmap_walk_ksm(struct page *page, struct rmap_walk_control *rwc)
1937{ 1937{
1938 struct stable_node *stable_node; 1938 struct stable_node *stable_node;
1939 struct rmap_item *rmap_item; 1939 struct rmap_item *rmap_item;
1940 int ret = SWAP_AGAIN;
1941 int search_new_forks = 0; 1940 int search_new_forks = 0;
1942 1941
1943 VM_BUG_ON_PAGE(!PageKsm(page), page); 1942 VM_BUG_ON_PAGE(!PageKsm(page), page);
@@ -1950,7 +1949,7 @@ int rmap_walk_ksm(struct page *page, struct rmap_walk_control *rwc)
1950 1949
1951 stable_node = page_stable_node(page); 1950 stable_node = page_stable_node(page);
1952 if (!stable_node) 1951 if (!stable_node)
1953 return ret; 1952 return;
1954again: 1953again:
1955 hlist_for_each_entry(rmap_item, &stable_node->hlist, hlist) { 1954 hlist_for_each_entry(rmap_item, &stable_node->hlist, hlist) {
1956 struct anon_vma *anon_vma = rmap_item->anon_vma; 1955 struct anon_vma *anon_vma = rmap_item->anon_vma;
@@ -1978,23 +1977,20 @@ again:
1978 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) 1977 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1979 continue; 1978 continue;
1980 1979
1981 ret = rwc->rmap_one(page, vma, 1980 if (!rwc->rmap_one(page, vma,
1982 rmap_item->address, rwc->arg); 1981 rmap_item->address, rwc->arg)) {
1983 if (ret != SWAP_AGAIN) {
1984 anon_vma_unlock_read(anon_vma); 1982 anon_vma_unlock_read(anon_vma);
1985 goto out; 1983 return;
1986 } 1984 }
1987 if (rwc->done && rwc->done(page)) { 1985 if (rwc->done && rwc->done(page)) {
1988 anon_vma_unlock_read(anon_vma); 1986 anon_vma_unlock_read(anon_vma);
1989 goto out; 1987 return;
1990 } 1988 }
1991 } 1989 }
1992 anon_vma_unlock_read(anon_vma); 1990 anon_vma_unlock_read(anon_vma);
1993 } 1991 }
1994 if (!search_new_forks++) 1992 if (!search_new_forks++)
1995 goto again; 1993 goto again;
1996out:
1997 return ret;
1998} 1994}
1999 1995
2000#ifdef CONFIG_MIGRATION 1996#ifdef CONFIG_MIGRATION
diff --git a/mm/madvise.c b/mm/madvise.c
index 7a2abf0127ae..25b78ee4fc2c 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -411,10 +411,9 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
411 ptent = pte_mkold(ptent); 411 ptent = pte_mkold(ptent);
412 ptent = pte_mkclean(ptent); 412 ptent = pte_mkclean(ptent);
413 set_pte_at(mm, addr, pte, ptent); 413 set_pte_at(mm, addr, pte, ptent);
414 if (PageActive(page))
415 deactivate_page(page);
416 tlb_remove_tlb_entry(tlb, pte, addr); 414 tlb_remove_tlb_entry(tlb, pte, addr);
417 } 415 }
416 mark_page_lazyfree(page);
418 } 417 }
419out: 418out:
420 if (nr_swap) { 419 if (nr_swap) {
@@ -606,34 +605,40 @@ static long madvise_remove(struct vm_area_struct *vma,
606/* 605/*
607 * Error injection support for memory error handling. 606 * Error injection support for memory error handling.
608 */ 607 */
609static int madvise_hwpoison(int bhv, unsigned long start, unsigned long end) 608static int madvise_inject_error(int behavior,
609 unsigned long start, unsigned long end)
610{ 610{
611 struct page *p; 611 struct page *page;
612
612 if (!capable(CAP_SYS_ADMIN)) 613 if (!capable(CAP_SYS_ADMIN))
613 return -EPERM; 614 return -EPERM;
615
614 for (; start < end; start += PAGE_SIZE << 616 for (; start < end; start += PAGE_SIZE <<
615 compound_order(compound_head(p))) { 617 compound_order(compound_head(page))) {
616 int ret; 618 int ret;
617 619
618 ret = get_user_pages_fast(start, 1, 0, &p); 620 ret = get_user_pages_fast(start, 1, 0, &page);
619 if (ret != 1) 621 if (ret != 1)
620 return ret; 622 return ret;
621 623
622 if (PageHWPoison(p)) { 624 if (PageHWPoison(page)) {
623 put_page(p); 625 put_page(page);
624 continue; 626 continue;
625 } 627 }
626 if (bhv == MADV_SOFT_OFFLINE) { 628
627 pr_info("Soft offlining page %#lx at %#lx\n", 629 if (behavior == MADV_SOFT_OFFLINE) {
628 page_to_pfn(p), start); 630 pr_info("Soft offlining pfn %#lx at process virtual address %#lx\n",
629 ret = soft_offline_page(p, MF_COUNT_INCREASED); 631 page_to_pfn(page), start);
632
633 ret = soft_offline_page(page, MF_COUNT_INCREASED);
630 if (ret) 634 if (ret)
631 return ret; 635 return ret;
632 continue; 636 continue;
633 } 637 }
634 pr_info("Injecting memory failure for page %#lx at %#lx\n", 638 pr_info("Injecting memory failure for pfn %#lx at process virtual address %#lx\n",
635 page_to_pfn(p), start); 639 page_to_pfn(page), start);
636 ret = memory_failure(page_to_pfn(p), 0, MF_COUNT_INCREASED); 640
641 ret = memory_failure(page_to_pfn(page), 0, MF_COUNT_INCREASED);
637 if (ret) 642 if (ret)
638 return ret; 643 return ret;
639 } 644 }
@@ -651,13 +656,7 @@ madvise_vma(struct vm_area_struct *vma, struct vm_area_struct **prev,
651 case MADV_WILLNEED: 656 case MADV_WILLNEED:
652 return madvise_willneed(vma, prev, start, end); 657 return madvise_willneed(vma, prev, start, end);
653 case MADV_FREE: 658 case MADV_FREE:
654 /* 659 return madvise_free(vma, prev, start, end);
655 * XXX: In this implementation, MADV_FREE works like
656 * MADV_DONTNEED on swapless system or full swap.
657 */
658 if (get_nr_swap_pages() > 0)
659 return madvise_free(vma, prev, start, end);
660 /* passthrough */
661 case MADV_DONTNEED: 660 case MADV_DONTNEED:
662 return madvise_dontneed(vma, prev, start, end); 661 return madvise_dontneed(vma, prev, start, end);
663 default: 662 default:
@@ -688,6 +687,10 @@ madvise_behavior_valid(int behavior)
688#endif 687#endif
689 case MADV_DONTDUMP: 688 case MADV_DONTDUMP:
690 case MADV_DODUMP: 689 case MADV_DODUMP:
690#ifdef CONFIG_MEMORY_FAILURE
691 case MADV_SOFT_OFFLINE:
692 case MADV_HWPOISON:
693#endif
691 return true; 694 return true;
692 695
693 default: 696 default:
@@ -761,10 +764,6 @@ SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
761 size_t len; 764 size_t len;
762 struct blk_plug plug; 765 struct blk_plug plug;
763 766
764#ifdef CONFIG_MEMORY_FAILURE
765 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
766 return madvise_hwpoison(behavior, start, start+len_in);
767#endif
768 if (!madvise_behavior_valid(behavior)) 767 if (!madvise_behavior_valid(behavior))
769 return error; 768 return error;
770 769
@@ -784,6 +783,11 @@ SYSCALL_DEFINE3(madvise, unsigned long, start, size_t, len_in, int, behavior)
784 if (end == start) 783 if (end == start)
785 return error; 784 return error;
786 785
786#ifdef CONFIG_MEMORY_FAILURE
787 if (behavior == MADV_HWPOISON || behavior == MADV_SOFT_OFFLINE)
788 return madvise_inject_error(behavior, start, start + len_in);
789#endif
790
787 write = madvise_need_mmap_write(behavior); 791 write = madvise_need_mmap_write(behavior);
788 if (write) { 792 if (write) {
789 if (down_write_killable(&current->mm->mmap_sem)) 793 if (down_write_killable(&current->mm->mmap_sem))
diff --git a/mm/memblock.c b/mm/memblock.c
index 696f06d17c4e..b049c9b2dba8 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -805,6 +805,18 @@ int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
805} 805}
806 806
807/** 807/**
808 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
809 * @base: the base phys addr of the region
810 * @size: the size of the region
811 *
812 * Return 0 on success, -errno on failure.
813 */
814int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
815{
816 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
817}
818
819/**
808 * __next_reserved_mem_region - next function for for_each_reserved_region() 820 * __next_reserved_mem_region - next function for for_each_reserved_region()
809 * @idx: pointer to u64 loop variable 821 * @idx: pointer to u64 loop variable
810 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL 822 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
@@ -1531,11 +1543,37 @@ void __init memblock_enforce_memory_limit(phys_addr_t limit)
1531 (phys_addr_t)ULLONG_MAX); 1543 (phys_addr_t)ULLONG_MAX);
1532} 1544}
1533 1545
1546void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1547{
1548 int start_rgn, end_rgn;
1549 int i, ret;
1550
1551 if (!size)
1552 return;
1553
1554 ret = memblock_isolate_range(&memblock.memory, base, size,
1555 &start_rgn, &end_rgn);
1556 if (ret)
1557 return;
1558
1559 /* remove all the MAP regions */
1560 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1561 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1562 memblock_remove_region(&memblock.memory, i);
1563
1564 for (i = start_rgn - 1; i >= 0; i--)
1565 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1566 memblock_remove_region(&memblock.memory, i);
1567
1568 /* truncate the reserved regions */
1569 memblock_remove_range(&memblock.reserved, 0, base);
1570 memblock_remove_range(&memblock.reserved,
1571 base + size, (phys_addr_t)ULLONG_MAX);
1572}
1573
1534void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1574void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1535{ 1575{
1536 struct memblock_type *type = &memblock.memory;
1537 phys_addr_t max_addr; 1576 phys_addr_t max_addr;
1538 int i, ret, start_rgn, end_rgn;
1539 1577
1540 if (!limit) 1578 if (!limit)
1541 return; 1579 return;
@@ -1546,19 +1584,7 @@ void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1546 if (max_addr == (phys_addr_t)ULLONG_MAX) 1584 if (max_addr == (phys_addr_t)ULLONG_MAX)
1547 return; 1585 return;
1548 1586
1549 ret = memblock_isolate_range(type, max_addr, (phys_addr_t)ULLONG_MAX, 1587 memblock_cap_memory_range(0, max_addr);
1550 &start_rgn, &end_rgn);
1551 if (ret)
1552 return;
1553
1554 /* remove all the MAP regions above the limit */
1555 for (i = end_rgn - 1; i >= start_rgn; i--) {
1556 if (!memblock_is_nomap(&type->regions[i]))
1557 memblock_remove_region(type, i);
1558 }
1559 /* truncate the reserved regions */
1560 memblock_remove_range(&memblock.reserved, max_addr,
1561 (phys_addr_t)ULLONG_MAX);
1562} 1588}
1563 1589
1564static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1590static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 2bd7541d7c11..ff73899af61a 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -100,24 +100,7 @@ static bool do_memsw_account(void)
100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; 100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
101} 101}
102 102
103static const char * const mem_cgroup_stat_names[] = { 103static const char *const mem_cgroup_lru_names[] = {
104 "cache",
105 "rss",
106 "rss_huge",
107 "mapped_file",
108 "dirty",
109 "writeback",
110 "swap",
111};
112
113static const char * const mem_cgroup_events_names[] = {
114 "pgpgin",
115 "pgpgout",
116 "pgfault",
117 "pgmajfault",
118};
119
120static const char * const mem_cgroup_lru_names[] = {
121 "inactive_anon", 104 "inactive_anon",
122 "active_anon", 105 "active_anon",
123 "inactive_file", 106 "inactive_file",
@@ -568,32 +551,15 @@ mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
568 * common workload, threshold and synchronization as vmstat[] should be 551 * common workload, threshold and synchronization as vmstat[] should be
569 * implemented. 552 * implemented.
570 */ 553 */
571static unsigned long
572mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
573{
574 long val = 0;
575 int cpu;
576
577 /* Per-cpu values can be negative, use a signed accumulator */
578 for_each_possible_cpu(cpu)
579 val += per_cpu(memcg->stat->count[idx], cpu);
580 /*
581 * Summing races with updates, so val may be negative. Avoid exposing
582 * transient negative values.
583 */
584 if (val < 0)
585 val = 0;
586 return val;
587}
588 554
589static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, 555static unsigned long memcg_sum_events(struct mem_cgroup *memcg,
590 enum mem_cgroup_events_index idx) 556 enum memcg_event_item event)
591{ 557{
592 unsigned long val = 0; 558 unsigned long val = 0;
593 int cpu; 559 int cpu;
594 560
595 for_each_possible_cpu(cpu) 561 for_each_possible_cpu(cpu)
596 val += per_cpu(memcg->stat->events[idx], cpu); 562 val += per_cpu(memcg->stat->events[event], cpu);
597 return val; 563 return val;
598} 564}
599 565
@@ -606,23 +572,23 @@ static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
606 * counted as CACHE even if it's on ANON LRU. 572 * counted as CACHE even if it's on ANON LRU.
607 */ 573 */
608 if (PageAnon(page)) 574 if (PageAnon(page))
609 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], 575 __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages);
610 nr_pages); 576 else {
611 else 577 __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages);
612 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], 578 if (PageSwapBacked(page))
613 nr_pages); 579 __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages);
580 }
614 581
615 if (compound) { 582 if (compound) {
616 VM_BUG_ON_PAGE(!PageTransHuge(page), page); 583 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
617 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], 584 __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages);
618 nr_pages);
619 } 585 }
620 586
621 /* pagein of a big page is an event. So, ignore page size */ 587 /* pagein of a big page is an event. So, ignore page size */
622 if (nr_pages > 0) 588 if (nr_pages > 0)
623 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); 589 __this_cpu_inc(memcg->stat->events[PGPGIN]);
624 else { 590 else {
625 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); 591 __this_cpu_inc(memcg->stat->events[PGPGOUT]);
626 nr_pages = -nr_pages; /* for event */ 592 nr_pages = -nr_pages; /* for event */
627 } 593 }
628 594
@@ -1144,6 +1110,28 @@ static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1144 return false; 1110 return false;
1145} 1111}
1146 1112
1113unsigned int memcg1_stats[] = {
1114 MEMCG_CACHE,
1115 MEMCG_RSS,
1116 MEMCG_RSS_HUGE,
1117 NR_SHMEM,
1118 NR_FILE_MAPPED,
1119 NR_FILE_DIRTY,
1120 NR_WRITEBACK,
1121 MEMCG_SWAP,
1122};
1123
1124static const char *const memcg1_stat_names[] = {
1125 "cache",
1126 "rss",
1127 "rss_huge",
1128 "shmem",
1129 "mapped_file",
1130 "dirty",
1131 "writeback",
1132 "swap",
1133};
1134
1147#define K(x) ((x) << (PAGE_SHIFT-10)) 1135#define K(x) ((x) << (PAGE_SHIFT-10))
1148/** 1136/**
1149 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. 1137 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
@@ -1188,11 +1176,11 @@ void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1188 pr_cont_cgroup_path(iter->css.cgroup); 1176 pr_cont_cgroup_path(iter->css.cgroup);
1189 pr_cont(":"); 1177 pr_cont(":");
1190 1178
1191 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { 1179 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
1192 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) 1180 if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account)
1193 continue; 1181 continue;
1194 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], 1182 pr_cont(" %s:%luKB", memcg1_stat_names[i],
1195 K(mem_cgroup_read_stat(iter, i))); 1183 K(memcg_page_state(iter, memcg1_stats[i])));
1196 } 1184 }
1197 1185
1198 for (i = 0; i < NR_LRU_LISTS; i++) 1186 for (i = 0; i < NR_LRU_LISTS; i++)
@@ -1837,7 +1825,7 @@ static void reclaim_high(struct mem_cgroup *memcg,
1837 do { 1825 do {
1838 if (page_counter_read(&memcg->memory) <= memcg->high) 1826 if (page_counter_read(&memcg->memory) <= memcg->high)
1839 continue; 1827 continue;
1840 mem_cgroup_events(memcg, MEMCG_HIGH, 1); 1828 mem_cgroup_event(memcg, MEMCG_HIGH);
1841 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); 1829 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1842 } while ((memcg = parent_mem_cgroup(memcg))); 1830 } while ((memcg = parent_mem_cgroup(memcg)));
1843} 1831}
@@ -1928,7 +1916,7 @@ retry:
1928 if (!gfpflags_allow_blocking(gfp_mask)) 1916 if (!gfpflags_allow_blocking(gfp_mask))
1929 goto nomem; 1917 goto nomem;
1930 1918
1931 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); 1919 mem_cgroup_event(mem_over_limit, MEMCG_MAX);
1932 1920
1933 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, 1921 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1934 gfp_mask, may_swap); 1922 gfp_mask, may_swap);
@@ -1971,7 +1959,7 @@ retry:
1971 if (fatal_signal_pending(current)) 1959 if (fatal_signal_pending(current))
1972 goto force; 1960 goto force;
1973 1961
1974 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); 1962 mem_cgroup_event(mem_over_limit, MEMCG_OOM);
1975 1963
1976 mem_cgroup_oom(mem_over_limit, gfp_mask, 1964 mem_cgroup_oom(mem_over_limit, gfp_mask,
1977 get_order(nr_pages * PAGE_SIZE)); 1965 get_order(nr_pages * PAGE_SIZE));
@@ -2381,7 +2369,7 @@ void mem_cgroup_split_huge_fixup(struct page *head)
2381 for (i = 1; i < HPAGE_PMD_NR; i++) 2369 for (i = 1; i < HPAGE_PMD_NR; i++)
2382 head[i].mem_cgroup = head->mem_cgroup; 2370 head[i].mem_cgroup = head->mem_cgroup;
2383 2371
2384 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], 2372 __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE],
2385 HPAGE_PMD_NR); 2373 HPAGE_PMD_NR);
2386} 2374}
2387#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 2375#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
@@ -2391,7 +2379,7 @@ static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2391 bool charge) 2379 bool charge)
2392{ 2380{
2393 int val = (charge) ? 1 : -1; 2381 int val = (charge) ? 1 : -1;
2394 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); 2382 this_cpu_add(memcg->stat->count[MEMCG_SWAP], val);
2395} 2383}
2396 2384
2397/** 2385/**
@@ -2725,7 +2713,7 @@ static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
2725 2713
2726 for_each_mem_cgroup_tree(iter, memcg) { 2714 for_each_mem_cgroup_tree(iter, memcg) {
2727 for (i = 0; i < MEMCG_NR_STAT; i++) 2715 for (i = 0; i < MEMCG_NR_STAT; i++)
2728 stat[i] += mem_cgroup_read_stat(iter, i); 2716 stat[i] += memcg_page_state(iter, i);
2729 } 2717 }
2730} 2718}
2731 2719
@@ -2738,7 +2726,7 @@ static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
2738 2726
2739 for_each_mem_cgroup_tree(iter, memcg) { 2727 for_each_mem_cgroup_tree(iter, memcg) {
2740 for (i = 0; i < MEMCG_NR_EVENTS; i++) 2728 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2741 events[i] += mem_cgroup_read_events(iter, i); 2729 events[i] += memcg_sum_events(iter, i);
2742 } 2730 }
2743} 2731}
2744 2732
@@ -2750,13 +2738,10 @@ static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2750 struct mem_cgroup *iter; 2738 struct mem_cgroup *iter;
2751 2739
2752 for_each_mem_cgroup_tree(iter, memcg) { 2740 for_each_mem_cgroup_tree(iter, memcg) {
2753 val += mem_cgroup_read_stat(iter, 2741 val += memcg_page_state(iter, MEMCG_CACHE);
2754 MEM_CGROUP_STAT_CACHE); 2742 val += memcg_page_state(iter, MEMCG_RSS);
2755 val += mem_cgroup_read_stat(iter,
2756 MEM_CGROUP_STAT_RSS);
2757 if (swap) 2743 if (swap)
2758 val += mem_cgroup_read_stat(iter, 2744 val += memcg_page_state(iter, MEMCG_SWAP);
2759 MEM_CGROUP_STAT_SWAP);
2760 } 2745 }
2761 } else { 2746 } else {
2762 if (!swap) 2747 if (!swap)
@@ -3131,6 +3116,21 @@ static int memcg_numa_stat_show(struct seq_file *m, void *v)
3131} 3116}
3132#endif /* CONFIG_NUMA */ 3117#endif /* CONFIG_NUMA */
3133 3118
3119/* Universal VM events cgroup1 shows, original sort order */
3120unsigned int memcg1_events[] = {
3121 PGPGIN,
3122 PGPGOUT,
3123 PGFAULT,
3124 PGMAJFAULT,
3125};
3126
3127static const char *const memcg1_event_names[] = {
3128 "pgpgin",
3129 "pgpgout",
3130 "pgfault",
3131 "pgmajfault",
3132};
3133
3134static int memcg_stat_show(struct seq_file *m, void *v) 3134static int memcg_stat_show(struct seq_file *m, void *v)
3135{ 3135{
3136 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); 3136 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
@@ -3138,22 +3138,20 @@ static int memcg_stat_show(struct seq_file *m, void *v)
3138 struct mem_cgroup *mi; 3138 struct mem_cgroup *mi;
3139 unsigned int i; 3139 unsigned int i;
3140 3140
3141 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != 3141 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
3142 MEM_CGROUP_STAT_NSTATS);
3143 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3144 MEM_CGROUP_EVENTS_NSTATS);
3145 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); 3142 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3146 3143
3147 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { 3144 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
3148 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) 3145 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
3149 continue; 3146 continue;
3150 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], 3147 seq_printf(m, "%s %lu\n", memcg1_stat_names[i],
3151 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); 3148 memcg_page_state(memcg, memcg1_stats[i]) *
3149 PAGE_SIZE);
3152 } 3150 }
3153 3151
3154 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) 3152 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
3155 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], 3153 seq_printf(m, "%s %lu\n", memcg1_event_names[i],
3156 mem_cgroup_read_events(memcg, i)); 3154 memcg_sum_events(memcg, memcg1_events[i]));
3157 3155
3158 for (i = 0; i < NR_LRU_LISTS; i++) 3156 for (i = 0; i < NR_LRU_LISTS; i++)
3159 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], 3157 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
@@ -3171,23 +3169,23 @@ static int memcg_stat_show(struct seq_file *m, void *v)
3171 seq_printf(m, "hierarchical_memsw_limit %llu\n", 3169 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3172 (u64)memsw * PAGE_SIZE); 3170 (u64)memsw * PAGE_SIZE);
3173 3171
3174 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { 3172 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
3175 unsigned long long val = 0; 3173 unsigned long long val = 0;
3176 3174
3177 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) 3175 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
3178 continue; 3176 continue;
3179 for_each_mem_cgroup_tree(mi, memcg) 3177 for_each_mem_cgroup_tree(mi, memcg)
3180 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; 3178 val += memcg_page_state(mi, memcg1_stats[i]) *
3181 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); 3179 PAGE_SIZE;
3180 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val);
3182 } 3181 }
3183 3182
3184 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { 3183 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) {
3185 unsigned long long val = 0; 3184 unsigned long long val = 0;
3186 3185
3187 for_each_mem_cgroup_tree(mi, memcg) 3186 for_each_mem_cgroup_tree(mi, memcg)
3188 val += mem_cgroup_read_events(mi, i); 3187 val += memcg_sum_events(mi, memcg1_events[i]);
3189 seq_printf(m, "total_%s %llu\n", 3188 seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val);
3190 mem_cgroup_events_names[i], val);
3191 } 3189 }
3192 3190
3193 for (i = 0; i < NR_LRU_LISTS; i++) { 3191 for (i = 0; i < NR_LRU_LISTS; i++) {
@@ -3652,10 +3650,10 @@ void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3652 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); 3650 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3653 struct mem_cgroup *parent; 3651 struct mem_cgroup *parent;
3654 3652
3655 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); 3653 *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
3656 3654
3657 /* this should eventually include NR_UNSTABLE_NFS */ 3655 /* this should eventually include NR_UNSTABLE_NFS */
3658 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); 3656 *pwriteback = memcg_page_state(memcg, NR_WRITEBACK);
3659 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | 3657 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3660 (1 << LRU_ACTIVE_FILE)); 3658 (1 << LRU_ACTIVE_FILE));
3661 *pheadroom = PAGE_COUNTER_MAX; 3659 *pheadroom = PAGE_COUNTER_MAX;
@@ -4511,33 +4509,29 @@ static int mem_cgroup_move_account(struct page *page,
4511 spin_lock_irqsave(&from->move_lock, flags); 4509 spin_lock_irqsave(&from->move_lock, flags);
4512 4510
4513 if (!anon && page_mapped(page)) { 4511 if (!anon && page_mapped(page)) {
4514 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], 4512 __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages);
4515 nr_pages); 4513 __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages);
4516 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4517 nr_pages);
4518 } 4514 }
4519 4515
4520 /* 4516 /*
4521 * move_lock grabbed above and caller set from->moving_account, so 4517 * move_lock grabbed above and caller set from->moving_account, so
4522 * mem_cgroup_update_page_stat() will serialize updates to PageDirty. 4518 * mod_memcg_page_state will serialize updates to PageDirty.
4523 * So mapping should be stable for dirty pages. 4519 * So mapping should be stable for dirty pages.
4524 */ 4520 */
4525 if (!anon && PageDirty(page)) { 4521 if (!anon && PageDirty(page)) {
4526 struct address_space *mapping = page_mapping(page); 4522 struct address_space *mapping = page_mapping(page);
4527 4523
4528 if (mapping_cap_account_dirty(mapping)) { 4524 if (mapping_cap_account_dirty(mapping)) {
4529 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], 4525 __this_cpu_sub(from->stat->count[NR_FILE_DIRTY],
4530 nr_pages); 4526 nr_pages);
4531 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], 4527 __this_cpu_add(to->stat->count[NR_FILE_DIRTY],
4532 nr_pages); 4528 nr_pages);
4533 } 4529 }
4534 } 4530 }
4535 4531
4536 if (PageWriteback(page)) { 4532 if (PageWriteback(page)) {
4537 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], 4533 __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages);
4538 nr_pages); 4534 __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages);
4539 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4540 nr_pages);
4541 } 4535 }
4542 4536
4543 /* 4537 /*
@@ -5154,7 +5148,7 @@ static ssize_t memory_max_write(struct kernfs_open_file *of,
5154 continue; 5148 continue;
5155 } 5149 }
5156 5150
5157 mem_cgroup_events(memcg, MEMCG_OOM, 1); 5151 mem_cgroup_event(memcg, MEMCG_OOM);
5158 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) 5152 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5159 break; 5153 break;
5160 } 5154 }
@@ -5167,10 +5161,10 @@ static int memory_events_show(struct seq_file *m, void *v)
5167{ 5161{
5168 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); 5162 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5169 5163
5170 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); 5164 seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW));
5171 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); 5165 seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH));
5172 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); 5166 seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX));
5173 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); 5167 seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM));
5174 5168
5175 return 0; 5169 return 0;
5176} 5170}
@@ -5197,9 +5191,9 @@ static int memory_stat_show(struct seq_file *m, void *v)
5197 tree_events(memcg, events); 5191 tree_events(memcg, events);
5198 5192
5199 seq_printf(m, "anon %llu\n", 5193 seq_printf(m, "anon %llu\n",
5200 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); 5194 (u64)stat[MEMCG_RSS] * PAGE_SIZE);
5201 seq_printf(m, "file %llu\n", 5195 seq_printf(m, "file %llu\n",
5202 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); 5196 (u64)stat[MEMCG_CACHE] * PAGE_SIZE);
5203 seq_printf(m, "kernel_stack %llu\n", 5197 seq_printf(m, "kernel_stack %llu\n",
5204 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); 5198 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
5205 seq_printf(m, "slab %llu\n", 5199 seq_printf(m, "slab %llu\n",
@@ -5208,12 +5202,14 @@ static int memory_stat_show(struct seq_file *m, void *v)
5208 seq_printf(m, "sock %llu\n", 5202 seq_printf(m, "sock %llu\n",
5209 (u64)stat[MEMCG_SOCK] * PAGE_SIZE); 5203 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
5210 5204
5205 seq_printf(m, "shmem %llu\n",
5206 (u64)stat[NR_SHMEM] * PAGE_SIZE);
5211 seq_printf(m, "file_mapped %llu\n", 5207 seq_printf(m, "file_mapped %llu\n",
5212 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); 5208 (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE);
5213 seq_printf(m, "file_dirty %llu\n", 5209 seq_printf(m, "file_dirty %llu\n",
5214 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); 5210 (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE);
5215 seq_printf(m, "file_writeback %llu\n", 5211 seq_printf(m, "file_writeback %llu\n",
5216 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); 5212 (u64)stat[NR_WRITEBACK] * PAGE_SIZE);
5217 5213
5218 for (i = 0; i < NR_LRU_LISTS; i++) { 5214 for (i = 0; i < NR_LRU_LISTS; i++) {
5219 struct mem_cgroup *mi; 5215 struct mem_cgroup *mi;
@@ -5232,10 +5228,15 @@ static int memory_stat_show(struct seq_file *m, void *v)
5232 5228
5233 /* Accumulated memory events */ 5229 /* Accumulated memory events */
5234 5230
5235 seq_printf(m, "pgfault %lu\n", 5231 seq_printf(m, "pgfault %lu\n", events[PGFAULT]);
5236 events[MEM_CGROUP_EVENTS_PGFAULT]); 5232 seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]);
5237 seq_printf(m, "pgmajfault %lu\n", 5233
5238 events[MEM_CGROUP_EVENTS_PGMAJFAULT]); 5234 seq_printf(m, "workingset_refault %lu\n",
5235 stat[WORKINGSET_REFAULT]);
5236 seq_printf(m, "workingset_activate %lu\n",
5237 stat[WORKINGSET_ACTIVATE]);
5238 seq_printf(m, "workingset_nodereclaim %lu\n",
5239 stat[WORKINGSET_NODERECLAIM]);
5239 5240
5240 return 0; 5241 return 0;
5241} 5242}
@@ -5476,8 +5477,8 @@ void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5476 5477
5477static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, 5478static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5478 unsigned long nr_anon, unsigned long nr_file, 5479 unsigned long nr_anon, unsigned long nr_file,
5479 unsigned long nr_huge, unsigned long nr_kmem, 5480 unsigned long nr_kmem, unsigned long nr_huge,
5480 struct page *dummy_page) 5481 unsigned long nr_shmem, struct page *dummy_page)
5481{ 5482{
5482 unsigned long nr_pages = nr_anon + nr_file + nr_kmem; 5483 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
5483 unsigned long flags; 5484 unsigned long flags;
@@ -5492,10 +5493,11 @@ static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5492 } 5493 }
5493 5494
5494 local_irq_save(flags); 5495 local_irq_save(flags);
5495 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); 5496 __this_cpu_sub(memcg->stat->count[MEMCG_RSS], nr_anon);
5496 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); 5497 __this_cpu_sub(memcg->stat->count[MEMCG_CACHE], nr_file);
5497 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); 5498 __this_cpu_sub(memcg->stat->count[MEMCG_RSS_HUGE], nr_huge);
5498 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); 5499 __this_cpu_sub(memcg->stat->count[NR_SHMEM], nr_shmem);
5500 __this_cpu_add(memcg->stat->events[PGPGOUT], pgpgout);
5499 __this_cpu_add(memcg->stat->nr_page_events, nr_pages); 5501 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5500 memcg_check_events(memcg, dummy_page); 5502 memcg_check_events(memcg, dummy_page);
5501 local_irq_restore(flags); 5503 local_irq_restore(flags);
@@ -5507,6 +5509,7 @@ static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5507static void uncharge_list(struct list_head *page_list) 5509static void uncharge_list(struct list_head *page_list)
5508{ 5510{
5509 struct mem_cgroup *memcg = NULL; 5511 struct mem_cgroup *memcg = NULL;
5512 unsigned long nr_shmem = 0;
5510 unsigned long nr_anon = 0; 5513 unsigned long nr_anon = 0;
5511 unsigned long nr_file = 0; 5514 unsigned long nr_file = 0;
5512 unsigned long nr_huge = 0; 5515 unsigned long nr_huge = 0;
@@ -5539,9 +5542,9 @@ static void uncharge_list(struct list_head *page_list)
5539 if (memcg != page->mem_cgroup) { 5542 if (memcg != page->mem_cgroup) {
5540 if (memcg) { 5543 if (memcg) {
5541 uncharge_batch(memcg, pgpgout, nr_anon, nr_file, 5544 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5542 nr_huge, nr_kmem, page); 5545 nr_kmem, nr_huge, nr_shmem, page);
5543 pgpgout = nr_anon = nr_file = 5546 pgpgout = nr_anon = nr_file = nr_kmem = 0;
5544 nr_huge = nr_kmem = 0; 5547 nr_huge = nr_shmem = 0;
5545 } 5548 }
5546 memcg = page->mem_cgroup; 5549 memcg = page->mem_cgroup;
5547 } 5550 }
@@ -5555,8 +5558,11 @@ static void uncharge_list(struct list_head *page_list)
5555 } 5558 }
5556 if (PageAnon(page)) 5559 if (PageAnon(page))
5557 nr_anon += nr_pages; 5560 nr_anon += nr_pages;
5558 else 5561 else {
5559 nr_file += nr_pages; 5562 nr_file += nr_pages;
5563 if (PageSwapBacked(page))
5564 nr_shmem += nr_pages;
5565 }
5560 pgpgout++; 5566 pgpgout++;
5561 } else { 5567 } else {
5562 nr_kmem += 1 << compound_order(page); 5568 nr_kmem += 1 << compound_order(page);
@@ -5568,7 +5574,7 @@ static void uncharge_list(struct list_head *page_list)
5568 5574
5569 if (memcg) 5575 if (memcg)
5570 uncharge_batch(memcg, pgpgout, nr_anon, nr_file, 5576 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5571 nr_huge, nr_kmem, page); 5577 nr_kmem, nr_huge, nr_shmem, page);
5572} 5578}
5573 5579
5574/** 5580/**
diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 27f7210e7fab..73066b80d14a 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -220,6 +220,9 @@ static int kill_proc(struct task_struct *t, unsigned long addr, int trapno,
220 */ 220 */
221void shake_page(struct page *p, int access) 221void shake_page(struct page *p, int access)
222{ 222{
223 if (PageHuge(p))
224 return;
225
223 if (!PageSlab(p)) { 226 if (!PageSlab(p)) {
224 lru_add_drain_all(); 227 lru_add_drain_all();
225 if (PageLRU(p)) 228 if (PageLRU(p))
@@ -322,7 +325,7 @@ static void add_to_kill(struct task_struct *tsk, struct page *p,
322 * wrong earlier. 325 * wrong earlier.
323 */ 326 */
324static void kill_procs(struct list_head *to_kill, int forcekill, int trapno, 327static void kill_procs(struct list_head *to_kill, int forcekill, int trapno,
325 int fail, struct page *page, unsigned long pfn, 328 bool fail, struct page *page, unsigned long pfn,
326 int flags) 329 int flags)
327{ 330{
328 struct to_kill *tk, *next; 331 struct to_kill *tk, *next;
@@ -904,35 +907,36 @@ EXPORT_SYMBOL_GPL(get_hwpoison_page);
904 * Do all that is necessary to remove user space mappings. Unmap 907 * Do all that is necessary to remove user space mappings. Unmap
905 * the pages and send SIGBUS to the processes if the data was dirty. 908 * the pages and send SIGBUS to the processes if the data was dirty.
906 */ 909 */
907static int hwpoison_user_mappings(struct page *p, unsigned long pfn, 910static bool hwpoison_user_mappings(struct page *p, unsigned long pfn,
908 int trapno, int flags, struct page **hpagep) 911 int trapno, int flags, struct page **hpagep)
909{ 912{
910 enum ttu_flags ttu = TTU_UNMAP | TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS; 913 enum ttu_flags ttu = TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS;
911 struct address_space *mapping; 914 struct address_space *mapping;
912 LIST_HEAD(tokill); 915 LIST_HEAD(tokill);
913 int ret; 916 bool unmap_success;
914 int kill = 1, forcekill; 917 int kill = 1, forcekill;
915 struct page *hpage = *hpagep; 918 struct page *hpage = *hpagep;
919 bool mlocked = PageMlocked(hpage);
916 920
917 /* 921 /*
918 * Here we are interested only in user-mapped pages, so skip any 922 * Here we are interested only in user-mapped pages, so skip any
919 * other types of pages. 923 * other types of pages.
920 */ 924 */
921 if (PageReserved(p) || PageSlab(p)) 925 if (PageReserved(p) || PageSlab(p))
922 return SWAP_SUCCESS; 926 return true;
923 if (!(PageLRU(hpage) || PageHuge(p))) 927 if (!(PageLRU(hpage) || PageHuge(p)))
924 return SWAP_SUCCESS; 928 return true;
925 929
926 /* 930 /*
927 * This check implies we don't kill processes if their pages 931 * This check implies we don't kill processes if their pages
928 * are in the swap cache early. Those are always late kills. 932 * are in the swap cache early. Those are always late kills.
929 */ 933 */
930 if (!page_mapped(hpage)) 934 if (!page_mapped(hpage))
931 return SWAP_SUCCESS; 935 return true;
932 936
933 if (PageKsm(p)) { 937 if (PageKsm(p)) {
934 pr_err("Memory failure: %#lx: can't handle KSM pages.\n", pfn); 938 pr_err("Memory failure: %#lx: can't handle KSM pages.\n", pfn);
935 return SWAP_FAIL; 939 return false;
936 } 940 }
937 941
938 if (PageSwapCache(p)) { 942 if (PageSwapCache(p)) {
@@ -971,12 +975,19 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
971 if (kill) 975 if (kill)
972 collect_procs(hpage, &tokill, flags & MF_ACTION_REQUIRED); 976 collect_procs(hpage, &tokill, flags & MF_ACTION_REQUIRED);
973 977
974 ret = try_to_unmap(hpage, ttu); 978 unmap_success = try_to_unmap(hpage, ttu);
975 if (ret != SWAP_SUCCESS) 979 if (!unmap_success)
976 pr_err("Memory failure: %#lx: failed to unmap page (mapcount=%d)\n", 980 pr_err("Memory failure: %#lx: failed to unmap page (mapcount=%d)\n",
977 pfn, page_mapcount(hpage)); 981 pfn, page_mapcount(hpage));
978 982
979 /* 983 /*
984 * try_to_unmap() might put mlocked page in lru cache, so call
985 * shake_page() again to ensure that it's flushed.
986 */
987 if (mlocked)
988 shake_page(hpage, 0);
989
990 /*
980 * Now that the dirty bit has been propagated to the 991 * Now that the dirty bit has been propagated to the
981 * struct page and all unmaps done we can decide if 992 * struct page and all unmaps done we can decide if
982 * killing is needed or not. Only kill when the page 993 * killing is needed or not. Only kill when the page
@@ -987,10 +998,9 @@ static int hwpoison_user_mappings(struct page *p, unsigned long pfn,
987 * any accesses to the poisoned memory. 998 * any accesses to the poisoned memory.
988 */ 999 */
989 forcekill = PageDirty(hpage) || (flags & MF_MUST_KILL); 1000 forcekill = PageDirty(hpage) || (flags & MF_MUST_KILL);
990 kill_procs(&tokill, forcekill, trapno, 1001 kill_procs(&tokill, forcekill, trapno, !unmap_success, p, pfn, flags);
991 ret != SWAP_SUCCESS, p, pfn, flags);
992 1002
993 return ret; 1003 return unmap_success;
994} 1004}
995 1005
996static void set_page_hwpoison_huge_page(struct page *hpage) 1006static void set_page_hwpoison_huge_page(struct page *hpage)
@@ -1138,22 +1148,14 @@ int memory_failure(unsigned long pfn, int trapno, int flags)
1138 * The check (unnecessarily) ignores LRU pages being isolated and 1148 * The check (unnecessarily) ignores LRU pages being isolated and
1139 * walked by the page reclaim code, however that's not a big loss. 1149 * walked by the page reclaim code, however that's not a big loss.
1140 */ 1150 */
1141 if (!PageHuge(p)) { 1151 shake_page(p, 0);
1142 if (!PageLRU(p)) 1152 /* shake_page could have turned it free. */
1143 shake_page(p, 0); 1153 if (!PageLRU(p) && is_free_buddy_page(p)) {
1144 if (!PageLRU(p)) { 1154 if (flags & MF_COUNT_INCREASED)
1145 /* 1155 action_result(pfn, MF_MSG_BUDDY, MF_DELAYED);
1146 * shake_page could have turned it free. 1156 else
1147 */ 1157 action_result(pfn, MF_MSG_BUDDY_2ND, MF_DELAYED);
1148 if (is_free_buddy_page(p)) { 1158 return 0;
1149 if (flags & MF_COUNT_INCREASED)
1150 action_result(pfn, MF_MSG_BUDDY, MF_DELAYED);
1151 else
1152 action_result(pfn, MF_MSG_BUDDY_2ND,
1153 MF_DELAYED);
1154 return 0;
1155 }
1156 }
1157 } 1159 }
1158 1160
1159 lock_page(hpage); 1161 lock_page(hpage);
@@ -1230,8 +1232,7 @@ int memory_failure(unsigned long pfn, int trapno, int flags)
1230 * When the raw error page is thp tail page, hpage points to the raw 1232 * When the raw error page is thp tail page, hpage points to the raw
1231 * page after thp split. 1233 * page after thp split.
1232 */ 1234 */
1233 if (hwpoison_user_mappings(p, pfn, trapno, flags, &hpage) 1235 if (!hwpoison_user_mappings(p, pfn, trapno, flags, &hpage)) {
1234 != SWAP_SUCCESS) {
1235 action_result(pfn, MF_MSG_UNMAP_FAILED, MF_IGNORED); 1236 action_result(pfn, MF_MSG_UNMAP_FAILED, MF_IGNORED);
1236 res = -EBUSY; 1237 res = -EBUSY;
1237 goto out; 1238 goto out;
@@ -1543,8 +1544,8 @@ static int get_any_page(struct page *page, unsigned long pfn, int flags)
1543 if (ret == 1 && !PageLRU(page)) { 1544 if (ret == 1 && !PageLRU(page)) {
1544 /* Drop page reference which is from __get_any_page() */ 1545 /* Drop page reference which is from __get_any_page() */
1545 put_hwpoison_page(page); 1546 put_hwpoison_page(page);
1546 pr_info("soft_offline: %#lx: unknown non LRU page type %lx\n", 1547 pr_info("soft_offline: %#lx: unknown non LRU page type %lx (%pGp)\n",
1547 pfn, page->flags); 1548 pfn, page->flags, &page->flags);
1548 return -EIO; 1549 return -EIO;
1549 } 1550 }
1550 } 1551 }
@@ -1585,8 +1586,8 @@ static int soft_offline_huge_page(struct page *page, int flags)
1585 ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL, 1586 ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL,
1586 MIGRATE_SYNC, MR_MEMORY_FAILURE); 1587 MIGRATE_SYNC, MR_MEMORY_FAILURE);
1587 if (ret) { 1588 if (ret) {
1588 pr_info("soft offline: %#lx: migration failed %d, type %lx\n", 1589 pr_info("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n",
1589 pfn, ret, page->flags); 1590 pfn, ret, page->flags, &page->flags);
1590 /* 1591 /*
1591 * We know that soft_offline_huge_page() tries to migrate 1592 * We know that soft_offline_huge_page() tries to migrate
1592 * only one hugepage pointed to by hpage, so we need not 1593 * only one hugepage pointed to by hpage, so we need not
@@ -1677,14 +1678,14 @@ static int __soft_offline_page(struct page *page, int flags)
1677 if (!list_empty(&pagelist)) 1678 if (!list_empty(&pagelist))
1678 putback_movable_pages(&pagelist); 1679 putback_movable_pages(&pagelist);
1679 1680
1680 pr_info("soft offline: %#lx: migration failed %d, type %lx\n", 1681 pr_info("soft offline: %#lx: migration failed %d, type %lx (%pGp)\n",
1681 pfn, ret, page->flags); 1682 pfn, ret, page->flags, &page->flags);
1682 if (ret > 0) 1683 if (ret > 0)
1683 ret = -EIO; 1684 ret = -EIO;
1684 } 1685 }
1685 } else { 1686 } else {
1686 pr_info("soft offline: %#lx: isolation failed: %d, page count %d, type %lx\n", 1687 pr_info("soft offline: %#lx: isolation failed: %d, page count %d, type %lx (%pGp)\n",
1687 pfn, ret, page_count(page), page->flags); 1688 pfn, ret, page_count(page), page->flags, &page->flags);
1688 } 1689 }
1689 return ret; 1690 return ret;
1690} 1691}
diff --git a/mm/memory.c b/mm/memory.c
index 235ba51b2fbf..6ff5d729ded0 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4298,7 +4298,7 @@ void __might_fault(const char *file, int line)
4298 * get paged out, therefore we'll never actually fault, and the 4298 * get paged out, therefore we'll never actually fault, and the
4299 * below annotations will generate false positives. 4299 * below annotations will generate false positives.
4300 */ 4300 */
4301 if (segment_eq(get_fs(), KERNEL_DS)) 4301 if (uaccess_kernel())
4302 return; 4302 return;
4303 if (pagefault_disabled()) 4303 if (pagefault_disabled())
4304 return; 4304 return;
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 6fa7208bcd56..b63d7d1239df 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1208,7 +1208,11 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1208 1208
1209 arch_refresh_nodedata(nid, pgdat); 1209 arch_refresh_nodedata(nid, pgdat);
1210 } else { 1210 } else {
1211 /* Reset the nr_zones, order and classzone_idx before reuse */ 1211 /*
1212 * Reset the nr_zones, order and classzone_idx before reuse.
1213 * Note that kswapd will init kswapd_classzone_idx properly
1214 * when it starts in the near future.
1215 */
1212 pgdat->nr_zones = 0; 1216 pgdat->nr_zones = 0;
1213 pgdat->kswapd_order = 0; 1217 pgdat->kswapd_order = 0;
1214 pgdat->kswapd_classzone_idx = 0; 1218 pgdat->kswapd_classzone_idx = 0;
diff --git a/mm/migrate.c b/mm/migrate.c
index ed97c2c14fa8..89a0a1707f4c 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -184,9 +184,9 @@ void putback_movable_pages(struct list_head *l)
184 unlock_page(page); 184 unlock_page(page);
185 put_page(page); 185 put_page(page);
186 } else { 186 } else {
187 putback_lru_page(page);
188 dec_node_page_state(page, NR_ISOLATED_ANON + 187 dec_node_page_state(page, NR_ISOLATED_ANON +
189 page_is_file_cache(page)); 188 page_is_file_cache(page));
189 putback_lru_page(page);
190 } 190 }
191 } 191 }
192} 192}
@@ -194,7 +194,7 @@ void putback_movable_pages(struct list_head *l)
194/* 194/*
195 * Restore a potential migration pte to a working pte entry 195 * Restore a potential migration pte to a working pte entry
196 */ 196 */
197static int remove_migration_pte(struct page *page, struct vm_area_struct *vma, 197static bool remove_migration_pte(struct page *page, struct vm_area_struct *vma,
198 unsigned long addr, void *old) 198 unsigned long addr, void *old)
199{ 199{
200 struct page_vma_mapped_walk pvmw = { 200 struct page_vma_mapped_walk pvmw = {
@@ -253,7 +253,7 @@ static int remove_migration_pte(struct page *page, struct vm_area_struct *vma,
253 update_mmu_cache(vma, pvmw.address, pvmw.pte); 253 update_mmu_cache(vma, pvmw.address, pvmw.pte);
254 } 254 }
255 255
256 return SWAP_AGAIN; 256 return true;
257} 257}
258 258
259/* 259/*
@@ -1722,9 +1722,6 @@ static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
1722{ 1722{
1723 int z; 1723 int z;
1724 1724
1725 if (!pgdat_reclaimable(pgdat))
1726 return false;
1727
1728 for (z = pgdat->nr_zones - 1; z >= 0; z--) { 1725 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1729 struct zone *zone = pgdat->node_zones + z; 1726 struct zone *zone = pgdat->node_zones + z;
1730 1727
@@ -1947,7 +1944,8 @@ int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1947 1944
1948 /* Prepare a page as a migration target */ 1945 /* Prepare a page as a migration target */
1949 __SetPageLocked(new_page); 1946 __SetPageLocked(new_page);
1950 __SetPageSwapBacked(new_page); 1947 if (PageSwapBacked(page))
1948 __SetPageSwapBacked(new_page);
1951 1949
1952 /* anon mapping, we can simply copy page->mapping to the new page: */ 1950 /* anon mapping, we can simply copy page->mapping to the new page: */
1953 new_page->mapping = page->mapping; 1951 new_page->mapping = page->mapping;
diff --git a/mm/mlock.c b/mm/mlock.c
index 0dd9ca18e19e..c483c5c20b4b 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -123,17 +123,15 @@ static bool __munlock_isolate_lru_page(struct page *page, bool getpage)
123 */ 123 */
124static void __munlock_isolated_page(struct page *page) 124static void __munlock_isolated_page(struct page *page)
125{ 125{
126 int ret = SWAP_AGAIN;
127
128 /* 126 /*
129 * Optimization: if the page was mapped just once, that's our mapping 127 * Optimization: if the page was mapped just once, that's our mapping
130 * and we don't need to check all the other vmas. 128 * and we don't need to check all the other vmas.
131 */ 129 */
132 if (page_mapcount(page) > 1) 130 if (page_mapcount(page) > 1)
133 ret = try_to_munlock(page); 131 try_to_munlock(page);
134 132
135 /* Did try_to_unlock() succeed or punt? */ 133 /* Did try_to_unlock() succeed or punt? */
136 if (ret != SWAP_MLOCK) 134 if (!PageMlocked(page))
137 count_vm_event(UNEVICTABLE_PGMUNLOCKED); 135 count_vm_event(UNEVICTABLE_PGMUNLOCKED);
138 136
139 putback_lru_page(page); 137 putback_lru_page(page);
diff --git a/mm/mmap.c b/mm/mmap.c
index bfbe8856d134..f82741e199c0 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -1479,7 +1479,7 @@ SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1479 struct user_struct *user = NULL; 1479 struct user_struct *user = NULL;
1480 struct hstate *hs; 1480 struct hstate *hs;
1481 1481
1482 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1482 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1483 if (!hs) 1483 if (!hs)
1484 return -EINVAL; 1484 return -EINVAL;
1485 1485
diff --git a/mm/nommu.c b/mm/nommu.c
index 2d131b97a851..fc184f597d59 100644
--- a/mm/nommu.c
+++ b/mm/nommu.c
@@ -237,12 +237,16 @@ void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
237} 237}
238EXPORT_SYMBOL(__vmalloc); 238EXPORT_SYMBOL(__vmalloc);
239 239
240void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
241{
242 return __vmalloc(size, flags, PAGE_KERNEL);
243}
244
240void *vmalloc_user(unsigned long size) 245void *vmalloc_user(unsigned long size)
241{ 246{
242 void *ret; 247 void *ret;
243 248
244 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 249 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
245 PAGE_KERNEL);
246 if (ret) { 250 if (ret) {
247 struct vm_area_struct *vma; 251 struct vm_area_struct *vma;
248 252
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index d083714a2bb9..04c9143a8625 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -685,6 +685,7 @@ void exit_oom_victim(void)
685void oom_killer_enable(void) 685void oom_killer_enable(void)
686{ 686{
687 oom_killer_disabled = false; 687 oom_killer_disabled = false;
688 pr_info("OOM killer enabled.\n");
688} 689}
689 690
690/** 691/**
@@ -721,6 +722,7 @@ bool oom_killer_disable(signed long timeout)
721 oom_killer_enable(); 722 oom_killer_enable();
722 return false; 723 return false;
723 } 724 }
725 pr_info("OOM killer disabled.\n");
724 726
725 return true; 727 return true;
726} 728}
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index d8ac2a7fb9e7..143c1c25d680 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -650,9 +650,8 @@ int wb_domain_init(struct wb_domain *dom, gfp_t gfp)
650 650
651 spin_lock_init(&dom->lock); 651 spin_lock_init(&dom->lock);
652 652
653 init_timer_deferrable(&dom->period_timer); 653 setup_deferrable_timer(&dom->period_timer, writeout_period,
654 dom->period_timer.function = writeout_period; 654 (unsigned long)dom);
655 dom->period_timer.data = (unsigned long)dom;
656 655
657 dom->dirty_limit_tstamp = jiffies; 656 dom->dirty_limit_tstamp = jiffies;
658 657
@@ -2353,10 +2352,16 @@ int do_writepages(struct address_space *mapping, struct writeback_control *wbc)
2353 2352
2354 if (wbc->nr_to_write <= 0) 2353 if (wbc->nr_to_write <= 0)
2355 return 0; 2354 return 0;
2356 if (mapping->a_ops->writepages) 2355 while (1) {
2357 ret = mapping->a_ops->writepages(mapping, wbc); 2356 if (mapping->a_ops->writepages)
2358 else 2357 ret = mapping->a_ops->writepages(mapping, wbc);
2359 ret = generic_writepages(mapping, wbc); 2358 else
2359 ret = generic_writepages(mapping, wbc);
2360 if ((ret != -ENOMEM) || (wbc->sync_mode != WB_SYNC_ALL))
2361 break;
2362 cond_resched();
2363 congestion_wait(BLK_RW_ASYNC, HZ/50);
2364 }
2360 return ret; 2365 return ret;
2361} 2366}
2362 2367
@@ -2428,7 +2433,7 @@ void account_page_dirtied(struct page *page, struct address_space *mapping)
2428 inode_attach_wb(inode, page); 2433 inode_attach_wb(inode, page);
2429 wb = inode_to_wb(inode); 2434 wb = inode_to_wb(inode);
2430 2435
2431 mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_DIRTY); 2436 inc_memcg_page_state(page, NR_FILE_DIRTY);
2432 __inc_node_page_state(page, NR_FILE_DIRTY); 2437 __inc_node_page_state(page, NR_FILE_DIRTY);
2433 __inc_zone_page_state(page, NR_ZONE_WRITE_PENDING); 2438 __inc_zone_page_state(page, NR_ZONE_WRITE_PENDING);
2434 __inc_node_page_state(page, NR_DIRTIED); 2439 __inc_node_page_state(page, NR_DIRTIED);
@@ -2450,7 +2455,7 @@ void account_page_cleaned(struct page *page, struct address_space *mapping,
2450 struct bdi_writeback *wb) 2455 struct bdi_writeback *wb)
2451{ 2456{
2452 if (mapping_cap_account_dirty(mapping)) { 2457 if (mapping_cap_account_dirty(mapping)) {
2453 mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_DIRTY); 2458 dec_memcg_page_state(page, NR_FILE_DIRTY);
2454 dec_node_page_state(page, NR_FILE_DIRTY); 2459 dec_node_page_state(page, NR_FILE_DIRTY);
2455 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING); 2460 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING);
2456 dec_wb_stat(wb, WB_RECLAIMABLE); 2461 dec_wb_stat(wb, WB_RECLAIMABLE);
@@ -2707,7 +2712,7 @@ int clear_page_dirty_for_io(struct page *page)
2707 */ 2712 */
2708 wb = unlocked_inode_to_wb_begin(inode, &locked); 2713 wb = unlocked_inode_to_wb_begin(inode, &locked);
2709 if (TestClearPageDirty(page)) { 2714 if (TestClearPageDirty(page)) {
2710 mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_DIRTY); 2715 dec_memcg_page_state(page, NR_FILE_DIRTY);
2711 dec_node_page_state(page, NR_FILE_DIRTY); 2716 dec_node_page_state(page, NR_FILE_DIRTY);
2712 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING); 2717 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING);
2713 dec_wb_stat(wb, WB_RECLAIMABLE); 2718 dec_wb_stat(wb, WB_RECLAIMABLE);
@@ -2754,7 +2759,7 @@ int test_clear_page_writeback(struct page *page)
2754 ret = TestClearPageWriteback(page); 2759 ret = TestClearPageWriteback(page);
2755 } 2760 }
2756 if (ret) { 2761 if (ret) {
2757 mem_cgroup_dec_page_stat(page, MEM_CGROUP_STAT_WRITEBACK); 2762 dec_memcg_page_state(page, NR_WRITEBACK);
2758 dec_node_page_state(page, NR_WRITEBACK); 2763 dec_node_page_state(page, NR_WRITEBACK);
2759 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING); 2764 dec_zone_page_state(page, NR_ZONE_WRITE_PENDING);
2760 inc_node_page_state(page, NR_WRITTEN); 2765 inc_node_page_state(page, NR_WRITTEN);
@@ -2809,7 +2814,7 @@ int __test_set_page_writeback(struct page *page, bool keep_write)
2809 ret = TestSetPageWriteback(page); 2814 ret = TestSetPageWriteback(page);
2810 } 2815 }
2811 if (!ret) { 2816 if (!ret) {
2812 mem_cgroup_inc_page_stat(page, MEM_CGROUP_STAT_WRITEBACK); 2817 inc_memcg_page_state(page, NR_WRITEBACK);
2813 inc_node_page_state(page, NR_WRITEBACK); 2818 inc_node_page_state(page, NR_WRITEBACK);
2814 inc_zone_page_state(page, NR_ZONE_WRITE_PENDING); 2819 inc_zone_page_state(page, NR_ZONE_WRITE_PENDING);
2815 } 2820 }
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index f3d603cef2c0..f9e450c6b6e4 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -65,6 +65,7 @@
65#include <linux/page_owner.h> 65#include <linux/page_owner.h>
66#include <linux/kthread.h> 66#include <linux/kthread.h>
67#include <linux/memcontrol.h> 67#include <linux/memcontrol.h>
68#include <linux/ftrace.h>
68 69
69#include <asm/sections.h> 70#include <asm/sections.h>
70#include <asm/tlbflush.h> 71#include <asm/tlbflush.h>
@@ -1090,14 +1091,10 @@ static void free_pcppages_bulk(struct zone *zone, int count,
1090{ 1091{
1091 int migratetype = 0; 1092 int migratetype = 0;
1092 int batch_free = 0; 1093 int batch_free = 0;
1093 unsigned long nr_scanned, flags;
1094 bool isolated_pageblocks; 1094 bool isolated_pageblocks;
1095 1095
1096 spin_lock_irqsave(&zone->lock, flags); 1096 spin_lock(&zone->lock);
1097 isolated_pageblocks = has_isolate_pageblock(zone); 1097 isolated_pageblocks = has_isolate_pageblock(zone);
1098 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1099 if (nr_scanned)
1100 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1101 1098
1102 while (count) { 1099 while (count) {
1103 struct page *page; 1100 struct page *page;
@@ -1142,7 +1139,7 @@ static void free_pcppages_bulk(struct zone *zone, int count,
1142 trace_mm_page_pcpu_drain(page, 0, mt); 1139 trace_mm_page_pcpu_drain(page, 0, mt);
1143 } while (--count && --batch_free && !list_empty(list)); 1140 } while (--count && --batch_free && !list_empty(list));
1144 } 1141 }
1145 spin_unlock_irqrestore(&zone->lock, flags); 1142 spin_unlock(&zone->lock);
1146} 1143}
1147 1144
1148static void free_one_page(struct zone *zone, 1145static void free_one_page(struct zone *zone,
@@ -1150,19 +1147,13 @@ static void free_one_page(struct zone *zone,
1150 unsigned int order, 1147 unsigned int order,
1151 int migratetype) 1148 int migratetype)
1152{ 1149{
1153 unsigned long nr_scanned, flags; 1150 spin_lock(&zone->lock);
1154 spin_lock_irqsave(&zone->lock, flags);
1155 __count_vm_events(PGFREE, 1 << order);
1156 nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1157 if (nr_scanned)
1158 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1159
1160 if (unlikely(has_isolate_pageblock(zone) || 1151 if (unlikely(has_isolate_pageblock(zone) ||
1161 is_migrate_isolate(migratetype))) { 1152 is_migrate_isolate(migratetype))) {
1162 migratetype = get_pfnblock_migratetype(page, pfn); 1153 migratetype = get_pfnblock_migratetype(page, pfn);
1163 } 1154 }
1164 __free_one_page(page, pfn, zone, order, migratetype); 1155 __free_one_page(page, pfn, zone, order, migratetype);
1165 spin_unlock_irqrestore(&zone->lock, flags); 1156 spin_unlock(&zone->lock);
1166} 1157}
1167 1158
1168static void __meminit __init_single_page(struct page *page, unsigned long pfn, 1159static void __meminit __init_single_page(struct page *page, unsigned long pfn,
@@ -1240,6 +1231,7 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1240 1231
1241static void __free_pages_ok(struct page *page, unsigned int order) 1232static void __free_pages_ok(struct page *page, unsigned int order)
1242{ 1233{
1234 unsigned long flags;
1243 int migratetype; 1235 int migratetype;
1244 unsigned long pfn = page_to_pfn(page); 1236 unsigned long pfn = page_to_pfn(page);
1245 1237
@@ -1247,7 +1239,10 @@ static void __free_pages_ok(struct page *page, unsigned int order)
1247 return; 1239 return;
1248 1240
1249 migratetype = get_pfnblock_migratetype(page, pfn); 1241 migratetype = get_pfnblock_migratetype(page, pfn);
1242 local_irq_save(flags);
1243 __count_vm_events(PGFREE, 1 << order);
1250 free_one_page(page_zone(page), page, pfn, order, migratetype); 1244 free_one_page(page_zone(page), page, pfn, order, migratetype);
1245 local_irq_restore(flags);
1251} 1246}
1252 1247
1253static void __init __free_pages_boot_core(struct page *page, unsigned int order) 1248static void __init __free_pages_boot_core(struct page *page, unsigned int order)
@@ -1695,10 +1690,10 @@ static inline int check_new_page(struct page *page)
1695 return 1; 1690 return 1;
1696} 1691}
1697 1692
1698static inline bool free_pages_prezeroed(bool poisoned) 1693static inline bool free_pages_prezeroed(void)
1699{ 1694{
1700 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) && 1695 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1701 page_poisoning_enabled() && poisoned; 1696 page_poisoning_enabled();
1702} 1697}
1703 1698
1704#ifdef CONFIG_DEBUG_VM 1699#ifdef CONFIG_DEBUG_VM
@@ -1752,17 +1747,10 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
1752 unsigned int alloc_flags) 1747 unsigned int alloc_flags)
1753{ 1748{
1754 int i; 1749 int i;
1755 bool poisoned = true;
1756
1757 for (i = 0; i < (1 << order); i++) {
1758 struct page *p = page + i;
1759 if (poisoned)
1760 poisoned &= page_is_poisoned(p);
1761 }
1762 1750
1763 post_alloc_hook(page, order, gfp_flags); 1751 post_alloc_hook(page, order, gfp_flags);
1764 1752
1765 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO)) 1753 if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
1766 for (i = 0; i < (1 << order); i++) 1754 for (i = 0; i < (1 << order); i++)
1767 clear_highpage(page + i); 1755 clear_highpage(page + i);
1768 1756
@@ -1844,9 +1832,9 @@ static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1844 * Note that start_page and end_pages are not aligned on a pageblock 1832 * Note that start_page and end_pages are not aligned on a pageblock
1845 * boundary. If alignment is required, use move_freepages_block() 1833 * boundary. If alignment is required, use move_freepages_block()
1846 */ 1834 */
1847int move_freepages(struct zone *zone, 1835static int move_freepages(struct zone *zone,
1848 struct page *start_page, struct page *end_page, 1836 struct page *start_page, struct page *end_page,
1849 int migratetype) 1837 int migratetype, int *num_movable)
1850{ 1838{
1851 struct page *page; 1839 struct page *page;
1852 unsigned int order; 1840 unsigned int order;
@@ -1863,6 +1851,9 @@ int move_freepages(struct zone *zone,
1863 VM_BUG_ON(page_zone(start_page) != page_zone(end_page)); 1851 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1864#endif 1852#endif
1865 1853
1854 if (num_movable)
1855 *num_movable = 0;
1856
1866 for (page = start_page; page <= end_page;) { 1857 for (page = start_page; page <= end_page;) {
1867 if (!pfn_valid_within(page_to_pfn(page))) { 1858 if (!pfn_valid_within(page_to_pfn(page))) {
1868 page++; 1859 page++;
@@ -1873,6 +1864,15 @@ int move_freepages(struct zone *zone,
1873 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page); 1864 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1874 1865
1875 if (!PageBuddy(page)) { 1866 if (!PageBuddy(page)) {
1867 /*
1868 * We assume that pages that could be isolated for
1869 * migration are movable. But we don't actually try
1870 * isolating, as that would be expensive.
1871 */
1872 if (num_movable &&
1873 (PageLRU(page) || __PageMovable(page)))
1874 (*num_movable)++;
1875
1876 page++; 1876 page++;
1877 continue; 1877 continue;
1878 } 1878 }
@@ -1888,7 +1888,7 @@ int move_freepages(struct zone *zone,
1888} 1888}
1889 1889
1890int move_freepages_block(struct zone *zone, struct page *page, 1890int move_freepages_block(struct zone *zone, struct page *page,
1891 int migratetype) 1891 int migratetype, int *num_movable)
1892{ 1892{
1893 unsigned long start_pfn, end_pfn; 1893 unsigned long start_pfn, end_pfn;
1894 struct page *start_page, *end_page; 1894 struct page *start_page, *end_page;
@@ -1905,7 +1905,8 @@ int move_freepages_block(struct zone *zone, struct page *page,
1905 if (!zone_spans_pfn(zone, end_pfn)) 1905 if (!zone_spans_pfn(zone, end_pfn))
1906 return 0; 1906 return 0;
1907 1907
1908 return move_freepages(zone, start_page, end_page, migratetype); 1908 return move_freepages(zone, start_page, end_page, migratetype,
1909 num_movable);
1909} 1910}
1910 1911
1911static void change_pageblock_range(struct page *pageblock_page, 1912static void change_pageblock_range(struct page *pageblock_page,
@@ -1955,28 +1956,79 @@ static bool can_steal_fallback(unsigned int order, int start_mt)
1955/* 1956/*
1956 * This function implements actual steal behaviour. If order is large enough, 1957 * This function implements actual steal behaviour. If order is large enough,
1957 * we can steal whole pageblock. If not, we first move freepages in this 1958 * we can steal whole pageblock. If not, we first move freepages in this
1958 * pageblock and check whether half of pages are moved or not. If half of 1959 * pageblock to our migratetype and determine how many already-allocated pages
1959 * pages are moved, we can change migratetype of pageblock and permanently 1960 * are there in the pageblock with a compatible migratetype. If at least half
1960 * use it's pages as requested migratetype in the future. 1961 * of pages are free or compatible, we can change migratetype of the pageblock
1962 * itself, so pages freed in the future will be put on the correct free list.
1961 */ 1963 */
1962static void steal_suitable_fallback(struct zone *zone, struct page *page, 1964static void steal_suitable_fallback(struct zone *zone, struct page *page,
1963 int start_type) 1965 int start_type, bool whole_block)
1964{ 1966{
1965 unsigned int current_order = page_order(page); 1967 unsigned int current_order = page_order(page);
1966 int pages; 1968 struct free_area *area;
1969 int free_pages, movable_pages, alike_pages;
1970 int old_block_type;
1971
1972 old_block_type = get_pageblock_migratetype(page);
1973
1974 /*
1975 * This can happen due to races and we want to prevent broken
1976 * highatomic accounting.
1977 */
1978 if (is_migrate_highatomic(old_block_type))
1979 goto single_page;
1967 1980
1968 /* Take ownership for orders >= pageblock_order */ 1981 /* Take ownership for orders >= pageblock_order */
1969 if (current_order >= pageblock_order) { 1982 if (current_order >= pageblock_order) {
1970 change_pageblock_range(page, current_order, start_type); 1983 change_pageblock_range(page, current_order, start_type);
1971 return; 1984 goto single_page;
1985 }
1986
1987 /* We are not allowed to try stealing from the whole block */
1988 if (!whole_block)
1989 goto single_page;
1990
1991 free_pages = move_freepages_block(zone, page, start_type,
1992 &movable_pages);
1993 /*
1994 * Determine how many pages are compatible with our allocation.
1995 * For movable allocation, it's the number of movable pages which
1996 * we just obtained. For other types it's a bit more tricky.
1997 */
1998 if (start_type == MIGRATE_MOVABLE) {
1999 alike_pages = movable_pages;
2000 } else {
2001 /*
2002 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2003 * to MOVABLE pageblock, consider all non-movable pages as
2004 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2005 * vice versa, be conservative since we can't distinguish the
2006 * exact migratetype of non-movable pages.
2007 */
2008 if (old_block_type == MIGRATE_MOVABLE)
2009 alike_pages = pageblock_nr_pages
2010 - (free_pages + movable_pages);
2011 else
2012 alike_pages = 0;
1972 } 2013 }
1973 2014
1974 pages = move_freepages_block(zone, page, start_type); 2015 /* moving whole block can fail due to zone boundary conditions */
2016 if (!free_pages)
2017 goto single_page;
1975 2018
1976 /* Claim the whole block if over half of it is free */ 2019 /*
1977 if (pages >= (1 << (pageblock_order-1)) || 2020 * If a sufficient number of pages in the block are either free or of
2021 * comparable migratability as our allocation, claim the whole block.
2022 */
2023 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
1978 page_group_by_mobility_disabled) 2024 page_group_by_mobility_disabled)
1979 set_pageblock_migratetype(page, start_type); 2025 set_pageblock_migratetype(page, start_type);
2026
2027 return;
2028
2029single_page:
2030 area = &zone->free_area[current_order];
2031 list_move(&page->lru, &area->free_list[start_type]);
1980} 2032}
1981 2033
1982/* 2034/*
@@ -2042,11 +2094,11 @@ static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2042 2094
2043 /* Yoink! */ 2095 /* Yoink! */
2044 mt = get_pageblock_migratetype(page); 2096 mt = get_pageblock_migratetype(page);
2045 if (mt != MIGRATE_HIGHATOMIC && 2097 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2046 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) { 2098 && !is_migrate_cma(mt)) {
2047 zone->nr_reserved_highatomic += pageblock_nr_pages; 2099 zone->nr_reserved_highatomic += pageblock_nr_pages;
2048 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC); 2100 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2049 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC); 2101 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2050 } 2102 }
2051 2103
2052out_unlock: 2104out_unlock:
@@ -2100,8 +2152,7 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2100 * from highatomic to ac->migratetype. So we should 2152 * from highatomic to ac->migratetype. So we should
2101 * adjust the count once. 2153 * adjust the count once.
2102 */ 2154 */
2103 if (get_pageblock_migratetype(page) == 2155 if (is_migrate_highatomic_page(page)) {
2104 MIGRATE_HIGHATOMIC) {
2105 /* 2156 /*
2106 * It should never happen but changes to 2157 * It should never happen but changes to
2107 * locking could inadvertently allow a per-cpu 2158 * locking could inadvertently allow a per-cpu
@@ -2124,7 +2175,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2124 * may increase. 2175 * may increase.
2125 */ 2176 */
2126 set_pageblock_migratetype(page, ac->migratetype); 2177 set_pageblock_migratetype(page, ac->migratetype);
2127 ret = move_freepages_block(zone, page, ac->migratetype); 2178 ret = move_freepages_block(zone, page, ac->migratetype,
2179 NULL);
2128 if (ret) { 2180 if (ret) {
2129 spin_unlock_irqrestore(&zone->lock, flags); 2181 spin_unlock_irqrestore(&zone->lock, flags);
2130 return ret; 2182 return ret;
@@ -2136,8 +2188,13 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2136 return false; 2188 return false;
2137} 2189}
2138 2190
2139/* Remove an element from the buddy allocator from the fallback list */ 2191/*
2140static inline struct page * 2192 * Try finding a free buddy page on the fallback list and put it on the free
2193 * list of requested migratetype, possibly along with other pages from the same
2194 * block, depending on fragmentation avoidance heuristics. Returns true if
2195 * fallback was found so that __rmqueue_smallest() can grab it.
2196 */
2197static inline bool
2141__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype) 2198__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
2142{ 2199{
2143 struct free_area *area; 2200 struct free_area *area;
@@ -2158,33 +2215,17 @@ __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
2158 2215
2159 page = list_first_entry(&area->free_list[fallback_mt], 2216 page = list_first_entry(&area->free_list[fallback_mt],
2160 struct page, lru); 2217 struct page, lru);
2161 if (can_steal &&
2162 get_pageblock_migratetype(page) != MIGRATE_HIGHATOMIC)
2163 steal_suitable_fallback(zone, page, start_migratetype);
2164
2165 /* Remove the page from the freelists */
2166 area->nr_free--;
2167 list_del(&page->lru);
2168 rmv_page_order(page);
2169 2218
2170 expand(zone, page, order, current_order, area, 2219 steal_suitable_fallback(zone, page, start_migratetype,
2171 start_migratetype); 2220 can_steal);
2172 /*
2173 * The pcppage_migratetype may differ from pageblock's
2174 * migratetype depending on the decisions in
2175 * find_suitable_fallback(). This is OK as long as it does not
2176 * differ for MIGRATE_CMA pageblocks. Those can be used as
2177 * fallback only via special __rmqueue_cma_fallback() function
2178 */
2179 set_pcppage_migratetype(page, start_migratetype);
2180 2221
2181 trace_mm_page_alloc_extfrag(page, order, current_order, 2222 trace_mm_page_alloc_extfrag(page, order, current_order,
2182 start_migratetype, fallback_mt); 2223 start_migratetype, fallback_mt);
2183 2224
2184 return page; 2225 return true;
2185 } 2226 }
2186 2227
2187 return NULL; 2228 return false;
2188} 2229}
2189 2230
2190/* 2231/*
@@ -2196,13 +2237,14 @@ static struct page *__rmqueue(struct zone *zone, unsigned int order,
2196{ 2237{
2197 struct page *page; 2238 struct page *page;
2198 2239
2240retry:
2199 page = __rmqueue_smallest(zone, order, migratetype); 2241 page = __rmqueue_smallest(zone, order, migratetype);
2200 if (unlikely(!page)) { 2242 if (unlikely(!page)) {
2201 if (migratetype == MIGRATE_MOVABLE) 2243 if (migratetype == MIGRATE_MOVABLE)
2202 page = __rmqueue_cma_fallback(zone, order); 2244 page = __rmqueue_cma_fallback(zone, order);
2203 2245
2204 if (!page) 2246 if (!page && __rmqueue_fallback(zone, order, migratetype))
2205 page = __rmqueue_fallback(zone, order, migratetype); 2247 goto retry;
2206 } 2248 }
2207 2249
2208 trace_mm_page_alloc_zone_locked(page, order, migratetype); 2250 trace_mm_page_alloc_zone_locked(page, order, migratetype);
@@ -2219,9 +2261,8 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
2219 int migratetype, bool cold) 2261 int migratetype, bool cold)
2220{ 2262{
2221 int i, alloced = 0; 2263 int i, alloced = 0;
2222 unsigned long flags;
2223 2264
2224 spin_lock_irqsave(&zone->lock, flags); 2265 spin_lock(&zone->lock);
2225 for (i = 0; i < count; ++i) { 2266 for (i = 0; i < count; ++i) {
2226 struct page *page = __rmqueue(zone, order, migratetype); 2267 struct page *page = __rmqueue(zone, order, migratetype);
2227 if (unlikely(page == NULL)) 2268 if (unlikely(page == NULL))
@@ -2257,7 +2298,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
2257 * pages added to the pcp list. 2298 * pages added to the pcp list.
2258 */ 2299 */
2259 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); 2300 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2260 spin_unlock_irqrestore(&zone->lock, flags); 2301 spin_unlock(&zone->lock);
2261 return alloced; 2302 return alloced;
2262} 2303}
2263 2304
@@ -2485,25 +2526,22 @@ void free_hot_cold_page(struct page *page, bool cold)
2485{ 2526{
2486 struct zone *zone = page_zone(page); 2527 struct zone *zone = page_zone(page);
2487 struct per_cpu_pages *pcp; 2528 struct per_cpu_pages *pcp;
2529 unsigned long flags;
2488 unsigned long pfn = page_to_pfn(page); 2530 unsigned long pfn = page_to_pfn(page);
2489 int migratetype; 2531 int migratetype;
2490 2532
2491 if (in_interrupt()) {
2492 __free_pages_ok(page, 0);
2493 return;
2494 }
2495
2496 if (!free_pcp_prepare(page)) 2533 if (!free_pcp_prepare(page))
2497 return; 2534 return;
2498 2535
2499 migratetype = get_pfnblock_migratetype(page, pfn); 2536 migratetype = get_pfnblock_migratetype(page, pfn);
2500 set_pcppage_migratetype(page, migratetype); 2537 set_pcppage_migratetype(page, migratetype);
2501 preempt_disable(); 2538 local_irq_save(flags);
2539 __count_vm_event(PGFREE);
2502 2540
2503 /* 2541 /*
2504 * We only track unmovable, reclaimable and movable on pcp lists. 2542 * We only track unmovable, reclaimable and movable on pcp lists.
2505 * Free ISOLATE pages back to the allocator because they are being 2543 * Free ISOLATE pages back to the allocator because they are being
2506 * offlined but treat RESERVE as movable pages so we can get those 2544 * offlined but treat HIGHATOMIC as movable pages so we can get those
2507 * areas back if necessary. Otherwise, we may have to free 2545 * areas back if necessary. Otherwise, we may have to free
2508 * excessively into the page allocator 2546 * excessively into the page allocator
2509 */ 2547 */
@@ -2515,7 +2553,6 @@ void free_hot_cold_page(struct page *page, bool cold)
2515 migratetype = MIGRATE_MOVABLE; 2553 migratetype = MIGRATE_MOVABLE;
2516 } 2554 }
2517 2555
2518 __count_vm_event(PGFREE);
2519 pcp = &this_cpu_ptr(zone->pageset)->pcp; 2556 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2520 if (!cold) 2557 if (!cold)
2521 list_add(&page->lru, &pcp->lists[migratetype]); 2558 list_add(&page->lru, &pcp->lists[migratetype]);
@@ -2529,7 +2566,7 @@ void free_hot_cold_page(struct page *page, bool cold)
2529 } 2566 }
2530 2567
2531out: 2568out:
2532 preempt_enable(); 2569 local_irq_restore(flags);
2533} 2570}
2534 2571
2535/* 2572/*
@@ -2614,7 +2651,7 @@ int __isolate_free_page(struct page *page, unsigned int order)
2614 for (; page < endpage; page += pageblock_nr_pages) { 2651 for (; page < endpage; page += pageblock_nr_pages) {
2615 int mt = get_pageblock_migratetype(page); 2652 int mt = get_pageblock_migratetype(page);
2616 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt) 2653 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
2617 && mt != MIGRATE_HIGHATOMIC) 2654 && !is_migrate_highatomic(mt))
2618 set_pageblock_migratetype(page, 2655 set_pageblock_migratetype(page,
2619 MIGRATE_MOVABLE); 2656 MIGRATE_MOVABLE);
2620 } 2657 }
@@ -2654,8 +2691,6 @@ static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
2654{ 2691{
2655 struct page *page; 2692 struct page *page;
2656 2693
2657 VM_BUG_ON(in_interrupt());
2658
2659 do { 2694 do {
2660 if (list_empty(list)) { 2695 if (list_empty(list)) {
2661 pcp->count += rmqueue_bulk(zone, 0, 2696 pcp->count += rmqueue_bulk(zone, 0,
@@ -2686,8 +2721,9 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2686 struct list_head *list; 2721 struct list_head *list;
2687 bool cold = ((gfp_flags & __GFP_COLD) != 0); 2722 bool cold = ((gfp_flags & __GFP_COLD) != 0);
2688 struct page *page; 2723 struct page *page;
2724 unsigned long flags;
2689 2725
2690 preempt_disable(); 2726 local_irq_save(flags);
2691 pcp = &this_cpu_ptr(zone->pageset)->pcp; 2727 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2692 list = &pcp->lists[migratetype]; 2728 list = &pcp->lists[migratetype];
2693 page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list); 2729 page = __rmqueue_pcplist(zone, migratetype, cold, pcp, list);
@@ -2695,7 +2731,7 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2695 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order); 2731 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2696 zone_statistics(preferred_zone, zone); 2732 zone_statistics(preferred_zone, zone);
2697 } 2733 }
2698 preempt_enable(); 2734 local_irq_restore(flags);
2699 return page; 2735 return page;
2700} 2736}
2701 2737
@@ -2711,7 +2747,7 @@ struct page *rmqueue(struct zone *preferred_zone,
2711 unsigned long flags; 2747 unsigned long flags;
2712 struct page *page; 2748 struct page *page;
2713 2749
2714 if (likely(order == 0) && !in_interrupt()) { 2750 if (likely(order == 0)) {
2715 page = rmqueue_pcplist(preferred_zone, zone, order, 2751 page = rmqueue_pcplist(preferred_zone, zone, order,
2716 gfp_flags, migratetype); 2752 gfp_flags, migratetype);
2717 goto out; 2753 goto out;
@@ -3113,8 +3149,7 @@ void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3113 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL, 3149 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3114 DEFAULT_RATELIMIT_BURST); 3150 DEFAULT_RATELIMIT_BURST);
3115 3151
3116 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || 3152 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3117 debug_guardpage_minorder() > 0)
3118 return; 3153 return;
3119 3154
3120 pr_warn("%s: ", current->comm); 3155 pr_warn("%s: ", current->comm);
@@ -3248,14 +3283,15 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3248 enum compact_priority prio, enum compact_result *compact_result) 3283 enum compact_priority prio, enum compact_result *compact_result)
3249{ 3284{
3250 struct page *page; 3285 struct page *page;
3286 unsigned int noreclaim_flag;
3251 3287
3252 if (!order) 3288 if (!order)
3253 return NULL; 3289 return NULL;
3254 3290
3255 current->flags |= PF_MEMALLOC; 3291 noreclaim_flag = memalloc_noreclaim_save();
3256 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, 3292 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3257 prio); 3293 prio);
3258 current->flags &= ~PF_MEMALLOC; 3294 memalloc_noreclaim_restore(noreclaim_flag);
3259 3295
3260 if (*compact_result <= COMPACT_INACTIVE) 3296 if (*compact_result <= COMPACT_INACTIVE)
3261 return NULL; 3297 return NULL;
@@ -3402,12 +3438,13 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3402{ 3438{
3403 struct reclaim_state reclaim_state; 3439 struct reclaim_state reclaim_state;
3404 int progress; 3440 int progress;
3441 unsigned int noreclaim_flag;
3405 3442
3406 cond_resched(); 3443 cond_resched();
3407 3444
3408 /* We now go into synchronous reclaim */ 3445 /* We now go into synchronous reclaim */
3409 cpuset_memory_pressure_bump(); 3446 cpuset_memory_pressure_bump();
3410 current->flags |= PF_MEMALLOC; 3447 noreclaim_flag = memalloc_noreclaim_save();
3411 lockdep_set_current_reclaim_state(gfp_mask); 3448 lockdep_set_current_reclaim_state(gfp_mask);
3412 reclaim_state.reclaimed_slab = 0; 3449 reclaim_state.reclaimed_slab = 0;
3413 current->reclaim_state = &reclaim_state; 3450 current->reclaim_state = &reclaim_state;
@@ -3417,7 +3454,7 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3417 3454
3418 current->reclaim_state = NULL; 3455 current->reclaim_state = NULL;
3419 lockdep_clear_current_reclaim_state(); 3456 lockdep_clear_current_reclaim_state();
3420 current->flags &= ~PF_MEMALLOC; 3457 memalloc_noreclaim_restore(noreclaim_flag);
3421 3458
3422 cond_resched(); 3459 cond_resched();
3423 3460
@@ -3525,19 +3562,12 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3525} 3562}
3526 3563
3527/* 3564/*
3528 * Maximum number of reclaim retries without any progress before OOM killer
3529 * is consider as the only way to move forward.
3530 */
3531#define MAX_RECLAIM_RETRIES 16
3532
3533/*
3534 * Checks whether it makes sense to retry the reclaim to make a forward progress 3565 * Checks whether it makes sense to retry the reclaim to make a forward progress
3535 * for the given allocation request. 3566 * for the given allocation request.
3536 * The reclaim feedback represented by did_some_progress (any progress during 3567 *
3537 * the last reclaim round) and no_progress_loops (number of reclaim rounds without 3568 * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3538 * any progress in a row) is considered as well as the reclaimable pages on the 3569 * without success, or when we couldn't even meet the watermark if we
3539 * applicable zone list (with a backoff mechanism which is a function of 3570 * reclaimed all remaining pages on the LRU lists.
3540 * no_progress_loops).
3541 * 3571 *
3542 * Returns true if a retry is viable or false to enter the oom path. 3572 * Returns true if a retry is viable or false to enter the oom path.
3543 */ 3573 */
@@ -3582,13 +3612,11 @@ should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3582 bool wmark; 3612 bool wmark;
3583 3613
3584 available = reclaimable = zone_reclaimable_pages(zone); 3614 available = reclaimable = zone_reclaimable_pages(zone);
3585 available -= DIV_ROUND_UP((*no_progress_loops) * available,
3586 MAX_RECLAIM_RETRIES);
3587 available += zone_page_state_snapshot(zone, NR_FREE_PAGES); 3615 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3588 3616
3589 /* 3617 /*
3590 * Would the allocation succeed if we reclaimed the whole 3618 * Would the allocation succeed if we reclaimed all
3591 * available? 3619 * reclaimable pages?
3592 */ 3620 */
3593 wmark = __zone_watermark_ok(zone, order, min_wmark, 3621 wmark = __zone_watermark_ok(zone, order, min_wmark,
3594 ac_classzone_idx(ac), alloc_flags, available); 3622 ac_classzone_idx(ac), alloc_flags, available);
@@ -3639,6 +3667,7 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3639 struct alloc_context *ac) 3667 struct alloc_context *ac)
3640{ 3668{
3641 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM; 3669 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3670 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
3642 struct page *page = NULL; 3671 struct page *page = NULL;
3643 unsigned int alloc_flags; 3672 unsigned int alloc_flags;
3644 unsigned long did_some_progress; 3673 unsigned long did_some_progress;
@@ -3706,12 +3735,17 @@ retry_cpuset:
3706 3735
3707 /* 3736 /*
3708 * For costly allocations, try direct compaction first, as it's likely 3737 * For costly allocations, try direct compaction first, as it's likely
3709 * that we have enough base pages and don't need to reclaim. Don't try 3738 * that we have enough base pages and don't need to reclaim. For non-
3710 * that for allocations that are allowed to ignore watermarks, as the 3739 * movable high-order allocations, do that as well, as compaction will
3711 * ALLOC_NO_WATERMARKS attempt didn't yet happen. 3740 * try prevent permanent fragmentation by migrating from blocks of the
3741 * same migratetype.
3742 * Don't try this for allocations that are allowed to ignore
3743 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
3712 */ 3744 */
3713 if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER && 3745 if (can_direct_reclaim &&
3714 !gfp_pfmemalloc_allowed(gfp_mask)) { 3746 (costly_order ||
3747 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
3748 && !gfp_pfmemalloc_allowed(gfp_mask)) {
3715 page = __alloc_pages_direct_compact(gfp_mask, order, 3749 page = __alloc_pages_direct_compact(gfp_mask, order,
3716 alloc_flags, ac, 3750 alloc_flags, ac,
3717 INIT_COMPACT_PRIORITY, 3751 INIT_COMPACT_PRIORITY,
@@ -3723,7 +3757,7 @@ retry_cpuset:
3723 * Checks for costly allocations with __GFP_NORETRY, which 3757 * Checks for costly allocations with __GFP_NORETRY, which
3724 * includes THP page fault allocations 3758 * includes THP page fault allocations
3725 */ 3759 */
3726 if (gfp_mask & __GFP_NORETRY) { 3760 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
3727 /* 3761 /*
3728 * If compaction is deferred for high-order allocations, 3762 * If compaction is deferred for high-order allocations,
3729 * it is because sync compaction recently failed. If 3763 * it is because sync compaction recently failed. If
@@ -3774,7 +3808,7 @@ retry:
3774 3808
3775 /* Make sure we know about allocations which stall for too long */ 3809 /* Make sure we know about allocations which stall for too long */
3776 if (time_after(jiffies, alloc_start + stall_timeout)) { 3810 if (time_after(jiffies, alloc_start + stall_timeout)) {
3777 warn_alloc(gfp_mask, ac->nodemask, 3811 warn_alloc(gfp_mask & ~__GFP_NOWARN, ac->nodemask,
3778 "page allocation stalls for %ums, order:%u", 3812 "page allocation stalls for %ums, order:%u",
3779 jiffies_to_msecs(jiffies-alloc_start), order); 3813 jiffies_to_msecs(jiffies-alloc_start), order);
3780 stall_timeout += 10 * HZ; 3814 stall_timeout += 10 * HZ;
@@ -3804,7 +3838,7 @@ retry:
3804 * Do not retry costly high order allocations unless they are 3838 * Do not retry costly high order allocations unless they are
3805 * __GFP_REPEAT 3839 * __GFP_REPEAT
3806 */ 3840 */
3807 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT)) 3841 if (costly_order && !(gfp_mask & __GFP_REPEAT))
3808 goto nopage; 3842 goto nopage;
3809 3843
3810 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, 3844 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
@@ -3974,10 +4008,12 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3974 goto out; 4008 goto out;
3975 4009
3976 /* 4010 /*
3977 * Runtime PM, block IO and its error handling path can deadlock 4011 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
3978 * because I/O on the device might not complete. 4012 * resp. GFP_NOIO which has to be inherited for all allocation requests
4013 * from a particular context which has been marked by
4014 * memalloc_no{fs,io}_{save,restore}.
3979 */ 4015 */
3980 alloc_mask = memalloc_noio_flags(gfp_mask); 4016 alloc_mask = current_gfp_context(gfp_mask);
3981 ac.spread_dirty_pages = false; 4017 ac.spread_dirty_pages = false;
3982 4018
3983 /* 4019 /*
@@ -4250,7 +4286,8 @@ EXPORT_SYMBOL(free_pages_exact);
4250 * nr_free_zone_pages() counts the number of counts pages which are beyond the 4286 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4251 * high watermark within all zones at or below a given zone index. For each 4287 * high watermark within all zones at or below a given zone index. For each
4252 * zone, the number of pages is calculated as: 4288 * zone, the number of pages is calculated as:
4253 * managed_pages - high_pages 4289 *
4290 * nr_free_zone_pages = managed_pages - high_pages
4254 */ 4291 */
4255static unsigned long nr_free_zone_pages(int offset) 4292static unsigned long nr_free_zone_pages(int offset)
4256{ 4293{
@@ -4512,7 +4549,6 @@ void show_free_areas(unsigned int filter, nodemask_t *nodemask)
4512#endif 4549#endif
4513 " writeback_tmp:%lukB" 4550 " writeback_tmp:%lukB"
4514 " unstable:%lukB" 4551 " unstable:%lukB"
4515 " pages_scanned:%lu"
4516 " all_unreclaimable? %s" 4552 " all_unreclaimable? %s"
4517 "\n", 4553 "\n",
4518 pgdat->node_id, 4554 pgdat->node_id,
@@ -4535,8 +4571,8 @@ void show_free_areas(unsigned int filter, nodemask_t *nodemask)
4535#endif 4571#endif
4536 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)), 4572 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4537 K(node_page_state(pgdat, NR_UNSTABLE_NFS)), 4573 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
4538 node_page_state(pgdat, NR_PAGES_SCANNED), 4574 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
4539 !pgdat_reclaimable(pgdat) ? "yes" : "no"); 4575 "yes" : "no");
4540 } 4576 }
4541 4577
4542 for_each_populated_zone(zone) { 4578 for_each_populated_zone(zone) {
@@ -7431,7 +7467,7 @@ int alloc_contig_range(unsigned long start, unsigned long end,
7431 .zone = page_zone(pfn_to_page(start)), 7467 .zone = page_zone(pfn_to_page(start)),
7432 .mode = MIGRATE_SYNC, 7468 .mode = MIGRATE_SYNC,
7433 .ignore_skip_hint = true, 7469 .ignore_skip_hint = true,
7434 .gfp_mask = memalloc_noio_flags(gfp_mask), 7470 .gfp_mask = current_gfp_context(gfp_mask),
7435 }; 7471 };
7436 INIT_LIST_HEAD(&cc.migratepages); 7472 INIT_LIST_HEAD(&cc.migratepages);
7437 7473
diff --git a/mm/page_ext.c b/mm/page_ext.c
index 121dcffc4ec1..88ccc044b09a 100644
--- a/mm/page_ext.c
+++ b/mm/page_ext.c
@@ -59,9 +59,6 @@
59 59
60static struct page_ext_operations *page_ext_ops[] = { 60static struct page_ext_operations *page_ext_ops[] = {
61 &debug_guardpage_ops, 61 &debug_guardpage_ops,
62#ifdef CONFIG_PAGE_POISONING
63 &page_poisoning_ops,
64#endif
65#ifdef CONFIG_PAGE_OWNER 62#ifdef CONFIG_PAGE_OWNER
66 &page_owner_ops, 63 &page_owner_ops,
67#endif 64#endif
@@ -127,15 +124,12 @@ struct page_ext *lookup_page_ext(struct page *page)
127 struct page_ext *base; 124 struct page_ext *base;
128 125
129 base = NODE_DATA(page_to_nid(page))->node_page_ext; 126 base = NODE_DATA(page_to_nid(page))->node_page_ext;
130#if defined(CONFIG_DEBUG_VM) || defined(CONFIG_PAGE_POISONING) 127#if defined(CONFIG_DEBUG_VM)
131 /* 128 /*
132 * The sanity checks the page allocator does upon freeing a 129 * The sanity checks the page allocator does upon freeing a
133 * page can reach here before the page_ext arrays are 130 * page can reach here before the page_ext arrays are
134 * allocated when feeding a range of pages to the allocator 131 * allocated when feeding a range of pages to the allocator
135 * for the first time during bootup or memory hotplug. 132 * for the first time during bootup or memory hotplug.
136 *
137 * This check is also necessary for ensuring page poisoning
138 * works as expected when enabled
139 */ 133 */
140 if (unlikely(!base)) 134 if (unlikely(!base))
141 return NULL; 135 return NULL;
@@ -204,15 +198,12 @@ struct page_ext *lookup_page_ext(struct page *page)
204{ 198{
205 unsigned long pfn = page_to_pfn(page); 199 unsigned long pfn = page_to_pfn(page);
206 struct mem_section *section = __pfn_to_section(pfn); 200 struct mem_section *section = __pfn_to_section(pfn);
207#if defined(CONFIG_DEBUG_VM) || defined(CONFIG_PAGE_POISONING) 201#if defined(CONFIG_DEBUG_VM)
208 /* 202 /*
209 * The sanity checks the page allocator does upon freeing a 203 * The sanity checks the page allocator does upon freeing a
210 * page can reach here before the page_ext arrays are 204 * page can reach here before the page_ext arrays are
211 * allocated when feeding a range of pages to the allocator 205 * allocated when feeding a range of pages to the allocator
212 * for the first time during bootup or memory hotplug. 206 * for the first time during bootup or memory hotplug.
213 *
214 * This check is also necessary for ensuring page poisoning
215 * works as expected when enabled
216 */ 207 */
217 if (!section->page_ext) 208 if (!section->page_ext)
218 return NULL; 209 return NULL;
diff --git a/mm/page_idle.c b/mm/page_idle.c
index b0ee56c56b58..1b0f48c62316 100644
--- a/mm/page_idle.c
+++ b/mm/page_idle.c
@@ -50,7 +50,7 @@ static struct page *page_idle_get_page(unsigned long pfn)
50 return page; 50 return page;
51} 51}
52 52
53static int page_idle_clear_pte_refs_one(struct page *page, 53static bool page_idle_clear_pte_refs_one(struct page *page,
54 struct vm_area_struct *vma, 54 struct vm_area_struct *vma,
55 unsigned long addr, void *arg) 55 unsigned long addr, void *arg)
56{ 56{
@@ -84,7 +84,7 @@ static int page_idle_clear_pte_refs_one(struct page *page,
84 */ 84 */
85 set_page_young(page); 85 set_page_young(page);
86 } 86 }
87 return SWAP_AGAIN; 87 return true;
88} 88}
89 89
90static void page_idle_clear_pte_refs(struct page *page) 90static void page_idle_clear_pte_refs(struct page *page)
diff --git a/mm/page_isolation.c b/mm/page_isolation.c
index f4e17a57926a..5092e4ef00c8 100644
--- a/mm/page_isolation.c
+++ b/mm/page_isolation.c
@@ -66,7 +66,8 @@ out:
66 66
67 set_pageblock_migratetype(page, MIGRATE_ISOLATE); 67 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
68 zone->nr_isolate_pageblock++; 68 zone->nr_isolate_pageblock++;
69 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE); 69 nr_pages = move_freepages_block(zone, page, MIGRATE_ISOLATE,
70 NULL);
70 71
71 __mod_zone_freepage_state(zone, -nr_pages, migratetype); 72 __mod_zone_freepage_state(zone, -nr_pages, migratetype);
72 } 73 }
@@ -88,7 +89,7 @@ static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
88 89
89 zone = page_zone(page); 90 zone = page_zone(page);
90 spin_lock_irqsave(&zone->lock, flags); 91 spin_lock_irqsave(&zone->lock, flags);
91 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) 92 if (!is_migrate_isolate_page(page))
92 goto out; 93 goto out;
93 94
94 /* 95 /*
@@ -120,7 +121,7 @@ static void unset_migratetype_isolate(struct page *page, unsigned migratetype)
120 * pageblock scanning for freepage moving. 121 * pageblock scanning for freepage moving.
121 */ 122 */
122 if (!isolated_page) { 123 if (!isolated_page) {
123 nr_pages = move_freepages_block(zone, page, migratetype); 124 nr_pages = move_freepages_block(zone, page, migratetype, NULL);
124 __mod_zone_freepage_state(zone, nr_pages, migratetype); 125 __mod_zone_freepage_state(zone, nr_pages, migratetype);
125 } 126 }
126 set_pageblock_migratetype(page, migratetype); 127 set_pageblock_migratetype(page, migratetype);
@@ -205,7 +206,7 @@ int undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
205 pfn < end_pfn; 206 pfn < end_pfn;
206 pfn += pageblock_nr_pages) { 207 pfn += pageblock_nr_pages) {
207 page = __first_valid_page(pfn, pageblock_nr_pages); 208 page = __first_valid_page(pfn, pageblock_nr_pages);
208 if (!page || get_pageblock_migratetype(page) != MIGRATE_ISOLATE) 209 if (!page || !is_migrate_isolate_page(page))
209 continue; 210 continue;
210 unset_migratetype_isolate(page, migratetype); 211 unset_migratetype_isolate(page, migratetype);
211 } 212 }
@@ -262,7 +263,7 @@ int test_pages_isolated(unsigned long start_pfn, unsigned long end_pfn,
262 */ 263 */
263 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { 264 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
264 page = __first_valid_page(pfn, pageblock_nr_pages); 265 page = __first_valid_page(pfn, pageblock_nr_pages);
265 if (page && get_pageblock_migratetype(page) != MIGRATE_ISOLATE) 266 if (page && !is_migrate_isolate_page(page))
266 break; 267 break;
267 } 268 }
268 page = __first_valid_page(start_pfn, end_pfn - start_pfn); 269 page = __first_valid_page(start_pfn, end_pfn - start_pfn);
diff --git a/mm/page_poison.c b/mm/page_poison.c
index 2e647c65916b..be19e989ccff 100644
--- a/mm/page_poison.c
+++ b/mm/page_poison.c
@@ -6,7 +6,6 @@
6#include <linux/poison.h> 6#include <linux/poison.h>
7#include <linux/ratelimit.h> 7#include <linux/ratelimit.h>
8 8
9static bool __page_poisoning_enabled __read_mostly;
10static bool want_page_poisoning __read_mostly; 9static bool want_page_poisoning __read_mostly;
11 10
12static int early_page_poison_param(char *buf) 11static int early_page_poison_param(char *buf)
@@ -19,74 +18,21 @@ early_param("page_poison", early_page_poison_param);
19 18
20bool page_poisoning_enabled(void) 19bool page_poisoning_enabled(void)
21{ 20{
22 return __page_poisoning_enabled;
23}
24
25static bool need_page_poisoning(void)
26{
27 return want_page_poisoning;
28}
29
30static void init_page_poisoning(void)
31{
32 /* 21 /*
33 * page poisoning is debug page alloc for some arches. If either 22 * Assumes that debug_pagealloc_enabled is set before
34 * of those options are enabled, enable poisoning 23 * free_all_bootmem.
24 * Page poisoning is debug page alloc for some arches. If
25 * either of those options are enabled, enable poisoning.
35 */ 26 */
36 if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC)) { 27 return (want_page_poisoning ||
37 if (!want_page_poisoning && !debug_pagealloc_enabled()) 28 (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
38 return; 29 debug_pagealloc_enabled()));
39 } else {
40 if (!want_page_poisoning)
41 return;
42 }
43
44 __page_poisoning_enabled = true;
45}
46
47struct page_ext_operations page_poisoning_ops = {
48 .need = need_page_poisoning,
49 .init = init_page_poisoning,
50};
51
52static inline void set_page_poison(struct page *page)
53{
54 struct page_ext *page_ext;
55
56 page_ext = lookup_page_ext(page);
57 if (unlikely(!page_ext))
58 return;
59
60 __set_bit(PAGE_EXT_DEBUG_POISON, &page_ext->flags);
61}
62
63static inline void clear_page_poison(struct page *page)
64{
65 struct page_ext *page_ext;
66
67 page_ext = lookup_page_ext(page);
68 if (unlikely(!page_ext))
69 return;
70
71 __clear_bit(PAGE_EXT_DEBUG_POISON, &page_ext->flags);
72}
73
74bool page_is_poisoned(struct page *page)
75{
76 struct page_ext *page_ext;
77
78 page_ext = lookup_page_ext(page);
79 if (unlikely(!page_ext))
80 return false;
81
82 return test_bit(PAGE_EXT_DEBUG_POISON, &page_ext->flags);
83} 30}
84 31
85static void poison_page(struct page *page) 32static void poison_page(struct page *page)
86{ 33{
87 void *addr = kmap_atomic(page); 34 void *addr = kmap_atomic(page);
88 35
89 set_page_poison(page);
90 memset(addr, PAGE_POISON, PAGE_SIZE); 36 memset(addr, PAGE_POISON, PAGE_SIZE);
91 kunmap_atomic(addr); 37 kunmap_atomic(addr);
92} 38}
@@ -140,12 +86,13 @@ static void unpoison_page(struct page *page)
140{ 86{
141 void *addr; 87 void *addr;
142 88
143 if (!page_is_poisoned(page))
144 return;
145
146 addr = kmap_atomic(page); 89 addr = kmap_atomic(page);
90 /*
91 * Page poisoning when enabled poisons each and every page
92 * that is freed to buddy. Thus no extra check is done to
93 * see if a page was posioned.
94 */
147 check_poison_mem(addr, PAGE_SIZE); 95 check_poison_mem(addr, PAGE_SIZE);
148 clear_page_poison(page);
149 kunmap_atomic(addr); 96 kunmap_atomic(addr);
150} 97}
151 98
diff --git a/mm/percpu.c b/mm/percpu.c
index 60a6488e9e6d..e0aa8ae7bde7 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -1284,18 +1284,7 @@ void free_percpu(void __percpu *ptr)
1284} 1284}
1285EXPORT_SYMBOL_GPL(free_percpu); 1285EXPORT_SYMBOL_GPL(free_percpu);
1286 1286
1287/** 1287bool __is_kernel_percpu_address(unsigned long addr, unsigned long *can_addr)
1288 * is_kernel_percpu_address - test whether address is from static percpu area
1289 * @addr: address to test
1290 *
1291 * Test whether @addr belongs to in-kernel static percpu area. Module
1292 * static percpu areas are not considered. For those, use
1293 * is_module_percpu_address().
1294 *
1295 * RETURNS:
1296 * %true if @addr is from in-kernel static percpu area, %false otherwise.
1297 */
1298bool is_kernel_percpu_address(unsigned long addr)
1299{ 1288{
1300#ifdef CONFIG_SMP 1289#ifdef CONFIG_SMP
1301 const size_t static_size = __per_cpu_end - __per_cpu_start; 1290 const size_t static_size = __per_cpu_end - __per_cpu_start;
@@ -1304,16 +1293,39 @@ bool is_kernel_percpu_address(unsigned long addr)
1304 1293
1305 for_each_possible_cpu(cpu) { 1294 for_each_possible_cpu(cpu) {
1306 void *start = per_cpu_ptr(base, cpu); 1295 void *start = per_cpu_ptr(base, cpu);
1296 void *va = (void *)addr;
1307 1297
1308 if ((void *)addr >= start && (void *)addr < start + static_size) 1298 if (va >= start && va < start + static_size) {
1299 if (can_addr) {
1300 *can_addr = (unsigned long) (va - start);
1301 *can_addr += (unsigned long)
1302 per_cpu_ptr(base, get_boot_cpu_id());
1303 }
1309 return true; 1304 return true;
1310 } 1305 }
1306 }
1311#endif 1307#endif
1312 /* on UP, can't distinguish from other static vars, always false */ 1308 /* on UP, can't distinguish from other static vars, always false */
1313 return false; 1309 return false;
1314} 1310}
1315 1311
1316/** 1312/**
1313 * is_kernel_percpu_address - test whether address is from static percpu area
1314 * @addr: address to test
1315 *
1316 * Test whether @addr belongs to in-kernel static percpu area. Module
1317 * static percpu areas are not considered. For those, use
1318 * is_module_percpu_address().
1319 *
1320 * RETURNS:
1321 * %true if @addr is from in-kernel static percpu area, %false otherwise.
1322 */
1323bool is_kernel_percpu_address(unsigned long addr)
1324{
1325 return __is_kernel_percpu_address(addr, NULL);
1326}
1327
1328/**
1317 * per_cpu_ptr_to_phys - convert translated percpu address to physical address 1329 * per_cpu_ptr_to_phys - convert translated percpu address to physical address
1318 * @addr: the address to be converted to physical address 1330 * @addr: the address to be converted to physical address
1319 * 1331 *
diff --git a/mm/rmap.c b/mm/rmap.c
index f6838015810f..3ff241f714eb 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -724,7 +724,7 @@ struct page_referenced_arg {
724/* 724/*
725 * arg: page_referenced_arg will be passed 725 * arg: page_referenced_arg will be passed
726 */ 726 */
727static int page_referenced_one(struct page *page, struct vm_area_struct *vma, 727static bool page_referenced_one(struct page *page, struct vm_area_struct *vma,
728 unsigned long address, void *arg) 728 unsigned long address, void *arg)
729{ 729{
730 struct page_referenced_arg *pra = arg; 730 struct page_referenced_arg *pra = arg;
@@ -741,7 +741,7 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
741 if (vma->vm_flags & VM_LOCKED) { 741 if (vma->vm_flags & VM_LOCKED) {
742 page_vma_mapped_walk_done(&pvmw); 742 page_vma_mapped_walk_done(&pvmw);
743 pra->vm_flags |= VM_LOCKED; 743 pra->vm_flags |= VM_LOCKED;
744 return SWAP_FAIL; /* To break the loop */ 744 return false; /* To break the loop */
745 } 745 }
746 746
747 if (pvmw.pte) { 747 if (pvmw.pte) {
@@ -781,9 +781,9 @@ static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
781 } 781 }
782 782
783 if (!pra->mapcount) 783 if (!pra->mapcount)
784 return SWAP_SUCCESS; /* To break the loop */ 784 return false; /* To break the loop */
785 785
786 return SWAP_AGAIN; 786 return true;
787} 787}
788 788
789static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) 789static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
@@ -812,7 +812,6 @@ int page_referenced(struct page *page,
812 struct mem_cgroup *memcg, 812 struct mem_cgroup *memcg,
813 unsigned long *vm_flags) 813 unsigned long *vm_flags)
814{ 814{
815 int ret;
816 int we_locked = 0; 815 int we_locked = 0;
817 struct page_referenced_arg pra = { 816 struct page_referenced_arg pra = {
818 .mapcount = total_mapcount(page), 817 .mapcount = total_mapcount(page),
@@ -846,7 +845,7 @@ int page_referenced(struct page *page,
846 rwc.invalid_vma = invalid_page_referenced_vma; 845 rwc.invalid_vma = invalid_page_referenced_vma;
847 } 846 }
848 847
849 ret = rmap_walk(page, &rwc); 848 rmap_walk(page, &rwc);
850 *vm_flags = pra.vm_flags; 849 *vm_flags = pra.vm_flags;
851 850
852 if (we_locked) 851 if (we_locked)
@@ -855,7 +854,7 @@ int page_referenced(struct page *page,
855 return pra.referenced; 854 return pra.referenced;
856} 855}
857 856
858static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, 857static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
859 unsigned long address, void *arg) 858 unsigned long address, void *arg)
860{ 859{
861 struct page_vma_mapped_walk pvmw = { 860 struct page_vma_mapped_walk pvmw = {
@@ -908,7 +907,7 @@ static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
908 } 907 }
909 } 908 }
910 909
911 return SWAP_AGAIN; 910 return true;
912} 911}
913 912
914static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) 913static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
@@ -1159,7 +1158,7 @@ void page_add_file_rmap(struct page *page, bool compound)
1159 goto out; 1158 goto out;
1160 } 1159 }
1161 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, nr); 1160 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, nr);
1162 mem_cgroup_update_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED, nr); 1161 mod_memcg_page_state(page, NR_FILE_MAPPED, nr);
1163out: 1162out:
1164 unlock_page_memcg(page); 1163 unlock_page_memcg(page);
1165} 1164}
@@ -1199,7 +1198,7 @@ static void page_remove_file_rmap(struct page *page, bool compound)
1199 * pte lock(a spinlock) is held, which implies preemption disabled. 1198 * pte lock(a spinlock) is held, which implies preemption disabled.
1200 */ 1199 */
1201 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, -nr); 1200 __mod_node_page_state(page_pgdat(page), NR_FILE_MAPPED, -nr);
1202 mem_cgroup_update_page_stat(page, MEM_CGROUP_STAT_FILE_MAPPED, -nr); 1201 mod_memcg_page_state(page, NR_FILE_MAPPED, -nr);
1203 1202
1204 if (unlikely(PageMlocked(page))) 1203 if (unlikely(PageMlocked(page)))
1205 clear_page_mlock(page); 1204 clear_page_mlock(page);
@@ -1288,15 +1287,10 @@ void page_remove_rmap(struct page *page, bool compound)
1288 */ 1287 */
1289} 1288}
1290 1289
1291struct rmap_private {
1292 enum ttu_flags flags;
1293 int lazyfreed;
1294};
1295
1296/* 1290/*
1297 * @arg: enum ttu_flags will be passed to this argument 1291 * @arg: enum ttu_flags will be passed to this argument
1298 */ 1292 */
1299static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, 1293static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1300 unsigned long address, void *arg) 1294 unsigned long address, void *arg)
1301{ 1295{
1302 struct mm_struct *mm = vma->vm_mm; 1296 struct mm_struct *mm = vma->vm_mm;
@@ -1307,13 +1301,12 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1307 }; 1301 };
1308 pte_t pteval; 1302 pte_t pteval;
1309 struct page *subpage; 1303 struct page *subpage;
1310 int ret = SWAP_AGAIN; 1304 bool ret = true;
1311 struct rmap_private *rp = arg; 1305 enum ttu_flags flags = (enum ttu_flags)arg;
1312 enum ttu_flags flags = rp->flags;
1313 1306
1314 /* munlock has nothing to gain from examining un-locked vmas */ 1307 /* munlock has nothing to gain from examining un-locked vmas */
1315 if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) 1308 if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
1316 return SWAP_AGAIN; 1309 return true;
1317 1310
1318 if (flags & TTU_SPLIT_HUGE_PMD) { 1311 if (flags & TTU_SPLIT_HUGE_PMD) {
1319 split_huge_pmd_address(vma, address, 1312 split_huge_pmd_address(vma, address,
@@ -1336,7 +1329,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1336 */ 1329 */
1337 mlock_vma_page(page); 1330 mlock_vma_page(page);
1338 } 1331 }
1339 ret = SWAP_MLOCK; 1332 ret = false;
1340 page_vma_mapped_walk_done(&pvmw); 1333 page_vma_mapped_walk_done(&pvmw);
1341 break; 1334 break;
1342 } 1335 }
@@ -1354,7 +1347,7 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1354 if (!(flags & TTU_IGNORE_ACCESS)) { 1347 if (!(flags & TTU_IGNORE_ACCESS)) {
1355 if (ptep_clear_flush_young_notify(vma, address, 1348 if (ptep_clear_flush_young_notify(vma, address,
1356 pvmw.pte)) { 1349 pvmw.pte)) {
1357 ret = SWAP_FAIL; 1350 ret = false;
1358 page_vma_mapped_walk_done(&pvmw); 1351 page_vma_mapped_walk_done(&pvmw);
1359 break; 1352 break;
1360 } 1353 }
@@ -1424,18 +1417,34 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1424 * Store the swap location in the pte. 1417 * Store the swap location in the pte.
1425 * See handle_pte_fault() ... 1418 * See handle_pte_fault() ...
1426 */ 1419 */
1427 VM_BUG_ON_PAGE(!PageSwapCache(page), page); 1420 if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) {
1421 WARN_ON_ONCE(1);
1422 ret = false;
1423 page_vma_mapped_walk_done(&pvmw);
1424 break;
1425 }
1426
1427 /* MADV_FREE page check */
1428 if (!PageSwapBacked(page)) {
1429 if (!PageDirty(page)) {
1430 dec_mm_counter(mm, MM_ANONPAGES);
1431 goto discard;
1432 }
1428 1433
1429 if (!PageDirty(page) && (flags & TTU_LZFREE)) { 1434 /*
1430 /* It's a freeable page by MADV_FREE */ 1435 * If the page was redirtied, it cannot be
1431 dec_mm_counter(mm, MM_ANONPAGES); 1436 * discarded. Remap the page to page table.
1432 rp->lazyfreed++; 1437 */
1433 goto discard; 1438 set_pte_at(mm, address, pvmw.pte, pteval);
1439 SetPageSwapBacked(page);
1440 ret = false;
1441 page_vma_mapped_walk_done(&pvmw);
1442 break;
1434 } 1443 }
1435 1444
1436 if (swap_duplicate(entry) < 0) { 1445 if (swap_duplicate(entry) < 0) {
1437 set_pte_at(mm, address, pvmw.pte, pteval); 1446 set_pte_at(mm, address, pvmw.pte, pteval);
1438 ret = SWAP_FAIL; 1447 ret = false;
1439 page_vma_mapped_walk_done(&pvmw); 1448 page_vma_mapped_walk_done(&pvmw);
1440 break; 1449 break;
1441 } 1450 }
@@ -1492,24 +1501,14 @@ static int page_mapcount_is_zero(struct page *page)
1492 * 1501 *
1493 * Tries to remove all the page table entries which are mapping this 1502 * Tries to remove all the page table entries which are mapping this
1494 * page, used in the pageout path. Caller must hold the page lock. 1503 * page, used in the pageout path. Caller must hold the page lock.
1495 * Return values are:
1496 * 1504 *
1497 * SWAP_SUCCESS - we succeeded in removing all mappings 1505 * If unmap is successful, return true. Otherwise, false.
1498 * SWAP_AGAIN - we missed a mapping, try again later
1499 * SWAP_FAIL - the page is unswappable
1500 * SWAP_MLOCK - page is mlocked.
1501 */ 1506 */
1502int try_to_unmap(struct page *page, enum ttu_flags flags) 1507bool try_to_unmap(struct page *page, enum ttu_flags flags)
1503{ 1508{
1504 int ret;
1505 struct rmap_private rp = {
1506 .flags = flags,
1507 .lazyfreed = 0,
1508 };
1509
1510 struct rmap_walk_control rwc = { 1509 struct rmap_walk_control rwc = {
1511 .rmap_one = try_to_unmap_one, 1510 .rmap_one = try_to_unmap_one,
1512 .arg = &rp, 1511 .arg = (void *)flags,
1513 .done = page_mapcount_is_zero, 1512 .done = page_mapcount_is_zero,
1514 .anon_lock = page_lock_anon_vma_read, 1513 .anon_lock = page_lock_anon_vma_read,
1515 }; 1514 };
@@ -1526,16 +1525,11 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
1526 rwc.invalid_vma = invalid_migration_vma; 1525 rwc.invalid_vma = invalid_migration_vma;
1527 1526
1528 if (flags & TTU_RMAP_LOCKED) 1527 if (flags & TTU_RMAP_LOCKED)
1529 ret = rmap_walk_locked(page, &rwc); 1528 rmap_walk_locked(page, &rwc);
1530 else 1529 else
1531 ret = rmap_walk(page, &rwc); 1530 rmap_walk(page, &rwc);
1532 1531
1533 if (ret != SWAP_MLOCK && !page_mapcount(page)) { 1532 return !page_mapcount(page) ? true : false;
1534 ret = SWAP_SUCCESS;
1535 if (rp.lazyfreed && !PageDirty(page))
1536 ret = SWAP_LZFREE;
1537 }
1538 return ret;
1539} 1533}
1540 1534
1541static int page_not_mapped(struct page *page) 1535static int page_not_mapped(struct page *page)
@@ -1550,34 +1544,22 @@ static int page_not_mapped(struct page *page)
1550 * Called from munlock code. Checks all of the VMAs mapping the page 1544 * Called from munlock code. Checks all of the VMAs mapping the page
1551 * to make sure nobody else has this page mlocked. The page will be 1545 * to make sure nobody else has this page mlocked. The page will be
1552 * returned with PG_mlocked cleared if no other vmas have it mlocked. 1546 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1553 *
1554 * Return values are:
1555 *
1556 * SWAP_AGAIN - no vma is holding page mlocked, or,
1557 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1558 * SWAP_FAIL - page cannot be located at present
1559 * SWAP_MLOCK - page is now mlocked.
1560 */ 1547 */
1561int try_to_munlock(struct page *page)
1562{
1563 int ret;
1564 struct rmap_private rp = {
1565 .flags = TTU_MUNLOCK,
1566 .lazyfreed = 0,
1567 };
1568 1548
1549void try_to_munlock(struct page *page)
1550{
1569 struct rmap_walk_control rwc = { 1551 struct rmap_walk_control rwc = {
1570 .rmap_one = try_to_unmap_one, 1552 .rmap_one = try_to_unmap_one,
1571 .arg = &rp, 1553 .arg = (void *)TTU_MUNLOCK,
1572 .done = page_not_mapped, 1554 .done = page_not_mapped,
1573 .anon_lock = page_lock_anon_vma_read, 1555 .anon_lock = page_lock_anon_vma_read,
1574 1556
1575 }; 1557 };
1576 1558
1577 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); 1559 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
1560 VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
1578 1561
1579 ret = rmap_walk(page, &rwc); 1562 rmap_walk(page, &rwc);
1580 return ret;
1581} 1563}
1582 1564
1583void __put_anon_vma(struct anon_vma *anon_vma) 1565void __put_anon_vma(struct anon_vma *anon_vma)
@@ -1625,13 +1607,12 @@ static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1625 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1607 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1626 * LOCKED. 1608 * LOCKED.
1627 */ 1609 */
1628static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, 1610static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1629 bool locked) 1611 bool locked)
1630{ 1612{
1631 struct anon_vma *anon_vma; 1613 struct anon_vma *anon_vma;
1632 pgoff_t pgoff_start, pgoff_end; 1614 pgoff_t pgoff_start, pgoff_end;
1633 struct anon_vma_chain *avc; 1615 struct anon_vma_chain *avc;
1634 int ret = SWAP_AGAIN;
1635 1616
1636 if (locked) { 1617 if (locked) {
1637 anon_vma = page_anon_vma(page); 1618 anon_vma = page_anon_vma(page);
@@ -1641,7 +1622,7 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1641 anon_vma = rmap_walk_anon_lock(page, rwc); 1622 anon_vma = rmap_walk_anon_lock(page, rwc);
1642 } 1623 }
1643 if (!anon_vma) 1624 if (!anon_vma)
1644 return ret; 1625 return;
1645 1626
1646 pgoff_start = page_to_pgoff(page); 1627 pgoff_start = page_to_pgoff(page);
1647 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; 1628 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
@@ -1655,8 +1636,7 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1655 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) 1636 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1656 continue; 1637 continue;
1657 1638
1658 ret = rwc->rmap_one(page, vma, address, rwc->arg); 1639 if (!rwc->rmap_one(page, vma, address, rwc->arg))
1659 if (ret != SWAP_AGAIN)
1660 break; 1640 break;
1661 if (rwc->done && rwc->done(page)) 1641 if (rwc->done && rwc->done(page))
1662 break; 1642 break;
@@ -1664,7 +1644,6 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1664 1644
1665 if (!locked) 1645 if (!locked)
1666 anon_vma_unlock_read(anon_vma); 1646 anon_vma_unlock_read(anon_vma);
1667 return ret;
1668} 1647}
1669 1648
1670/* 1649/*
@@ -1680,13 +1659,12 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1680 * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1659 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1681 * LOCKED. 1660 * LOCKED.
1682 */ 1661 */
1683static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, 1662static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
1684 bool locked) 1663 bool locked)
1685{ 1664{
1686 struct address_space *mapping = page_mapping(page); 1665 struct address_space *mapping = page_mapping(page);
1687 pgoff_t pgoff_start, pgoff_end; 1666 pgoff_t pgoff_start, pgoff_end;
1688 struct vm_area_struct *vma; 1667 struct vm_area_struct *vma;
1689 int ret = SWAP_AGAIN;
1690 1668
1691 /* 1669 /*
1692 * The page lock not only makes sure that page->mapping cannot 1670 * The page lock not only makes sure that page->mapping cannot
@@ -1697,7 +1675,7 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
1697 VM_BUG_ON_PAGE(!PageLocked(page), page); 1675 VM_BUG_ON_PAGE(!PageLocked(page), page);
1698 1676
1699 if (!mapping) 1677 if (!mapping)
1700 return ret; 1678 return;
1701 1679
1702 pgoff_start = page_to_pgoff(page); 1680 pgoff_start = page_to_pgoff(page);
1703 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; 1681 pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
@@ -1712,8 +1690,7 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
1712 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) 1690 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1713 continue; 1691 continue;
1714 1692
1715 ret = rwc->rmap_one(page, vma, address, rwc->arg); 1693 if (!rwc->rmap_one(page, vma, address, rwc->arg))
1716 if (ret != SWAP_AGAIN)
1717 goto done; 1694 goto done;
1718 if (rwc->done && rwc->done(page)) 1695 if (rwc->done && rwc->done(page))
1719 goto done; 1696 goto done;
@@ -1722,28 +1699,27 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
1722done: 1699done:
1723 if (!locked) 1700 if (!locked)
1724 i_mmap_unlock_read(mapping); 1701 i_mmap_unlock_read(mapping);
1725 return ret;
1726} 1702}
1727 1703
1728int rmap_walk(struct page *page, struct rmap_walk_control *rwc) 1704void rmap_walk(struct page *page, struct rmap_walk_control *rwc)
1729{ 1705{
1730 if (unlikely(PageKsm(page))) 1706 if (unlikely(PageKsm(page)))
1731 return rmap_walk_ksm(page, rwc); 1707 rmap_walk_ksm(page, rwc);
1732 else if (PageAnon(page)) 1708 else if (PageAnon(page))
1733 return rmap_walk_anon(page, rwc, false); 1709 rmap_walk_anon(page, rwc, false);
1734 else 1710 else
1735 return rmap_walk_file(page, rwc, false); 1711 rmap_walk_file(page, rwc, false);
1736} 1712}
1737 1713
1738/* Like rmap_walk, but caller holds relevant rmap lock */ 1714/* Like rmap_walk, but caller holds relevant rmap lock */
1739int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc) 1715void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc)
1740{ 1716{
1741 /* no ksm support for now */ 1717 /* no ksm support for now */
1742 VM_BUG_ON_PAGE(PageKsm(page), page); 1718 VM_BUG_ON_PAGE(PageKsm(page), page);
1743 if (PageAnon(page)) 1719 if (PageAnon(page))
1744 return rmap_walk_anon(page, rwc, true); 1720 rmap_walk_anon(page, rwc, true);
1745 else 1721 else
1746 return rmap_walk_file(page, rwc, true); 1722 rmap_walk_file(page, rwc, true);
1747} 1723}
1748 1724
1749#ifdef CONFIG_HUGETLB_PAGE 1725#ifdef CONFIG_HUGETLB_PAGE
diff --git a/mm/rodata_test.c b/mm/rodata_test.c
index 0fd21670b513..6bb4deb12e78 100644
--- a/mm/rodata_test.c
+++ b/mm/rodata_test.c
@@ -9,11 +9,12 @@
9 * as published by the Free Software Foundation; version 2 9 * as published by the Free Software Foundation; version 2
10 * of the License. 10 * of the License.
11 */ 11 */
12#define pr_fmt(fmt) "rodata_test: " fmt
13
12#include <linux/uaccess.h> 14#include <linux/uaccess.h>
13#include <asm/sections.h> 15#include <asm/sections.h>
14 16
15const int rodata_test_data = 0xC3; 17const int rodata_test_data = 0xC3;
16EXPORT_SYMBOL_GPL(rodata_test_data);
17 18
18void rodata_test(void) 19void rodata_test(void)
19{ 20{
@@ -23,20 +24,20 @@ void rodata_test(void)
23 /* test 1: read the value */ 24 /* test 1: read the value */
24 /* If this test fails, some previous testrun has clobbered the state */ 25 /* If this test fails, some previous testrun has clobbered the state */
25 if (!rodata_test_data) { 26 if (!rodata_test_data) {
26 pr_err("rodata_test: test 1 fails (start data)\n"); 27 pr_err("test 1 fails (start data)\n");
27 return; 28 return;
28 } 29 }
29 30
30 /* test 2: write to the variable; this should fault */ 31 /* test 2: write to the variable; this should fault */
31 if (!probe_kernel_write((void *)&rodata_test_data, 32 if (!probe_kernel_write((void *)&rodata_test_data,
32 (void *)&zero, sizeof(zero))) { 33 (void *)&zero, sizeof(zero))) {
33 pr_err("rodata_test: test data was not read only\n"); 34 pr_err("test data was not read only\n");
34 return; 35 return;
35 } 36 }
36 37
37 /* test 3: check the value hasn't changed */ 38 /* test 3: check the value hasn't changed */
38 if (rodata_test_data == zero) { 39 if (rodata_test_data == zero) {
39 pr_err("rodata_test: test data was changed\n"); 40 pr_err("test data was changed\n");
40 return; 41 return;
41 } 42 }
42 43
@@ -44,13 +45,13 @@ void rodata_test(void)
44 start = (unsigned long)__start_rodata; 45 start = (unsigned long)__start_rodata;
45 end = (unsigned long)__end_rodata; 46 end = (unsigned long)__end_rodata;
46 if (start & (PAGE_SIZE - 1)) { 47 if (start & (PAGE_SIZE - 1)) {
47 pr_err("rodata_test: start of .rodata is not page size aligned\n"); 48 pr_err("start of .rodata is not page size aligned\n");
48 return; 49 return;
49 } 50 }
50 if (end & (PAGE_SIZE - 1)) { 51 if (end & (PAGE_SIZE - 1)) {
51 pr_err("rodata_test: end of .rodata is not page size aligned\n"); 52 pr_err("end of .rodata is not page size aligned\n");
52 return; 53 return;
53 } 54 }
54 55
55 pr_info("rodata_test: all tests were successful\n"); 56 pr_info("all tests were successful\n");
56} 57}
diff --git a/mm/slab.c b/mm/slab.c
index 807d86c76908..1880d482a0cb 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -3879,7 +3879,12 @@ static int __do_tune_cpucache(struct kmem_cache *cachep, int limit,
3879 3879
3880 prev = cachep->cpu_cache; 3880 prev = cachep->cpu_cache;
3881 cachep->cpu_cache = cpu_cache; 3881 cachep->cpu_cache = cpu_cache;
3882 kick_all_cpus_sync(); 3882 /*
3883 * Without a previous cpu_cache there's no need to synchronize remote
3884 * cpus, so skip the IPIs.
3885 */
3886 if (prev)
3887 kick_all_cpus_sync();
3883 3888
3884 check_irq_on(); 3889 check_irq_on();
3885 cachep->batchcount = batchcount; 3890 cachep->batchcount = batchcount;
diff --git a/mm/sparse.c b/mm/sparse.c
index db6bf3c97ea2..6903c8fc3085 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -248,10 +248,7 @@ static int __meminit sparse_init_one_section(struct mem_section *ms,
248 248
249unsigned long usemap_size(void) 249unsigned long usemap_size(void)
250{ 250{
251 unsigned long size_bytes; 251 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
252 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
253 size_bytes = roundup(size_bytes, sizeof(unsigned long));
254 return size_bytes;
255} 252}
256 253
257#ifdef CONFIG_MEMORY_HOTPLUG 254#ifdef CONFIG_MEMORY_HOTPLUG
diff --git a/mm/swap.c b/mm/swap.c
index 5dabf444d724..98d08b4579fa 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -46,7 +46,7 @@ int page_cluster;
46static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); 46static DEFINE_PER_CPU(struct pagevec, lru_add_pvec);
47static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); 47static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
48static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs); 48static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs);
49static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs); 49static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs);
50#ifdef CONFIG_SMP 50#ifdef CONFIG_SMP
51static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); 51static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs);
52#endif 52#endif
@@ -97,6 +97,16 @@ static void __put_compound_page(struct page *page)
97 97
98void __put_page(struct page *page) 98void __put_page(struct page *page)
99{ 99{
100 if (is_zone_device_page(page)) {
101 put_dev_pagemap(page->pgmap);
102
103 /*
104 * The page belongs to the device that created pgmap. Do
105 * not return it to page allocator.
106 */
107 return;
108 }
109
100 if (unlikely(PageCompound(page))) 110 if (unlikely(PageCompound(page)))
101 __put_compound_page(page); 111 __put_compound_page(page);
102 else 112 else
@@ -561,20 +571,27 @@ static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec,
561} 571}
562 572
563 573
564static void lru_deactivate_fn(struct page *page, struct lruvec *lruvec, 574static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec,
565 void *arg) 575 void *arg)
566{ 576{
567 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { 577 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
568 int file = page_is_file_cache(page); 578 !PageUnevictable(page)) {
569 int lru = page_lru_base_type(page); 579 bool active = PageActive(page);
570 580
571 del_page_from_lru_list(page, lruvec, lru + LRU_ACTIVE); 581 del_page_from_lru_list(page, lruvec,
582 LRU_INACTIVE_ANON + active);
572 ClearPageActive(page); 583 ClearPageActive(page);
573 ClearPageReferenced(page); 584 ClearPageReferenced(page);
574 add_page_to_lru_list(page, lruvec, lru); 585 /*
586 * lazyfree pages are clean anonymous pages. They have
587 * SwapBacked flag cleared to distinguish normal anonymous
588 * pages
589 */
590 ClearPageSwapBacked(page);
591 add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE);
575 592
576 __count_vm_event(PGDEACTIVATE); 593 __count_vm_events(PGLAZYFREE, hpage_nr_pages(page));
577 update_page_reclaim_stat(lruvec, file, 0); 594 update_page_reclaim_stat(lruvec, 1, 0);
578 } 595 }
579} 596}
580 597
@@ -604,9 +621,9 @@ void lru_add_drain_cpu(int cpu)
604 if (pagevec_count(pvec)) 621 if (pagevec_count(pvec))
605 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); 622 pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL);
606 623
607 pvec = &per_cpu(lru_deactivate_pvecs, cpu); 624 pvec = &per_cpu(lru_lazyfree_pvecs, cpu);
608 if (pagevec_count(pvec)) 625 if (pagevec_count(pvec))
609 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); 626 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
610 627
611 activate_page_drain(cpu); 628 activate_page_drain(cpu);
612} 629}
@@ -638,22 +655,22 @@ void deactivate_file_page(struct page *page)
638} 655}
639 656
640/** 657/**
641 * deactivate_page - deactivate a page 658 * mark_page_lazyfree - make an anon page lazyfree
642 * @page: page to deactivate 659 * @page: page to deactivate
643 * 660 *
644 * deactivate_page() moves @page to the inactive list if @page was on the active 661 * mark_page_lazyfree() moves @page to the inactive file list.
645 * list and was not an unevictable page. This is done to accelerate the reclaim 662 * This is done to accelerate the reclaim of @page.
646 * of @page.
647 */ 663 */
648void deactivate_page(struct page *page) 664void mark_page_lazyfree(struct page *page)
649{ 665{
650 if (PageLRU(page) && PageActive(page) && !PageUnevictable(page)) { 666 if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) &&
651 struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs); 667 !PageUnevictable(page)) {
668 struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs);
652 669
653 get_page(page); 670 get_page(page);
654 if (!pagevec_add(pvec, page) || PageCompound(page)) 671 if (!pagevec_add(pvec, page) || PageCompound(page))
655 pagevec_lru_move_fn(pvec, lru_deactivate_fn, NULL); 672 pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL);
656 put_cpu_var(lru_deactivate_pvecs); 673 put_cpu_var(lru_lazyfree_pvecs);
657 } 674 }
658} 675}
659 676
@@ -693,7 +710,7 @@ void lru_add_drain_all(void)
693 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || 710 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
694 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || 711 pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
695 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) || 712 pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
696 pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) || 713 pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
697 need_activate_page_drain(cpu)) { 714 need_activate_page_drain(cpu)) {
698 INIT_WORK(work, lru_add_drain_per_cpu); 715 INIT_WORK(work, lru_add_drain_per_cpu);
699 queue_work_on(cpu, mm_percpu_wq, work); 716 queue_work_on(cpu, mm_percpu_wq, work);
diff --git a/mm/swap_slots.c b/mm/swap_slots.c
index b1ccb58ad397..58f6c78f1dad 100644
--- a/mm/swap_slots.c
+++ b/mm/swap_slots.c
@@ -31,6 +31,7 @@
31#include <linux/cpumask.h> 31#include <linux/cpumask.h>
32#include <linux/vmalloc.h> 32#include <linux/vmalloc.h>
33#include <linux/mutex.h> 33#include <linux/mutex.h>
34#include <linux/mm.h>
34 35
35#ifdef CONFIG_SWAP 36#ifdef CONFIG_SWAP
36 37
@@ -119,16 +120,18 @@ static int alloc_swap_slot_cache(unsigned int cpu)
119 120
120 /* 121 /*
121 * Do allocation outside swap_slots_cache_mutex 122 * Do allocation outside swap_slots_cache_mutex
122 * as vzalloc could trigger reclaim and get_swap_page, 123 * as kvzalloc could trigger reclaim and get_swap_page,
123 * which can lock swap_slots_cache_mutex. 124 * which can lock swap_slots_cache_mutex.
124 */ 125 */
125 slots = vzalloc(sizeof(swp_entry_t) * SWAP_SLOTS_CACHE_SIZE); 126 slots = kvzalloc(sizeof(swp_entry_t) * SWAP_SLOTS_CACHE_SIZE,
127 GFP_KERNEL);
126 if (!slots) 128 if (!slots)
127 return -ENOMEM; 129 return -ENOMEM;
128 130
129 slots_ret = vzalloc(sizeof(swp_entry_t) * SWAP_SLOTS_CACHE_SIZE); 131 slots_ret = kvzalloc(sizeof(swp_entry_t) * SWAP_SLOTS_CACHE_SIZE,
132 GFP_KERNEL);
130 if (!slots_ret) { 133 if (!slots_ret) {
131 vfree(slots); 134 kvfree(slots);
132 return -ENOMEM; 135 return -ENOMEM;
133 } 136 }
134 137
@@ -152,9 +155,9 @@ static int alloc_swap_slot_cache(unsigned int cpu)
152out: 155out:
153 mutex_unlock(&swap_slots_cache_mutex); 156 mutex_unlock(&swap_slots_cache_mutex);
154 if (slots) 157 if (slots)
155 vfree(slots); 158 kvfree(slots);
156 if (slots_ret) 159 if (slots_ret)
157 vfree(slots_ret); 160 kvfree(slots_ret);
158 return 0; 161 return 0;
159} 162}
160 163
@@ -171,7 +174,7 @@ static void drain_slots_cache_cpu(unsigned int cpu, unsigned int type,
171 cache->cur = 0; 174 cache->cur = 0;
172 cache->nr = 0; 175 cache->nr = 0;
173 if (free_slots && cache->slots) { 176 if (free_slots && cache->slots) {
174 vfree(cache->slots); 177 kvfree(cache->slots);
175 cache->slots = NULL; 178 cache->slots = NULL;
176 } 179 }
177 mutex_unlock(&cache->alloc_lock); 180 mutex_unlock(&cache->alloc_lock);
@@ -186,7 +189,7 @@ static void drain_slots_cache_cpu(unsigned int cpu, unsigned int type,
186 } 189 }
187 spin_unlock_irq(&cache->free_lock); 190 spin_unlock_irq(&cache->free_lock);
188 if (slots) 191 if (slots)
189 vfree(slots); 192 kvfree(slots);
190 } 193 }
191} 194}
192 195
@@ -241,8 +244,10 @@ int enable_swap_slots_cache(void)
241 244
242 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "swap_slots_cache", 245 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "swap_slots_cache",
243 alloc_swap_slot_cache, free_slot_cache); 246 alloc_swap_slot_cache, free_slot_cache);
244 if (ret < 0) 247 if (WARN_ONCE(ret < 0, "Cache allocation failed (%s), operating "
248 "without swap slots cache.\n", __func__))
245 goto out_unlock; 249 goto out_unlock;
250
246 swap_slot_cache_initialized = true; 251 swap_slot_cache_initialized = true;
247 __reenable_swap_slots_cache(); 252 __reenable_swap_slots_cache();
248out_unlock: 253out_unlock:
diff --git a/mm/swap_state.c b/mm/swap_state.c
index 473b71e052a8..539b8885e3d1 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -360,17 +360,7 @@ struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
360 /* 360 /*
361 * We might race against get_swap_page() and stumble 361 * We might race against get_swap_page() and stumble
362 * across a SWAP_HAS_CACHE swap_map entry whose page 362 * across a SWAP_HAS_CACHE swap_map entry whose page
363 * has not been brought into the swapcache yet, while 363 * has not been brought into the swapcache yet.
364 * the other end is scheduled away waiting on discard
365 * I/O completion at scan_swap_map().
366 *
367 * In order to avoid turning this transitory state
368 * into a permanent loop around this -EEXIST case
369 * if !CONFIG_PREEMPT and the I/O completion happens
370 * to be waiting on the CPU waitqueue where we are now
371 * busy looping, we just conditionally invoke the
372 * scheduler here, if there are some more important
373 * tasks to run.
374 */ 364 */
375 cond_resched(); 365 cond_resched();
376 continue; 366 continue;
@@ -533,7 +523,7 @@ int init_swap_address_space(unsigned int type, unsigned long nr_pages)
533 unsigned int i, nr; 523 unsigned int i, nr;
534 524
535 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); 525 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
536 spaces = vzalloc(sizeof(struct address_space) * nr); 526 spaces = kvzalloc(sizeof(struct address_space) * nr, GFP_KERNEL);
537 if (!spaces) 527 if (!spaces)
538 return -ENOMEM; 528 return -ENOMEM;
539 for (i = 0; i < nr; i++) { 529 for (i = 0; i < nr; i++) {
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 178130880b90..4f6cba1b6632 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -335,7 +335,7 @@ static void cluster_list_add_tail(struct swap_cluster_list *list,
335 ci_tail = ci + tail; 335 ci_tail = ci + tail;
336 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); 336 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
337 cluster_set_next(ci_tail, idx); 337 cluster_set_next(ci_tail, idx);
338 unlock_cluster(ci_tail); 338 spin_unlock(&ci_tail->lock);
339 cluster_set_next_flag(&list->tail, idx, 0); 339 cluster_set_next_flag(&list->tail, idx, 0);
340 } 340 }
341} 341}
@@ -672,6 +672,9 @@ checks:
672 else 672 else
673 goto done; 673 goto done;
674 } 674 }
675 si->swap_map[offset] = usage;
676 inc_cluster_info_page(si, si->cluster_info, offset);
677 unlock_cluster(ci);
675 678
676 if (offset == si->lowest_bit) 679 if (offset == si->lowest_bit)
677 si->lowest_bit++; 680 si->lowest_bit++;
@@ -685,9 +688,6 @@ checks:
685 plist_del(&si->avail_list, &swap_avail_head); 688 plist_del(&si->avail_list, &swap_avail_head);
686 spin_unlock(&swap_avail_lock); 689 spin_unlock(&swap_avail_lock);
687 } 690 }
688 si->swap_map[offset] = usage;
689 inc_cluster_info_page(si, si->cluster_info, offset);
690 unlock_cluster(ci);
691 si->cluster_next = offset + 1; 691 si->cluster_next = offset + 1;
692 slots[n_ret++] = swp_entry(si->type, offset); 692 slots[n_ret++] = swp_entry(si->type, offset);
693 693
@@ -1079,8 +1079,6 @@ void swapcache_free_entries(swp_entry_t *entries, int n)
1079 p = swap_info_get_cont(entries[i], prev); 1079 p = swap_info_get_cont(entries[i], prev);
1080 if (p) 1080 if (p)
1081 swap_entry_free(p, entries[i]); 1081 swap_entry_free(p, entries[i]);
1082 else
1083 break;
1084 prev = p; 1082 prev = p;
1085 } 1083 }
1086 if (p) 1084 if (p)
@@ -1111,6 +1109,18 @@ int page_swapcount(struct page *page)
1111 return count; 1109 return count;
1112} 1110}
1113 1111
1112static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1113{
1114 int count = 0;
1115 pgoff_t offset = swp_offset(entry);
1116 struct swap_cluster_info *ci;
1117
1118 ci = lock_cluster_or_swap_info(si, offset);
1119 count = swap_count(si->swap_map[offset]);
1120 unlock_cluster_or_swap_info(si, ci);
1121 return count;
1122}
1123
1114/* 1124/*
1115 * How many references to @entry are currently swapped out? 1125 * How many references to @entry are currently swapped out?
1116 * This does not give an exact answer when swap count is continued, 1126 * This does not give an exact answer when swap count is continued,
@@ -1119,17 +1129,11 @@ int page_swapcount(struct page *page)
1119int __swp_swapcount(swp_entry_t entry) 1129int __swp_swapcount(swp_entry_t entry)
1120{ 1130{
1121 int count = 0; 1131 int count = 0;
1122 pgoff_t offset;
1123 struct swap_info_struct *si; 1132 struct swap_info_struct *si;
1124 struct swap_cluster_info *ci;
1125 1133
1126 si = __swap_info_get(entry); 1134 si = __swap_info_get(entry);
1127 if (si) { 1135 if (si)
1128 offset = swp_offset(entry); 1136 count = swap_swapcount(si, entry);
1129 ci = lock_cluster_or_swap_info(si, offset);
1130 count = swap_count(si->swap_map[offset]);
1131 unlock_cluster_or_swap_info(si, ci);
1132 }
1133 return count; 1137 return count;
1134} 1138}
1135 1139
@@ -1291,7 +1295,8 @@ int free_swap_and_cache(swp_entry_t entry)
1291 * Also recheck PageSwapCache now page is locked (above). 1295 * Also recheck PageSwapCache now page is locked (above).
1292 */ 1296 */
1293 if (PageSwapCache(page) && !PageWriteback(page) && 1297 if (PageSwapCache(page) && !PageWriteback(page) &&
1294 (!page_mapped(page) || mem_cgroup_swap_full(page))) { 1298 (!page_mapped(page) || mem_cgroup_swap_full(page)) &&
1299 !swap_swapcount(p, entry)) {
1295 delete_from_swap_cache(page); 1300 delete_from_swap_cache(page);
1296 SetPageDirty(page); 1301 SetPageDirty(page);
1297 } 1302 }
@@ -2265,8 +2270,8 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2265 free_percpu(p->percpu_cluster); 2270 free_percpu(p->percpu_cluster);
2266 p->percpu_cluster = NULL; 2271 p->percpu_cluster = NULL;
2267 vfree(swap_map); 2272 vfree(swap_map);
2268 vfree(cluster_info); 2273 kvfree(cluster_info);
2269 vfree(frontswap_map); 2274 kvfree(frontswap_map);
2270 /* Destroy swap account information */ 2275 /* Destroy swap account information */
2271 swap_cgroup_swapoff(p->type); 2276 swap_cgroup_swapoff(p->type);
2272 exit_swap_address_space(p->type); 2277 exit_swap_address_space(p->type);
@@ -2789,7 +2794,8 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2789 p->cluster_next = 1 + (prandom_u32() % p->highest_bit); 2794 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2790 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); 2795 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2791 2796
2792 cluster_info = vzalloc(nr_cluster * sizeof(*cluster_info)); 2797 cluster_info = kvzalloc(nr_cluster * sizeof(*cluster_info),
2798 GFP_KERNEL);
2793 if (!cluster_info) { 2799 if (!cluster_info) {
2794 error = -ENOMEM; 2800 error = -ENOMEM;
2795 goto bad_swap; 2801 goto bad_swap;
@@ -2822,7 +2828,8 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2822 } 2828 }
2823 /* frontswap enabled? set up bit-per-page map for frontswap */ 2829 /* frontswap enabled? set up bit-per-page map for frontswap */
2824 if (IS_ENABLED(CONFIG_FRONTSWAP)) 2830 if (IS_ENABLED(CONFIG_FRONTSWAP))
2825 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long)); 2831 frontswap_map = kvzalloc(BITS_TO_LONGS(maxpages) * sizeof(long),
2832 GFP_KERNEL);
2826 2833
2827 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { 2834 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2828 /* 2835 /*
diff --git a/mm/truncate.c b/mm/truncate.c
index 6263affdef88..83a059e8cd1d 100644
--- a/mm/truncate.c
+++ b/mm/truncate.c
@@ -266,9 +266,8 @@ void truncate_inode_pages_range(struct address_space *mapping,
266 pgoff_t index; 266 pgoff_t index;
267 int i; 267 int i;
268 268
269 cleancache_invalidate_inode(mapping);
270 if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 269 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
271 return; 270 goto out;
272 271
273 /* Offsets within partial pages */ 272 /* Offsets within partial pages */
274 partial_start = lstart & (PAGE_SIZE - 1); 273 partial_start = lstart & (PAGE_SIZE - 1);
@@ -363,7 +362,7 @@ void truncate_inode_pages_range(struct address_space *mapping,
363 * will be released, just zeroed, so we can bail out now. 362 * will be released, just zeroed, so we can bail out now.
364 */ 363 */
365 if (start >= end) 364 if (start >= end)
366 return; 365 goto out;
367 366
368 index = start; 367 index = start;
369 for ( ; ; ) { 368 for ( ; ; ) {
@@ -410,6 +409,8 @@ void truncate_inode_pages_range(struct address_space *mapping,
410 pagevec_release(&pvec); 409 pagevec_release(&pvec);
411 index++; 410 index++;
412 } 411 }
412
413out:
413 cleancache_invalidate_inode(mapping); 414 cleancache_invalidate_inode(mapping);
414} 415}
415EXPORT_SYMBOL(truncate_inode_pages_range); 416EXPORT_SYMBOL(truncate_inode_pages_range);
@@ -623,7 +624,9 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
623 int ret2 = 0; 624 int ret2 = 0;
624 int did_range_unmap = 0; 625 int did_range_unmap = 0;
625 626
626 cleancache_invalidate_inode(mapping); 627 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
628 goto out;
629
627 pagevec_init(&pvec, 0); 630 pagevec_init(&pvec, 0);
628 index = start; 631 index = start;
629 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 632 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
@@ -686,6 +689,8 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
686 cond_resched(); 689 cond_resched();
687 index++; 690 index++;
688 } 691 }
692
693out:
689 cleancache_invalidate_inode(mapping); 694 cleancache_invalidate_inode(mapping);
690 return ret; 695 return ret;
691} 696}
diff --git a/mm/usercopy.c b/mm/usercopy.c
index d155e12563b1..a9852b24715d 100644
--- a/mm/usercopy.c
+++ b/mm/usercopy.c
@@ -19,15 +19,9 @@
19#include <linux/sched.h> 19#include <linux/sched.h>
20#include <linux/sched/task.h> 20#include <linux/sched/task.h>
21#include <linux/sched/task_stack.h> 21#include <linux/sched/task_stack.h>
22#include <linux/thread_info.h>
22#include <asm/sections.h> 23#include <asm/sections.h>
23 24
24enum {
25 BAD_STACK = -1,
26 NOT_STACK = 0,
27 GOOD_FRAME,
28 GOOD_STACK,
29};
30
31/* 25/*
32 * Checks if a given pointer and length is contained by the current 26 * Checks if a given pointer and length is contained by the current
33 * stack frame (if possible). 27 * stack frame (if possible).
@@ -206,17 +200,6 @@ static inline const char *check_heap_object(const void *ptr, unsigned long n,
206{ 200{
207 struct page *page; 201 struct page *page;
208 202
209 /*
210 * Some architectures (arm64) return true for virt_addr_valid() on
211 * vmalloced addresses. Work around this by checking for vmalloc
212 * first.
213 *
214 * We also need to check for module addresses explicitly since we
215 * may copy static data from modules to userspace
216 */
217 if (is_vmalloc_or_module_addr(ptr))
218 return NULL;
219
220 if (!virt_addr_valid(ptr)) 203 if (!virt_addr_valid(ptr))
221 return NULL; 204 return NULL;
222 205
diff --git a/mm/util.c b/mm/util.c
index 656dc5e37a87..718154debc87 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -329,6 +329,63 @@ unsigned long vm_mmap(struct file *file, unsigned long addr,
329} 329}
330EXPORT_SYMBOL(vm_mmap); 330EXPORT_SYMBOL(vm_mmap);
331 331
332/**
333 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
334 * failure, fall back to non-contiguous (vmalloc) allocation.
335 * @size: size of the request.
336 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
337 * @node: numa node to allocate from
338 *
339 * Uses kmalloc to get the memory but if the allocation fails then falls back
340 * to the vmalloc allocator. Use kvfree for freeing the memory.
341 *
342 * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported. __GFP_REPEAT
343 * is supported only for large (>32kB) allocations, and it should be used only if
344 * kmalloc is preferable to the vmalloc fallback, due to visible performance drawbacks.
345 *
346 * Any use of gfp flags outside of GFP_KERNEL should be consulted with mm people.
347 */
348void *kvmalloc_node(size_t size, gfp_t flags, int node)
349{
350 gfp_t kmalloc_flags = flags;
351 void *ret;
352
353 /*
354 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
355 * so the given set of flags has to be compatible.
356 */
357 WARN_ON_ONCE((flags & GFP_KERNEL) != GFP_KERNEL);
358
359 /*
360 * Make sure that larger requests are not too disruptive - no OOM
361 * killer and no allocation failure warnings as we have a fallback
362 */
363 if (size > PAGE_SIZE) {
364 kmalloc_flags |= __GFP_NOWARN;
365
366 /*
367 * We have to override __GFP_REPEAT by __GFP_NORETRY for !costly
368 * requests because there is no other way to tell the allocator
369 * that we want to fail rather than retry endlessly.
370 */
371 if (!(kmalloc_flags & __GFP_REPEAT) ||
372 (size <= PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
373 kmalloc_flags |= __GFP_NORETRY;
374 }
375
376 ret = kmalloc_node(size, kmalloc_flags, node);
377
378 /*
379 * It doesn't really make sense to fallback to vmalloc for sub page
380 * requests
381 */
382 if (ret || size <= PAGE_SIZE)
383 return ret;
384
385 return __vmalloc_node_flags(size, node, flags);
386}
387EXPORT_SYMBOL(kvmalloc_node);
388
332void kvfree(const void *addr) 389void kvfree(const void *addr)
333{ 390{
334 if (is_vmalloc_addr(addr)) 391 if (is_vmalloc_addr(addr))
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 0b057628a7ba..1dda6d8a200a 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -1579,7 +1579,7 @@ void vfree_atomic(const void *addr)
1579 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling 1579 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1580 * conventions for vfree() arch-depenedent would be a really bad idea) 1580 * conventions for vfree() arch-depenedent would be a really bad idea)
1581 * 1581 *
1582 * NOTE: assumes that the object at *addr has a size >= sizeof(llist_node) 1582 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1583 */ 1583 */
1584void vfree(const void *addr) 1584void vfree(const void *addr)
1585{ 1585{
@@ -1649,16 +1649,13 @@ void *vmap(struct page **pages, unsigned int count,
1649} 1649}
1650EXPORT_SYMBOL(vmap); 1650EXPORT_SYMBOL(vmap);
1651 1651
1652static void *__vmalloc_node(unsigned long size, unsigned long align,
1653 gfp_t gfp_mask, pgprot_t prot,
1654 int node, const void *caller);
1655static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask, 1652static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1656 pgprot_t prot, int node) 1653 pgprot_t prot, int node)
1657{ 1654{
1658 struct page **pages; 1655 struct page **pages;
1659 unsigned int nr_pages, array_size, i; 1656 unsigned int nr_pages, array_size, i;
1660 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO; 1657 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1661 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN; 1658 const gfp_t alloc_mask = gfp_mask | __GFP_HIGHMEM | __GFP_NOWARN;
1662 1659
1663 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT; 1660 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1664 array_size = (nr_pages * sizeof(struct page *)); 1661 array_size = (nr_pages * sizeof(struct page *));
@@ -1786,8 +1783,15 @@ fail:
1786 * Allocate enough pages to cover @size from the page level 1783 * Allocate enough pages to cover @size from the page level
1787 * allocator with @gfp_mask flags. Map them into contiguous 1784 * allocator with @gfp_mask flags. Map them into contiguous
1788 * kernel virtual space, using a pagetable protection of @prot. 1785 * kernel virtual space, using a pagetable protection of @prot.
1786 *
1787 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_REPEAT
1788 * and __GFP_NOFAIL are not supported
1789 *
1790 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1791 * with mm people.
1792 *
1789 */ 1793 */
1790static void *__vmalloc_node(unsigned long size, unsigned long align, 1794void *__vmalloc_node(unsigned long size, unsigned long align,
1791 gfp_t gfp_mask, pgprot_t prot, 1795 gfp_t gfp_mask, pgprot_t prot,
1792 int node, const void *caller) 1796 int node, const void *caller)
1793{ 1797{
@@ -1802,13 +1806,6 @@ void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1802} 1806}
1803EXPORT_SYMBOL(__vmalloc); 1807EXPORT_SYMBOL(__vmalloc);
1804 1808
1805static inline void *__vmalloc_node_flags(unsigned long size,
1806 int node, gfp_t flags)
1807{
1808 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1809 node, __builtin_return_address(0));
1810}
1811
1812/** 1809/**
1813 * vmalloc - allocate virtually contiguous memory 1810 * vmalloc - allocate virtually contiguous memory
1814 * @size: allocation size 1811 * @size: allocation size
@@ -1821,7 +1818,7 @@ static inline void *__vmalloc_node_flags(unsigned long size,
1821void *vmalloc(unsigned long size) 1818void *vmalloc(unsigned long size)
1822{ 1819{
1823 return __vmalloc_node_flags(size, NUMA_NO_NODE, 1820 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1824 GFP_KERNEL | __GFP_HIGHMEM); 1821 GFP_KERNEL);
1825} 1822}
1826EXPORT_SYMBOL(vmalloc); 1823EXPORT_SYMBOL(vmalloc);
1827 1824
@@ -1838,7 +1835,7 @@ EXPORT_SYMBOL(vmalloc);
1838void *vzalloc(unsigned long size) 1835void *vzalloc(unsigned long size)
1839{ 1836{
1840 return __vmalloc_node_flags(size, NUMA_NO_NODE, 1837 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1841 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); 1838 GFP_KERNEL | __GFP_ZERO);
1842} 1839}
1843EXPORT_SYMBOL(vzalloc); 1840EXPORT_SYMBOL(vzalloc);
1844 1841
@@ -1855,7 +1852,7 @@ void *vmalloc_user(unsigned long size)
1855 void *ret; 1852 void *ret;
1856 1853
1857 ret = __vmalloc_node(size, SHMLBA, 1854 ret = __vmalloc_node(size, SHMLBA,
1858 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 1855 GFP_KERNEL | __GFP_ZERO,
1859 PAGE_KERNEL, NUMA_NO_NODE, 1856 PAGE_KERNEL, NUMA_NO_NODE,
1860 __builtin_return_address(0)); 1857 __builtin_return_address(0));
1861 if (ret) { 1858 if (ret) {
@@ -1879,7 +1876,7 @@ EXPORT_SYMBOL(vmalloc_user);
1879 */ 1876 */
1880void *vmalloc_node(unsigned long size, int node) 1877void *vmalloc_node(unsigned long size, int node)
1881{ 1878{
1882 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, 1879 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
1883 node, __builtin_return_address(0)); 1880 node, __builtin_return_address(0));
1884} 1881}
1885EXPORT_SYMBOL(vmalloc_node); 1882EXPORT_SYMBOL(vmalloc_node);
@@ -1899,7 +1896,7 @@ EXPORT_SYMBOL(vmalloc_node);
1899void *vzalloc_node(unsigned long size, int node) 1896void *vzalloc_node(unsigned long size, int node)
1900{ 1897{
1901 return __vmalloc_node_flags(size, node, 1898 return __vmalloc_node_flags(size, node,
1902 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO); 1899 GFP_KERNEL | __GFP_ZERO);
1903} 1900}
1904EXPORT_SYMBOL(vzalloc_node); 1901EXPORT_SYMBOL(vzalloc_node);
1905 1902
@@ -1921,7 +1918,7 @@ EXPORT_SYMBOL(vzalloc_node);
1921 1918
1922void *vmalloc_exec(unsigned long size) 1919void *vmalloc_exec(unsigned long size)
1923{ 1920{
1924 return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC, 1921 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1925 NUMA_NO_NODE, __builtin_return_address(0)); 1922 NUMA_NO_NODE, __builtin_return_address(0));
1926} 1923}
1927 1924
diff --git a/mm/vmscan.c b/mm/vmscan.c
index bc8031ef994d..2f45c0520f43 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -97,8 +97,13 @@ struct scan_control {
97 /* Can pages be swapped as part of reclaim? */ 97 /* Can pages be swapped as part of reclaim? */
98 unsigned int may_swap:1; 98 unsigned int may_swap:1;
99 99
100 /* Can cgroups be reclaimed below their normal consumption range? */ 100 /*
101 unsigned int may_thrash:1; 101 * Cgroups are not reclaimed below their configured memory.low,
102 * unless we threaten to OOM. If any cgroups are skipped due to
103 * memory.low and nothing was reclaimed, go back for memory.low.
104 */
105 unsigned int memcg_low_reclaim:1;
106 unsigned int memcg_low_skipped:1;
102 107
103 unsigned int hibernation_mode:1; 108 unsigned int hibernation_mode:1;
104 109
@@ -230,12 +235,6 @@ unsigned long pgdat_reclaimable_pages(struct pglist_data *pgdat)
230 return nr; 235 return nr;
231} 236}
232 237
233bool pgdat_reclaimable(struct pglist_data *pgdat)
234{
235 return node_page_state_snapshot(pgdat, NR_PAGES_SCANNED) <
236 pgdat_reclaimable_pages(pgdat) * 6;
237}
238
239/** 238/**
240 * lruvec_lru_size - Returns the number of pages on the given LRU list. 239 * lruvec_lru_size - Returns the number of pages on the given LRU list.
241 * @lruvec: lru vector 240 * @lruvec: lru vector
@@ -912,7 +911,8 @@ static void page_check_dirty_writeback(struct page *page,
912 * Anonymous pages are not handled by flushers and must be written 911 * Anonymous pages are not handled by flushers and must be written
913 * from reclaim context. Do not stall reclaim based on them 912 * from reclaim context. Do not stall reclaim based on them
914 */ 913 */
915 if (!page_is_file_cache(page)) { 914 if (!page_is_file_cache(page) ||
915 (PageAnon(page) && !PageSwapBacked(page))) {
916 *dirty = false; 916 *dirty = false;
917 *writeback = false; 917 *writeback = false;
918 return; 918 return;
@@ -972,8 +972,6 @@ static unsigned long shrink_page_list(struct list_head *page_list,
972 int may_enter_fs; 972 int may_enter_fs;
973 enum page_references references = PAGEREF_RECLAIM_CLEAN; 973 enum page_references references = PAGEREF_RECLAIM_CLEAN;
974 bool dirty, writeback; 974 bool dirty, writeback;
975 bool lazyfree = false;
976 int ret = SWAP_SUCCESS;
977 975
978 cond_resched(); 976 cond_resched();
979 977
@@ -988,13 +986,14 @@ static unsigned long shrink_page_list(struct list_head *page_list,
988 sc->nr_scanned++; 986 sc->nr_scanned++;
989 987
990 if (unlikely(!page_evictable(page))) 988 if (unlikely(!page_evictable(page)))
991 goto cull_mlocked; 989 goto activate_locked;
992 990
993 if (!sc->may_unmap && page_mapped(page)) 991 if (!sc->may_unmap && page_mapped(page))
994 goto keep_locked; 992 goto keep_locked;
995 993
996 /* Double the slab pressure for mapped and swapcache pages */ 994 /* Double the slab pressure for mapped and swapcache pages */
997 if (page_mapped(page) || PageSwapCache(page)) 995 if ((page_mapped(page) || PageSwapCache(page)) &&
996 !(PageAnon(page) && !PageSwapBacked(page)))
998 sc->nr_scanned++; 997 sc->nr_scanned++;
999 998
1000 may_enter_fs = (sc->gfp_mask & __GFP_FS) || 999 may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
@@ -1120,13 +1119,14 @@ static unsigned long shrink_page_list(struct list_head *page_list,
1120 /* 1119 /*
1121 * Anonymous process memory has backing store? 1120 * Anonymous process memory has backing store?
1122 * Try to allocate it some swap space here. 1121 * Try to allocate it some swap space here.
1122 * Lazyfree page could be freed directly
1123 */ 1123 */
1124 if (PageAnon(page) && !PageSwapCache(page)) { 1124 if (PageAnon(page) && PageSwapBacked(page) &&
1125 !PageSwapCache(page)) {
1125 if (!(sc->gfp_mask & __GFP_IO)) 1126 if (!(sc->gfp_mask & __GFP_IO))
1126 goto keep_locked; 1127 goto keep_locked;
1127 if (!add_to_swap(page, page_list)) 1128 if (!add_to_swap(page, page_list))
1128 goto activate_locked; 1129 goto activate_locked;
1129 lazyfree = true;
1130 may_enter_fs = 1; 1130 may_enter_fs = 1;
1131 1131
1132 /* Adding to swap updated mapping */ 1132 /* Adding to swap updated mapping */
@@ -1143,21 +1143,10 @@ static unsigned long shrink_page_list(struct list_head *page_list,
1143 * The page is mapped into the page tables of one or more 1143 * The page is mapped into the page tables of one or more
1144 * processes. Try to unmap it here. 1144 * processes. Try to unmap it here.
1145 */ 1145 */
1146 if (page_mapped(page) && mapping) { 1146 if (page_mapped(page)) {
1147 switch (ret = try_to_unmap(page, lazyfree ? 1147 if (!try_to_unmap(page, ttu_flags | TTU_BATCH_FLUSH)) {
1148 (ttu_flags | TTU_BATCH_FLUSH | TTU_LZFREE) :
1149 (ttu_flags | TTU_BATCH_FLUSH))) {
1150 case SWAP_FAIL:
1151 nr_unmap_fail++; 1148 nr_unmap_fail++;
1152 goto activate_locked; 1149 goto activate_locked;
1153 case SWAP_AGAIN:
1154 goto keep_locked;
1155 case SWAP_MLOCK:
1156 goto cull_mlocked;
1157 case SWAP_LZFREE:
1158 goto lazyfree;
1159 case SWAP_SUCCESS:
1160 ; /* try to free the page below */
1161 } 1150 }
1162 } 1151 }
1163 1152
@@ -1267,10 +1256,18 @@ static unsigned long shrink_page_list(struct list_head *page_list,
1267 } 1256 }
1268 } 1257 }
1269 1258
1270lazyfree: 1259 if (PageAnon(page) && !PageSwapBacked(page)) {
1271 if (!mapping || !__remove_mapping(mapping, page, true)) 1260 /* follow __remove_mapping for reference */
1272 goto keep_locked; 1261 if (!page_ref_freeze(page, 1))
1262 goto keep_locked;
1263 if (PageDirty(page)) {
1264 page_ref_unfreeze(page, 1);
1265 goto keep_locked;
1266 }
1273 1267
1268 count_vm_event(PGLAZYFREED);
1269 } else if (!mapping || !__remove_mapping(mapping, page, true))
1270 goto keep_locked;
1274 /* 1271 /*
1275 * At this point, we have no other references and there is 1272 * At this point, we have no other references and there is
1276 * no way to pick any more up (removed from LRU, removed 1273 * no way to pick any more up (removed from LRU, removed
@@ -1280,9 +1277,6 @@ lazyfree:
1280 */ 1277 */
1281 __ClearPageLocked(page); 1278 __ClearPageLocked(page);
1282free_it: 1279free_it:
1283 if (ret == SWAP_LZFREE)
1284 count_vm_event(PGLAZYFREED);
1285
1286 nr_reclaimed++; 1280 nr_reclaimed++;
1287 1281
1288 /* 1282 /*
@@ -1292,20 +1286,16 @@ free_it:
1292 list_add(&page->lru, &free_pages); 1286 list_add(&page->lru, &free_pages);
1293 continue; 1287 continue;
1294 1288
1295cull_mlocked:
1296 if (PageSwapCache(page))
1297 try_to_free_swap(page);
1298 unlock_page(page);
1299 list_add(&page->lru, &ret_pages);
1300 continue;
1301
1302activate_locked: 1289activate_locked:
1303 /* Not a candidate for swapping, so reclaim swap space. */ 1290 /* Not a candidate for swapping, so reclaim swap space. */
1304 if (PageSwapCache(page) && mem_cgroup_swap_full(page)) 1291 if (PageSwapCache(page) && (mem_cgroup_swap_full(page) ||
1292 PageMlocked(page)))
1305 try_to_free_swap(page); 1293 try_to_free_swap(page);
1306 VM_BUG_ON_PAGE(PageActive(page), page); 1294 VM_BUG_ON_PAGE(PageActive(page), page);
1307 SetPageActive(page); 1295 if (!PageMlocked(page)) {
1308 pgactivate++; 1296 SetPageActive(page);
1297 pgactivate++;
1298 }
1309keep_locked: 1299keep_locked:
1310 unlock_page(page); 1300 unlock_page(page);
1311keep: 1301keep:
@@ -1354,7 +1344,7 @@ unsigned long reclaim_clean_pages_from_list(struct zone *zone,
1354 } 1344 }
1355 1345
1356 ret = shrink_page_list(&clean_pages, zone->zone_pgdat, &sc, 1346 ret = shrink_page_list(&clean_pages, zone->zone_pgdat, &sc,
1357 TTU_UNMAP|TTU_IGNORE_ACCESS, NULL, true); 1347 TTU_IGNORE_ACCESS, NULL, true);
1358 list_splice(&clean_pages, page_list); 1348 list_splice(&clean_pages, page_list);
1359 mod_node_page_state(zone->zone_pgdat, NR_ISOLATED_FILE, -ret); 1349 mod_node_page_state(zone->zone_pgdat, NR_ISOLATED_FILE, -ret);
1360 return ret; 1350 return ret;
@@ -1478,12 +1468,12 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1478 unsigned long nr_taken = 0; 1468 unsigned long nr_taken = 0;
1479 unsigned long nr_zone_taken[MAX_NR_ZONES] = { 0 }; 1469 unsigned long nr_zone_taken[MAX_NR_ZONES] = { 0 };
1480 unsigned long nr_skipped[MAX_NR_ZONES] = { 0, }; 1470 unsigned long nr_skipped[MAX_NR_ZONES] = { 0, };
1481 unsigned long skipped = 0, total_skipped = 0; 1471 unsigned long skipped = 0;
1482 unsigned long scan, nr_pages; 1472 unsigned long scan, nr_pages;
1483 LIST_HEAD(pages_skipped); 1473 LIST_HEAD(pages_skipped);
1484 1474
1485 for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan && 1475 for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan &&
1486 !list_empty(src);) { 1476 !list_empty(src); scan++) {
1487 struct page *page; 1477 struct page *page;
1488 1478
1489 page = lru_to_page(src); 1479 page = lru_to_page(src);
@@ -1497,12 +1487,6 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1497 continue; 1487 continue;
1498 } 1488 }
1499 1489
1500 /*
1501 * Account for scanned and skipped separetly to avoid the pgdat
1502 * being prematurely marked unreclaimable by pgdat_reclaimable.
1503 */
1504 scan++;
1505
1506 switch (__isolate_lru_page(page, mode)) { 1490 switch (__isolate_lru_page(page, mode)) {
1507 case 0: 1491 case 0:
1508 nr_pages = hpage_nr_pages(page); 1492 nr_pages = hpage_nr_pages(page);
@@ -1531,6 +1515,7 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1531 if (!list_empty(&pages_skipped)) { 1515 if (!list_empty(&pages_skipped)) {
1532 int zid; 1516 int zid;
1533 1517
1518 list_splice(&pages_skipped, src);
1534 for (zid = 0; zid < MAX_NR_ZONES; zid++) { 1519 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1535 if (!nr_skipped[zid]) 1520 if (!nr_skipped[zid])
1536 continue; 1521 continue;
@@ -1538,17 +1523,8 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
1538 __count_zid_vm_events(PGSCAN_SKIP, zid, nr_skipped[zid]); 1523 __count_zid_vm_events(PGSCAN_SKIP, zid, nr_skipped[zid]);
1539 skipped += nr_skipped[zid]; 1524 skipped += nr_skipped[zid];
1540 } 1525 }
1541
1542 /*
1543 * Account skipped pages as a partial scan as the pgdat may be
1544 * close to unreclaimable. If the LRU list is empty, account
1545 * skipped pages as a full scan.
1546 */
1547 total_skipped = list_empty(src) ? skipped : skipped >> 2;
1548
1549 list_splice(&pages_skipped, src);
1550 } 1526 }
1551 *nr_scanned = scan + total_skipped; 1527 *nr_scanned = scan;
1552 trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, nr_to_scan, 1528 trace_mm_vmscan_lru_isolate(sc->reclaim_idx, sc->order, nr_to_scan,
1553 scan, skipped, nr_taken, mode, lru); 1529 scan, skipped, nr_taken, mode, lru);
1554 update_lru_sizes(lruvec, lru, nr_zone_taken); 1530 update_lru_sizes(lruvec, lru, nr_zone_taken);
@@ -1750,7 +1726,6 @@ shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec,
1750 reclaim_stat->recent_scanned[file] += nr_taken; 1726 reclaim_stat->recent_scanned[file] += nr_taken;
1751 1727
1752 if (global_reclaim(sc)) { 1728 if (global_reclaim(sc)) {
1753 __mod_node_page_state(pgdat, NR_PAGES_SCANNED, nr_scanned);
1754 if (current_is_kswapd()) 1729 if (current_is_kswapd())
1755 __count_vm_events(PGSCAN_KSWAPD, nr_scanned); 1730 __count_vm_events(PGSCAN_KSWAPD, nr_scanned);
1756 else 1731 else
@@ -1761,7 +1736,7 @@ shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec,
1761 if (nr_taken == 0) 1736 if (nr_taken == 0)
1762 return 0; 1737 return 0;
1763 1738
1764 nr_reclaimed = shrink_page_list(&page_list, pgdat, sc, TTU_UNMAP, 1739 nr_reclaimed = shrink_page_list(&page_list, pgdat, sc, 0,
1765 &stat, false); 1740 &stat, false);
1766 1741
1767 spin_lock_irq(&pgdat->lru_lock); 1742 spin_lock_irq(&pgdat->lru_lock);
@@ -1953,8 +1928,6 @@ static void shrink_active_list(unsigned long nr_to_scan,
1953 __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken); 1928 __mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, nr_taken);
1954 reclaim_stat->recent_scanned[file] += nr_taken; 1929 reclaim_stat->recent_scanned[file] += nr_taken;
1955 1930
1956 if (global_reclaim(sc))
1957 __mod_node_page_state(pgdat, NR_PAGES_SCANNED, nr_scanned);
1958 __count_vm_events(PGREFILL, nr_scanned); 1931 __count_vm_events(PGREFILL, nr_scanned);
1959 1932
1960 spin_unlock_irq(&pgdat->lru_lock); 1933 spin_unlock_irq(&pgdat->lru_lock);
@@ -2033,6 +2006,8 @@ static void shrink_active_list(unsigned long nr_to_scan,
2033 * Both inactive lists should also be large enough that each inactive 2006 * Both inactive lists should also be large enough that each inactive
2034 * page has a chance to be referenced again before it is reclaimed. 2007 * page has a chance to be referenced again before it is reclaimed.
2035 * 2008 *
2009 * If that fails and refaulting is observed, the inactive list grows.
2010 *
2036 * The inactive_ratio is the target ratio of ACTIVE to INACTIVE pages 2011 * The inactive_ratio is the target ratio of ACTIVE to INACTIVE pages
2037 * on this LRU, maintained by the pageout code. A zone->inactive_ratio 2012 * on this LRU, maintained by the pageout code. A zone->inactive_ratio
2038 * of 3 means 3:1 or 25% of the pages are kept on the inactive list. 2013 * of 3 means 3:1 or 25% of the pages are kept on the inactive list.
@@ -2049,12 +2024,15 @@ static void shrink_active_list(unsigned long nr_to_scan,
2049 * 10TB 320 32GB 2024 * 10TB 320 32GB
2050 */ 2025 */
2051static bool inactive_list_is_low(struct lruvec *lruvec, bool file, 2026static bool inactive_list_is_low(struct lruvec *lruvec, bool file,
2052 struct scan_control *sc, bool trace) 2027 struct mem_cgroup *memcg,
2028 struct scan_control *sc, bool actual_reclaim)
2053{ 2029{
2054 unsigned long inactive_ratio;
2055 unsigned long inactive, active;
2056 enum lru_list inactive_lru = file * LRU_FILE;
2057 enum lru_list active_lru = file * LRU_FILE + LRU_ACTIVE; 2030 enum lru_list active_lru = file * LRU_FILE + LRU_ACTIVE;
2031 struct pglist_data *pgdat = lruvec_pgdat(lruvec);
2032 enum lru_list inactive_lru = file * LRU_FILE;
2033 unsigned long inactive, active;
2034 unsigned long inactive_ratio;
2035 unsigned long refaults;
2058 unsigned long gb; 2036 unsigned long gb;
2059 2037
2060 /* 2038 /*
@@ -2067,27 +2045,42 @@ static bool inactive_list_is_low(struct lruvec *lruvec, bool file,
2067 inactive = lruvec_lru_size(lruvec, inactive_lru, sc->reclaim_idx); 2045 inactive = lruvec_lru_size(lruvec, inactive_lru, sc->reclaim_idx);
2068 active = lruvec_lru_size(lruvec, active_lru, sc->reclaim_idx); 2046 active = lruvec_lru_size(lruvec, active_lru, sc->reclaim_idx);
2069 2047
2070 gb = (inactive + active) >> (30 - PAGE_SHIFT); 2048 if (memcg)
2071 if (gb) 2049 refaults = memcg_page_state(memcg, WORKINGSET_ACTIVATE);
2072 inactive_ratio = int_sqrt(10 * gb);
2073 else 2050 else
2074 inactive_ratio = 1; 2051 refaults = node_page_state(pgdat, WORKINGSET_ACTIVATE);
2052
2053 /*
2054 * When refaults are being observed, it means a new workingset
2055 * is being established. Disable active list protection to get
2056 * rid of the stale workingset quickly.
2057 */
2058 if (file && actual_reclaim && lruvec->refaults != refaults) {
2059 inactive_ratio = 0;
2060 } else {
2061 gb = (inactive + active) >> (30 - PAGE_SHIFT);
2062 if (gb)
2063 inactive_ratio = int_sqrt(10 * gb);
2064 else
2065 inactive_ratio = 1;
2066 }
2075 2067
2076 if (trace) 2068 if (actual_reclaim)
2077 trace_mm_vmscan_inactive_list_is_low(lruvec_pgdat(lruvec)->node_id, 2069 trace_mm_vmscan_inactive_list_is_low(pgdat->node_id, sc->reclaim_idx,
2078 sc->reclaim_idx, 2070 lruvec_lru_size(lruvec, inactive_lru, MAX_NR_ZONES), inactive,
2079 lruvec_lru_size(lruvec, inactive_lru, MAX_NR_ZONES), inactive, 2071 lruvec_lru_size(lruvec, active_lru, MAX_NR_ZONES), active,
2080 lruvec_lru_size(lruvec, active_lru, MAX_NR_ZONES), active, 2072 inactive_ratio, file);
2081 inactive_ratio, file);
2082 2073
2083 return inactive * inactive_ratio < active; 2074 return inactive * inactive_ratio < active;
2084} 2075}
2085 2076
2086static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, 2077static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
2087 struct lruvec *lruvec, struct scan_control *sc) 2078 struct lruvec *lruvec, struct mem_cgroup *memcg,
2079 struct scan_control *sc)
2088{ 2080{
2089 if (is_active_lru(lru)) { 2081 if (is_active_lru(lru)) {
2090 if (inactive_list_is_low(lruvec, is_file_lru(lru), sc, true)) 2082 if (inactive_list_is_low(lruvec, is_file_lru(lru),
2083 memcg, sc, true))
2091 shrink_active_list(nr_to_scan, lruvec, sc, lru); 2084 shrink_active_list(nr_to_scan, lruvec, sc, lru);
2092 return 0; 2085 return 0;
2093 } 2086 }
@@ -2123,30 +2116,8 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg,
2123 unsigned long anon_prio, file_prio; 2116 unsigned long anon_prio, file_prio;
2124 enum scan_balance scan_balance; 2117 enum scan_balance scan_balance;
2125 unsigned long anon, file; 2118 unsigned long anon, file;
2126 bool force_scan = false;
2127 unsigned long ap, fp; 2119 unsigned long ap, fp;
2128 enum lru_list lru; 2120 enum lru_list lru;
2129 bool some_scanned;
2130 int pass;
2131
2132 /*
2133 * If the zone or memcg is small, nr[l] can be 0. This
2134 * results in no scanning on this priority and a potential
2135 * priority drop. Global direct reclaim can go to the next
2136 * zone and tends to have no problems. Global kswapd is for
2137 * zone balancing and it needs to scan a minimum amount. When
2138 * reclaiming for a memcg, a priority drop can cause high
2139 * latencies, so it's better to scan a minimum amount there as
2140 * well.
2141 */
2142 if (current_is_kswapd()) {
2143 if (!pgdat_reclaimable(pgdat))
2144 force_scan = true;
2145 if (!mem_cgroup_online(memcg))
2146 force_scan = true;
2147 }
2148 if (!global_reclaim(sc))
2149 force_scan = true;
2150 2121
2151 /* If we have no swap space, do not bother scanning anon pages. */ 2122 /* If we have no swap space, do not bother scanning anon pages. */
2152 if (!sc->may_swap || mem_cgroup_get_nr_swap_pages(memcg) <= 0) { 2123 if (!sc->may_swap || mem_cgroup_get_nr_swap_pages(memcg) <= 0) {
@@ -2218,7 +2189,7 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg,
2218 * lruvec even if it has plenty of old anonymous pages unless the 2189 * lruvec even if it has plenty of old anonymous pages unless the
2219 * system is under heavy pressure. 2190 * system is under heavy pressure.
2220 */ 2191 */
2221 if (!inactive_list_is_low(lruvec, true, sc, false) && 2192 if (!inactive_list_is_low(lruvec, true, memcg, sc, false) &&
2222 lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, sc->reclaim_idx) >> sc->priority) { 2193 lruvec_lru_size(lruvec, LRU_INACTIVE_FILE, sc->reclaim_idx) >> sc->priority) {
2223 scan_balance = SCAN_FILE; 2194 scan_balance = SCAN_FILE;
2224 goto out; 2195 goto out;
@@ -2277,55 +2248,48 @@ static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg,
2277 fraction[1] = fp; 2248 fraction[1] = fp;
2278 denominator = ap + fp + 1; 2249 denominator = ap + fp + 1;
2279out: 2250out:
2280 some_scanned = false; 2251 *lru_pages = 0;
2281 /* Only use force_scan on second pass. */ 2252 for_each_evictable_lru(lru) {
2282 for (pass = 0; !some_scanned && pass < 2; pass++) { 2253 int file = is_file_lru(lru);
2283 *lru_pages = 0; 2254 unsigned long size;
2284 for_each_evictable_lru(lru) { 2255 unsigned long scan;
2285 int file = is_file_lru(lru);
2286 unsigned long size;
2287 unsigned long scan;
2288
2289 size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx);
2290 scan = size >> sc->priority;
2291
2292 if (!scan && pass && force_scan)
2293 scan = min(size, SWAP_CLUSTER_MAX);
2294
2295 switch (scan_balance) {
2296 case SCAN_EQUAL:
2297 /* Scan lists relative to size */
2298 break;
2299 case SCAN_FRACT:
2300 /*
2301 * Scan types proportional to swappiness and
2302 * their relative recent reclaim efficiency.
2303 */
2304 scan = div64_u64(scan * fraction[file],
2305 denominator);
2306 break;
2307 case SCAN_FILE:
2308 case SCAN_ANON:
2309 /* Scan one type exclusively */
2310 if ((scan_balance == SCAN_FILE) != file) {
2311 size = 0;
2312 scan = 0;
2313 }
2314 break;
2315 default:
2316 /* Look ma, no brain */
2317 BUG();
2318 }
2319 2256
2320 *lru_pages += size; 2257 size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx);
2321 nr[lru] = scan; 2258 scan = size >> sc->priority;
2259 /*
2260 * If the cgroup's already been deleted, make sure to
2261 * scrape out the remaining cache.
2262 */
2263 if (!scan && !mem_cgroup_online(memcg))
2264 scan = min(size, SWAP_CLUSTER_MAX);
2322 2265
2266 switch (scan_balance) {
2267 case SCAN_EQUAL:
2268 /* Scan lists relative to size */
2269 break;
2270 case SCAN_FRACT:
2323 /* 2271 /*
2324 * Skip the second pass and don't force_scan, 2272 * Scan types proportional to swappiness and
2325 * if we found something to scan. 2273 * their relative recent reclaim efficiency.
2326 */ 2274 */
2327 some_scanned |= !!scan; 2275 scan = div64_u64(scan * fraction[file],
2276 denominator);
2277 break;
2278 case SCAN_FILE:
2279 case SCAN_ANON:
2280 /* Scan one type exclusively */
2281 if ((scan_balance == SCAN_FILE) != file) {
2282 size = 0;
2283 scan = 0;
2284 }
2285 break;
2286 default:
2287 /* Look ma, no brain */
2288 BUG();
2328 } 2289 }
2290
2291 *lru_pages += size;
2292 nr[lru] = scan;
2329 } 2293 }
2330} 2294}
2331 2295
@@ -2376,7 +2340,7 @@ static void shrink_node_memcg(struct pglist_data *pgdat, struct mem_cgroup *memc
2376 nr[lru] -= nr_to_scan; 2340 nr[lru] -= nr_to_scan;
2377 2341
2378 nr_reclaimed += shrink_list(lru, nr_to_scan, 2342 nr_reclaimed += shrink_list(lru, nr_to_scan,
2379 lruvec, sc); 2343 lruvec, memcg, sc);
2380 } 2344 }
2381 } 2345 }
2382 2346
@@ -2443,7 +2407,7 @@ static void shrink_node_memcg(struct pglist_data *pgdat, struct mem_cgroup *memc
2443 * Even if we did not try to evict anon pages at all, we want to 2407 * Even if we did not try to evict anon pages at all, we want to
2444 * rebalance the anon lru active/inactive ratio. 2408 * rebalance the anon lru active/inactive ratio.
2445 */ 2409 */
2446 if (inactive_list_is_low(lruvec, false, sc, true)) 2410 if (inactive_list_is_low(lruvec, false, memcg, sc, true))
2447 shrink_active_list(SWAP_CLUSTER_MAX, lruvec, 2411 shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
2448 sc, LRU_ACTIVE_ANON); 2412 sc, LRU_ACTIVE_ANON);
2449} 2413}
@@ -2557,9 +2521,11 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
2557 unsigned long scanned; 2521 unsigned long scanned;
2558 2522
2559 if (mem_cgroup_low(root, memcg)) { 2523 if (mem_cgroup_low(root, memcg)) {
2560 if (!sc->may_thrash) 2524 if (!sc->memcg_low_reclaim) {
2525 sc->memcg_low_skipped = 1;
2561 continue; 2526 continue;
2562 mem_cgroup_events(memcg, MEMCG_LOW, 1); 2527 }
2528 mem_cgroup_event(memcg, MEMCG_LOW);
2563 } 2529 }
2564 2530
2565 reclaimed = sc->nr_reclaimed; 2531 reclaimed = sc->nr_reclaimed;
@@ -2620,6 +2586,15 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc)
2620 } while (should_continue_reclaim(pgdat, sc->nr_reclaimed - nr_reclaimed, 2586 } while (should_continue_reclaim(pgdat, sc->nr_reclaimed - nr_reclaimed,
2621 sc->nr_scanned - nr_scanned, sc)); 2587 sc->nr_scanned - nr_scanned, sc));
2622 2588
2589 /*
2590 * Kswapd gives up on balancing particular nodes after too
2591 * many failures to reclaim anything from them and goes to
2592 * sleep. On reclaim progress, reset the failure counter. A
2593 * successful direct reclaim run will revive a dormant kswapd.
2594 */
2595 if (reclaimable)
2596 pgdat->kswapd_failures = 0;
2597
2623 return reclaimable; 2598 return reclaimable;
2624} 2599}
2625 2600
@@ -2694,10 +2669,6 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
2694 GFP_KERNEL | __GFP_HARDWALL)) 2669 GFP_KERNEL | __GFP_HARDWALL))
2695 continue; 2670 continue;
2696 2671
2697 if (sc->priority != DEF_PRIORITY &&
2698 !pgdat_reclaimable(zone->zone_pgdat))
2699 continue; /* Let kswapd poll it */
2700
2701 /* 2672 /*
2702 * If we already have plenty of memory free for 2673 * If we already have plenty of memory free for
2703 * compaction in this zone, don't free any more. 2674 * compaction in this zone, don't free any more.
@@ -2752,6 +2723,25 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
2752 sc->gfp_mask = orig_mask; 2723 sc->gfp_mask = orig_mask;
2753} 2724}
2754 2725
2726static void snapshot_refaults(struct mem_cgroup *root_memcg, pg_data_t *pgdat)
2727{
2728 struct mem_cgroup *memcg;
2729
2730 memcg = mem_cgroup_iter(root_memcg, NULL, NULL);
2731 do {
2732 unsigned long refaults;
2733 struct lruvec *lruvec;
2734
2735 if (memcg)
2736 refaults = memcg_page_state(memcg, WORKINGSET_ACTIVATE);
2737 else
2738 refaults = node_page_state(pgdat, WORKINGSET_ACTIVATE);
2739
2740 lruvec = mem_cgroup_lruvec(pgdat, memcg);
2741 lruvec->refaults = refaults;
2742 } while ((memcg = mem_cgroup_iter(root_memcg, memcg, NULL)));
2743}
2744
2755/* 2745/*
2756 * This is the main entry point to direct page reclaim. 2746 * This is the main entry point to direct page reclaim.
2757 * 2747 *
@@ -2772,6 +2762,9 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
2772 struct scan_control *sc) 2762 struct scan_control *sc)
2773{ 2763{
2774 int initial_priority = sc->priority; 2764 int initial_priority = sc->priority;
2765 pg_data_t *last_pgdat;
2766 struct zoneref *z;
2767 struct zone *zone;
2775retry: 2768retry:
2776 delayacct_freepages_start(); 2769 delayacct_freepages_start();
2777 2770
@@ -2798,6 +2791,15 @@ retry:
2798 sc->may_writepage = 1; 2791 sc->may_writepage = 1;
2799 } while (--sc->priority >= 0); 2792 } while (--sc->priority >= 0);
2800 2793
2794 last_pgdat = NULL;
2795 for_each_zone_zonelist_nodemask(zone, z, zonelist, sc->reclaim_idx,
2796 sc->nodemask) {
2797 if (zone->zone_pgdat == last_pgdat)
2798 continue;
2799 last_pgdat = zone->zone_pgdat;
2800 snapshot_refaults(sc->target_mem_cgroup, zone->zone_pgdat);
2801 }
2802
2801 delayacct_freepages_end(); 2803 delayacct_freepages_end();
2802 2804
2803 if (sc->nr_reclaimed) 2805 if (sc->nr_reclaimed)
@@ -2808,16 +2810,17 @@ retry:
2808 return 1; 2810 return 1;
2809 2811
2810 /* Untapped cgroup reserves? Don't OOM, retry. */ 2812 /* Untapped cgroup reserves? Don't OOM, retry. */
2811 if (!sc->may_thrash) { 2813 if (sc->memcg_low_skipped) {
2812 sc->priority = initial_priority; 2814 sc->priority = initial_priority;
2813 sc->may_thrash = 1; 2815 sc->memcg_low_reclaim = 1;
2816 sc->memcg_low_skipped = 0;
2814 goto retry; 2817 goto retry;
2815 } 2818 }
2816 2819
2817 return 0; 2820 return 0;
2818} 2821}
2819 2822
2820static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) 2823static bool allow_direct_reclaim(pg_data_t *pgdat)
2821{ 2824{
2822 struct zone *zone; 2825 struct zone *zone;
2823 unsigned long pfmemalloc_reserve = 0; 2826 unsigned long pfmemalloc_reserve = 0;
@@ -2825,10 +2828,15 @@ static bool pfmemalloc_watermark_ok(pg_data_t *pgdat)
2825 int i; 2828 int i;
2826 bool wmark_ok; 2829 bool wmark_ok;
2827 2830
2831 if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES)
2832 return true;
2833
2828 for (i = 0; i <= ZONE_NORMAL; i++) { 2834 for (i = 0; i <= ZONE_NORMAL; i++) {
2829 zone = &pgdat->node_zones[i]; 2835 zone = &pgdat->node_zones[i];
2830 if (!managed_zone(zone) || 2836 if (!managed_zone(zone))
2831 pgdat_reclaimable_pages(pgdat) == 0) 2837 continue;
2838
2839 if (!zone_reclaimable_pages(zone))
2832 continue; 2840 continue;
2833 2841
2834 pfmemalloc_reserve += min_wmark_pages(zone); 2842 pfmemalloc_reserve += min_wmark_pages(zone);
@@ -2905,7 +2913,7 @@ static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
2905 2913
2906 /* Throttle based on the first usable node */ 2914 /* Throttle based on the first usable node */
2907 pgdat = zone->zone_pgdat; 2915 pgdat = zone->zone_pgdat;
2908 if (pfmemalloc_watermark_ok(pgdat)) 2916 if (allow_direct_reclaim(pgdat))
2909 goto out; 2917 goto out;
2910 break; 2918 break;
2911 } 2919 }
@@ -2927,14 +2935,14 @@ static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
2927 */ 2935 */
2928 if (!(gfp_mask & __GFP_FS)) { 2936 if (!(gfp_mask & __GFP_FS)) {
2929 wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, 2937 wait_event_interruptible_timeout(pgdat->pfmemalloc_wait,
2930 pfmemalloc_watermark_ok(pgdat), HZ); 2938 allow_direct_reclaim(pgdat), HZ);
2931 2939
2932 goto check_pending; 2940 goto check_pending;
2933 } 2941 }
2934 2942
2935 /* Throttle until kswapd wakes the process */ 2943 /* Throttle until kswapd wakes the process */
2936 wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, 2944 wait_event_killable(zone->zone_pgdat->pfmemalloc_wait,
2937 pfmemalloc_watermark_ok(pgdat)); 2945 allow_direct_reclaim(pgdat));
2938 2946
2939check_pending: 2947check_pending:
2940 if (fatal_signal_pending(current)) 2948 if (fatal_signal_pending(current))
@@ -2950,7 +2958,7 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
2950 unsigned long nr_reclaimed; 2958 unsigned long nr_reclaimed;
2951 struct scan_control sc = { 2959 struct scan_control sc = {
2952 .nr_to_reclaim = SWAP_CLUSTER_MAX, 2960 .nr_to_reclaim = SWAP_CLUSTER_MAX,
2953 .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), 2961 .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)),
2954 .reclaim_idx = gfp_zone(gfp_mask), 2962 .reclaim_idx = gfp_zone(gfp_mask),
2955 .order = order, 2963 .order = order,
2956 .nodemask = nodemask, 2964 .nodemask = nodemask,
@@ -3028,9 +3036,10 @@ unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
3028 struct zonelist *zonelist; 3036 struct zonelist *zonelist;
3029 unsigned long nr_reclaimed; 3037 unsigned long nr_reclaimed;
3030 int nid; 3038 int nid;
3039 unsigned int noreclaim_flag;
3031 struct scan_control sc = { 3040 struct scan_control sc = {
3032 .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), 3041 .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX),
3033 .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | 3042 .gfp_mask = (current_gfp_context(gfp_mask) & GFP_RECLAIM_MASK) |
3034 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), 3043 (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK),
3035 .reclaim_idx = MAX_NR_ZONES - 1, 3044 .reclaim_idx = MAX_NR_ZONES - 1,
3036 .target_mem_cgroup = memcg, 3045 .target_mem_cgroup = memcg,
@@ -3054,9 +3063,9 @@ unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
3054 sc.gfp_mask, 3063 sc.gfp_mask,
3055 sc.reclaim_idx); 3064 sc.reclaim_idx);
3056 3065
3057 current->flags |= PF_MEMALLOC; 3066 noreclaim_flag = memalloc_noreclaim_save();
3058 nr_reclaimed = do_try_to_free_pages(zonelist, &sc); 3067 nr_reclaimed = do_try_to_free_pages(zonelist, &sc);
3059 current->flags &= ~PF_MEMALLOC; 3068 memalloc_noreclaim_restore(noreclaim_flag);
3060 3069
3061 trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); 3070 trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
3062 3071
@@ -3076,7 +3085,7 @@ static void age_active_anon(struct pglist_data *pgdat,
3076 do { 3085 do {
3077 struct lruvec *lruvec = mem_cgroup_lruvec(pgdat, memcg); 3086 struct lruvec *lruvec = mem_cgroup_lruvec(pgdat, memcg);
3078 3087
3079 if (inactive_list_is_low(lruvec, false, sc, true)) 3088 if (inactive_list_is_low(lruvec, false, memcg, sc, true))
3080 shrink_active_list(SWAP_CLUSTER_MAX, lruvec, 3089 shrink_active_list(SWAP_CLUSTER_MAX, lruvec,
3081 sc, LRU_ACTIVE_ANON); 3090 sc, LRU_ACTIVE_ANON);
3082 3091
@@ -3084,22 +3093,44 @@ static void age_active_anon(struct pglist_data *pgdat,
3084 } while (memcg); 3093 } while (memcg);
3085} 3094}
3086 3095
3087static bool zone_balanced(struct zone *zone, int order, int classzone_idx) 3096/*
3097 * Returns true if there is an eligible zone balanced for the request order
3098 * and classzone_idx
3099 */
3100static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx)
3088{ 3101{
3089 unsigned long mark = high_wmark_pages(zone); 3102 int i;
3103 unsigned long mark = -1;
3104 struct zone *zone;
3090 3105
3091 if (!zone_watermark_ok_safe(zone, order, mark, classzone_idx)) 3106 for (i = 0; i <= classzone_idx; i++) {
3092 return false; 3107 zone = pgdat->node_zones + i;
3108
3109 if (!managed_zone(zone))
3110 continue;
3111
3112 mark = high_wmark_pages(zone);
3113 if (zone_watermark_ok_safe(zone, order, mark, classzone_idx))
3114 return true;
3115 }
3093 3116
3094 /* 3117 /*
3095 * If any eligible zone is balanced then the node is not considered 3118 * If a node has no populated zone within classzone_idx, it does not
3096 * to be congested or dirty 3119 * need balancing by definition. This can happen if a zone-restricted
3120 * allocation tries to wake a remote kswapd.
3097 */ 3121 */
3098 clear_bit(PGDAT_CONGESTED, &zone->zone_pgdat->flags); 3122 if (mark == -1)
3099 clear_bit(PGDAT_DIRTY, &zone->zone_pgdat->flags); 3123 return true;
3100 clear_bit(PGDAT_WRITEBACK, &zone->zone_pgdat->flags);
3101 3124
3102 return true; 3125 return false;
3126}
3127
3128/* Clear pgdat state for congested, dirty or under writeback. */
3129static void clear_pgdat_congested(pg_data_t *pgdat)
3130{
3131 clear_bit(PGDAT_CONGESTED, &pgdat->flags);
3132 clear_bit(PGDAT_DIRTY, &pgdat->flags);
3133 clear_bit(PGDAT_WRITEBACK, &pgdat->flags);
3103} 3134}
3104 3135
3105/* 3136/*
@@ -3110,11 +3141,9 @@ static bool zone_balanced(struct zone *zone, int order, int classzone_idx)
3110 */ 3141 */
3111static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, int classzone_idx) 3142static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, int classzone_idx)
3112{ 3143{
3113 int i;
3114
3115 /* 3144 /*
3116 * The throttled processes are normally woken up in balance_pgdat() as 3145 * The throttled processes are normally woken up in balance_pgdat() as
3117 * soon as pfmemalloc_watermark_ok() is true. But there is a potential 3146 * soon as allow_direct_reclaim() is true. But there is a potential
3118 * race between when kswapd checks the watermarks and a process gets 3147 * race between when kswapd checks the watermarks and a process gets
3119 * throttled. There is also a potential race if processes get 3148 * throttled. There is also a potential race if processes get
3120 * throttled, kswapd wakes, a large process exits thereby balancing the 3149 * throttled, kswapd wakes, a large process exits thereby balancing the
@@ -3128,17 +3157,16 @@ static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, int classzone_idx)
3128 if (waitqueue_active(&pgdat->pfmemalloc_wait)) 3157 if (waitqueue_active(&pgdat->pfmemalloc_wait))
3129 wake_up_all(&pgdat->pfmemalloc_wait); 3158 wake_up_all(&pgdat->pfmemalloc_wait);
3130 3159
3131 for (i = 0; i <= classzone_idx; i++) { 3160 /* Hopeless node, leave it to direct reclaim */
3132 struct zone *zone = pgdat->node_zones + i; 3161 if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES)
3133 3162 return true;
3134 if (!managed_zone(zone))
3135 continue;
3136 3163
3137 if (!zone_balanced(zone, order, classzone_idx)) 3164 if (pgdat_balanced(pgdat, order, classzone_idx)) {
3138 return false; 3165 clear_pgdat_congested(pgdat);
3166 return true;
3139 } 3167 }
3140 3168
3141 return true; 3169 return false;
3142} 3170}
3143 3171
3144/* 3172/*
@@ -3214,9 +3242,9 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
3214 count_vm_event(PAGEOUTRUN); 3242 count_vm_event(PAGEOUTRUN);
3215 3243
3216 do { 3244 do {
3245 unsigned long nr_reclaimed = sc.nr_reclaimed;
3217 bool raise_priority = true; 3246 bool raise_priority = true;
3218 3247
3219 sc.nr_reclaimed = 0;
3220 sc.reclaim_idx = classzone_idx; 3248 sc.reclaim_idx = classzone_idx;
3221 3249
3222 /* 3250 /*
@@ -3241,23 +3269,12 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
3241 } 3269 }
3242 3270
3243 /* 3271 /*
3244 * Only reclaim if there are no eligible zones. Check from 3272 * Only reclaim if there are no eligible zones. Note that
3245 * high to low zone as allocations prefer higher zones. 3273 * sc.reclaim_idx is not used as buffer_heads_over_limit may
3246 * Scanning from low to high zone would allow congestion to be 3274 * have adjusted it.
3247 * cleared during a very small window when a small low
3248 * zone was balanced even under extreme pressure when the
3249 * overall node may be congested. Note that sc.reclaim_idx
3250 * is not used as buffer_heads_over_limit may have adjusted
3251 * it.
3252 */ 3275 */
3253 for (i = classzone_idx; i >= 0; i--) { 3276 if (pgdat_balanced(pgdat, sc.order, classzone_idx))
3254 zone = pgdat->node_zones + i; 3277 goto out;
3255 if (!managed_zone(zone))
3256 continue;
3257
3258 if (zone_balanced(zone, sc.order, classzone_idx))
3259 goto out;
3260 }
3261 3278
3262 /* 3279 /*
3263 * Do some background aging of the anon list, to give 3280 * Do some background aging of the anon list, to give
@@ -3271,7 +3288,7 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
3271 * If we're getting trouble reclaiming, start doing writepage 3288 * If we're getting trouble reclaiming, start doing writepage
3272 * even in laptop mode. 3289 * even in laptop mode.
3273 */ 3290 */
3274 if (sc.priority < DEF_PRIORITY - 2 || !pgdat_reclaimable(pgdat)) 3291 if (sc.priority < DEF_PRIORITY - 2)
3275 sc.may_writepage = 1; 3292 sc.may_writepage = 1;
3276 3293
3277 /* Call soft limit reclaim before calling shrink_node. */ 3294 /* Call soft limit reclaim before calling shrink_node. */
@@ -3295,7 +3312,7 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
3295 * able to safely make forward progress. Wake them 3312 * able to safely make forward progress. Wake them
3296 */ 3313 */
3297 if (waitqueue_active(&pgdat->pfmemalloc_wait) && 3314 if (waitqueue_active(&pgdat->pfmemalloc_wait) &&
3298 pfmemalloc_watermark_ok(pgdat)) 3315 allow_direct_reclaim(pgdat))
3299 wake_up_all(&pgdat->pfmemalloc_wait); 3316 wake_up_all(&pgdat->pfmemalloc_wait);
3300 3317
3301 /* Check if kswapd should be suspending */ 3318 /* Check if kswapd should be suspending */
@@ -3306,11 +3323,16 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
3306 * Raise priority if scanning rate is too low or there was no 3323 * Raise priority if scanning rate is too low or there was no
3307 * progress in reclaiming pages 3324 * progress in reclaiming pages
3308 */ 3325 */
3309 if (raise_priority || !sc.nr_reclaimed) 3326 nr_reclaimed = sc.nr_reclaimed - nr_reclaimed;
3327 if (raise_priority || !nr_reclaimed)
3310 sc.priority--; 3328 sc.priority--;
3311 } while (sc.priority >= 1); 3329 } while (sc.priority >= 1);
3312 3330
3331 if (!sc.nr_reclaimed)
3332 pgdat->kswapd_failures++;
3333
3313out: 3334out:
3335 snapshot_refaults(NULL, pgdat);
3314 /* 3336 /*
3315 * Return the order kswapd stopped reclaiming at as 3337 * Return the order kswapd stopped reclaiming at as
3316 * prepare_kswapd_sleep() takes it into account. If another caller 3338 * prepare_kswapd_sleep() takes it into account. If another caller
@@ -3320,6 +3342,22 @@ out:
3320 return sc.order; 3342 return sc.order;
3321} 3343}
3322 3344
3345/*
3346 * pgdat->kswapd_classzone_idx is the highest zone index that a recent
3347 * allocation request woke kswapd for. When kswapd has not woken recently,
3348 * the value is MAX_NR_ZONES which is not a valid index. This compares a
3349 * given classzone and returns it or the highest classzone index kswapd
3350 * was recently woke for.
3351 */
3352static enum zone_type kswapd_classzone_idx(pg_data_t *pgdat,
3353 enum zone_type classzone_idx)
3354{
3355 if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
3356 return classzone_idx;
3357
3358 return max(pgdat->kswapd_classzone_idx, classzone_idx);
3359}
3360
3323static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order, 3361static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order,
3324 unsigned int classzone_idx) 3362 unsigned int classzone_idx)
3325{ 3363{
@@ -3331,7 +3369,13 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
3331 3369
3332 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); 3370 prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
3333 3371
3334 /* Try to sleep for a short interval */ 3372 /*
3373 * Try to sleep for a short interval. Note that kcompactd will only be
3374 * woken if it is possible to sleep for a short interval. This is
3375 * deliberate on the assumption that if reclaim cannot keep an
3376 * eligible zone balanced that it's also unlikely that compaction will
3377 * succeed.
3378 */
3335 if (prepare_kswapd_sleep(pgdat, reclaim_order, classzone_idx)) { 3379 if (prepare_kswapd_sleep(pgdat, reclaim_order, classzone_idx)) {
3336 /* 3380 /*
3337 * Compaction records what page blocks it recently failed to 3381 * Compaction records what page blocks it recently failed to
@@ -3355,7 +3399,7 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
3355 * the previous request that slept prematurely. 3399 * the previous request that slept prematurely.
3356 */ 3400 */
3357 if (remaining) { 3401 if (remaining) {
3358 pgdat->kswapd_classzone_idx = max(pgdat->kswapd_classzone_idx, classzone_idx); 3402 pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx);
3359 pgdat->kswapd_order = max(pgdat->kswapd_order, reclaim_order); 3403 pgdat->kswapd_order = max(pgdat->kswapd_order, reclaim_order);
3360 } 3404 }
3361 3405
@@ -3409,7 +3453,8 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_o
3409 */ 3453 */
3410static int kswapd(void *p) 3454static int kswapd(void *p)
3411{ 3455{
3412 unsigned int alloc_order, reclaim_order, classzone_idx; 3456 unsigned int alloc_order, reclaim_order;
3457 unsigned int classzone_idx = MAX_NR_ZONES - 1;
3413 pg_data_t *pgdat = (pg_data_t*)p; 3458 pg_data_t *pgdat = (pg_data_t*)p;
3414 struct task_struct *tsk = current; 3459 struct task_struct *tsk = current;
3415 3460
@@ -3439,20 +3484,23 @@ static int kswapd(void *p)
3439 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; 3484 tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
3440 set_freezable(); 3485 set_freezable();
3441 3486
3442 pgdat->kswapd_order = alloc_order = reclaim_order = 0; 3487 pgdat->kswapd_order = 0;
3443 pgdat->kswapd_classzone_idx = classzone_idx = 0; 3488 pgdat->kswapd_classzone_idx = MAX_NR_ZONES;
3444 for ( ; ; ) { 3489 for ( ; ; ) {
3445 bool ret; 3490 bool ret;
3446 3491
3492 alloc_order = reclaim_order = pgdat->kswapd_order;
3493 classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx);
3494
3447kswapd_try_sleep: 3495kswapd_try_sleep:
3448 kswapd_try_to_sleep(pgdat, alloc_order, reclaim_order, 3496 kswapd_try_to_sleep(pgdat, alloc_order, reclaim_order,
3449 classzone_idx); 3497 classzone_idx);
3450 3498
3451 /* Read the new order and classzone_idx */ 3499 /* Read the new order and classzone_idx */
3452 alloc_order = reclaim_order = pgdat->kswapd_order; 3500 alloc_order = reclaim_order = pgdat->kswapd_order;
3453 classzone_idx = pgdat->kswapd_classzone_idx; 3501 classzone_idx = kswapd_classzone_idx(pgdat, 0);
3454 pgdat->kswapd_order = 0; 3502 pgdat->kswapd_order = 0;
3455 pgdat->kswapd_classzone_idx = 0; 3503 pgdat->kswapd_classzone_idx = MAX_NR_ZONES;
3456 3504
3457 ret = try_to_freeze(); 3505 ret = try_to_freeze();
3458 if (kthread_should_stop()) 3506 if (kthread_should_stop())
@@ -3478,9 +3526,6 @@ kswapd_try_sleep:
3478 reclaim_order = balance_pgdat(pgdat, alloc_order, classzone_idx); 3526 reclaim_order = balance_pgdat(pgdat, alloc_order, classzone_idx);
3479 if (reclaim_order < alloc_order) 3527 if (reclaim_order < alloc_order)
3480 goto kswapd_try_sleep; 3528 goto kswapd_try_sleep;
3481
3482 alloc_order = reclaim_order = pgdat->kswapd_order;
3483 classzone_idx = pgdat->kswapd_classzone_idx;
3484 } 3529 }
3485 3530
3486 tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); 3531 tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD);
@@ -3496,7 +3541,6 @@ kswapd_try_sleep:
3496void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) 3541void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
3497{ 3542{
3498 pg_data_t *pgdat; 3543 pg_data_t *pgdat;
3499 int z;
3500 3544
3501 if (!managed_zone(zone)) 3545 if (!managed_zone(zone))
3502 return; 3546 return;
@@ -3504,22 +3548,20 @@ void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
3504 if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL)) 3548 if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL))
3505 return; 3549 return;
3506 pgdat = zone->zone_pgdat; 3550 pgdat = zone->zone_pgdat;
3507 pgdat->kswapd_classzone_idx = max(pgdat->kswapd_classzone_idx, classzone_idx); 3551 pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat,
3552 classzone_idx);
3508 pgdat->kswapd_order = max(pgdat->kswapd_order, order); 3553 pgdat->kswapd_order = max(pgdat->kswapd_order, order);
3509 if (!waitqueue_active(&pgdat->kswapd_wait)) 3554 if (!waitqueue_active(&pgdat->kswapd_wait))
3510 return; 3555 return;
3511 3556
3512 /* Only wake kswapd if all zones are unbalanced */ 3557 /* Hopeless node, leave it to direct reclaim */
3513 for (z = 0; z <= classzone_idx; z++) { 3558 if (pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES)
3514 zone = pgdat->node_zones + z; 3559 return;
3515 if (!managed_zone(zone))
3516 continue;
3517 3560
3518 if (zone_balanced(zone, order, classzone_idx)) 3561 if (pgdat_balanced(pgdat, order, classzone_idx))
3519 return; 3562 return;
3520 }
3521 3563
3522 trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); 3564 trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, classzone_idx, order);
3523 wake_up_interruptible(&pgdat->kswapd_wait); 3565 wake_up_interruptible(&pgdat->kswapd_wait);
3524} 3566}
3525 3567
@@ -3548,8 +3590,9 @@ unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
3548 struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); 3590 struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
3549 struct task_struct *p = current; 3591 struct task_struct *p = current;
3550 unsigned long nr_reclaimed; 3592 unsigned long nr_reclaimed;
3593 unsigned int noreclaim_flag;
3551 3594
3552 p->flags |= PF_MEMALLOC; 3595 noreclaim_flag = memalloc_noreclaim_save();
3553 lockdep_set_current_reclaim_state(sc.gfp_mask); 3596 lockdep_set_current_reclaim_state(sc.gfp_mask);
3554 reclaim_state.reclaimed_slab = 0; 3597 reclaim_state.reclaimed_slab = 0;
3555 p->reclaim_state = &reclaim_state; 3598 p->reclaim_state = &reclaim_state;
@@ -3558,7 +3601,7 @@ unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
3558 3601
3559 p->reclaim_state = NULL; 3602 p->reclaim_state = NULL;
3560 lockdep_clear_current_reclaim_state(); 3603 lockdep_clear_current_reclaim_state();
3561 p->flags &= ~PF_MEMALLOC; 3604 memalloc_noreclaim_restore(noreclaim_flag);
3562 3605
3563 return nr_reclaimed; 3606 return nr_reclaimed;
3564} 3607}
@@ -3723,9 +3766,10 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
3723 struct task_struct *p = current; 3766 struct task_struct *p = current;
3724 struct reclaim_state reclaim_state; 3767 struct reclaim_state reclaim_state;
3725 int classzone_idx = gfp_zone(gfp_mask); 3768 int classzone_idx = gfp_zone(gfp_mask);
3769 unsigned int noreclaim_flag;
3726 struct scan_control sc = { 3770 struct scan_control sc = {
3727 .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), 3771 .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX),
3728 .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), 3772 .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)),
3729 .order = order, 3773 .order = order,
3730 .priority = NODE_RECLAIM_PRIORITY, 3774 .priority = NODE_RECLAIM_PRIORITY,
3731 .may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE), 3775 .may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE),
@@ -3740,7 +3784,8 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
3740 * and we also need to be able to write out pages for RECLAIM_WRITE 3784 * and we also need to be able to write out pages for RECLAIM_WRITE
3741 * and RECLAIM_UNMAP. 3785 * and RECLAIM_UNMAP.
3742 */ 3786 */
3743 p->flags |= PF_MEMALLOC | PF_SWAPWRITE; 3787 noreclaim_flag = memalloc_noreclaim_save();
3788 p->flags |= PF_SWAPWRITE;
3744 lockdep_set_current_reclaim_state(gfp_mask); 3789 lockdep_set_current_reclaim_state(gfp_mask);
3745 reclaim_state.reclaimed_slab = 0; 3790 reclaim_state.reclaimed_slab = 0;
3746 p->reclaim_state = &reclaim_state; 3791 p->reclaim_state = &reclaim_state;
@@ -3756,7 +3801,8 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
3756 } 3801 }
3757 3802
3758 p->reclaim_state = NULL; 3803 p->reclaim_state = NULL;
3759 current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); 3804 current->flags &= ~PF_SWAPWRITE;
3805 memalloc_noreclaim_restore(noreclaim_flag);
3760 lockdep_clear_current_reclaim_state(); 3806 lockdep_clear_current_reclaim_state();
3761 return sc.nr_reclaimed >= nr_pages; 3807 return sc.nr_reclaimed >= nr_pages;
3762} 3808}
@@ -3779,9 +3825,6 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order)
3779 sum_zone_node_page_state(pgdat->node_id, NR_SLAB_RECLAIMABLE) <= pgdat->min_slab_pages) 3825 sum_zone_node_page_state(pgdat->node_id, NR_SLAB_RECLAIMABLE) <= pgdat->min_slab_pages)
3780 return NODE_RECLAIM_FULL; 3826 return NODE_RECLAIM_FULL;
3781 3827
3782 if (!pgdat_reclaimable(pgdat))
3783 return NODE_RECLAIM_FULL;
3784
3785 /* 3828 /*
3786 * Do not scan if the allocation should not be delayed. 3829 * Do not scan if the allocation should not be delayed.
3787 */ 3830 */
diff --git a/mm/vmstat.c b/mm/vmstat.c
index 809025ed97ea..f5fa1bd1eb16 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -954,7 +954,6 @@ const char * const vmstat_text[] = {
954 "nr_unevictable", 954 "nr_unevictable",
955 "nr_isolated_anon", 955 "nr_isolated_anon",
956 "nr_isolated_file", 956 "nr_isolated_file",
957 "nr_pages_scanned",
958 "workingset_refault", 957 "workingset_refault",
959 "workingset_activate", 958 "workingset_activate",
960 "workingset_nodereclaim", 959 "workingset_nodereclaim",
@@ -992,6 +991,7 @@ const char * const vmstat_text[] = {
992 "pgfree", 991 "pgfree",
993 "pgactivate", 992 "pgactivate",
994 "pgdeactivate", 993 "pgdeactivate",
994 "pglazyfree",
995 995
996 "pgfault", 996 "pgfault",
997 "pgmajfault", 997 "pgmajfault",
@@ -1124,8 +1124,12 @@ static void frag_stop(struct seq_file *m, void *arg)
1124{ 1124{
1125} 1125}
1126 1126
1127/* Walk all the zones in a node and print using a callback */ 1127/*
1128 * Walk zones in a node and print using a callback.
1129 * If @assert_populated is true, only use callback for zones that are populated.
1130 */
1128static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat, 1131static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1132 bool assert_populated,
1129 void (*print)(struct seq_file *m, pg_data_t *, struct zone *)) 1133 void (*print)(struct seq_file *m, pg_data_t *, struct zone *))
1130{ 1134{
1131 struct zone *zone; 1135 struct zone *zone;
@@ -1133,7 +1137,7 @@ static void walk_zones_in_node(struct seq_file *m, pg_data_t *pgdat,
1133 unsigned long flags; 1137 unsigned long flags;
1134 1138
1135 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { 1139 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1136 if (!populated_zone(zone)) 1140 if (assert_populated && !populated_zone(zone))
1137 continue; 1141 continue;
1138 1142
1139 spin_lock_irqsave(&zone->lock, flags); 1143 spin_lock_irqsave(&zone->lock, flags);
@@ -1161,7 +1165,7 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
1161static int frag_show(struct seq_file *m, void *arg) 1165static int frag_show(struct seq_file *m, void *arg)
1162{ 1166{
1163 pg_data_t *pgdat = (pg_data_t *)arg; 1167 pg_data_t *pgdat = (pg_data_t *)arg;
1164 walk_zones_in_node(m, pgdat, frag_show_print); 1168 walk_zones_in_node(m, pgdat, true, frag_show_print);
1165 return 0; 1169 return 0;
1166} 1170}
1167 1171
@@ -1202,7 +1206,7 @@ static int pagetypeinfo_showfree(struct seq_file *m, void *arg)
1202 seq_printf(m, "%6d ", order); 1206 seq_printf(m, "%6d ", order);
1203 seq_putc(m, '\n'); 1207 seq_putc(m, '\n');
1204 1208
1205 walk_zones_in_node(m, pgdat, pagetypeinfo_showfree_print); 1209 walk_zones_in_node(m, pgdat, true, pagetypeinfo_showfree_print);
1206 1210
1207 return 0; 1211 return 0;
1208} 1212}
@@ -1254,7 +1258,7 @@ static int pagetypeinfo_showblockcount(struct seq_file *m, void *arg)
1254 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) 1258 for (mtype = 0; mtype < MIGRATE_TYPES; mtype++)
1255 seq_printf(m, "%12s ", migratetype_names[mtype]); 1259 seq_printf(m, "%12s ", migratetype_names[mtype]);
1256 seq_putc(m, '\n'); 1260 seq_putc(m, '\n');
1257 walk_zones_in_node(m, pgdat, pagetypeinfo_showblockcount_print); 1261 walk_zones_in_node(m, pgdat, true, pagetypeinfo_showblockcount_print);
1258 1262
1259 return 0; 1263 return 0;
1260} 1264}
@@ -1280,7 +1284,7 @@ static void pagetypeinfo_showmixedcount(struct seq_file *m, pg_data_t *pgdat)
1280 seq_printf(m, "%12s ", migratetype_names[mtype]); 1284 seq_printf(m, "%12s ", migratetype_names[mtype]);
1281 seq_putc(m, '\n'); 1285 seq_putc(m, '\n');
1282 1286
1283 walk_zones_in_node(m, pgdat, pagetypeinfo_showmixedcount_print); 1287 walk_zones_in_node(m, pgdat, true, pagetypeinfo_showmixedcount_print);
1284#endif /* CONFIG_PAGE_OWNER */ 1288#endif /* CONFIG_PAGE_OWNER */
1285} 1289}
1286 1290
@@ -1378,7 +1382,6 @@ static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1378 "\n min %lu" 1382 "\n min %lu"
1379 "\n low %lu" 1383 "\n low %lu"
1380 "\n high %lu" 1384 "\n high %lu"
1381 "\n node_scanned %lu"
1382 "\n spanned %lu" 1385 "\n spanned %lu"
1383 "\n present %lu" 1386 "\n present %lu"
1384 "\n managed %lu", 1387 "\n managed %lu",
@@ -1386,23 +1389,28 @@ static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1386 min_wmark_pages(zone), 1389 min_wmark_pages(zone),
1387 low_wmark_pages(zone), 1390 low_wmark_pages(zone),
1388 high_wmark_pages(zone), 1391 high_wmark_pages(zone),
1389 node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED),
1390 zone->spanned_pages, 1392 zone->spanned_pages,
1391 zone->present_pages, 1393 zone->present_pages,
1392 zone->managed_pages); 1394 zone->managed_pages);
1393 1395
1394 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1395 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1396 zone_page_state(zone, i));
1397
1398 seq_printf(m, 1396 seq_printf(m,
1399 "\n protection: (%ld", 1397 "\n protection: (%ld",
1400 zone->lowmem_reserve[0]); 1398 zone->lowmem_reserve[0]);
1401 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) 1399 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
1402 seq_printf(m, ", %ld", zone->lowmem_reserve[i]); 1400 seq_printf(m, ", %ld", zone->lowmem_reserve[i]);
1403 seq_printf(m, 1401 seq_putc(m, ')');
1404 ")" 1402
1405 "\n pagesets"); 1403 /* If unpopulated, no other information is useful */
1404 if (!populated_zone(zone)) {
1405 seq_putc(m, '\n');
1406 return;
1407 }
1408
1409 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++)
1410 seq_printf(m, "\n %-12s %lu", vmstat_text[i],
1411 zone_page_state(zone, i));
1412
1413 seq_printf(m, "\n pagesets");
1406 for_each_online_cpu(i) { 1414 for_each_online_cpu(i) {
1407 struct per_cpu_pageset *pageset; 1415 struct per_cpu_pageset *pageset;
1408 1416
@@ -1425,19 +1433,22 @@ static void zoneinfo_show_print(struct seq_file *m, pg_data_t *pgdat,
1425 "\n node_unreclaimable: %u" 1433 "\n node_unreclaimable: %u"
1426 "\n start_pfn: %lu" 1434 "\n start_pfn: %lu"
1427 "\n node_inactive_ratio: %u", 1435 "\n node_inactive_ratio: %u",
1428 !pgdat_reclaimable(zone->zone_pgdat), 1436 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES,
1429 zone->zone_start_pfn, 1437 zone->zone_start_pfn,
1430 zone->zone_pgdat->inactive_ratio); 1438 zone->zone_pgdat->inactive_ratio);
1431 seq_putc(m, '\n'); 1439 seq_putc(m, '\n');
1432} 1440}
1433 1441
1434/* 1442/*
1435 * Output information about zones in @pgdat. 1443 * Output information about zones in @pgdat. All zones are printed regardless
1444 * of whether they are populated or not: lowmem_reserve_ratio operates on the
1445 * set of all zones and userspace would not be aware of such zones if they are
1446 * suppressed here (zoneinfo displays the effect of lowmem_reserve_ratio).
1436 */ 1447 */
1437static int zoneinfo_show(struct seq_file *m, void *arg) 1448static int zoneinfo_show(struct seq_file *m, void *arg)
1438{ 1449{
1439 pg_data_t *pgdat = (pg_data_t *)arg; 1450 pg_data_t *pgdat = (pg_data_t *)arg;
1440 walk_zones_in_node(m, pgdat, zoneinfo_show_print); 1451 walk_zones_in_node(m, pgdat, false, zoneinfo_show_print);
1441 return 0; 1452 return 0;
1442} 1453}
1443 1454
@@ -1586,22 +1597,9 @@ int vmstat_refresh(struct ctl_table *table, int write,
1586 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { 1597 for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) {
1587 val = atomic_long_read(&vm_zone_stat[i]); 1598 val = atomic_long_read(&vm_zone_stat[i]);
1588 if (val < 0) { 1599 if (val < 0) {
1589 switch (i) { 1600 pr_warn("%s: %s %ld\n",
1590 case NR_PAGES_SCANNED: 1601 __func__, vmstat_text[i], val);
1591 /* 1602 err = -EINVAL;
1592 * This is often seen to go negative in
1593 * recent kernels, but not to go permanently
1594 * negative. Whilst it would be nicer not to
1595 * have exceptions, rooting them out would be
1596 * another task, of rather low priority.
1597 */
1598 break;
1599 default:
1600 pr_warn("%s: %s %ld\n",
1601 __func__, vmstat_text[i], val);
1602 err = -EINVAL;
1603 break;
1604 }
1605 } 1603 }
1606 } 1604 }
1607 if (err) 1605 if (err)
@@ -1768,8 +1766,7 @@ void __init init_mm_internals(void)
1768{ 1766{
1769 int ret __maybe_unused; 1767 int ret __maybe_unused;
1770 1768
1771 mm_percpu_wq = alloc_workqueue("mm_percpu_wq", 1769 mm_percpu_wq = alloc_workqueue("mm_percpu_wq", WQ_MEM_RECLAIM, 0);
1772 WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
1773 1770
1774#ifdef CONFIG_SMP 1771#ifdef CONFIG_SMP
1775 ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead", 1772 ret = cpuhp_setup_state_nocalls(CPUHP_MM_VMSTAT_DEAD, "mm/vmstat:dead",
@@ -1857,7 +1854,7 @@ static int unusable_show(struct seq_file *m, void *arg)
1857 if (!node_state(pgdat->node_id, N_MEMORY)) 1854 if (!node_state(pgdat->node_id, N_MEMORY))
1858 return 0; 1855 return 0;
1859 1856
1860 walk_zones_in_node(m, pgdat, unusable_show_print); 1857 walk_zones_in_node(m, pgdat, true, unusable_show_print);
1861 1858
1862 return 0; 1859 return 0;
1863} 1860}
@@ -1909,7 +1906,7 @@ static int extfrag_show(struct seq_file *m, void *arg)
1909{ 1906{
1910 pg_data_t *pgdat = (pg_data_t *)arg; 1907 pg_data_t *pgdat = (pg_data_t *)arg;
1911 1908
1912 walk_zones_in_node(m, pgdat, extfrag_show_print); 1909 walk_zones_in_node(m, pgdat, true, extfrag_show_print);
1913 1910
1914 return 0; 1911 return 0;
1915} 1912}
diff --git a/mm/workingset.c b/mm/workingset.c
index eda05c71fa49..b8c9ab678479 100644
--- a/mm/workingset.c
+++ b/mm/workingset.c
@@ -269,7 +269,6 @@ bool workingset_refault(void *shadow)
269 lruvec = mem_cgroup_lruvec(pgdat, memcg); 269 lruvec = mem_cgroup_lruvec(pgdat, memcg);
270 refault = atomic_long_read(&lruvec->inactive_age); 270 refault = atomic_long_read(&lruvec->inactive_age);
271 active_file = lruvec_lru_size(lruvec, LRU_ACTIVE_FILE, MAX_NR_ZONES); 271 active_file = lruvec_lru_size(lruvec, LRU_ACTIVE_FILE, MAX_NR_ZONES);
272 rcu_read_unlock();
273 272
274 /* 273 /*
275 * The unsigned subtraction here gives an accurate distance 274 * The unsigned subtraction here gives an accurate distance
@@ -290,11 +289,15 @@ bool workingset_refault(void *shadow)
290 refault_distance = (refault - eviction) & EVICTION_MASK; 289 refault_distance = (refault - eviction) & EVICTION_MASK;
291 290
292 inc_node_state(pgdat, WORKINGSET_REFAULT); 291 inc_node_state(pgdat, WORKINGSET_REFAULT);
292 inc_memcg_state(memcg, WORKINGSET_REFAULT);
293 293
294 if (refault_distance <= active_file) { 294 if (refault_distance <= active_file) {
295 inc_node_state(pgdat, WORKINGSET_ACTIVATE); 295 inc_node_state(pgdat, WORKINGSET_ACTIVATE);
296 inc_memcg_state(memcg, WORKINGSET_ACTIVATE);
297 rcu_read_unlock();
296 return true; 298 return true;
297 } 299 }
300 rcu_read_unlock();
298 return false; 301 return false;
299} 302}
300 303
@@ -472,6 +475,7 @@ static enum lru_status shadow_lru_isolate(struct list_head *item,
472 if (WARN_ON_ONCE(node->exceptional)) 475 if (WARN_ON_ONCE(node->exceptional))
473 goto out_invalid; 476 goto out_invalid;
474 inc_node_state(page_pgdat(virt_to_page(node)), WORKINGSET_NODERECLAIM); 477 inc_node_state(page_pgdat(virt_to_page(node)), WORKINGSET_NODERECLAIM);
478 inc_memcg_page_state(virt_to_page(node), WORKINGSET_NODERECLAIM);
475 __radix_tree_delete_node(&mapping->page_tree, node, 479 __radix_tree_delete_node(&mapping->page_tree, node,
476 workingset_update_node, mapping); 480 workingset_update_node, mapping);
477 481
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-17 13:30:50 -0400