diff options
author | Benjamin Herrenschmidt <benh@kernel.crashing.org> | 2010-05-06 21:29:25 -0400 |
---|---|---|
committer | Benjamin Herrenschmidt <benh@kernel.crashing.org> | 2010-05-06 21:29:25 -0400 |
commit | 1ed31d6db90d51010545921e59d369d2f92b7ac2 (patch) | |
tree | 358a0b346bc8135cd5e53700eb44308b1a7c8c5b /mm | |
parent | ceba1abcb00b0ef0b1efcd715285f6e05523edef (diff) | |
parent | 722154e4cacf015161efe60009ae9be23d492296 (diff) |
Merge commit 'origin/master' into next
Diffstat (limited to 'mm')
-rw-r--r-- | mm/backing-dev.c | 37 | ||||
-rw-r--r-- | mm/bootmem.c | 17 | ||||
-rw-r--r-- | mm/hugetlb.c | 5 | ||||
-rw-r--r-- | mm/ksm.c | 12 | ||||
-rw-r--r-- | mm/memcontrol.c | 22 | ||||
-rw-r--r-- | mm/memory.c | 3 | ||||
-rw-r--r-- | mm/mmap.c | 113 | ||||
-rw-r--r-- | mm/pagewalk.c | 47 | ||||
-rw-r--r-- | mm/readahead.c | 2 | ||||
-rw-r--r-- | mm/rmap.c | 33 | ||||
-rw-r--r-- | mm/slab.c | 13 | ||||
-rw-r--r-- | mm/slub.c | 5 | ||||
-rw-r--r-- | mm/util.c | 21 | ||||
-rw-r--r-- | mm/vmscan.c | 23 |
14 files changed, 244 insertions, 109 deletions
diff --git a/mm/backing-dev.c b/mm/backing-dev.c index 0e8ca0347707..707d0dc6da0f 100644 --- a/mm/backing-dev.c +++ b/mm/backing-dev.c | |||
@@ -11,6 +11,8 @@ | |||
11 | #include <linux/writeback.h> | 11 | #include <linux/writeback.h> |
12 | #include <linux/device.h> | 12 | #include <linux/device.h> |
13 | 13 | ||
14 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); | ||
15 | |||
14 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) | 16 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
15 | { | 17 | { |
16 | } | 18 | } |
@@ -25,6 +27,11 @@ struct backing_dev_info default_backing_dev_info = { | |||
25 | }; | 27 | }; |
26 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | 28 | EXPORT_SYMBOL_GPL(default_backing_dev_info); |
27 | 29 | ||
30 | struct backing_dev_info noop_backing_dev_info = { | ||
31 | .name = "noop", | ||
32 | }; | ||
33 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | ||
34 | |||
28 | static struct class *bdi_class; | 35 | static struct class *bdi_class; |
29 | 36 | ||
30 | /* | 37 | /* |
@@ -227,6 +234,9 @@ static struct device_attribute bdi_dev_attrs[] = { | |||
227 | static __init int bdi_class_init(void) | 234 | static __init int bdi_class_init(void) |
228 | { | 235 | { |
229 | bdi_class = class_create(THIS_MODULE, "bdi"); | 236 | bdi_class = class_create(THIS_MODULE, "bdi"); |
237 | if (IS_ERR(bdi_class)) | ||
238 | return PTR_ERR(bdi_class); | ||
239 | |||
230 | bdi_class->dev_attrs = bdi_dev_attrs; | 240 | bdi_class->dev_attrs = bdi_dev_attrs; |
231 | bdi_debug_init(); | 241 | bdi_debug_init(); |
232 | return 0; | 242 | return 0; |
@@ -712,6 +722,33 @@ void bdi_destroy(struct backing_dev_info *bdi) | |||
712 | } | 722 | } |
713 | EXPORT_SYMBOL(bdi_destroy); | 723 | EXPORT_SYMBOL(bdi_destroy); |
714 | 724 | ||
725 | /* | ||
726 | * For use from filesystems to quickly init and register a bdi associated | ||
727 | * with dirty writeback | ||
728 | */ | ||
729 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | ||
730 | unsigned int cap) | ||
731 | { | ||
732 | char tmp[32]; | ||
733 | int err; | ||
734 | |||
735 | bdi->name = name; | ||
736 | bdi->capabilities = cap; | ||
737 | err = bdi_init(bdi); | ||
738 | if (err) | ||
739 | return err; | ||
740 | |||
741 | sprintf(tmp, "%.28s%s", name, "-%d"); | ||
742 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | ||
743 | if (err) { | ||
744 | bdi_destroy(bdi); | ||
745 | return err; | ||
746 | } | ||
747 | |||
748 | return 0; | ||
749 | } | ||
750 | EXPORT_SYMBOL(bdi_setup_and_register); | ||
751 | |||
715 | static wait_queue_head_t congestion_wqh[2] = { | 752 | static wait_queue_head_t congestion_wqh[2] = { |
716 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | 753 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
717 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | 754 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
diff --git a/mm/bootmem.c b/mm/bootmem.c index eff224220571..58c66cc5056a 100644 --- a/mm/bootmem.c +++ b/mm/bootmem.c | |||
@@ -304,9 +304,22 @@ unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) | |||
304 | unsigned long __init free_all_bootmem(void) | 304 | unsigned long __init free_all_bootmem(void) |
305 | { | 305 | { |
306 | #ifdef CONFIG_NO_BOOTMEM | 306 | #ifdef CONFIG_NO_BOOTMEM |
307 | return free_all_memory_core_early(NODE_DATA(0)->node_id); | 307 | /* |
308 | * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id | ||
309 | * because in some case like Node0 doesnt have RAM installed | ||
310 | * low ram will be on Node1 | ||
311 | * Use MAX_NUMNODES will make sure all ranges in early_node_map[] | ||
312 | * will be used instead of only Node0 related | ||
313 | */ | ||
314 | return free_all_memory_core_early(MAX_NUMNODES); | ||
308 | #else | 315 | #else |
309 | return free_all_bootmem_core(NODE_DATA(0)->bdata); | 316 | unsigned long total_pages = 0; |
317 | bootmem_data_t *bdata; | ||
318 | |||
319 | list_for_each_entry(bdata, &bdata_list, list) | ||
320 | total_pages += free_all_bootmem_core(bdata); | ||
321 | |||
322 | return total_pages; | ||
310 | #endif | 323 | #endif |
311 | } | 324 | } |
312 | 325 | ||
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 6034dc9e9796..ffbdfc86aedf 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c | |||
@@ -546,6 +546,7 @@ static void free_huge_page(struct page *page) | |||
546 | 546 | ||
547 | mapping = (struct address_space *) page_private(page); | 547 | mapping = (struct address_space *) page_private(page); |
548 | set_page_private(page, 0); | 548 | set_page_private(page, 0); |
549 | page->mapping = NULL; | ||
549 | BUG_ON(page_count(page)); | 550 | BUG_ON(page_count(page)); |
550 | INIT_LIST_HEAD(&page->lru); | 551 | INIT_LIST_HEAD(&page->lru); |
551 | 552 | ||
@@ -2447,8 +2448,10 @@ retry: | |||
2447 | spin_lock(&inode->i_lock); | 2448 | spin_lock(&inode->i_lock); |
2448 | inode->i_blocks += blocks_per_huge_page(h); | 2449 | inode->i_blocks += blocks_per_huge_page(h); |
2449 | spin_unlock(&inode->i_lock); | 2450 | spin_unlock(&inode->i_lock); |
2450 | } else | 2451 | } else { |
2451 | lock_page(page); | 2452 | lock_page(page); |
2453 | page->mapping = HUGETLB_POISON; | ||
2454 | } | ||
2452 | } | 2455 | } |
2453 | 2456 | ||
2454 | /* | 2457 | /* |
@@ -365,7 +365,7 @@ static int break_ksm(struct vm_area_struct *vma, unsigned long addr) | |||
365 | do { | 365 | do { |
366 | cond_resched(); | 366 | cond_resched(); |
367 | page = follow_page(vma, addr, FOLL_GET); | 367 | page = follow_page(vma, addr, FOLL_GET); |
368 | if (!page) | 368 | if (IS_ERR_OR_NULL(page)) |
369 | break; | 369 | break; |
370 | if (PageKsm(page)) | 370 | if (PageKsm(page)) |
371 | ret = handle_mm_fault(vma->vm_mm, vma, addr, | 371 | ret = handle_mm_fault(vma->vm_mm, vma, addr, |
@@ -447,7 +447,7 @@ static struct page *get_mergeable_page(struct rmap_item *rmap_item) | |||
447 | goto out; | 447 | goto out; |
448 | 448 | ||
449 | page = follow_page(vma, addr, FOLL_GET); | 449 | page = follow_page(vma, addr, FOLL_GET); |
450 | if (!page) | 450 | if (IS_ERR_OR_NULL(page)) |
451 | goto out; | 451 | goto out; |
452 | if (PageAnon(page)) { | 452 | if (PageAnon(page)) { |
453 | flush_anon_page(vma, page, addr); | 453 | flush_anon_page(vma, page, addr); |
@@ -1086,7 +1086,7 @@ struct rmap_item *unstable_tree_search_insert(struct rmap_item *rmap_item, | |||
1086 | cond_resched(); | 1086 | cond_resched(); |
1087 | tree_rmap_item = rb_entry(*new, struct rmap_item, node); | 1087 | tree_rmap_item = rb_entry(*new, struct rmap_item, node); |
1088 | tree_page = get_mergeable_page(tree_rmap_item); | 1088 | tree_page = get_mergeable_page(tree_rmap_item); |
1089 | if (!tree_page) | 1089 | if (IS_ERR_OR_NULL(tree_page)) |
1090 | return NULL; | 1090 | return NULL; |
1091 | 1091 | ||
1092 | /* | 1092 | /* |
@@ -1294,7 +1294,7 @@ next_mm: | |||
1294 | if (ksm_test_exit(mm)) | 1294 | if (ksm_test_exit(mm)) |
1295 | break; | 1295 | break; |
1296 | *page = follow_page(vma, ksm_scan.address, FOLL_GET); | 1296 | *page = follow_page(vma, ksm_scan.address, FOLL_GET); |
1297 | if (*page && PageAnon(*page)) { | 1297 | if (!IS_ERR_OR_NULL(*page) && PageAnon(*page)) { |
1298 | flush_anon_page(vma, *page, ksm_scan.address); | 1298 | flush_anon_page(vma, *page, ksm_scan.address); |
1299 | flush_dcache_page(*page); | 1299 | flush_dcache_page(*page); |
1300 | rmap_item = get_next_rmap_item(slot, | 1300 | rmap_item = get_next_rmap_item(slot, |
@@ -1308,7 +1308,7 @@ next_mm: | |||
1308 | up_read(&mm->mmap_sem); | 1308 | up_read(&mm->mmap_sem); |
1309 | return rmap_item; | 1309 | return rmap_item; |
1310 | } | 1310 | } |
1311 | if (*page) | 1311 | if (!IS_ERR_OR_NULL(*page)) |
1312 | put_page(*page); | 1312 | put_page(*page); |
1313 | ksm_scan.address += PAGE_SIZE; | 1313 | ksm_scan.address += PAGE_SIZE; |
1314 | cond_resched(); | 1314 | cond_resched(); |
@@ -1367,7 +1367,7 @@ next_mm: | |||
1367 | static void ksm_do_scan(unsigned int scan_npages) | 1367 | static void ksm_do_scan(unsigned int scan_npages) |
1368 | { | 1368 | { |
1369 | struct rmap_item *rmap_item; | 1369 | struct rmap_item *rmap_item; |
1370 | struct page *page; | 1370 | struct page *uninitialized_var(page); |
1371 | 1371 | ||
1372 | while (scan_npages--) { | 1372 | while (scan_npages--) { |
1373 | cond_resched(); | 1373 | cond_resched(); |
diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 9ed760dc7448..6c755de385f7 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c | |||
@@ -1359,16 +1359,19 @@ void mem_cgroup_update_file_mapped(struct page *page, int val) | |||
1359 | 1359 | ||
1360 | lock_page_cgroup(pc); | 1360 | lock_page_cgroup(pc); |
1361 | mem = pc->mem_cgroup; | 1361 | mem = pc->mem_cgroup; |
1362 | if (!mem) | 1362 | if (!mem || !PageCgroupUsed(pc)) |
1363 | goto done; | ||
1364 | |||
1365 | if (!PageCgroupUsed(pc)) | ||
1366 | goto done; | 1363 | goto done; |
1367 | 1364 | ||
1368 | /* | 1365 | /* |
1369 | * Preemption is already disabled. We can use __this_cpu_xxx | 1366 | * Preemption is already disabled. We can use __this_cpu_xxx |
1370 | */ | 1367 | */ |
1371 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], val); | 1368 | if (val > 0) { |
1369 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | ||
1370 | SetPageCgroupFileMapped(pc); | ||
1371 | } else { | ||
1372 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | ||
1373 | ClearPageCgroupFileMapped(pc); | ||
1374 | } | ||
1372 | 1375 | ||
1373 | done: | 1376 | done: |
1374 | unlock_page_cgroup(pc); | 1377 | unlock_page_cgroup(pc); |
@@ -1801,16 +1804,13 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |||
1801 | static void __mem_cgroup_move_account(struct page_cgroup *pc, | 1804 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
1802 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) | 1805 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
1803 | { | 1806 | { |
1804 | struct page *page; | ||
1805 | |||
1806 | VM_BUG_ON(from == to); | 1807 | VM_BUG_ON(from == to); |
1807 | VM_BUG_ON(PageLRU(pc->page)); | 1808 | VM_BUG_ON(PageLRU(pc->page)); |
1808 | VM_BUG_ON(!PageCgroupLocked(pc)); | 1809 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1809 | VM_BUG_ON(!PageCgroupUsed(pc)); | 1810 | VM_BUG_ON(!PageCgroupUsed(pc)); |
1810 | VM_BUG_ON(pc->mem_cgroup != from); | 1811 | VM_BUG_ON(pc->mem_cgroup != from); |
1811 | 1812 | ||
1812 | page = pc->page; | 1813 | if (PageCgroupFileMapped(pc)) { |
1813 | if (page_mapped(page) && !PageAnon(page)) { | ||
1814 | /* Update mapped_file data for mem_cgroup */ | 1814 | /* Update mapped_file data for mem_cgroup */ |
1815 | preempt_disable(); | 1815 | preempt_disable(); |
1816 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | 1816 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); |
@@ -2429,11 +2429,11 @@ int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) | |||
2429 | } | 2429 | } |
2430 | unlock_page_cgroup(pc); | 2430 | unlock_page_cgroup(pc); |
2431 | 2431 | ||
2432 | *ptr = mem; | ||
2432 | if (mem) { | 2433 | if (mem) { |
2433 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); | 2434 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false); |
2434 | css_put(&mem->css); | 2435 | css_put(&mem->css); |
2435 | } | 2436 | } |
2436 | *ptr = mem; | ||
2437 | return ret; | 2437 | return ret; |
2438 | } | 2438 | } |
2439 | 2439 | ||
diff --git a/mm/memory.c b/mm/memory.c index 1d2ea39260e5..833952d8b74d 100644 --- a/mm/memory.c +++ b/mm/memory.c | |||
@@ -125,13 +125,12 @@ core_initcall(init_zero_pfn); | |||
125 | 125 | ||
126 | #if defined(SPLIT_RSS_COUNTING) | 126 | #if defined(SPLIT_RSS_COUNTING) |
127 | 127 | ||
128 | void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm) | 128 | static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm) |
129 | { | 129 | { |
130 | int i; | 130 | int i; |
131 | 131 | ||
132 | for (i = 0; i < NR_MM_COUNTERS; i++) { | 132 | for (i = 0; i < NR_MM_COUNTERS; i++) { |
133 | if (task->rss_stat.count[i]) { | 133 | if (task->rss_stat.count[i]) { |
134 | BUG_ON(!mm); | ||
135 | add_mm_counter(mm, i, task->rss_stat.count[i]); | 134 | add_mm_counter(mm, i, task->rss_stat.count[i]); |
136 | task->rss_stat.count[i] = 0; | 135 | task->rss_stat.count[i] = 0; |
137 | } | 136 | } |
@@ -507,11 +507,12 @@ int vma_adjust(struct vm_area_struct *vma, unsigned long start, | |||
507 | struct address_space *mapping = NULL; | 507 | struct address_space *mapping = NULL; |
508 | struct prio_tree_root *root = NULL; | 508 | struct prio_tree_root *root = NULL; |
509 | struct file *file = vma->vm_file; | 509 | struct file *file = vma->vm_file; |
510 | struct anon_vma *anon_vma = NULL; | ||
511 | long adjust_next = 0; | 510 | long adjust_next = 0; |
512 | int remove_next = 0; | 511 | int remove_next = 0; |
513 | 512 | ||
514 | if (next && !insert) { | 513 | if (next && !insert) { |
514 | struct vm_area_struct *exporter = NULL; | ||
515 | |||
515 | if (end >= next->vm_end) { | 516 | if (end >= next->vm_end) { |
516 | /* | 517 | /* |
517 | * vma expands, overlapping all the next, and | 518 | * vma expands, overlapping all the next, and |
@@ -519,7 +520,7 @@ int vma_adjust(struct vm_area_struct *vma, unsigned long start, | |||
519 | */ | 520 | */ |
520 | again: remove_next = 1 + (end > next->vm_end); | 521 | again: remove_next = 1 + (end > next->vm_end); |
521 | end = next->vm_end; | 522 | end = next->vm_end; |
522 | anon_vma = next->anon_vma; | 523 | exporter = next; |
523 | importer = vma; | 524 | importer = vma; |
524 | } else if (end > next->vm_start) { | 525 | } else if (end > next->vm_start) { |
525 | /* | 526 | /* |
@@ -527,7 +528,7 @@ again: remove_next = 1 + (end > next->vm_end); | |||
527 | * mprotect case 5 shifting the boundary up. | 528 | * mprotect case 5 shifting the boundary up. |
528 | */ | 529 | */ |
529 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; | 530 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; |
530 | anon_vma = next->anon_vma; | 531 | exporter = next; |
531 | importer = vma; | 532 | importer = vma; |
532 | } else if (end < vma->vm_end) { | 533 | } else if (end < vma->vm_end) { |
533 | /* | 534 | /* |
@@ -536,28 +537,19 @@ again: remove_next = 1 + (end > next->vm_end); | |||
536 | * mprotect case 4 shifting the boundary down. | 537 | * mprotect case 4 shifting the boundary down. |
537 | */ | 538 | */ |
538 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); | 539 | adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); |
539 | anon_vma = next->anon_vma; | 540 | exporter = vma; |
540 | importer = next; | 541 | importer = next; |
541 | } | 542 | } |
542 | } | ||
543 | 543 | ||
544 | /* | ||
545 | * When changing only vma->vm_end, we don't really need anon_vma lock. | ||
546 | */ | ||
547 | if (vma->anon_vma && (insert || importer || start != vma->vm_start)) | ||
548 | anon_vma = vma->anon_vma; | ||
549 | if (anon_vma) { | ||
550 | /* | 544 | /* |
551 | * Easily overlooked: when mprotect shifts the boundary, | 545 | * Easily overlooked: when mprotect shifts the boundary, |
552 | * make sure the expanding vma has anon_vma set if the | 546 | * make sure the expanding vma has anon_vma set if the |
553 | * shrinking vma had, to cover any anon pages imported. | 547 | * shrinking vma had, to cover any anon pages imported. |
554 | */ | 548 | */ |
555 | if (importer && !importer->anon_vma) { | 549 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
556 | /* Block reverse map lookups until things are set up. */ | 550 | if (anon_vma_clone(importer, exporter)) |
557 | if (anon_vma_clone(importer, vma)) { | ||
558 | return -ENOMEM; | 551 | return -ENOMEM; |
559 | } | 552 | importer->anon_vma = exporter->anon_vma; |
560 | importer->anon_vma = anon_vma; | ||
561 | } | 553 | } |
562 | } | 554 | } |
563 | 555 | ||
@@ -825,6 +817,61 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm, | |||
825 | } | 817 | } |
826 | 818 | ||
827 | /* | 819 | /* |
820 | * Rough compatbility check to quickly see if it's even worth looking | ||
821 | * at sharing an anon_vma. | ||
822 | * | ||
823 | * They need to have the same vm_file, and the flags can only differ | ||
824 | * in things that mprotect may change. | ||
825 | * | ||
826 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that | ||
827 | * we can merge the two vma's. For example, we refuse to merge a vma if | ||
828 | * there is a vm_ops->close() function, because that indicates that the | ||
829 | * driver is doing some kind of reference counting. But that doesn't | ||
830 | * really matter for the anon_vma sharing case. | ||
831 | */ | ||
832 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) | ||
833 | { | ||
834 | return a->vm_end == b->vm_start && | ||
835 | mpol_equal(vma_policy(a), vma_policy(b)) && | ||
836 | a->vm_file == b->vm_file && | ||
837 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && | ||
838 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); | ||
839 | } | ||
840 | |||
841 | /* | ||
842 | * Do some basic sanity checking to see if we can re-use the anon_vma | ||
843 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be | ||
844 | * the same as 'old', the other will be the new one that is trying | ||
845 | * to share the anon_vma. | ||
846 | * | ||
847 | * NOTE! This runs with mm_sem held for reading, so it is possible that | ||
848 | * the anon_vma of 'old' is concurrently in the process of being set up | ||
849 | * by another page fault trying to merge _that_. But that's ok: if it | ||
850 | * is being set up, that automatically means that it will be a singleton | ||
851 | * acceptable for merging, so we can do all of this optimistically. But | ||
852 | * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. | ||
853 | * | ||
854 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only | ||
855 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid | ||
856 | * is to return an anon_vma that is "complex" due to having gone through | ||
857 | * a fork). | ||
858 | * | ||
859 | * We also make sure that the two vma's are compatible (adjacent, | ||
860 | * and with the same memory policies). That's all stable, even with just | ||
861 | * a read lock on the mm_sem. | ||
862 | */ | ||
863 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) | ||
864 | { | ||
865 | if (anon_vma_compatible(a, b)) { | ||
866 | struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); | ||
867 | |||
868 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) | ||
869 | return anon_vma; | ||
870 | } | ||
871 | return NULL; | ||
872 | } | ||
873 | |||
874 | /* | ||
828 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check | 875 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check |
829 | * neighbouring vmas for a suitable anon_vma, before it goes off | 876 | * neighbouring vmas for a suitable anon_vma, before it goes off |
830 | * to allocate a new anon_vma. It checks because a repetitive | 877 | * to allocate a new anon_vma. It checks because a repetitive |
@@ -834,28 +881,16 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm, | |||
834 | */ | 881 | */ |
835 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) | 882 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) |
836 | { | 883 | { |
884 | struct anon_vma *anon_vma; | ||
837 | struct vm_area_struct *near; | 885 | struct vm_area_struct *near; |
838 | unsigned long vm_flags; | ||
839 | 886 | ||
840 | near = vma->vm_next; | 887 | near = vma->vm_next; |
841 | if (!near) | 888 | if (!near) |
842 | goto try_prev; | 889 | goto try_prev; |
843 | 890 | ||
844 | /* | 891 | anon_vma = reusable_anon_vma(near, vma, near); |
845 | * Since only mprotect tries to remerge vmas, match flags | 892 | if (anon_vma) |
846 | * which might be mprotected into each other later on. | 893 | return anon_vma; |
847 | * Neither mlock nor madvise tries to remerge at present, | ||
848 | * so leave their flags as obstructing a merge. | ||
849 | */ | ||
850 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | ||
851 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | ||
852 | |||
853 | if (near->anon_vma && vma->vm_end == near->vm_start && | ||
854 | mpol_equal(vma_policy(vma), vma_policy(near)) && | ||
855 | can_vma_merge_before(near, vm_flags, | ||
856 | NULL, vma->vm_file, vma->vm_pgoff + | ||
857 | ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) | ||
858 | return near->anon_vma; | ||
859 | try_prev: | 894 | try_prev: |
860 | /* | 895 | /* |
861 | * It is potentially slow to have to call find_vma_prev here. | 896 | * It is potentially slow to have to call find_vma_prev here. |
@@ -868,14 +903,9 @@ try_prev: | |||
868 | if (!near) | 903 | if (!near) |
869 | goto none; | 904 | goto none; |
870 | 905 | ||
871 | vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); | 906 | anon_vma = reusable_anon_vma(near, near, vma); |
872 | vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); | 907 | if (anon_vma) |
873 | 908 | return anon_vma; | |
874 | if (near->anon_vma && near->vm_end == vma->vm_start && | ||
875 | mpol_equal(vma_policy(near), vma_policy(vma)) && | ||
876 | can_vma_merge_after(near, vm_flags, | ||
877 | NULL, vma->vm_file, vma->vm_pgoff)) | ||
878 | return near->anon_vma; | ||
879 | none: | 909 | none: |
880 | /* | 910 | /* |
881 | * There's no absolute need to look only at touching neighbours: | 911 | * There's no absolute need to look only at touching neighbours: |
@@ -1947,7 +1977,8 @@ static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, | |||
1947 | return 0; | 1977 | return 0; |
1948 | 1978 | ||
1949 | /* Clean everything up if vma_adjust failed. */ | 1979 | /* Clean everything up if vma_adjust failed. */ |
1950 | new->vm_ops->close(new); | 1980 | if (new->vm_ops && new->vm_ops->close) |
1981 | new->vm_ops->close(new); | ||
1951 | if (new->vm_file) { | 1982 | if (new->vm_file) { |
1952 | if (vma->vm_flags & VM_EXECUTABLE) | 1983 | if (vma->vm_flags & VM_EXECUTABLE) |
1953 | removed_exe_file_vma(mm); | 1984 | removed_exe_file_vma(mm); |
diff --git a/mm/pagewalk.c b/mm/pagewalk.c index 7b47a57b6646..8b1a2ce21ee5 100644 --- a/mm/pagewalk.c +++ b/mm/pagewalk.c | |||
@@ -80,6 +80,37 @@ static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end, | |||
80 | return err; | 80 | return err; |
81 | } | 81 | } |
82 | 82 | ||
83 | #ifdef CONFIG_HUGETLB_PAGE | ||
84 | static unsigned long hugetlb_entry_end(struct hstate *h, unsigned long addr, | ||
85 | unsigned long end) | ||
86 | { | ||
87 | unsigned long boundary = (addr & huge_page_mask(h)) + huge_page_size(h); | ||
88 | return boundary < end ? boundary : end; | ||
89 | } | ||
90 | |||
91 | static int walk_hugetlb_range(struct vm_area_struct *vma, | ||
92 | unsigned long addr, unsigned long end, | ||
93 | struct mm_walk *walk) | ||
94 | { | ||
95 | struct hstate *h = hstate_vma(vma); | ||
96 | unsigned long next; | ||
97 | unsigned long hmask = huge_page_mask(h); | ||
98 | pte_t *pte; | ||
99 | int err = 0; | ||
100 | |||
101 | do { | ||
102 | next = hugetlb_entry_end(h, addr, end); | ||
103 | pte = huge_pte_offset(walk->mm, addr & hmask); | ||
104 | if (pte && walk->hugetlb_entry) | ||
105 | err = walk->hugetlb_entry(pte, hmask, addr, next, walk); | ||
106 | if (err) | ||
107 | return err; | ||
108 | } while (addr = next, addr != end); | ||
109 | |||
110 | return 0; | ||
111 | } | ||
112 | #endif | ||
113 | |||
83 | /** | 114 | /** |
84 | * walk_page_range - walk a memory map's page tables with a callback | 115 | * walk_page_range - walk a memory map's page tables with a callback |
85 | * @mm: memory map to walk | 116 | * @mm: memory map to walk |
@@ -128,20 +159,16 @@ int walk_page_range(unsigned long addr, unsigned long end, | |||
128 | vma = find_vma(walk->mm, addr); | 159 | vma = find_vma(walk->mm, addr); |
129 | #ifdef CONFIG_HUGETLB_PAGE | 160 | #ifdef CONFIG_HUGETLB_PAGE |
130 | if (vma && is_vm_hugetlb_page(vma)) { | 161 | if (vma && is_vm_hugetlb_page(vma)) { |
131 | pte_t *pte; | ||
132 | struct hstate *hs; | ||
133 | |||
134 | if (vma->vm_end < next) | 162 | if (vma->vm_end < next) |
135 | next = vma->vm_end; | 163 | next = vma->vm_end; |
136 | hs = hstate_vma(vma); | 164 | /* |
137 | pte = huge_pte_offset(walk->mm, | 165 | * Hugepage is very tightly coupled with vma, so |
138 | addr & huge_page_mask(hs)); | 166 | * walk through hugetlb entries within a given vma. |
139 | if (pte && !huge_pte_none(huge_ptep_get(pte)) | 167 | */ |
140 | && walk->hugetlb_entry) | 168 | err = walk_hugetlb_range(vma, addr, next, walk); |
141 | err = walk->hugetlb_entry(pte, addr, | ||
142 | next, walk); | ||
143 | if (err) | 169 | if (err) |
144 | break; | 170 | break; |
171 | pgd = pgd_offset(walk->mm, next); | ||
145 | continue; | 172 | continue; |
146 | } | 173 | } |
147 | #endif | 174 | #endif |
diff --git a/mm/readahead.c b/mm/readahead.c index 999b54bb462f..dfa9a1a03a11 100644 --- a/mm/readahead.c +++ b/mm/readahead.c | |||
@@ -503,7 +503,7 @@ void page_cache_sync_readahead(struct address_space *mapping, | |||
503 | return; | 503 | return; |
504 | 504 | ||
505 | /* be dumb */ | 505 | /* be dumb */ |
506 | if (filp->f_mode & FMODE_RANDOM) { | 506 | if (filp && (filp->f_mode & FMODE_RANDOM)) { |
507 | force_page_cache_readahead(mapping, filp, offset, req_size); | 507 | force_page_cache_readahead(mapping, filp, offset, req_size); |
508 | return; | 508 | return; |
509 | } | 509 | } |
@@ -133,8 +133,8 @@ int anon_vma_prepare(struct vm_area_struct *vma) | |||
133 | goto out_enomem_free_avc; | 133 | goto out_enomem_free_avc; |
134 | allocated = anon_vma; | 134 | allocated = anon_vma; |
135 | } | 135 | } |
136 | spin_lock(&anon_vma->lock); | ||
137 | 136 | ||
137 | spin_lock(&anon_vma->lock); | ||
138 | /* page_table_lock to protect against threads */ | 138 | /* page_table_lock to protect against threads */ |
139 | spin_lock(&mm->page_table_lock); | 139 | spin_lock(&mm->page_table_lock); |
140 | if (likely(!vma->anon_vma)) { | 140 | if (likely(!vma->anon_vma)) { |
@@ -144,14 +144,15 @@ int anon_vma_prepare(struct vm_area_struct *vma) | |||
144 | list_add(&avc->same_vma, &vma->anon_vma_chain); | 144 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
145 | list_add(&avc->same_anon_vma, &anon_vma->head); | 145 | list_add(&avc->same_anon_vma, &anon_vma->head); |
146 | allocated = NULL; | 146 | allocated = NULL; |
147 | avc = NULL; | ||
147 | } | 148 | } |
148 | spin_unlock(&mm->page_table_lock); | 149 | spin_unlock(&mm->page_table_lock); |
149 | |||
150 | spin_unlock(&anon_vma->lock); | 150 | spin_unlock(&anon_vma->lock); |
151 | if (unlikely(allocated)) { | 151 | |
152 | if (unlikely(allocated)) | ||
152 | anon_vma_free(allocated); | 153 | anon_vma_free(allocated); |
154 | if (unlikely(avc)) | ||
153 | anon_vma_chain_free(avc); | 155 | anon_vma_chain_free(avc); |
154 | } | ||
155 | } | 156 | } |
156 | return 0; | 157 | return 0; |
157 | 158 | ||
@@ -182,7 +183,7 @@ int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |||
182 | { | 183 | { |
183 | struct anon_vma_chain *avc, *pavc; | 184 | struct anon_vma_chain *avc, *pavc; |
184 | 185 | ||
185 | list_for_each_entry(pavc, &src->anon_vma_chain, same_vma) { | 186 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
186 | avc = anon_vma_chain_alloc(); | 187 | avc = anon_vma_chain_alloc(); |
187 | if (!avc) | 188 | if (!avc) |
188 | goto enomem_failure; | 189 | goto enomem_failure; |
@@ -730,13 +731,29 @@ void page_move_anon_rmap(struct page *page, | |||
730 | * @page: the page to add the mapping to | 731 | * @page: the page to add the mapping to |
731 | * @vma: the vm area in which the mapping is added | 732 | * @vma: the vm area in which the mapping is added |
732 | * @address: the user virtual address mapped | 733 | * @address: the user virtual address mapped |
734 | * @exclusive: the page is exclusively owned by the current process | ||
733 | */ | 735 | */ |
734 | static void __page_set_anon_rmap(struct page *page, | 736 | static void __page_set_anon_rmap(struct page *page, |
735 | struct vm_area_struct *vma, unsigned long address) | 737 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
736 | { | 738 | { |
737 | struct anon_vma *anon_vma = vma->anon_vma; | 739 | struct anon_vma *anon_vma = vma->anon_vma; |
738 | 740 | ||
739 | BUG_ON(!anon_vma); | 741 | BUG_ON(!anon_vma); |
742 | |||
743 | /* | ||
744 | * If the page isn't exclusively mapped into this vma, | ||
745 | * we must use the _oldest_ possible anon_vma for the | ||
746 | * page mapping! | ||
747 | * | ||
748 | * So take the last AVC chain entry in the vma, which is | ||
749 | * the deepest ancestor, and use the anon_vma from that. | ||
750 | */ | ||
751 | if (!exclusive) { | ||
752 | struct anon_vma_chain *avc; | ||
753 | avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma); | ||
754 | anon_vma = avc->anon_vma; | ||
755 | } | ||
756 | |||
740 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | 757 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
741 | page->mapping = (struct address_space *) anon_vma; | 758 | page->mapping = (struct address_space *) anon_vma; |
742 | page->index = linear_page_index(vma, address); | 759 | page->index = linear_page_index(vma, address); |
@@ -791,7 +808,7 @@ void page_add_anon_rmap(struct page *page, | |||
791 | VM_BUG_ON(!PageLocked(page)); | 808 | VM_BUG_ON(!PageLocked(page)); |
792 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | 809 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
793 | if (first) | 810 | if (first) |
794 | __page_set_anon_rmap(page, vma, address); | 811 | __page_set_anon_rmap(page, vma, address, 0); |
795 | else | 812 | else |
796 | __page_check_anon_rmap(page, vma, address); | 813 | __page_check_anon_rmap(page, vma, address); |
797 | } | 814 | } |
@@ -813,7 +830,7 @@ void page_add_new_anon_rmap(struct page *page, | |||
813 | SetPageSwapBacked(page); | 830 | SetPageSwapBacked(page); |
814 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | 831 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ |
815 | __inc_zone_page_state(page, NR_ANON_PAGES); | 832 | __inc_zone_page_state(page, NR_ANON_PAGES); |
816 | __page_set_anon_rmap(page, vma, address); | 833 | __page_set_anon_rmap(page, vma, address, 1); |
817 | if (page_evictable(page, vma)) | 834 | if (page_evictable(page, vma)) |
818 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); | 835 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
819 | else | 836 | else |
@@ -3602,21 +3602,10 @@ EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |||
3602 | */ | 3602 | */ |
3603 | int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) | 3603 | int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) |
3604 | { | 3604 | { |
3605 | unsigned long addr = (unsigned long)ptr; | ||
3606 | unsigned long min_addr = PAGE_OFFSET; | ||
3607 | unsigned long align_mask = BYTES_PER_WORD - 1; | ||
3608 | unsigned long size = cachep->buffer_size; | 3605 | unsigned long size = cachep->buffer_size; |
3609 | struct page *page; | 3606 | struct page *page; |
3610 | 3607 | ||
3611 | if (unlikely(addr < min_addr)) | 3608 | if (unlikely(!kern_ptr_validate(ptr, size))) |
3612 | goto out; | ||
3613 | if (unlikely(addr > (unsigned long)high_memory - size)) | ||
3614 | goto out; | ||
3615 | if (unlikely(addr & align_mask)) | ||
3616 | goto out; | ||
3617 | if (unlikely(!kern_addr_valid(addr))) | ||
3618 | goto out; | ||
3619 | if (unlikely(!kern_addr_valid(addr + size - 1))) | ||
3620 | goto out; | 3609 | goto out; |
3621 | page = virt_to_page(ptr); | 3610 | page = virt_to_page(ptr); |
3622 | if (unlikely(!PageSlab(page))) | 3611 | if (unlikely(!PageSlab(page))) |
@@ -2153,7 +2153,7 @@ static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |||
2153 | int local_node; | 2153 | int local_node; |
2154 | 2154 | ||
2155 | if (slab_state >= UP && (s < kmalloc_caches || | 2155 | if (slab_state >= UP && (s < kmalloc_caches || |
2156 | s > kmalloc_caches + KMALLOC_CACHES)) | 2156 | s >= kmalloc_caches + KMALLOC_CACHES)) |
2157 | local_node = page_to_nid(virt_to_page(s)); | 2157 | local_node = page_to_nid(virt_to_page(s)); |
2158 | else | 2158 | else |
2159 | local_node = 0; | 2159 | local_node = 0; |
@@ -2386,6 +2386,9 @@ int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |||
2386 | { | 2386 | { |
2387 | struct page *page; | 2387 | struct page *page; |
2388 | 2388 | ||
2389 | if (!kern_ptr_validate(object, s->size)) | ||
2390 | return 0; | ||
2391 | |||
2389 | page = get_object_page(object); | 2392 | page = get_object_page(object); |
2390 | 2393 | ||
2391 | if (!page || s != page->slab) | 2394 | if (!page || s != page->slab) |
@@ -186,6 +186,27 @@ void kzfree(const void *p) | |||
186 | } | 186 | } |
187 | EXPORT_SYMBOL(kzfree); | 187 | EXPORT_SYMBOL(kzfree); |
188 | 188 | ||
189 | int kern_ptr_validate(const void *ptr, unsigned long size) | ||
190 | { | ||
191 | unsigned long addr = (unsigned long)ptr; | ||
192 | unsigned long min_addr = PAGE_OFFSET; | ||
193 | unsigned long align_mask = sizeof(void *) - 1; | ||
194 | |||
195 | if (unlikely(addr < min_addr)) | ||
196 | goto out; | ||
197 | if (unlikely(addr > (unsigned long)high_memory - size)) | ||
198 | goto out; | ||
199 | if (unlikely(addr & align_mask)) | ||
200 | goto out; | ||
201 | if (unlikely(!kern_addr_valid(addr))) | ||
202 | goto out; | ||
203 | if (unlikely(!kern_addr_valid(addr + size - 1))) | ||
204 | goto out; | ||
205 | return 1; | ||
206 | out: | ||
207 | return 0; | ||
208 | } | ||
209 | |||
189 | /* | 210 | /* |
190 | * strndup_user - duplicate an existing string from user space | 211 | * strndup_user - duplicate an existing string from user space |
191 | * @s: The string to duplicate | 212 | * @s: The string to duplicate |
diff --git a/mm/vmscan.c b/mm/vmscan.c index e0e5f15bb726..3ff3311447f5 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c | |||
@@ -1535,13 +1535,6 @@ static void get_scan_ratio(struct zone *zone, struct scan_control *sc, | |||
1535 | unsigned long ap, fp; | 1535 | unsigned long ap, fp; |
1536 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); | 1536 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
1537 | 1537 | ||
1538 | /* If we have no swap space, do not bother scanning anon pages. */ | ||
1539 | if (!sc->may_swap || (nr_swap_pages <= 0)) { | ||
1540 | percent[0] = 0; | ||
1541 | percent[1] = 100; | ||
1542 | return; | ||
1543 | } | ||
1544 | |||
1545 | anon = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_ANON) + | 1538 | anon = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_ANON) + |
1546 | zone_nr_lru_pages(zone, sc, LRU_INACTIVE_ANON); | 1539 | zone_nr_lru_pages(zone, sc, LRU_INACTIVE_ANON); |
1547 | file = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_FILE) + | 1540 | file = zone_nr_lru_pages(zone, sc, LRU_ACTIVE_FILE) + |
@@ -1639,20 +1632,22 @@ static void shrink_zone(int priority, struct zone *zone, | |||
1639 | unsigned long nr_reclaimed = sc->nr_reclaimed; | 1632 | unsigned long nr_reclaimed = sc->nr_reclaimed; |
1640 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | 1633 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
1641 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); | 1634 | struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc); |
1635 | int noswap = 0; | ||
1642 | 1636 | ||
1643 | get_scan_ratio(zone, sc, percent); | 1637 | /* If we have no swap space, do not bother scanning anon pages. */ |
1638 | if (!sc->may_swap || (nr_swap_pages <= 0)) { | ||
1639 | noswap = 1; | ||
1640 | percent[0] = 0; | ||
1641 | percent[1] = 100; | ||
1642 | } else | ||
1643 | get_scan_ratio(zone, sc, percent); | ||
1644 | 1644 | ||
1645 | for_each_evictable_lru(l) { | 1645 | for_each_evictable_lru(l) { |
1646 | int file = is_file_lru(l); | 1646 | int file = is_file_lru(l); |
1647 | unsigned long scan; | 1647 | unsigned long scan; |
1648 | 1648 | ||
1649 | if (percent[file] == 0) { | ||
1650 | nr[l] = 0; | ||
1651 | continue; | ||
1652 | } | ||
1653 | |||
1654 | scan = zone_nr_lru_pages(zone, sc, l); | 1649 | scan = zone_nr_lru_pages(zone, sc, l); |
1655 | if (priority) { | 1650 | if (priority || noswap) { |
1656 | scan >>= priority; | 1651 | scan >>= priority; |
1657 | scan = (scan * percent[file]) / 100; | 1652 | scan = (scan * percent[file]) / 100; |
1658 | } | 1653 | } |