diff options
author | Rik van Riel <riel@redhat.com> | 2010-08-09 20:18:40 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2010-08-09 23:44:55 -0400 |
commit | 012f18004da33ba672e3c60838cc4898126174d3 (patch) | |
tree | 990382f9f8c0d885463ac9195b8e9a18043f716d /mm | |
parent | 5c341ee1dfc8fe69d66b1c8b19e463c6d7201ae1 (diff) |
mm: always lock the root (oldest) anon_vma
Always (and only) lock the root (oldest) anon_vma whenever we do something
in an anon_vma. The recently introduced anon_vma scalability is due to
the rmap code scanning only the VMAs that need to be scanned. Many common
operations still took the anon_vma lock on the root anon_vma, so always
taking that lock is not expected to introduce any scalability issues.
However, always taking the same lock does mean we only need to take one
lock, which means rmap_walk on pages from any anon_vma in the vma is
excluded from occurring during an munmap, expand_stack or other operation
that needs to exclude rmap_walk and similar functions.
Also add the proper locking to vma_adjust.
Signed-off-by: Rik van Riel <riel@redhat.com>
Tested-by: Larry Woodman <lwoodman@redhat.com>
Acked-by: Larry Woodman <lwoodman@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'mm')
-rw-r--r-- | mm/ksm.c | 2 | ||||
-rw-r--r-- | mm/migrate.c | 2 | ||||
-rw-r--r-- | mm/mmap.c | 30 |
3 files changed, 24 insertions, 10 deletions
@@ -325,7 +325,7 @@ static void drop_anon_vma(struct rmap_item *rmap_item) | |||
325 | { | 325 | { |
326 | struct anon_vma *anon_vma = rmap_item->anon_vma; | 326 | struct anon_vma *anon_vma = rmap_item->anon_vma; |
327 | 327 | ||
328 | if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->lock)) { | 328 | if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { |
329 | int empty = list_empty(&anon_vma->head); | 329 | int empty = list_empty(&anon_vma->head); |
330 | anon_vma_unlock(anon_vma); | 330 | anon_vma_unlock(anon_vma); |
331 | if (empty) | 331 | if (empty) |
diff --git a/mm/migrate.c b/mm/migrate.c index 1855f869917d..5208fa1d9712 100644 --- a/mm/migrate.c +++ b/mm/migrate.c | |||
@@ -682,7 +682,7 @@ skip_unmap: | |||
682 | rcu_unlock: | 682 | rcu_unlock: |
683 | 683 | ||
684 | /* Drop an anon_vma reference if we took one */ | 684 | /* Drop an anon_vma reference if we took one */ |
685 | if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->lock)) { | 685 | if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { |
686 | int empty = list_empty(&anon_vma->head); | 686 | int empty = list_empty(&anon_vma->head); |
687 | anon_vma_unlock(anon_vma); | 687 | anon_vma_unlock(anon_vma); |
688 | if (empty) | 688 | if (empty) |
@@ -506,6 +506,7 @@ int vma_adjust(struct vm_area_struct *vma, unsigned long start, | |||
506 | struct vm_area_struct *importer = NULL; | 506 | struct vm_area_struct *importer = NULL; |
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 anon_vma *anon_vma = NULL; | ||
509 | struct file *file = vma->vm_file; | 510 | struct file *file = vma->vm_file; |
510 | long adjust_next = 0; | 511 | long adjust_next = 0; |
511 | int remove_next = 0; | 512 | int remove_next = 0; |
@@ -578,6 +579,17 @@ again: remove_next = 1 + (end > next->vm_end); | |||
578 | } | 579 | } |
579 | } | 580 | } |
580 | 581 | ||
582 | /* | ||
583 | * When changing only vma->vm_end, we don't really need anon_vma | ||
584 | * lock. This is a fairly rare case by itself, but the anon_vma | ||
585 | * lock may be shared between many sibling processes. Skipping | ||
586 | * the lock for brk adjustments makes a difference sometimes. | ||
587 | */ | ||
588 | if (vma->anon_vma && (insert || importer || start != vma->vm_start)) { | ||
589 | anon_vma = vma->anon_vma; | ||
590 | anon_vma_lock(anon_vma); | ||
591 | } | ||
592 | |||
581 | if (root) { | 593 | if (root) { |
582 | flush_dcache_mmap_lock(mapping); | 594 | flush_dcache_mmap_lock(mapping); |
583 | vma_prio_tree_remove(vma, root); | 595 | vma_prio_tree_remove(vma, root); |
@@ -617,6 +629,8 @@ again: remove_next = 1 + (end > next->vm_end); | |||
617 | __insert_vm_struct(mm, insert); | 629 | __insert_vm_struct(mm, insert); |
618 | } | 630 | } |
619 | 631 | ||
632 | if (anon_vma) | ||
633 | anon_vma_unlock(anon_vma); | ||
620 | if (mapping) | 634 | if (mapping) |
621 | spin_unlock(&mapping->i_mmap_lock); | 635 | spin_unlock(&mapping->i_mmap_lock); |
622 | 636 | ||
@@ -2470,23 +2484,23 @@ static DEFINE_MUTEX(mm_all_locks_mutex); | |||
2470 | 2484 | ||
2471 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) | 2485 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
2472 | { | 2486 | { |
2473 | if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) { | 2487 | if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { |
2474 | /* | 2488 | /* |
2475 | * The LSB of head.next can't change from under us | 2489 | * The LSB of head.next can't change from under us |
2476 | * because we hold the mm_all_locks_mutex. | 2490 | * because we hold the mm_all_locks_mutex. |
2477 | */ | 2491 | */ |
2478 | spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem); | 2492 | spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem); |
2479 | /* | 2493 | /* |
2480 | * We can safely modify head.next after taking the | 2494 | * We can safely modify head.next after taking the |
2481 | * anon_vma->lock. If some other vma in this mm shares | 2495 | * anon_vma->root->lock. If some other vma in this mm shares |
2482 | * the same anon_vma we won't take it again. | 2496 | * the same anon_vma we won't take it again. |
2483 | * | 2497 | * |
2484 | * No need of atomic instructions here, head.next | 2498 | * No need of atomic instructions here, head.next |
2485 | * can't change from under us thanks to the | 2499 | * can't change from under us thanks to the |
2486 | * anon_vma->lock. | 2500 | * anon_vma->root->lock. |
2487 | */ | 2501 | */ |
2488 | if (__test_and_set_bit(0, (unsigned long *) | 2502 | if (__test_and_set_bit(0, (unsigned long *) |
2489 | &anon_vma->head.next)) | 2503 | &anon_vma->root->head.next)) |
2490 | BUG(); | 2504 | BUG(); |
2491 | } | 2505 | } |
2492 | } | 2506 | } |
@@ -2577,7 +2591,7 @@ out_unlock: | |||
2577 | 2591 | ||
2578 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | 2592 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) |
2579 | { | 2593 | { |
2580 | if (test_bit(0, (unsigned long *) &anon_vma->head.next)) { | 2594 | if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { |
2581 | /* | 2595 | /* |
2582 | * The LSB of head.next can't change to 0 from under | 2596 | * The LSB of head.next can't change to 0 from under |
2583 | * us because we hold the mm_all_locks_mutex. | 2597 | * us because we hold the mm_all_locks_mutex. |
@@ -2588,10 +2602,10 @@ static void vm_unlock_anon_vma(struct anon_vma *anon_vma) | |||
2588 | * | 2602 | * |
2589 | * No need of atomic instructions here, head.next | 2603 | * No need of atomic instructions here, head.next |
2590 | * can't change from under us until we release the | 2604 | * can't change from under us until we release the |
2591 | * anon_vma->lock. | 2605 | * anon_vma->root->lock. |
2592 | */ | 2606 | */ |
2593 | if (!__test_and_clear_bit(0, (unsigned long *) | 2607 | if (!__test_and_clear_bit(0, (unsigned long *) |
2594 | &anon_vma->head.next)) | 2608 | &anon_vma->root->head.next)) |
2595 | BUG(); | 2609 | BUG(); |
2596 | anon_vma_unlock(anon_vma); | 2610 | anon_vma_unlock(anon_vma); |
2597 | } | 2611 | } |