diff options
Diffstat (limited to 'mm')
-rw-r--r-- | mm/memory.c | 7 | ||||
-rw-r--r-- | mm/mempolicy.c | 167 | ||||
-rw-r--r-- | mm/oom_kill.c | 1 | ||||
-rw-r--r-- | mm/page_alloc.c | 6 | ||||
-rw-r--r-- | mm/rmap.c | 51 | ||||
-rw-r--r-- | mm/shmem.c | 8 | ||||
-rw-r--r-- | mm/slab.c | 639 | ||||
-rw-r--r-- | mm/swap_state.c | 1 | ||||
-rw-r--r-- | mm/swapfile.c | 16 | ||||
-rw-r--r-- | mm/vmscan.c | 343 |
10 files changed, 898 insertions, 341 deletions
diff --git a/mm/memory.c b/mm/memory.c index 7a11ddd5060f..2bee1f21aa8a 100644 --- a/mm/memory.c +++ b/mm/memory.c | |||
@@ -1871,6 +1871,7 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, | |||
1871 | goto out; | 1871 | goto out; |
1872 | 1872 | ||
1873 | entry = pte_to_swp_entry(orig_pte); | 1873 | entry = pte_to_swp_entry(orig_pte); |
1874 | again: | ||
1874 | page = lookup_swap_cache(entry); | 1875 | page = lookup_swap_cache(entry); |
1875 | if (!page) { | 1876 | if (!page) { |
1876 | swapin_readahead(entry, address, vma); | 1877 | swapin_readahead(entry, address, vma); |
@@ -1894,6 +1895,12 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, | |||
1894 | 1895 | ||
1895 | mark_page_accessed(page); | 1896 | mark_page_accessed(page); |
1896 | lock_page(page); | 1897 | lock_page(page); |
1898 | if (!PageSwapCache(page)) { | ||
1899 | /* Page migration has occured */ | ||
1900 | unlock_page(page); | ||
1901 | page_cache_release(page); | ||
1902 | goto again; | ||
1903 | } | ||
1897 | 1904 | ||
1898 | /* | 1905 | /* |
1899 | * Back out if somebody else already faulted in this pte. | 1906 | * Back out if somebody else already faulted in this pte. |
diff --git a/mm/mempolicy.c b/mm/mempolicy.c index 73790188b0eb..27da6d5c77ba 100644 --- a/mm/mempolicy.c +++ b/mm/mempolicy.c | |||
@@ -95,6 +95,9 @@ | |||
95 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ | 95 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
96 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ | 96 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
97 | 97 | ||
98 | /* The number of pages to migrate per call to migrate_pages() */ | ||
99 | #define MIGRATE_CHUNK_SIZE 256 | ||
100 | |||
98 | static kmem_cache_t *policy_cache; | 101 | static kmem_cache_t *policy_cache; |
99 | static kmem_cache_t *sn_cache; | 102 | static kmem_cache_t *sn_cache; |
100 | 103 | ||
@@ -543,24 +546,91 @@ static void migrate_page_add(struct page *page, struct list_head *pagelist, | |||
543 | } | 546 | } |
544 | } | 547 | } |
545 | 548 | ||
546 | static int swap_pages(struct list_head *pagelist) | 549 | /* |
550 | * Migrate the list 'pagelist' of pages to a certain destination. | ||
551 | * | ||
552 | * Specify destination with either non-NULL vma or dest_node >= 0 | ||
553 | * Return the number of pages not migrated or error code | ||
554 | */ | ||
555 | static int migrate_pages_to(struct list_head *pagelist, | ||
556 | struct vm_area_struct *vma, int dest) | ||
547 | { | 557 | { |
558 | LIST_HEAD(newlist); | ||
548 | LIST_HEAD(moved); | 559 | LIST_HEAD(moved); |
549 | LIST_HEAD(failed); | 560 | LIST_HEAD(failed); |
550 | int n; | 561 | int err = 0; |
562 | int nr_pages; | ||
563 | struct page *page; | ||
564 | struct list_head *p; | ||
551 | 565 | ||
552 | n = migrate_pages(pagelist, NULL, &moved, &failed); | 566 | redo: |
553 | putback_lru_pages(&failed); | 567 | nr_pages = 0; |
554 | putback_lru_pages(&moved); | 568 | list_for_each(p, pagelist) { |
569 | if (vma) | ||
570 | page = alloc_page_vma(GFP_HIGHUSER, vma, vma->vm_start); | ||
571 | else | ||
572 | page = alloc_pages_node(dest, GFP_HIGHUSER, 0); | ||
555 | 573 | ||
556 | return n; | 574 | if (!page) { |
575 | err = -ENOMEM; | ||
576 | goto out; | ||
577 | } | ||
578 | list_add(&page->lru, &newlist); | ||
579 | nr_pages++; | ||
580 | if (nr_pages > MIGRATE_CHUNK_SIZE); | ||
581 | break; | ||
582 | } | ||
583 | err = migrate_pages(pagelist, &newlist, &moved, &failed); | ||
584 | |||
585 | putback_lru_pages(&moved); /* Call release pages instead ?? */ | ||
586 | |||
587 | if (err >= 0 && list_empty(&newlist) && !list_empty(pagelist)) | ||
588 | goto redo; | ||
589 | out: | ||
590 | /* Return leftover allocated pages */ | ||
591 | while (!list_empty(&newlist)) { | ||
592 | page = list_entry(newlist.next, struct page, lru); | ||
593 | list_del(&page->lru); | ||
594 | __free_page(page); | ||
595 | } | ||
596 | list_splice(&failed, pagelist); | ||
597 | if (err < 0) | ||
598 | return err; | ||
599 | |||
600 | /* Calculate number of leftover pages */ | ||
601 | nr_pages = 0; | ||
602 | list_for_each(p, pagelist) | ||
603 | nr_pages++; | ||
604 | return nr_pages; | ||
605 | } | ||
606 | |||
607 | /* | ||
608 | * Migrate pages from one node to a target node. | ||
609 | * Returns error or the number of pages not migrated. | ||
610 | */ | ||
611 | int migrate_to_node(struct mm_struct *mm, int source, int dest, int flags) | ||
612 | { | ||
613 | nodemask_t nmask; | ||
614 | LIST_HEAD(pagelist); | ||
615 | int err = 0; | ||
616 | |||
617 | nodes_clear(nmask); | ||
618 | node_set(source, nmask); | ||
619 | |||
620 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, | ||
621 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | ||
622 | |||
623 | if (!list_empty(&pagelist)) { | ||
624 | err = migrate_pages_to(&pagelist, NULL, dest); | ||
625 | if (!list_empty(&pagelist)) | ||
626 | putback_lru_pages(&pagelist); | ||
627 | } | ||
628 | return err; | ||
557 | } | 629 | } |
558 | 630 | ||
559 | /* | 631 | /* |
560 | * For now migrate_pages simply swaps out the pages from nodes that are in | 632 | * Move pages between the two nodesets so as to preserve the physical |
561 | * the source set but not in the target set. In the future, we would | 633 | * layout as much as possible. |
562 | * want a function that moves pages between the two nodesets in such | ||
563 | * a way as to preserve the physical layout as much as possible. | ||
564 | * | 634 | * |
565 | * Returns the number of page that could not be moved. | 635 | * Returns the number of page that could not be moved. |
566 | */ | 636 | */ |
@@ -568,22 +638,76 @@ int do_migrate_pages(struct mm_struct *mm, | |||
568 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | 638 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) |
569 | { | 639 | { |
570 | LIST_HEAD(pagelist); | 640 | LIST_HEAD(pagelist); |
571 | int count = 0; | 641 | int busy = 0; |
572 | nodemask_t nodes; | 642 | int err = 0; |
643 | nodemask_t tmp; | ||
573 | 644 | ||
574 | nodes_andnot(nodes, *from_nodes, *to_nodes); | 645 | down_read(&mm->mmap_sem); |
575 | 646 | ||
576 | down_read(&mm->mmap_sem); | 647 | /* |
577 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nodes, | 648 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' |
578 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | 649 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' |
650 | * bit in 'tmp', and return that <source, dest> pair for migration. | ||
651 | * The pair of nodemasks 'to' and 'from' define the map. | ||
652 | * | ||
653 | * If no pair of bits is found that way, fallback to picking some | ||
654 | * pair of 'source' and 'dest' bits that are not the same. If the | ||
655 | * 'source' and 'dest' bits are the same, this represents a node | ||
656 | * that will be migrating to itself, so no pages need move. | ||
657 | * | ||
658 | * If no bits are left in 'tmp', or if all remaining bits left | ||
659 | * in 'tmp' correspond to the same bit in 'to', return false | ||
660 | * (nothing left to migrate). | ||
661 | * | ||
662 | * This lets us pick a pair of nodes to migrate between, such that | ||
663 | * if possible the dest node is not already occupied by some other | ||
664 | * source node, minimizing the risk of overloading the memory on a | ||
665 | * node that would happen if we migrated incoming memory to a node | ||
666 | * before migrating outgoing memory source that same node. | ||
667 | * | ||
668 | * A single scan of tmp is sufficient. As we go, we remember the | ||
669 | * most recent <s, d> pair that moved (s != d). If we find a pair | ||
670 | * that not only moved, but what's better, moved to an empty slot | ||
671 | * (d is not set in tmp), then we break out then, with that pair. | ||
672 | * Otherwise when we finish scannng from_tmp, we at least have the | ||
673 | * most recent <s, d> pair that moved. If we get all the way through | ||
674 | * the scan of tmp without finding any node that moved, much less | ||
675 | * moved to an empty node, then there is nothing left worth migrating. | ||
676 | */ | ||
579 | 677 | ||
580 | if (!list_empty(&pagelist)) { | 678 | tmp = *from_nodes; |
581 | count = swap_pages(&pagelist); | 679 | while (!nodes_empty(tmp)) { |
582 | putback_lru_pages(&pagelist); | 680 | int s,d; |
681 | int source = -1; | ||
682 | int dest = 0; | ||
683 | |||
684 | for_each_node_mask(s, tmp) { | ||
685 | d = node_remap(s, *from_nodes, *to_nodes); | ||
686 | if (s == d) | ||
687 | continue; | ||
688 | |||
689 | source = s; /* Node moved. Memorize */ | ||
690 | dest = d; | ||
691 | |||
692 | /* dest not in remaining from nodes? */ | ||
693 | if (!node_isset(dest, tmp)) | ||
694 | break; | ||
695 | } | ||
696 | if (source == -1) | ||
697 | break; | ||
698 | |||
699 | node_clear(source, tmp); | ||
700 | err = migrate_to_node(mm, source, dest, flags); | ||
701 | if (err > 0) | ||
702 | busy += err; | ||
703 | if (err < 0) | ||
704 | break; | ||
583 | } | 705 | } |
584 | 706 | ||
585 | up_read(&mm->mmap_sem); | 707 | up_read(&mm->mmap_sem); |
586 | return count; | 708 | if (err < 0) |
709 | return err; | ||
710 | return busy; | ||
587 | } | 711 | } |
588 | 712 | ||
589 | long do_mbind(unsigned long start, unsigned long len, | 713 | long do_mbind(unsigned long start, unsigned long len, |
@@ -643,8 +767,9 @@ long do_mbind(unsigned long start, unsigned long len, | |||
643 | int nr_failed = 0; | 767 | int nr_failed = 0; |
644 | 768 | ||
645 | err = mbind_range(vma, start, end, new); | 769 | err = mbind_range(vma, start, end, new); |
770 | |||
646 | if (!list_empty(&pagelist)) | 771 | if (!list_empty(&pagelist)) |
647 | nr_failed = swap_pages(&pagelist); | 772 | nr_failed = migrate_pages_to(&pagelist, vma, -1); |
648 | 773 | ||
649 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | 774 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) |
650 | err = -EIO; | 775 | err = -EIO; |
diff --git a/mm/oom_kill.c b/mm/oom_kill.c index 14bd4ec79597..b05ab8f2a562 100644 --- a/mm/oom_kill.c +++ b/mm/oom_kill.c | |||
@@ -271,6 +271,7 @@ void out_of_memory(gfp_t gfp_mask, int order) | |||
271 | if (printk_ratelimit()) { | 271 | if (printk_ratelimit()) { |
272 | printk("oom-killer: gfp_mask=0x%x, order=%d\n", | 272 | printk("oom-killer: gfp_mask=0x%x, order=%d\n", |
273 | gfp_mask, order); | 273 | gfp_mask, order); |
274 | dump_stack(); | ||
274 | show_mem(); | 275 | show_mem(); |
275 | } | 276 | } |
276 | 277 | ||
diff --git a/mm/page_alloc.c b/mm/page_alloc.c index df54e2fc8ee0..44b4eb4202d9 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c | |||
@@ -1799,7 +1799,7 @@ void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn, | |||
1799 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | 1799 | memmap_init_zone((size), (nid), (zone), (start_pfn)) |
1800 | #endif | 1800 | #endif |
1801 | 1801 | ||
1802 | static int __meminit zone_batchsize(struct zone *zone) | 1802 | static int __cpuinit zone_batchsize(struct zone *zone) |
1803 | { | 1803 | { |
1804 | int batch; | 1804 | int batch; |
1805 | 1805 | ||
@@ -1893,7 +1893,7 @@ static struct per_cpu_pageset | |||
1893 | * Dynamically allocate memory for the | 1893 | * Dynamically allocate memory for the |
1894 | * per cpu pageset array in struct zone. | 1894 | * per cpu pageset array in struct zone. |
1895 | */ | 1895 | */ |
1896 | static int __meminit process_zones(int cpu) | 1896 | static int __cpuinit process_zones(int cpu) |
1897 | { | 1897 | { |
1898 | struct zone *zone, *dzone; | 1898 | struct zone *zone, *dzone; |
1899 | 1899 | ||
@@ -1934,7 +1934,7 @@ static inline void free_zone_pagesets(int cpu) | |||
1934 | } | 1934 | } |
1935 | } | 1935 | } |
1936 | 1936 | ||
1937 | static int __meminit pageset_cpuup_callback(struct notifier_block *nfb, | 1937 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
1938 | unsigned long action, | 1938 | unsigned long action, |
1939 | void *hcpu) | 1939 | void *hcpu) |
1940 | { | 1940 | { |
@@ -52,6 +52,7 @@ | |||
52 | #include <linux/init.h> | 52 | #include <linux/init.h> |
53 | #include <linux/rmap.h> | 53 | #include <linux/rmap.h> |
54 | #include <linux/rcupdate.h> | 54 | #include <linux/rcupdate.h> |
55 | #include <linux/module.h> | ||
55 | 56 | ||
56 | #include <asm/tlbflush.h> | 57 | #include <asm/tlbflush.h> |
57 | 58 | ||
@@ -205,6 +206,36 @@ out: | |||
205 | return anon_vma; | 206 | return anon_vma; |
206 | } | 207 | } |
207 | 208 | ||
209 | #ifdef CONFIG_MIGRATION | ||
210 | /* | ||
211 | * Remove an anonymous page from swap replacing the swap pte's | ||
212 | * through real pte's pointing to valid pages and then releasing | ||
213 | * the page from the swap cache. | ||
214 | * | ||
215 | * Must hold page lock on page. | ||
216 | */ | ||
217 | void remove_from_swap(struct page *page) | ||
218 | { | ||
219 | struct anon_vma *anon_vma; | ||
220 | struct vm_area_struct *vma; | ||
221 | |||
222 | if (!PageAnon(page) || !PageSwapCache(page)) | ||
223 | return; | ||
224 | |||
225 | anon_vma = page_lock_anon_vma(page); | ||
226 | if (!anon_vma) | ||
227 | return; | ||
228 | |||
229 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) | ||
230 | remove_vma_swap(vma, page); | ||
231 | |||
232 | spin_unlock(&anon_vma->lock); | ||
233 | |||
234 | delete_from_swap_cache(page); | ||
235 | } | ||
236 | EXPORT_SYMBOL(remove_from_swap); | ||
237 | #endif | ||
238 | |||
208 | /* | 239 | /* |
209 | * At what user virtual address is page expected in vma? | 240 | * At what user virtual address is page expected in vma? |
210 | */ | 241 | */ |
@@ -541,7 +572,8 @@ void page_remove_rmap(struct page *page) | |||
541 | * Subfunctions of try_to_unmap: try_to_unmap_one called | 572 | * Subfunctions of try_to_unmap: try_to_unmap_one called |
542 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | 573 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. |
543 | */ | 574 | */ |
544 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma) | 575 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
576 | int ignore_refs) | ||
545 | { | 577 | { |
546 | struct mm_struct *mm = vma->vm_mm; | 578 | struct mm_struct *mm = vma->vm_mm; |
547 | unsigned long address; | 579 | unsigned long address; |
@@ -564,7 +596,8 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma) | |||
564 | * skipped over this mm) then we should reactivate it. | 596 | * skipped over this mm) then we should reactivate it. |
565 | */ | 597 | */ |
566 | if ((vma->vm_flags & VM_LOCKED) || | 598 | if ((vma->vm_flags & VM_LOCKED) || |
567 | ptep_clear_flush_young(vma, address, pte)) { | 599 | (ptep_clear_flush_young(vma, address, pte) |
600 | && !ignore_refs)) { | ||
568 | ret = SWAP_FAIL; | 601 | ret = SWAP_FAIL; |
569 | goto out_unmap; | 602 | goto out_unmap; |
570 | } | 603 | } |
@@ -698,7 +731,7 @@ static void try_to_unmap_cluster(unsigned long cursor, | |||
698 | pte_unmap_unlock(pte - 1, ptl); | 731 | pte_unmap_unlock(pte - 1, ptl); |
699 | } | 732 | } |
700 | 733 | ||
701 | static int try_to_unmap_anon(struct page *page) | 734 | static int try_to_unmap_anon(struct page *page, int ignore_refs) |
702 | { | 735 | { |
703 | struct anon_vma *anon_vma; | 736 | struct anon_vma *anon_vma; |
704 | struct vm_area_struct *vma; | 737 | struct vm_area_struct *vma; |
@@ -709,7 +742,7 @@ static int try_to_unmap_anon(struct page *page) | |||
709 | return ret; | 742 | return ret; |
710 | 743 | ||
711 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | 744 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { |
712 | ret = try_to_unmap_one(page, vma); | 745 | ret = try_to_unmap_one(page, vma, ignore_refs); |
713 | if (ret == SWAP_FAIL || !page_mapped(page)) | 746 | if (ret == SWAP_FAIL || !page_mapped(page)) |
714 | break; | 747 | break; |
715 | } | 748 | } |
@@ -726,7 +759,7 @@ static int try_to_unmap_anon(struct page *page) | |||
726 | * | 759 | * |
727 | * This function is only called from try_to_unmap for object-based pages. | 760 | * This function is only called from try_to_unmap for object-based pages. |
728 | */ | 761 | */ |
729 | static int try_to_unmap_file(struct page *page) | 762 | static int try_to_unmap_file(struct page *page, int ignore_refs) |
730 | { | 763 | { |
731 | struct address_space *mapping = page->mapping; | 764 | struct address_space *mapping = page->mapping; |
732 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | 765 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
@@ -740,7 +773,7 @@ static int try_to_unmap_file(struct page *page) | |||
740 | 773 | ||
741 | spin_lock(&mapping->i_mmap_lock); | 774 | spin_lock(&mapping->i_mmap_lock); |
742 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | 775 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
743 | ret = try_to_unmap_one(page, vma); | 776 | ret = try_to_unmap_one(page, vma, ignore_refs); |
744 | if (ret == SWAP_FAIL || !page_mapped(page)) | 777 | if (ret == SWAP_FAIL || !page_mapped(page)) |
745 | goto out; | 778 | goto out; |
746 | } | 779 | } |
@@ -825,16 +858,16 @@ out: | |||
825 | * SWAP_AGAIN - we missed a mapping, try again later | 858 | * SWAP_AGAIN - we missed a mapping, try again later |
826 | * SWAP_FAIL - the page is unswappable | 859 | * SWAP_FAIL - the page is unswappable |
827 | */ | 860 | */ |
828 | int try_to_unmap(struct page *page) | 861 | int try_to_unmap(struct page *page, int ignore_refs) |
829 | { | 862 | { |
830 | int ret; | 863 | int ret; |
831 | 864 | ||
832 | BUG_ON(!PageLocked(page)); | 865 | BUG_ON(!PageLocked(page)); |
833 | 866 | ||
834 | if (PageAnon(page)) | 867 | if (PageAnon(page)) |
835 | ret = try_to_unmap_anon(page); | 868 | ret = try_to_unmap_anon(page, ignore_refs); |
836 | else | 869 | else |
837 | ret = try_to_unmap_file(page); | 870 | ret = try_to_unmap_file(page, ignore_refs); |
838 | 871 | ||
839 | if (!page_mapped(page)) | 872 | if (!page_mapped(page)) |
840 | ret = SWAP_SUCCESS; | 873 | ret = SWAP_SUCCESS; |
diff --git a/mm/shmem.c b/mm/shmem.c index ce501bce1c2e..f7ac7b812f92 100644 --- a/mm/shmem.c +++ b/mm/shmem.c | |||
@@ -1028,6 +1028,14 @@ repeat: | |||
1028 | page_cache_release(swappage); | 1028 | page_cache_release(swappage); |
1029 | goto repeat; | 1029 | goto repeat; |
1030 | } | 1030 | } |
1031 | if (!PageSwapCache(swappage)) { | ||
1032 | /* Page migration has occured */ | ||
1033 | shmem_swp_unmap(entry); | ||
1034 | spin_unlock(&info->lock); | ||
1035 | unlock_page(swappage); | ||
1036 | page_cache_release(swappage); | ||
1037 | goto repeat; | ||
1038 | } | ||
1031 | if (PageWriteback(swappage)) { | 1039 | if (PageWriteback(swappage)) { |
1032 | shmem_swp_unmap(entry); | 1040 | shmem_swp_unmap(entry); |
1033 | spin_unlock(&info->lock); | 1041 | spin_unlock(&info->lock); |
@@ -55,7 +55,7 @@ | |||
55 | * | 55 | * |
56 | * SMP synchronization: | 56 | * SMP synchronization: |
57 | * constructors and destructors are called without any locking. | 57 | * constructors and destructors are called without any locking. |
58 | * Several members in kmem_cache_t and struct slab never change, they | 58 | * Several members in struct kmem_cache and struct slab never change, they |
59 | * are accessed without any locking. | 59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | 60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, |
61 | * and local interrupts are disabled so slab code is preempt-safe. | 61 | * and local interrupts are disabled so slab code is preempt-safe. |
@@ -244,7 +244,7 @@ struct slab { | |||
244 | */ | 244 | */ |
245 | struct slab_rcu { | 245 | struct slab_rcu { |
246 | struct rcu_head head; | 246 | struct rcu_head head; |
247 | kmem_cache_t *cachep; | 247 | struct kmem_cache *cachep; |
248 | void *addr; | 248 | void *addr; |
249 | }; | 249 | }; |
250 | 250 | ||
@@ -316,6 +316,8 @@ struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; | |||
316 | */ | 316 | */ |
317 | static __always_inline int index_of(const size_t size) | 317 | static __always_inline int index_of(const size_t size) |
318 | { | 318 | { |
319 | extern void __bad_size(void); | ||
320 | |||
319 | if (__builtin_constant_p(size)) { | 321 | if (__builtin_constant_p(size)) { |
320 | int i = 0; | 322 | int i = 0; |
321 | 323 | ||
@@ -326,19 +328,16 @@ static __always_inline int index_of(const size_t size) | |||
326 | i++; | 328 | i++; |
327 | #include "linux/kmalloc_sizes.h" | 329 | #include "linux/kmalloc_sizes.h" |
328 | #undef CACHE | 330 | #undef CACHE |
329 | { | 331 | __bad_size(); |
330 | extern void __bad_size(void); | ||
331 | __bad_size(); | ||
332 | } | ||
333 | } else | 332 | } else |
334 | BUG(); | 333 | __bad_size(); |
335 | return 0; | 334 | return 0; |
336 | } | 335 | } |
337 | 336 | ||
338 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) | 337 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) |
339 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) | 338 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) |
340 | 339 | ||
341 | static inline void kmem_list3_init(struct kmem_list3 *parent) | 340 | static void kmem_list3_init(struct kmem_list3 *parent) |
342 | { | 341 | { |
343 | INIT_LIST_HEAD(&parent->slabs_full); | 342 | INIT_LIST_HEAD(&parent->slabs_full); |
344 | INIT_LIST_HEAD(&parent->slabs_partial); | 343 | INIT_LIST_HEAD(&parent->slabs_partial); |
@@ -364,7 +363,7 @@ static inline void kmem_list3_init(struct kmem_list3 *parent) | |||
364 | } while (0) | 363 | } while (0) |
365 | 364 | ||
366 | /* | 365 | /* |
367 | * kmem_cache_t | 366 | * struct kmem_cache |
368 | * | 367 | * |
369 | * manages a cache. | 368 | * manages a cache. |
370 | */ | 369 | */ |
@@ -375,7 +374,7 @@ struct kmem_cache { | |||
375 | unsigned int batchcount; | 374 | unsigned int batchcount; |
376 | unsigned int limit; | 375 | unsigned int limit; |
377 | unsigned int shared; | 376 | unsigned int shared; |
378 | unsigned int objsize; | 377 | unsigned int buffer_size; |
379 | /* 2) touched by every alloc & free from the backend */ | 378 | /* 2) touched by every alloc & free from the backend */ |
380 | struct kmem_list3 *nodelists[MAX_NUMNODES]; | 379 | struct kmem_list3 *nodelists[MAX_NUMNODES]; |
381 | unsigned int flags; /* constant flags */ | 380 | unsigned int flags; /* constant flags */ |
@@ -392,15 +391,15 @@ struct kmem_cache { | |||
392 | size_t colour; /* cache colouring range */ | 391 | size_t colour; /* cache colouring range */ |
393 | unsigned int colour_off; /* colour offset */ | 392 | unsigned int colour_off; /* colour offset */ |
394 | unsigned int colour_next; /* cache colouring */ | 393 | unsigned int colour_next; /* cache colouring */ |
395 | kmem_cache_t *slabp_cache; | 394 | struct kmem_cache *slabp_cache; |
396 | unsigned int slab_size; | 395 | unsigned int slab_size; |
397 | unsigned int dflags; /* dynamic flags */ | 396 | unsigned int dflags; /* dynamic flags */ |
398 | 397 | ||
399 | /* constructor func */ | 398 | /* constructor func */ |
400 | void (*ctor) (void *, kmem_cache_t *, unsigned long); | 399 | void (*ctor) (void *, struct kmem_cache *, unsigned long); |
401 | 400 | ||
402 | /* de-constructor func */ | 401 | /* de-constructor func */ |
403 | void (*dtor) (void *, kmem_cache_t *, unsigned long); | 402 | void (*dtor) (void *, struct kmem_cache *, unsigned long); |
404 | 403 | ||
405 | /* 4) cache creation/removal */ | 404 | /* 4) cache creation/removal */ |
406 | const char *name; | 405 | const char *name; |
@@ -423,8 +422,14 @@ struct kmem_cache { | |||
423 | atomic_t freemiss; | 422 | atomic_t freemiss; |
424 | #endif | 423 | #endif |
425 | #if DEBUG | 424 | #if DEBUG |
426 | int dbghead; | 425 | /* |
427 | int reallen; | 426 | * If debugging is enabled, then the allocator can add additional |
427 | * fields and/or padding to every object. buffer_size contains the total | ||
428 | * object size including these internal fields, the following two | ||
429 | * variables contain the offset to the user object and its size. | ||
430 | */ | ||
431 | int obj_offset; | ||
432 | int obj_size; | ||
428 | #endif | 433 | #endif |
429 | }; | 434 | }; |
430 | 435 | ||
@@ -495,50 +500,50 @@ struct kmem_cache { | |||
495 | 500 | ||
496 | /* memory layout of objects: | 501 | /* memory layout of objects: |
497 | * 0 : objp | 502 | * 0 : objp |
498 | * 0 .. cachep->dbghead - BYTES_PER_WORD - 1: padding. This ensures that | 503 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
499 | * the end of an object is aligned with the end of the real | 504 | * the end of an object is aligned with the end of the real |
500 | * allocation. Catches writes behind the end of the allocation. | 505 | * allocation. Catches writes behind the end of the allocation. |
501 | * cachep->dbghead - BYTES_PER_WORD .. cachep->dbghead - 1: | 506 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
502 | * redzone word. | 507 | * redzone word. |
503 | * cachep->dbghead: The real object. | 508 | * cachep->obj_offset: The real object. |
504 | * cachep->objsize - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] | 509 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
505 | * cachep->objsize - 1* BYTES_PER_WORD: last caller address [BYTES_PER_WORD long] | 510 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address [BYTES_PER_WORD long] |
506 | */ | 511 | */ |
507 | static int obj_dbghead(kmem_cache_t *cachep) | 512 | static int obj_offset(struct kmem_cache *cachep) |
508 | { | 513 | { |
509 | return cachep->dbghead; | 514 | return cachep->obj_offset; |
510 | } | 515 | } |
511 | 516 | ||
512 | static int obj_reallen(kmem_cache_t *cachep) | 517 | static int obj_size(struct kmem_cache *cachep) |
513 | { | 518 | { |
514 | return cachep->reallen; | 519 | return cachep->obj_size; |
515 | } | 520 | } |
516 | 521 | ||
517 | static unsigned long *dbg_redzone1(kmem_cache_t *cachep, void *objp) | 522 | static unsigned long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
518 | { | 523 | { |
519 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | 524 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
520 | return (unsigned long*) (objp+obj_dbghead(cachep)-BYTES_PER_WORD); | 525 | return (unsigned long*) (objp+obj_offset(cachep)-BYTES_PER_WORD); |
521 | } | 526 | } |
522 | 527 | ||
523 | static unsigned long *dbg_redzone2(kmem_cache_t *cachep, void *objp) | 528 | static unsigned long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
524 | { | 529 | { |
525 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | 530 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); |
526 | if (cachep->flags & SLAB_STORE_USER) | 531 | if (cachep->flags & SLAB_STORE_USER) |
527 | return (unsigned long *)(objp + cachep->objsize - | 532 | return (unsigned long *)(objp + cachep->buffer_size - |
528 | 2 * BYTES_PER_WORD); | 533 | 2 * BYTES_PER_WORD); |
529 | return (unsigned long *)(objp + cachep->objsize - BYTES_PER_WORD); | 534 | return (unsigned long *)(objp + cachep->buffer_size - BYTES_PER_WORD); |
530 | } | 535 | } |
531 | 536 | ||
532 | static void **dbg_userword(kmem_cache_t *cachep, void *objp) | 537 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
533 | { | 538 | { |
534 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | 539 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); |
535 | return (void **)(objp + cachep->objsize - BYTES_PER_WORD); | 540 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
536 | } | 541 | } |
537 | 542 | ||
538 | #else | 543 | #else |
539 | 544 | ||
540 | #define obj_dbghead(x) 0 | 545 | #define obj_offset(x) 0 |
541 | #define obj_reallen(cachep) (cachep->objsize) | 546 | #define obj_size(cachep) (cachep->buffer_size) |
542 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long *)NULL;}) | 547 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long *)NULL;}) |
543 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long *)NULL;}) | 548 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long *)NULL;}) |
544 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) | 549 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
@@ -591,6 +596,18 @@ static inline struct slab *page_get_slab(struct page *page) | |||
591 | return (struct slab *)page->lru.prev; | 596 | return (struct slab *)page->lru.prev; |
592 | } | 597 | } |
593 | 598 | ||
599 | static inline struct kmem_cache *virt_to_cache(const void *obj) | ||
600 | { | ||
601 | struct page *page = virt_to_page(obj); | ||
602 | return page_get_cache(page); | ||
603 | } | ||
604 | |||
605 | static inline struct slab *virt_to_slab(const void *obj) | ||
606 | { | ||
607 | struct page *page = virt_to_page(obj); | ||
608 | return page_get_slab(page); | ||
609 | } | ||
610 | |||
594 | /* These are the default caches for kmalloc. Custom caches can have other sizes. */ | 611 | /* These are the default caches for kmalloc. Custom caches can have other sizes. */ |
595 | struct cache_sizes malloc_sizes[] = { | 612 | struct cache_sizes malloc_sizes[] = { |
596 | #define CACHE(x) { .cs_size = (x) }, | 613 | #define CACHE(x) { .cs_size = (x) }, |
@@ -619,16 +636,16 @@ static struct arraycache_init initarray_generic = | |||
619 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; | 636 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
620 | 637 | ||
621 | /* internal cache of cache description objs */ | 638 | /* internal cache of cache description objs */ |
622 | static kmem_cache_t cache_cache = { | 639 | static struct kmem_cache cache_cache = { |
623 | .batchcount = 1, | 640 | .batchcount = 1, |
624 | .limit = BOOT_CPUCACHE_ENTRIES, | 641 | .limit = BOOT_CPUCACHE_ENTRIES, |
625 | .shared = 1, | 642 | .shared = 1, |
626 | .objsize = sizeof(kmem_cache_t), | 643 | .buffer_size = sizeof(struct kmem_cache), |
627 | .flags = SLAB_NO_REAP, | 644 | .flags = SLAB_NO_REAP, |
628 | .spinlock = SPIN_LOCK_UNLOCKED, | 645 | .spinlock = SPIN_LOCK_UNLOCKED, |
629 | .name = "kmem_cache", | 646 | .name = "kmem_cache", |
630 | #if DEBUG | 647 | #if DEBUG |
631 | .reallen = sizeof(kmem_cache_t), | 648 | .obj_size = sizeof(struct kmem_cache), |
632 | #endif | 649 | #endif |
633 | }; | 650 | }; |
634 | 651 | ||
@@ -657,17 +674,17 @@ static enum { | |||
657 | 674 | ||
658 | static DEFINE_PER_CPU(struct work_struct, reap_work); | 675 | static DEFINE_PER_CPU(struct work_struct, reap_work); |
659 | 676 | ||
660 | static void free_block(kmem_cache_t *cachep, void **objpp, int len, int node); | 677 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, int node); |
661 | static void enable_cpucache(kmem_cache_t *cachep); | 678 | static void enable_cpucache(struct kmem_cache *cachep); |
662 | static void cache_reap(void *unused); | 679 | static void cache_reap(void *unused); |
663 | static int __node_shrink(kmem_cache_t *cachep, int node); | 680 | static int __node_shrink(struct kmem_cache *cachep, int node); |
664 | 681 | ||
665 | static inline struct array_cache *ac_data(kmem_cache_t *cachep) | 682 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
666 | { | 683 | { |
667 | return cachep->array[smp_processor_id()]; | 684 | return cachep->array[smp_processor_id()]; |
668 | } | 685 | } |
669 | 686 | ||
670 | static inline kmem_cache_t *__find_general_cachep(size_t size, gfp_t gfpflags) | 687 | static inline struct kmem_cache *__find_general_cachep(size_t size, gfp_t gfpflags) |
671 | { | 688 | { |
672 | struct cache_sizes *csizep = malloc_sizes; | 689 | struct cache_sizes *csizep = malloc_sizes; |
673 | 690 | ||
@@ -691,43 +708,80 @@ static inline kmem_cache_t *__find_general_cachep(size_t size, gfp_t gfpflags) | |||
691 | return csizep->cs_cachep; | 708 | return csizep->cs_cachep; |
692 | } | 709 | } |
693 | 710 | ||
694 | kmem_cache_t *kmem_find_general_cachep(size_t size, gfp_t gfpflags) | 711 | struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) |
695 | { | 712 | { |
696 | return __find_general_cachep(size, gfpflags); | 713 | return __find_general_cachep(size, gfpflags); |
697 | } | 714 | } |
698 | EXPORT_SYMBOL(kmem_find_general_cachep); | 715 | EXPORT_SYMBOL(kmem_find_general_cachep); |
699 | 716 | ||
700 | /* Cal the num objs, wastage, and bytes left over for a given slab size. */ | 717 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
701 | static void cache_estimate(unsigned long gfporder, size_t size, size_t align, | ||
702 | int flags, size_t *left_over, unsigned int *num) | ||
703 | { | 718 | { |
704 | int i; | 719 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
705 | size_t wastage = PAGE_SIZE << gfporder; | 720 | } |
706 | size_t extra = 0; | ||
707 | size_t base = 0; | ||
708 | 721 | ||
709 | if (!(flags & CFLGS_OFF_SLAB)) { | 722 | /* Calculate the number of objects and left-over bytes for a given |
710 | base = sizeof(struct slab); | 723 | buffer size. */ |
711 | extra = sizeof(kmem_bufctl_t); | 724 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
712 | } | 725 | size_t align, int flags, size_t *left_over, |
713 | i = 0; | 726 | unsigned int *num) |
714 | while (i * size + ALIGN(base + i * extra, align) <= wastage) | 727 | { |
715 | i++; | 728 | int nr_objs; |
716 | if (i > 0) | 729 | size_t mgmt_size; |
717 | i--; | 730 | size_t slab_size = PAGE_SIZE << gfporder; |
731 | |||
732 | /* | ||
733 | * The slab management structure can be either off the slab or | ||
734 | * on it. For the latter case, the memory allocated for a | ||
735 | * slab is used for: | ||
736 | * | ||
737 | * - The struct slab | ||
738 | * - One kmem_bufctl_t for each object | ||
739 | * - Padding to respect alignment of @align | ||
740 | * - @buffer_size bytes for each object | ||
741 | * | ||
742 | * If the slab management structure is off the slab, then the | ||
743 | * alignment will already be calculated into the size. Because | ||
744 | * the slabs are all pages aligned, the objects will be at the | ||
745 | * correct alignment when allocated. | ||
746 | */ | ||
747 | if (flags & CFLGS_OFF_SLAB) { | ||
748 | mgmt_size = 0; | ||
749 | nr_objs = slab_size / buffer_size; | ||
718 | 750 | ||
719 | if (i > SLAB_LIMIT) | 751 | if (nr_objs > SLAB_LIMIT) |
720 | i = SLAB_LIMIT; | 752 | nr_objs = SLAB_LIMIT; |
753 | } else { | ||
754 | /* | ||
755 | * Ignore padding for the initial guess. The padding | ||
756 | * is at most @align-1 bytes, and @buffer_size is at | ||
757 | * least @align. In the worst case, this result will | ||
758 | * be one greater than the number of objects that fit | ||
759 | * into the memory allocation when taking the padding | ||
760 | * into account. | ||
761 | */ | ||
762 | nr_objs = (slab_size - sizeof(struct slab)) / | ||
763 | (buffer_size + sizeof(kmem_bufctl_t)); | ||
764 | |||
765 | /* | ||
766 | * This calculated number will be either the right | ||
767 | * amount, or one greater than what we want. | ||
768 | */ | ||
769 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | ||
770 | > slab_size) | ||
771 | nr_objs--; | ||
721 | 772 | ||
722 | *num = i; | 773 | if (nr_objs > SLAB_LIMIT) |
723 | wastage -= i * size; | 774 | nr_objs = SLAB_LIMIT; |
724 | wastage -= ALIGN(base + i * extra, align); | 775 | |
725 | *left_over = wastage; | 776 | mgmt_size = slab_mgmt_size(nr_objs, align); |
777 | } | ||
778 | *num = nr_objs; | ||
779 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | ||
726 | } | 780 | } |
727 | 781 | ||
728 | #define slab_error(cachep, msg) __slab_error(__FUNCTION__, cachep, msg) | 782 | #define slab_error(cachep, msg) __slab_error(__FUNCTION__, cachep, msg) |
729 | 783 | ||
730 | static void __slab_error(const char *function, kmem_cache_t *cachep, char *msg) | 784 | static void __slab_error(const char *function, struct kmem_cache *cachep, char *msg) |
731 | { | 785 | { |
732 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | 786 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", |
733 | function, cachep->name, msg); | 787 | function, cachep->name, msg); |
@@ -774,9 +828,9 @@ static struct array_cache *alloc_arraycache(int node, int entries, | |||
774 | } | 828 | } |
775 | 829 | ||
776 | #ifdef CONFIG_NUMA | 830 | #ifdef CONFIG_NUMA |
777 | static void *__cache_alloc_node(kmem_cache_t *, gfp_t, int); | 831 | static void *__cache_alloc_node(struct kmem_cache *, gfp_t, int); |
778 | 832 | ||
779 | static inline struct array_cache **alloc_alien_cache(int node, int limit) | 833 | static struct array_cache **alloc_alien_cache(int node, int limit) |
780 | { | 834 | { |
781 | struct array_cache **ac_ptr; | 835 | struct array_cache **ac_ptr; |
782 | int memsize = sizeof(void *) * MAX_NUMNODES; | 836 | int memsize = sizeof(void *) * MAX_NUMNODES; |
@@ -803,7 +857,7 @@ static inline struct array_cache **alloc_alien_cache(int node, int limit) | |||
803 | return ac_ptr; | 857 | return ac_ptr; |
804 | } | 858 | } |
805 | 859 | ||
806 | static inline void free_alien_cache(struct array_cache **ac_ptr) | 860 | static void free_alien_cache(struct array_cache **ac_ptr) |
807 | { | 861 | { |
808 | int i; | 862 | int i; |
809 | 863 | ||
@@ -816,8 +870,8 @@ static inline void free_alien_cache(struct array_cache **ac_ptr) | |||
816 | kfree(ac_ptr); | 870 | kfree(ac_ptr); |
817 | } | 871 | } |
818 | 872 | ||
819 | static inline void __drain_alien_cache(kmem_cache_t *cachep, | 873 | static void __drain_alien_cache(struct kmem_cache *cachep, |
820 | struct array_cache *ac, int node) | 874 | struct array_cache *ac, int node) |
821 | { | 875 | { |
822 | struct kmem_list3 *rl3 = cachep->nodelists[node]; | 876 | struct kmem_list3 *rl3 = cachep->nodelists[node]; |
823 | 877 | ||
@@ -829,7 +883,7 @@ static inline void __drain_alien_cache(kmem_cache_t *cachep, | |||
829 | } | 883 | } |
830 | } | 884 | } |
831 | 885 | ||
832 | static void drain_alien_cache(kmem_cache_t *cachep, struct kmem_list3 *l3) | 886 | static void drain_alien_cache(struct kmem_cache *cachep, struct kmem_list3 *l3) |
833 | { | 887 | { |
834 | int i = 0; | 888 | int i = 0; |
835 | struct array_cache *ac; | 889 | struct array_cache *ac; |
@@ -854,7 +908,7 @@ static int __devinit cpuup_callback(struct notifier_block *nfb, | |||
854 | unsigned long action, void *hcpu) | 908 | unsigned long action, void *hcpu) |
855 | { | 909 | { |
856 | long cpu = (long)hcpu; | 910 | long cpu = (long)hcpu; |
857 | kmem_cache_t *cachep; | 911 | struct kmem_cache *cachep; |
858 | struct kmem_list3 *l3 = NULL; | 912 | struct kmem_list3 *l3 = NULL; |
859 | int node = cpu_to_node(cpu); | 913 | int node = cpu_to_node(cpu); |
860 | int memsize = sizeof(struct kmem_list3); | 914 | int memsize = sizeof(struct kmem_list3); |
@@ -992,7 +1046,7 @@ static struct notifier_block cpucache_notifier = { &cpuup_callback, NULL, 0 }; | |||
992 | /* | 1046 | /* |
993 | * swap the static kmem_list3 with kmalloced memory | 1047 | * swap the static kmem_list3 with kmalloced memory |
994 | */ | 1048 | */ |
995 | static void init_list(kmem_cache_t *cachep, struct kmem_list3 *list, int nodeid) | 1049 | static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, int nodeid) |
996 | { | 1050 | { |
997 | struct kmem_list3 *ptr; | 1051 | struct kmem_list3 *ptr; |
998 | 1052 | ||
@@ -1032,14 +1086,14 @@ void __init kmem_cache_init(void) | |||
1032 | 1086 | ||
1033 | /* Bootstrap is tricky, because several objects are allocated | 1087 | /* Bootstrap is tricky, because several objects are allocated |
1034 | * from caches that do not exist yet: | 1088 | * from caches that do not exist yet: |
1035 | * 1) initialize the cache_cache cache: it contains the kmem_cache_t | 1089 | * 1) initialize the cache_cache cache: it contains the struct kmem_cache |
1036 | * structures of all caches, except cache_cache itself: cache_cache | 1090 | * structures of all caches, except cache_cache itself: cache_cache |
1037 | * is statically allocated. | 1091 | * is statically allocated. |
1038 | * Initially an __init data area is used for the head array and the | 1092 | * Initially an __init data area is used for the head array and the |
1039 | * kmem_list3 structures, it's replaced with a kmalloc allocated | 1093 | * kmem_list3 structures, it's replaced with a kmalloc allocated |
1040 | * array at the end of the bootstrap. | 1094 | * array at the end of the bootstrap. |
1041 | * 2) Create the first kmalloc cache. | 1095 | * 2) Create the first kmalloc cache. |
1042 | * The kmem_cache_t for the new cache is allocated normally. | 1096 | * The struct kmem_cache for the new cache is allocated normally. |
1043 | * An __init data area is used for the head array. | 1097 | * An __init data area is used for the head array. |
1044 | * 3) Create the remaining kmalloc caches, with minimally sized | 1098 | * 3) Create the remaining kmalloc caches, with minimally sized |
1045 | * head arrays. | 1099 | * head arrays. |
@@ -1057,9 +1111,9 @@ void __init kmem_cache_init(void) | |||
1057 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; | 1111 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; |
1058 | cache_cache.nodelists[numa_node_id()] = &initkmem_list3[CACHE_CACHE]; | 1112 | cache_cache.nodelists[numa_node_id()] = &initkmem_list3[CACHE_CACHE]; |
1059 | 1113 | ||
1060 | cache_cache.objsize = ALIGN(cache_cache.objsize, cache_line_size()); | 1114 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, cache_line_size()); |
1061 | 1115 | ||
1062 | cache_estimate(0, cache_cache.objsize, cache_line_size(), 0, | 1116 | cache_estimate(0, cache_cache.buffer_size, cache_line_size(), 0, |
1063 | &left_over, &cache_cache.num); | 1117 | &left_over, &cache_cache.num); |
1064 | if (!cache_cache.num) | 1118 | if (!cache_cache.num) |
1065 | BUG(); | 1119 | BUG(); |
@@ -1132,8 +1186,8 @@ void __init kmem_cache_init(void) | |||
1132 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); | 1186 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
1133 | 1187 | ||
1134 | local_irq_disable(); | 1188 | local_irq_disable(); |
1135 | BUG_ON(ac_data(&cache_cache) != &initarray_cache.cache); | 1189 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
1136 | memcpy(ptr, ac_data(&cache_cache), | 1190 | memcpy(ptr, cpu_cache_get(&cache_cache), |
1137 | sizeof(struct arraycache_init)); | 1191 | sizeof(struct arraycache_init)); |
1138 | cache_cache.array[smp_processor_id()] = ptr; | 1192 | cache_cache.array[smp_processor_id()] = ptr; |
1139 | local_irq_enable(); | 1193 | local_irq_enable(); |
@@ -1141,9 +1195,9 @@ void __init kmem_cache_init(void) | |||
1141 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); | 1195 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
1142 | 1196 | ||
1143 | local_irq_disable(); | 1197 | local_irq_disable(); |
1144 | BUG_ON(ac_data(malloc_sizes[INDEX_AC].cs_cachep) | 1198 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
1145 | != &initarray_generic.cache); | 1199 | != &initarray_generic.cache); |
1146 | memcpy(ptr, ac_data(malloc_sizes[INDEX_AC].cs_cachep), | 1200 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
1147 | sizeof(struct arraycache_init)); | 1201 | sizeof(struct arraycache_init)); |
1148 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = | 1202 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
1149 | ptr; | 1203 | ptr; |
@@ -1170,7 +1224,7 @@ void __init kmem_cache_init(void) | |||
1170 | 1224 | ||
1171 | /* 6) resize the head arrays to their final sizes */ | 1225 | /* 6) resize the head arrays to their final sizes */ |
1172 | { | 1226 | { |
1173 | kmem_cache_t *cachep; | 1227 | struct kmem_cache *cachep; |
1174 | mutex_lock(&cache_chain_mutex); | 1228 | mutex_lock(&cache_chain_mutex); |
1175 | list_for_each_entry(cachep, &cache_chain, next) | 1229 | list_for_each_entry(cachep, &cache_chain, next) |
1176 | enable_cpucache(cachep); | 1230 | enable_cpucache(cachep); |
@@ -1181,7 +1235,7 @@ void __init kmem_cache_init(void) | |||
1181 | g_cpucache_up = FULL; | 1235 | g_cpucache_up = FULL; |
1182 | 1236 | ||
1183 | /* Register a cpu startup notifier callback | 1237 | /* Register a cpu startup notifier callback |
1184 | * that initializes ac_data for all new cpus | 1238 | * that initializes cpu_cache_get for all new cpus |
1185 | */ | 1239 | */ |
1186 | register_cpu_notifier(&cpucache_notifier); | 1240 | register_cpu_notifier(&cpucache_notifier); |
1187 | 1241 | ||
@@ -1213,7 +1267,7 @@ __initcall(cpucache_init); | |||
1213 | * did not request dmaable memory, we might get it, but that | 1267 | * did not request dmaable memory, we might get it, but that |
1214 | * would be relatively rare and ignorable. | 1268 | * would be relatively rare and ignorable. |
1215 | */ | 1269 | */ |
1216 | static void *kmem_getpages(kmem_cache_t *cachep, gfp_t flags, int nodeid) | 1270 | static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1217 | { | 1271 | { |
1218 | struct page *page; | 1272 | struct page *page; |
1219 | void *addr; | 1273 | void *addr; |
@@ -1239,7 +1293,7 @@ static void *kmem_getpages(kmem_cache_t *cachep, gfp_t flags, int nodeid) | |||
1239 | /* | 1293 | /* |
1240 | * Interface to system's page release. | 1294 | * Interface to system's page release. |
1241 | */ | 1295 | */ |
1242 | static void kmem_freepages(kmem_cache_t *cachep, void *addr) | 1296 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
1243 | { | 1297 | { |
1244 | unsigned long i = (1 << cachep->gfporder); | 1298 | unsigned long i = (1 << cachep->gfporder); |
1245 | struct page *page = virt_to_page(addr); | 1299 | struct page *page = virt_to_page(addr); |
@@ -1261,7 +1315,7 @@ static void kmem_freepages(kmem_cache_t *cachep, void *addr) | |||
1261 | static void kmem_rcu_free(struct rcu_head *head) | 1315 | static void kmem_rcu_free(struct rcu_head *head) |
1262 | { | 1316 | { |
1263 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; | 1317 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
1264 | kmem_cache_t *cachep = slab_rcu->cachep; | 1318 | struct kmem_cache *cachep = slab_rcu->cachep; |
1265 | 1319 | ||
1266 | kmem_freepages(cachep, slab_rcu->addr); | 1320 | kmem_freepages(cachep, slab_rcu->addr); |
1267 | if (OFF_SLAB(cachep)) | 1321 | if (OFF_SLAB(cachep)) |
@@ -1271,12 +1325,12 @@ static void kmem_rcu_free(struct rcu_head *head) | |||
1271 | #if DEBUG | 1325 | #if DEBUG |
1272 | 1326 | ||
1273 | #ifdef CONFIG_DEBUG_PAGEALLOC | 1327 | #ifdef CONFIG_DEBUG_PAGEALLOC |
1274 | static void store_stackinfo(kmem_cache_t *cachep, unsigned long *addr, | 1328 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
1275 | unsigned long caller) | 1329 | unsigned long caller) |
1276 | { | 1330 | { |
1277 | int size = obj_reallen(cachep); | 1331 | int size = obj_size(cachep); |
1278 | 1332 | ||
1279 | addr = (unsigned long *)&((char *)addr)[obj_dbghead(cachep)]; | 1333 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1280 | 1334 | ||
1281 | if (size < 5 * sizeof(unsigned long)) | 1335 | if (size < 5 * sizeof(unsigned long)) |
1282 | return; | 1336 | return; |
@@ -1304,10 +1358,10 @@ static void store_stackinfo(kmem_cache_t *cachep, unsigned long *addr, | |||
1304 | } | 1358 | } |
1305 | #endif | 1359 | #endif |
1306 | 1360 | ||
1307 | static void poison_obj(kmem_cache_t *cachep, void *addr, unsigned char val) | 1361 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1308 | { | 1362 | { |
1309 | int size = obj_reallen(cachep); | 1363 | int size = obj_size(cachep); |
1310 | addr = &((char *)addr)[obj_dbghead(cachep)]; | 1364 | addr = &((char *)addr)[obj_offset(cachep)]; |
1311 | 1365 | ||
1312 | memset(addr, val, size); | 1366 | memset(addr, val, size); |
1313 | *(unsigned char *)(addr + size - 1) = POISON_END; | 1367 | *(unsigned char *)(addr + size - 1) = POISON_END; |
@@ -1326,7 +1380,7 @@ static void dump_line(char *data, int offset, int limit) | |||
1326 | 1380 | ||
1327 | #if DEBUG | 1381 | #if DEBUG |
1328 | 1382 | ||
1329 | static void print_objinfo(kmem_cache_t *cachep, void *objp, int lines) | 1383 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1330 | { | 1384 | { |
1331 | int i, size; | 1385 | int i, size; |
1332 | char *realobj; | 1386 | char *realobj; |
@@ -1344,8 +1398,8 @@ static void print_objinfo(kmem_cache_t *cachep, void *objp, int lines) | |||
1344 | (unsigned long)*dbg_userword(cachep, objp)); | 1398 | (unsigned long)*dbg_userword(cachep, objp)); |
1345 | printk("\n"); | 1399 | printk("\n"); |
1346 | } | 1400 | } |
1347 | realobj = (char *)objp + obj_dbghead(cachep); | 1401 | realobj = (char *)objp + obj_offset(cachep); |
1348 | size = obj_reallen(cachep); | 1402 | size = obj_size(cachep); |
1349 | for (i = 0; i < size && lines; i += 16, lines--) { | 1403 | for (i = 0; i < size && lines; i += 16, lines--) { |
1350 | int limit; | 1404 | int limit; |
1351 | limit = 16; | 1405 | limit = 16; |
@@ -1355,14 +1409,14 @@ static void print_objinfo(kmem_cache_t *cachep, void *objp, int lines) | |||
1355 | } | 1409 | } |
1356 | } | 1410 | } |
1357 | 1411 | ||
1358 | static void check_poison_obj(kmem_cache_t *cachep, void *objp) | 1412 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1359 | { | 1413 | { |
1360 | char *realobj; | 1414 | char *realobj; |
1361 | int size, i; | 1415 | int size, i; |
1362 | int lines = 0; | 1416 | int lines = 0; |
1363 | 1417 | ||
1364 | realobj = (char *)objp + obj_dbghead(cachep); | 1418 | realobj = (char *)objp + obj_offset(cachep); |
1365 | size = obj_reallen(cachep); | 1419 | size = obj_size(cachep); |
1366 | 1420 | ||
1367 | for (i = 0; i < size; i++) { | 1421 | for (i = 0; i < size; i++) { |
1368 | char exp = POISON_FREE; | 1422 | char exp = POISON_FREE; |
@@ -1395,20 +1449,20 @@ static void check_poison_obj(kmem_cache_t *cachep, void *objp) | |||
1395 | /* Print some data about the neighboring objects, if they | 1449 | /* Print some data about the neighboring objects, if they |
1396 | * exist: | 1450 | * exist: |
1397 | */ | 1451 | */ |
1398 | struct slab *slabp = page_get_slab(virt_to_page(objp)); | 1452 | struct slab *slabp = virt_to_slab(objp); |
1399 | int objnr; | 1453 | int objnr; |
1400 | 1454 | ||
1401 | objnr = (objp - slabp->s_mem) / cachep->objsize; | 1455 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
1402 | if (objnr) { | 1456 | if (objnr) { |
1403 | objp = slabp->s_mem + (objnr - 1) * cachep->objsize; | 1457 | objp = slabp->s_mem + (objnr - 1) * cachep->buffer_size; |
1404 | realobj = (char *)objp + obj_dbghead(cachep); | 1458 | realobj = (char *)objp + obj_offset(cachep); |
1405 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", | 1459 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
1406 | realobj, size); | 1460 | realobj, size); |
1407 | print_objinfo(cachep, objp, 2); | 1461 | print_objinfo(cachep, objp, 2); |
1408 | } | 1462 | } |
1409 | if (objnr + 1 < cachep->num) { | 1463 | if (objnr + 1 < cachep->num) { |
1410 | objp = slabp->s_mem + (objnr + 1) * cachep->objsize; | 1464 | objp = slabp->s_mem + (objnr + 1) * cachep->buffer_size; |
1411 | realobj = (char *)objp + obj_dbghead(cachep); | 1465 | realobj = (char *)objp + obj_offset(cachep); |
1412 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", | 1466 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
1413 | realobj, size); | 1467 | realobj, size); |
1414 | print_objinfo(cachep, objp, 2); | 1468 | print_objinfo(cachep, objp, 2); |
@@ -1417,25 +1471,23 @@ static void check_poison_obj(kmem_cache_t *cachep, void *objp) | |||
1417 | } | 1471 | } |
1418 | #endif | 1472 | #endif |
1419 | 1473 | ||
1420 | /* Destroy all the objs in a slab, and release the mem back to the system. | 1474 | #if DEBUG |
1421 | * Before calling the slab must have been unlinked from the cache. | 1475 | /** |
1422 | * The cache-lock is not held/needed. | 1476 | * slab_destroy_objs - call the registered destructor for each object in |
1477 | * a slab that is to be destroyed. | ||
1423 | */ | 1478 | */ |
1424 | static void slab_destroy(kmem_cache_t *cachep, struct slab *slabp) | 1479 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
1425 | { | 1480 | { |
1426 | void *addr = slabp->s_mem - slabp->colouroff; | ||
1427 | |||
1428 | #if DEBUG | ||
1429 | int i; | 1481 | int i; |
1430 | for (i = 0; i < cachep->num; i++) { | 1482 | for (i = 0; i < cachep->num; i++) { |
1431 | void *objp = slabp->s_mem + cachep->objsize * i; | 1483 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
1432 | 1484 | ||
1433 | if (cachep->flags & SLAB_POISON) { | 1485 | if (cachep->flags & SLAB_POISON) { |
1434 | #ifdef CONFIG_DEBUG_PAGEALLOC | 1486 | #ifdef CONFIG_DEBUG_PAGEALLOC |
1435 | if ((cachep->objsize % PAGE_SIZE) == 0 | 1487 | if ((cachep->buffer_size % PAGE_SIZE) == 0 |
1436 | && OFF_SLAB(cachep)) | 1488 | && OFF_SLAB(cachep)) |
1437 | kernel_map_pages(virt_to_page(objp), | 1489 | kernel_map_pages(virt_to_page(objp), |
1438 | cachep->objsize / PAGE_SIZE, | 1490 | cachep->buffer_size / PAGE_SIZE, |
1439 | 1); | 1491 | 1); |
1440 | else | 1492 | else |
1441 | check_poison_obj(cachep, objp); | 1493 | check_poison_obj(cachep, objp); |
@@ -1452,18 +1504,32 @@ static void slab_destroy(kmem_cache_t *cachep, struct slab *slabp) | |||
1452 | "was overwritten"); | 1504 | "was overwritten"); |
1453 | } | 1505 | } |
1454 | if (cachep->dtor && !(cachep->flags & SLAB_POISON)) | 1506 | if (cachep->dtor && !(cachep->flags & SLAB_POISON)) |
1455 | (cachep->dtor) (objp + obj_dbghead(cachep), cachep, 0); | 1507 | (cachep->dtor) (objp + obj_offset(cachep), cachep, 0); |
1456 | } | 1508 | } |
1509 | } | ||
1457 | #else | 1510 | #else |
1511 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) | ||
1512 | { | ||
1458 | if (cachep->dtor) { | 1513 | if (cachep->dtor) { |
1459 | int i; | 1514 | int i; |
1460 | for (i = 0; i < cachep->num; i++) { | 1515 | for (i = 0; i < cachep->num; i++) { |
1461 | void *objp = slabp->s_mem + cachep->objsize * i; | 1516 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
1462 | (cachep->dtor) (objp, cachep, 0); | 1517 | (cachep->dtor) (objp, cachep, 0); |
1463 | } | 1518 | } |
1464 | } | 1519 | } |
1520 | } | ||
1465 | #endif | 1521 | #endif |
1466 | 1522 | ||
1523 | /** | ||
1524 | * Destroy all the objs in a slab, and release the mem back to the system. | ||
1525 | * Before calling the slab must have been unlinked from the cache. | ||
1526 | * The cache-lock is not held/needed. | ||
1527 | */ | ||
1528 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) | ||
1529 | { | ||
1530 | void *addr = slabp->s_mem - slabp->colouroff; | ||
1531 | |||
1532 | slab_destroy_objs(cachep, slabp); | ||
1467 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { | 1533 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
1468 | struct slab_rcu *slab_rcu; | 1534 | struct slab_rcu *slab_rcu; |
1469 | 1535 | ||
@@ -1478,9 +1544,9 @@ static void slab_destroy(kmem_cache_t *cachep, struct slab *slabp) | |||
1478 | } | 1544 | } |
1479 | } | 1545 | } |
1480 | 1546 | ||
1481 | /* For setting up all the kmem_list3s for cache whose objsize is same | 1547 | /* For setting up all the kmem_list3s for cache whose buffer_size is same |
1482 | as size of kmem_list3. */ | 1548 | as size of kmem_list3. */ |
1483 | static inline void set_up_list3s(kmem_cache_t *cachep, int index) | 1549 | static void set_up_list3s(struct kmem_cache *cachep, int index) |
1484 | { | 1550 | { |
1485 | int node; | 1551 | int node; |
1486 | 1552 | ||
@@ -1493,15 +1559,20 @@ static inline void set_up_list3s(kmem_cache_t *cachep, int index) | |||
1493 | } | 1559 | } |
1494 | 1560 | ||
1495 | /** | 1561 | /** |
1496 | * calculate_slab_order - calculate size (page order) of slabs and the number | 1562 | * calculate_slab_order - calculate size (page order) of slabs |
1497 | * of objects per slab. | 1563 | * @cachep: pointer to the cache that is being created |
1564 | * @size: size of objects to be created in this cache. | ||
1565 | * @align: required alignment for the objects. | ||
1566 | * @flags: slab allocation flags | ||
1567 | * | ||
1568 | * Also calculates the number of objects per slab. | ||
1498 | * | 1569 | * |
1499 | * This could be made much more intelligent. For now, try to avoid using | 1570 | * This could be made much more intelligent. For now, try to avoid using |
1500 | * high order pages for slabs. When the gfp() functions are more friendly | 1571 | * high order pages for slabs. When the gfp() functions are more friendly |
1501 | * towards high-order requests, this should be changed. | 1572 | * towards high-order requests, this should be changed. |
1502 | */ | 1573 | */ |
1503 | static inline size_t calculate_slab_order(kmem_cache_t *cachep, size_t size, | 1574 | static inline size_t calculate_slab_order(struct kmem_cache *cachep, |
1504 | size_t align, gfp_t flags) | 1575 | size_t size, size_t align, unsigned long flags) |
1505 | { | 1576 | { |
1506 | size_t left_over = 0; | 1577 | size_t left_over = 0; |
1507 | 1578 | ||
@@ -1572,13 +1643,13 @@ static inline size_t calculate_slab_order(kmem_cache_t *cachep, size_t size, | |||
1572 | * cacheline. This can be beneficial if you're counting cycles as closely | 1643 | * cacheline. This can be beneficial if you're counting cycles as closely |
1573 | * as davem. | 1644 | * as davem. |
1574 | */ | 1645 | */ |
1575 | kmem_cache_t * | 1646 | struct kmem_cache * |
1576 | kmem_cache_create (const char *name, size_t size, size_t align, | 1647 | kmem_cache_create (const char *name, size_t size, size_t align, |
1577 | unsigned long flags, void (*ctor)(void*, kmem_cache_t *, unsigned long), | 1648 | unsigned long flags, void (*ctor)(void*, struct kmem_cache *, unsigned long), |
1578 | void (*dtor)(void*, kmem_cache_t *, unsigned long)) | 1649 | void (*dtor)(void*, struct kmem_cache *, unsigned long)) |
1579 | { | 1650 | { |
1580 | size_t left_over, slab_size, ralign; | 1651 | size_t left_over, slab_size, ralign; |
1581 | kmem_cache_t *cachep = NULL; | 1652 | struct kmem_cache *cachep = NULL; |
1582 | struct list_head *p; | 1653 | struct list_head *p; |
1583 | 1654 | ||
1584 | /* | 1655 | /* |
@@ -1596,7 +1667,7 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1596 | mutex_lock(&cache_chain_mutex); | 1667 | mutex_lock(&cache_chain_mutex); |
1597 | 1668 | ||
1598 | list_for_each(p, &cache_chain) { | 1669 | list_for_each(p, &cache_chain) { |
1599 | kmem_cache_t *pc = list_entry(p, kmem_cache_t, next); | 1670 | struct kmem_cache *pc = list_entry(p, struct kmem_cache, next); |
1600 | mm_segment_t old_fs = get_fs(); | 1671 | mm_segment_t old_fs = get_fs(); |
1601 | char tmp; | 1672 | char tmp; |
1602 | int res; | 1673 | int res; |
@@ -1611,7 +1682,7 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1611 | set_fs(old_fs); | 1682 | set_fs(old_fs); |
1612 | if (res) { | 1683 | if (res) { |
1613 | printk("SLAB: cache with size %d has lost its name\n", | 1684 | printk("SLAB: cache with size %d has lost its name\n", |
1614 | pc->objsize); | 1685 | pc->buffer_size); |
1615 | continue; | 1686 | continue; |
1616 | } | 1687 | } |
1617 | 1688 | ||
@@ -1696,20 +1767,20 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1696 | align = ralign; | 1767 | align = ralign; |
1697 | 1768 | ||
1698 | /* Get cache's description obj. */ | 1769 | /* Get cache's description obj. */ |
1699 | cachep = (kmem_cache_t *) kmem_cache_alloc(&cache_cache, SLAB_KERNEL); | 1770 | cachep = kmem_cache_alloc(&cache_cache, SLAB_KERNEL); |
1700 | if (!cachep) | 1771 | if (!cachep) |
1701 | goto oops; | 1772 | goto oops; |
1702 | memset(cachep, 0, sizeof(kmem_cache_t)); | 1773 | memset(cachep, 0, sizeof(struct kmem_cache)); |
1703 | 1774 | ||
1704 | #if DEBUG | 1775 | #if DEBUG |
1705 | cachep->reallen = size; | 1776 | cachep->obj_size = size; |
1706 | 1777 | ||
1707 | if (flags & SLAB_RED_ZONE) { | 1778 | if (flags & SLAB_RED_ZONE) { |
1708 | /* redzoning only works with word aligned caches */ | 1779 | /* redzoning only works with word aligned caches */ |
1709 | align = BYTES_PER_WORD; | 1780 | align = BYTES_PER_WORD; |
1710 | 1781 | ||
1711 | /* add space for red zone words */ | 1782 | /* add space for red zone words */ |
1712 | cachep->dbghead += BYTES_PER_WORD; | 1783 | cachep->obj_offset += BYTES_PER_WORD; |
1713 | size += 2 * BYTES_PER_WORD; | 1784 | size += 2 * BYTES_PER_WORD; |
1714 | } | 1785 | } |
1715 | if (flags & SLAB_STORE_USER) { | 1786 | if (flags & SLAB_STORE_USER) { |
@@ -1722,8 +1793,8 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1722 | } | 1793 | } |
1723 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | 1794 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) |
1724 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size | 1795 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
1725 | && cachep->reallen > cache_line_size() && size < PAGE_SIZE) { | 1796 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
1726 | cachep->dbghead += PAGE_SIZE - size; | 1797 | cachep->obj_offset += PAGE_SIZE - size; |
1727 | size = PAGE_SIZE; | 1798 | size = PAGE_SIZE; |
1728 | } | 1799 | } |
1729 | #endif | 1800 | #endif |
@@ -1786,7 +1857,7 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1786 | if (flags & SLAB_CACHE_DMA) | 1857 | if (flags & SLAB_CACHE_DMA) |
1787 | cachep->gfpflags |= GFP_DMA; | 1858 | cachep->gfpflags |= GFP_DMA; |
1788 | spin_lock_init(&cachep->spinlock); | 1859 | spin_lock_init(&cachep->spinlock); |
1789 | cachep->objsize = size; | 1860 | cachep->buffer_size = size; |
1790 | 1861 | ||
1791 | if (flags & CFLGS_OFF_SLAB) | 1862 | if (flags & CFLGS_OFF_SLAB) |
1792 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); | 1863 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
@@ -1843,11 +1914,11 @@ kmem_cache_create (const char *name, size_t size, size_t align, | |||
1843 | jiffies + REAPTIMEOUT_LIST3 + | 1914 | jiffies + REAPTIMEOUT_LIST3 + |
1844 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | 1915 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
1845 | 1916 | ||
1846 | BUG_ON(!ac_data(cachep)); | 1917 | BUG_ON(!cpu_cache_get(cachep)); |
1847 | ac_data(cachep)->avail = 0; | 1918 | cpu_cache_get(cachep)->avail = 0; |
1848 | ac_data(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | 1919 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; |
1849 | ac_data(cachep)->batchcount = 1; | 1920 | cpu_cache_get(cachep)->batchcount = 1; |
1850 | ac_data(cachep)->touched = 0; | 1921 | cpu_cache_get(cachep)->touched = 0; |
1851 | cachep->batchcount = 1; | 1922 | cachep->batchcount = 1; |
1852 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | 1923 | cachep->limit = BOOT_CPUCACHE_ENTRIES; |
1853 | } | 1924 | } |
@@ -1875,7 +1946,7 @@ static void check_irq_on(void) | |||
1875 | BUG_ON(irqs_disabled()); | 1946 | BUG_ON(irqs_disabled()); |
1876 | } | 1947 | } |
1877 | 1948 | ||
1878 | static void check_spinlock_acquired(kmem_cache_t *cachep) | 1949 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1879 | { | 1950 | { |
1880 | #ifdef CONFIG_SMP | 1951 | #ifdef CONFIG_SMP |
1881 | check_irq_off(); | 1952 | check_irq_off(); |
@@ -1883,7 +1954,7 @@ static void check_spinlock_acquired(kmem_cache_t *cachep) | |||
1883 | #endif | 1954 | #endif |
1884 | } | 1955 | } |
1885 | 1956 | ||
1886 | static inline void check_spinlock_acquired_node(kmem_cache_t *cachep, int node) | 1957 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
1887 | { | 1958 | { |
1888 | #ifdef CONFIG_SMP | 1959 | #ifdef CONFIG_SMP |
1889 | check_irq_off(); | 1960 | check_irq_off(); |
@@ -1916,24 +1987,24 @@ static void smp_call_function_all_cpus(void (*func)(void *arg), void *arg) | |||
1916 | preempt_enable(); | 1987 | preempt_enable(); |
1917 | } | 1988 | } |
1918 | 1989 | ||
1919 | static void drain_array_locked(kmem_cache_t *cachep, struct array_cache *ac, | 1990 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
1920 | int force, int node); | 1991 | int force, int node); |
1921 | 1992 | ||
1922 | static void do_drain(void *arg) | 1993 | static void do_drain(void *arg) |
1923 | { | 1994 | { |
1924 | kmem_cache_t *cachep = (kmem_cache_t *) arg; | 1995 | struct kmem_cache *cachep = (struct kmem_cache *) arg; |
1925 | struct array_cache *ac; | 1996 | struct array_cache *ac; |
1926 | int node = numa_node_id(); | 1997 | int node = numa_node_id(); |
1927 | 1998 | ||
1928 | check_irq_off(); | 1999 | check_irq_off(); |
1929 | ac = ac_data(cachep); | 2000 | ac = cpu_cache_get(cachep); |
1930 | spin_lock(&cachep->nodelists[node]->list_lock); | 2001 | spin_lock(&cachep->nodelists[node]->list_lock); |
1931 | free_block(cachep, ac->entry, ac->avail, node); | 2002 | free_block(cachep, ac->entry, ac->avail, node); |
1932 | spin_unlock(&cachep->nodelists[node]->list_lock); | 2003 | spin_unlock(&cachep->nodelists[node]->list_lock); |
1933 | ac->avail = 0; | 2004 | ac->avail = 0; |
1934 | } | 2005 | } |
1935 | 2006 | ||
1936 | static void drain_cpu_caches(kmem_cache_t *cachep) | 2007 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1937 | { | 2008 | { |
1938 | struct kmem_list3 *l3; | 2009 | struct kmem_list3 *l3; |
1939 | int node; | 2010 | int node; |
@@ -1954,7 +2025,7 @@ static void drain_cpu_caches(kmem_cache_t *cachep) | |||
1954 | spin_unlock_irq(&cachep->spinlock); | 2025 | spin_unlock_irq(&cachep->spinlock); |
1955 | } | 2026 | } |
1956 | 2027 | ||
1957 | static int __node_shrink(kmem_cache_t *cachep, int node) | 2028 | static int __node_shrink(struct kmem_cache *cachep, int node) |
1958 | { | 2029 | { |
1959 | struct slab *slabp; | 2030 | struct slab *slabp; |
1960 | struct kmem_list3 *l3 = cachep->nodelists[node]; | 2031 | struct kmem_list3 *l3 = cachep->nodelists[node]; |
@@ -1983,7 +2054,7 @@ static int __node_shrink(kmem_cache_t *cachep, int node) | |||
1983 | return ret; | 2054 | return ret; |
1984 | } | 2055 | } |
1985 | 2056 | ||
1986 | static int __cache_shrink(kmem_cache_t *cachep) | 2057 | static int __cache_shrink(struct kmem_cache *cachep) |
1987 | { | 2058 | { |
1988 | int ret = 0, i = 0; | 2059 | int ret = 0, i = 0; |
1989 | struct kmem_list3 *l3; | 2060 | struct kmem_list3 *l3; |
@@ -2009,7 +2080,7 @@ static int __cache_shrink(kmem_cache_t *cachep) | |||
2009 | * Releases as many slabs as possible for a cache. | 2080 | * Releases as many slabs as possible for a cache. |
2010 | * To help debugging, a zero exit status indicates all slabs were released. | 2081 | * To help debugging, a zero exit status indicates all slabs were released. |
2011 | */ | 2082 | */ |
2012 | int kmem_cache_shrink(kmem_cache_t *cachep) | 2083 | int kmem_cache_shrink(struct kmem_cache *cachep) |
2013 | { | 2084 | { |
2014 | if (!cachep || in_interrupt()) | 2085 | if (!cachep || in_interrupt()) |
2015 | BUG(); | 2086 | BUG(); |
@@ -2022,7 +2093,7 @@ EXPORT_SYMBOL(kmem_cache_shrink); | |||
2022 | * kmem_cache_destroy - delete a cache | 2093 | * kmem_cache_destroy - delete a cache |
2023 | * @cachep: the cache to destroy | 2094 | * @cachep: the cache to destroy |
2024 | * | 2095 | * |
2025 | * Remove a kmem_cache_t object from the slab cache. | 2096 | * Remove a struct kmem_cache object from the slab cache. |
2026 | * Returns 0 on success. | 2097 | * Returns 0 on success. |
2027 | * | 2098 | * |
2028 | * It is expected this function will be called by a module when it is | 2099 | * It is expected this function will be called by a module when it is |
@@ -2035,7 +2106,7 @@ EXPORT_SYMBOL(kmem_cache_shrink); | |||
2035 | * The caller must guarantee that noone will allocate memory from the cache | 2106 | * The caller must guarantee that noone will allocate memory from the cache |
2036 | * during the kmem_cache_destroy(). | 2107 | * during the kmem_cache_destroy(). |
2037 | */ | 2108 | */ |
2038 | int kmem_cache_destroy(kmem_cache_t *cachep) | 2109 | int kmem_cache_destroy(struct kmem_cache *cachep) |
2039 | { | 2110 | { |
2040 | int i; | 2111 | int i; |
2041 | struct kmem_list3 *l3; | 2112 | struct kmem_list3 *l3; |
@@ -2086,7 +2157,7 @@ int kmem_cache_destroy(kmem_cache_t *cachep) | |||
2086 | EXPORT_SYMBOL(kmem_cache_destroy); | 2157 | EXPORT_SYMBOL(kmem_cache_destroy); |
2087 | 2158 | ||
2088 | /* Get the memory for a slab management obj. */ | 2159 | /* Get the memory for a slab management obj. */ |
2089 | static struct slab *alloc_slabmgmt(kmem_cache_t *cachep, void *objp, | 2160 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
2090 | int colour_off, gfp_t local_flags) | 2161 | int colour_off, gfp_t local_flags) |
2091 | { | 2162 | { |
2092 | struct slab *slabp; | 2163 | struct slab *slabp; |
@@ -2112,13 +2183,13 @@ static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) | |||
2112 | return (kmem_bufctl_t *) (slabp + 1); | 2183 | return (kmem_bufctl_t *) (slabp + 1); |
2113 | } | 2184 | } |
2114 | 2185 | ||
2115 | static void cache_init_objs(kmem_cache_t *cachep, | 2186 | static void cache_init_objs(struct kmem_cache *cachep, |
2116 | struct slab *slabp, unsigned long ctor_flags) | 2187 | struct slab *slabp, unsigned long ctor_flags) |
2117 | { | 2188 | { |
2118 | int i; | 2189 | int i; |
2119 | 2190 | ||
2120 | for (i = 0; i < cachep->num; i++) { | 2191 | for (i = 0; i < cachep->num; i++) { |
2121 | void *objp = slabp->s_mem + cachep->objsize * i; | 2192 | void *objp = slabp->s_mem + cachep->buffer_size * i; |
2122 | #if DEBUG | 2193 | #if DEBUG |
2123 | /* need to poison the objs? */ | 2194 | /* need to poison the objs? */ |
2124 | if (cachep->flags & SLAB_POISON) | 2195 | if (cachep->flags & SLAB_POISON) |
@@ -2136,7 +2207,7 @@ static void cache_init_objs(kmem_cache_t *cachep, | |||
2136 | * Otherwise, deadlock. They must also be threaded. | 2207 | * Otherwise, deadlock. They must also be threaded. |
2137 | */ | 2208 | */ |
2138 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | 2209 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) |
2139 | cachep->ctor(objp + obj_dbghead(cachep), cachep, | 2210 | cachep->ctor(objp + obj_offset(cachep), cachep, |
2140 | ctor_flags); | 2211 | ctor_flags); |
2141 | 2212 | ||
2142 | if (cachep->flags & SLAB_RED_ZONE) { | 2213 | if (cachep->flags & SLAB_RED_ZONE) { |
@@ -2147,10 +2218,10 @@ static void cache_init_objs(kmem_cache_t *cachep, | |||
2147 | slab_error(cachep, "constructor overwrote the" | 2218 | slab_error(cachep, "constructor overwrote the" |
2148 | " start of an object"); | 2219 | " start of an object"); |
2149 | } | 2220 | } |
2150 | if ((cachep->objsize % PAGE_SIZE) == 0 && OFF_SLAB(cachep) | 2221 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep) |
2151 | && cachep->flags & SLAB_POISON) | 2222 | && cachep->flags & SLAB_POISON) |
2152 | kernel_map_pages(virt_to_page(objp), | 2223 | kernel_map_pages(virt_to_page(objp), |
2153 | cachep->objsize / PAGE_SIZE, 0); | 2224 | cachep->buffer_size / PAGE_SIZE, 0); |
2154 | #else | 2225 | #else |
2155 | if (cachep->ctor) | 2226 | if (cachep->ctor) |
2156 | cachep->ctor(objp, cachep, ctor_flags); | 2227 | cachep->ctor(objp, cachep, ctor_flags); |
@@ -2161,7 +2232,7 @@ static void cache_init_objs(kmem_cache_t *cachep, | |||
2161 | slabp->free = 0; | 2232 | slabp->free = 0; |
2162 | } | 2233 | } |
2163 | 2234 | ||
2164 | static void kmem_flagcheck(kmem_cache_t *cachep, gfp_t flags) | 2235 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
2165 | { | 2236 | { |
2166 | if (flags & SLAB_DMA) { | 2237 | if (flags & SLAB_DMA) { |
2167 | if (!(cachep->gfpflags & GFP_DMA)) | 2238 | if (!(cachep->gfpflags & GFP_DMA)) |
@@ -2172,7 +2243,43 @@ static void kmem_flagcheck(kmem_cache_t *cachep, gfp_t flags) | |||
2172 | } | 2243 | } |
2173 | } | 2244 | } |
2174 | 2245 | ||
2175 | static void set_slab_attr(kmem_cache_t *cachep, struct slab *slabp, void *objp) | 2246 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, int nodeid) |
2247 | { | ||
2248 | void *objp = slabp->s_mem + (slabp->free * cachep->buffer_size); | ||
2249 | kmem_bufctl_t next; | ||
2250 | |||
2251 | slabp->inuse++; | ||
2252 | next = slab_bufctl(slabp)[slabp->free]; | ||
2253 | #if DEBUG | ||
2254 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | ||
2255 | WARN_ON(slabp->nodeid != nodeid); | ||
2256 | #endif | ||
2257 | slabp->free = next; | ||
2258 | |||
2259 | return objp; | ||
2260 | } | ||
2261 | |||
2262 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, void *objp, | ||
2263 | int nodeid) | ||
2264 | { | ||
2265 | unsigned int objnr = (unsigned)(objp-slabp->s_mem) / cachep->buffer_size; | ||
2266 | |||
2267 | #if DEBUG | ||
2268 | /* Verify that the slab belongs to the intended node */ | ||
2269 | WARN_ON(slabp->nodeid != nodeid); | ||
2270 | |||
2271 | if (slab_bufctl(slabp)[objnr] != BUFCTL_FREE) { | ||
2272 | printk(KERN_ERR "slab: double free detected in cache " | ||
2273 | "'%s', objp %p\n", cachep->name, objp); | ||
2274 | BUG(); | ||
2275 | } | ||
2276 | #endif | ||
2277 | slab_bufctl(slabp)[objnr] = slabp->free; | ||
2278 | slabp->free = objnr; | ||
2279 | slabp->inuse--; | ||
2280 | } | ||
2281 | |||
2282 | static void set_slab_attr(struct kmem_cache *cachep, struct slab *slabp, void *objp) | ||
2176 | { | 2283 | { |
2177 | int i; | 2284 | int i; |
2178 | struct page *page; | 2285 | struct page *page; |
@@ -2191,7 +2298,7 @@ static void set_slab_attr(kmem_cache_t *cachep, struct slab *slabp, void *objp) | |||
2191 | * Grow (by 1) the number of slabs within a cache. This is called by | 2298 | * Grow (by 1) the number of slabs within a cache. This is called by |
2192 | * kmem_cache_alloc() when there are no active objs left in a cache. | 2299 | * kmem_cache_alloc() when there are no active objs left in a cache. |
2193 | */ | 2300 | */ |
2194 | static int cache_grow(kmem_cache_t *cachep, gfp_t flags, int nodeid) | 2301 | static int cache_grow(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
2195 | { | 2302 | { |
2196 | struct slab *slabp; | 2303 | struct slab *slabp; |
2197 | void *objp; | 2304 | void *objp; |
@@ -2302,14 +2409,14 @@ static void kfree_debugcheck(const void *objp) | |||
2302 | } | 2409 | } |
2303 | } | 2410 | } |
2304 | 2411 | ||
2305 | static void *cache_free_debugcheck(kmem_cache_t *cachep, void *objp, | 2412 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
2306 | void *caller) | 2413 | void *caller) |
2307 | { | 2414 | { |
2308 | struct page *page; | 2415 | struct page *page; |
2309 | unsigned int objnr; | 2416 | unsigned int objnr; |
2310 | struct slab *slabp; | 2417 | struct slab *slabp; |
2311 | 2418 | ||
2312 | objp -= obj_dbghead(cachep); | 2419 | objp -= obj_offset(cachep); |
2313 | kfree_debugcheck(objp); | 2420 | kfree_debugcheck(objp); |
2314 | page = virt_to_page(objp); | 2421 | page = virt_to_page(objp); |
2315 | 2422 | ||
@@ -2341,31 +2448,31 @@ static void *cache_free_debugcheck(kmem_cache_t *cachep, void *objp, | |||
2341 | if (cachep->flags & SLAB_STORE_USER) | 2448 | if (cachep->flags & SLAB_STORE_USER) |
2342 | *dbg_userword(cachep, objp) = caller; | 2449 | *dbg_userword(cachep, objp) = caller; |
2343 | 2450 | ||
2344 | objnr = (objp - slabp->s_mem) / cachep->objsize; | 2451 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
2345 | 2452 | ||
2346 | BUG_ON(objnr >= cachep->num); | 2453 | BUG_ON(objnr >= cachep->num); |
2347 | BUG_ON(objp != slabp->s_mem + objnr * cachep->objsize); | 2454 | BUG_ON(objp != slabp->s_mem + objnr * cachep->buffer_size); |
2348 | 2455 | ||
2349 | if (cachep->flags & SLAB_DEBUG_INITIAL) { | 2456 | if (cachep->flags & SLAB_DEBUG_INITIAL) { |
2350 | /* Need to call the slab's constructor so the | 2457 | /* Need to call the slab's constructor so the |
2351 | * caller can perform a verify of its state (debugging). | 2458 | * caller can perform a verify of its state (debugging). |
2352 | * Called without the cache-lock held. | 2459 | * Called without the cache-lock held. |
2353 | */ | 2460 | */ |
2354 | cachep->ctor(objp + obj_dbghead(cachep), | 2461 | cachep->ctor(objp + obj_offset(cachep), |
2355 | cachep, SLAB_CTOR_CONSTRUCTOR | SLAB_CTOR_VERIFY); | 2462 | cachep, SLAB_CTOR_CONSTRUCTOR | SLAB_CTOR_VERIFY); |
2356 | } | 2463 | } |
2357 | if (cachep->flags & SLAB_POISON && cachep->dtor) { | 2464 | if (cachep->flags & SLAB_POISON && cachep->dtor) { |
2358 | /* we want to cache poison the object, | 2465 | /* we want to cache poison the object, |
2359 | * call the destruction callback | 2466 | * call the destruction callback |
2360 | */ | 2467 | */ |
2361 | cachep->dtor(objp + obj_dbghead(cachep), cachep, 0); | 2468 | cachep->dtor(objp + obj_offset(cachep), cachep, 0); |
2362 | } | 2469 | } |
2363 | if (cachep->flags & SLAB_POISON) { | 2470 | if (cachep->flags & SLAB_POISON) { |
2364 | #ifdef CONFIG_DEBUG_PAGEALLOC | 2471 | #ifdef CONFIG_DEBUG_PAGEALLOC |
2365 | if ((cachep->objsize % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) { | 2472 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) { |
2366 | store_stackinfo(cachep, objp, (unsigned long)caller); | 2473 | store_stackinfo(cachep, objp, (unsigned long)caller); |
2367 | kernel_map_pages(virt_to_page(objp), | 2474 | kernel_map_pages(virt_to_page(objp), |
2368 | cachep->objsize / PAGE_SIZE, 0); | 2475 | cachep->buffer_size / PAGE_SIZE, 0); |
2369 | } else { | 2476 | } else { |
2370 | poison_obj(cachep, objp, POISON_FREE); | 2477 | poison_obj(cachep, objp, POISON_FREE); |
2371 | } | 2478 | } |
@@ -2376,7 +2483,7 @@ static void *cache_free_debugcheck(kmem_cache_t *cachep, void *objp, | |||
2376 | return objp; | 2483 | return objp; |
2377 | } | 2484 | } |
2378 | 2485 | ||
2379 | static void check_slabp(kmem_cache_t *cachep, struct slab *slabp) | 2486 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
2380 | { | 2487 | { |
2381 | kmem_bufctl_t i; | 2488 | kmem_bufctl_t i; |
2382 | int entries = 0; | 2489 | int entries = 0; |
@@ -2409,14 +2516,14 @@ static void check_slabp(kmem_cache_t *cachep, struct slab *slabp) | |||
2409 | #define check_slabp(x,y) do { } while(0) | 2516 | #define check_slabp(x,y) do { } while(0) |
2410 | #endif | 2517 | #endif |
2411 | 2518 | ||
2412 | static void *cache_alloc_refill(kmem_cache_t *cachep, gfp_t flags) | 2519 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
2413 | { | 2520 | { |
2414 | int batchcount; | 2521 | int batchcount; |
2415 | struct kmem_list3 *l3; | 2522 | struct kmem_list3 *l3; |
2416 | struct array_cache *ac; | 2523 | struct array_cache *ac; |
2417 | 2524 | ||
2418 | check_irq_off(); | 2525 | check_irq_off(); |
2419 | ac = ac_data(cachep); | 2526 | ac = cpu_cache_get(cachep); |
2420 | retry: | 2527 | retry: |
2421 | batchcount = ac->batchcount; | 2528 | batchcount = ac->batchcount; |
2422 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | 2529 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { |
@@ -2461,22 +2568,12 @@ static void *cache_alloc_refill(kmem_cache_t *cachep, gfp_t flags) | |||
2461 | check_slabp(cachep, slabp); | 2568 | check_slabp(cachep, slabp); |
2462 | check_spinlock_acquired(cachep); | 2569 | check_spinlock_acquired(cachep); |
2463 | while (slabp->inuse < cachep->num && batchcount--) { | 2570 | while (slabp->inuse < cachep->num && batchcount--) { |
2464 | kmem_bufctl_t next; | ||
2465 | STATS_INC_ALLOCED(cachep); | 2571 | STATS_INC_ALLOCED(cachep); |
2466 | STATS_INC_ACTIVE(cachep); | 2572 | STATS_INC_ACTIVE(cachep); |
2467 | STATS_SET_HIGH(cachep); | 2573 | STATS_SET_HIGH(cachep); |
2468 | 2574 | ||
2469 | /* get obj pointer */ | 2575 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
2470 | ac->entry[ac->avail++] = slabp->s_mem + | 2576 | numa_node_id()); |
2471 | slabp->free * cachep->objsize; | ||
2472 | |||
2473 | slabp->inuse++; | ||
2474 | next = slab_bufctl(slabp)[slabp->free]; | ||
2475 | #if DEBUG | ||
2476 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | ||
2477 | WARN_ON(numa_node_id() != slabp->nodeid); | ||
2478 | #endif | ||
2479 | slabp->free = next; | ||
2480 | } | 2577 | } |
2481 | check_slabp(cachep, slabp); | 2578 | check_slabp(cachep, slabp); |
2482 | 2579 | ||
@@ -2498,7 +2595,7 @@ static void *cache_alloc_refill(kmem_cache_t *cachep, gfp_t flags) | |||
2498 | x = cache_grow(cachep, flags, numa_node_id()); | 2595 | x = cache_grow(cachep, flags, numa_node_id()); |
2499 | 2596 | ||
2500 | // cache_grow can reenable interrupts, then ac could change. | 2597 | // cache_grow can reenable interrupts, then ac could change. |
2501 | ac = ac_data(cachep); | 2598 | ac = cpu_cache_get(cachep); |
2502 | if (!x && ac->avail == 0) // no objects in sight? abort | 2599 | if (!x && ac->avail == 0) // no objects in sight? abort |
2503 | return NULL; | 2600 | return NULL; |
2504 | 2601 | ||
@@ -2510,7 +2607,7 @@ static void *cache_alloc_refill(kmem_cache_t *cachep, gfp_t flags) | |||
2510 | } | 2607 | } |
2511 | 2608 | ||
2512 | static inline void | 2609 | static inline void |
2513 | cache_alloc_debugcheck_before(kmem_cache_t *cachep, gfp_t flags) | 2610 | cache_alloc_debugcheck_before(struct kmem_cache *cachep, gfp_t flags) |
2514 | { | 2611 | { |
2515 | might_sleep_if(flags & __GFP_WAIT); | 2612 | might_sleep_if(flags & __GFP_WAIT); |
2516 | #if DEBUG | 2613 | #if DEBUG |
@@ -2519,16 +2616,16 @@ cache_alloc_debugcheck_before(kmem_cache_t *cachep, gfp_t flags) | |||
2519 | } | 2616 | } |
2520 | 2617 | ||
2521 | #if DEBUG | 2618 | #if DEBUG |
2522 | static void *cache_alloc_debugcheck_after(kmem_cache_t *cachep, gfp_t flags, | 2619 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, gfp_t flags, |
2523 | void *objp, void *caller) | 2620 | void *objp, void *caller) |
2524 | { | 2621 | { |
2525 | if (!objp) | 2622 | if (!objp) |
2526 | return objp; | 2623 | return objp; |
2527 | if (cachep->flags & SLAB_POISON) { | 2624 | if (cachep->flags & SLAB_POISON) { |
2528 | #ifdef CONFIG_DEBUG_PAGEALLOC | 2625 | #ifdef CONFIG_DEBUG_PAGEALLOC |
2529 | if ((cachep->objsize % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) | 2626 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
2530 | kernel_map_pages(virt_to_page(objp), | 2627 | kernel_map_pages(virt_to_page(objp), |
2531 | cachep->objsize / PAGE_SIZE, 1); | 2628 | cachep->buffer_size / PAGE_SIZE, 1); |
2532 | else | 2629 | else |
2533 | check_poison_obj(cachep, objp); | 2630 | check_poison_obj(cachep, objp); |
2534 | #else | 2631 | #else |
@@ -2553,7 +2650,7 @@ static void *cache_alloc_debugcheck_after(kmem_cache_t *cachep, gfp_t flags, | |||
2553 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | 2650 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; |
2554 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | 2651 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; |
2555 | } | 2652 | } |
2556 | objp += obj_dbghead(cachep); | 2653 | objp += obj_offset(cachep); |
2557 | if (cachep->ctor && cachep->flags & SLAB_POISON) { | 2654 | if (cachep->ctor && cachep->flags & SLAB_POISON) { |
2558 | unsigned long ctor_flags = SLAB_CTOR_CONSTRUCTOR; | 2655 | unsigned long ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
2559 | 2656 | ||
@@ -2568,7 +2665,7 @@ static void *cache_alloc_debugcheck_after(kmem_cache_t *cachep, gfp_t flags, | |||
2568 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | 2665 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) |
2569 | #endif | 2666 | #endif |
2570 | 2667 | ||
2571 | static inline void *____cache_alloc(kmem_cache_t *cachep, gfp_t flags) | 2668 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
2572 | { | 2669 | { |
2573 | void *objp; | 2670 | void *objp; |
2574 | struct array_cache *ac; | 2671 | struct array_cache *ac; |
@@ -2583,7 +2680,7 @@ static inline void *____cache_alloc(kmem_cache_t *cachep, gfp_t flags) | |||
2583 | #endif | 2680 | #endif |
2584 | 2681 | ||
2585 | check_irq_off(); | 2682 | check_irq_off(); |
2586 | ac = ac_data(cachep); | 2683 | ac = cpu_cache_get(cachep); |
2587 | if (likely(ac->avail)) { | 2684 | if (likely(ac->avail)) { |
2588 | STATS_INC_ALLOCHIT(cachep); | 2685 | STATS_INC_ALLOCHIT(cachep); |
2589 | ac->touched = 1; | 2686 | ac->touched = 1; |
@@ -2595,7 +2692,8 @@ static inline void *____cache_alloc(kmem_cache_t *cachep, gfp_t flags) | |||
2595 | return objp; | 2692 | return objp; |
2596 | } | 2693 | } |
2597 | 2694 | ||
2598 | static inline void *__cache_alloc(kmem_cache_t *cachep, gfp_t flags) | 2695 | static __always_inline void * |
2696 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) | ||
2599 | { | 2697 | { |
2600 | unsigned long save_flags; | 2698 | unsigned long save_flags; |
2601 | void *objp; | 2699 | void *objp; |
@@ -2606,7 +2704,7 @@ static inline void *__cache_alloc(kmem_cache_t *cachep, gfp_t flags) | |||
2606 | objp = ____cache_alloc(cachep, flags); | 2704 | objp = ____cache_alloc(cachep, flags); |
2607 | local_irq_restore(save_flags); | 2705 | local_irq_restore(save_flags); |
2608 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, | 2706 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, |
2609 | __builtin_return_address(0)); | 2707 | caller); |
2610 | prefetchw(objp); | 2708 | prefetchw(objp); |
2611 | return objp; | 2709 | return objp; |
2612 | } | 2710 | } |
@@ -2615,13 +2713,12 @@ static inline void *__cache_alloc(kmem_cache_t *cachep, gfp_t flags) | |||
2615 | /* | 2713 | /* |
2616 | * A interface to enable slab creation on nodeid | 2714 | * A interface to enable slab creation on nodeid |
2617 | */ | 2715 | */ |
2618 | static void *__cache_alloc_node(kmem_cache_t *cachep, gfp_t flags, int nodeid) | 2716 | static void *__cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
2619 | { | 2717 | { |
2620 | struct list_head *entry; | 2718 | struct list_head *entry; |
2621 | struct slab *slabp; | 2719 | struct slab *slabp; |
2622 | struct kmem_list3 *l3; | 2720 | struct kmem_list3 *l3; |
2623 | void *obj; | 2721 | void *obj; |
2624 | kmem_bufctl_t next; | ||
2625 | int x; | 2722 | int x; |
2626 | 2723 | ||
2627 | l3 = cachep->nodelists[nodeid]; | 2724 | l3 = cachep->nodelists[nodeid]; |
@@ -2647,14 +2744,7 @@ static void *__cache_alloc_node(kmem_cache_t *cachep, gfp_t flags, int nodeid) | |||
2647 | 2744 | ||
2648 | BUG_ON(slabp->inuse == cachep->num); | 2745 | BUG_ON(slabp->inuse == cachep->num); |
2649 | 2746 | ||
2650 | /* get obj pointer */ | 2747 | obj = slab_get_obj(cachep, slabp, nodeid); |
2651 | obj = slabp->s_mem + slabp->free * cachep->objsize; | ||
2652 | slabp->inuse++; | ||
2653 | next = slab_bufctl(slabp)[slabp->free]; | ||
2654 | #if DEBUG | ||
2655 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | ||
2656 | #endif | ||
2657 | slabp->free = next; | ||
2658 | check_slabp(cachep, slabp); | 2748 | check_slabp(cachep, slabp); |
2659 | l3->free_objects--; | 2749 | l3->free_objects--; |
2660 | /* move slabp to correct slabp list: */ | 2750 | /* move slabp to correct slabp list: */ |
@@ -2685,7 +2775,7 @@ static void *__cache_alloc_node(kmem_cache_t *cachep, gfp_t flags, int nodeid) | |||
2685 | /* | 2775 | /* |
2686 | * Caller needs to acquire correct kmem_list's list_lock | 2776 | * Caller needs to acquire correct kmem_list's list_lock |
2687 | */ | 2777 | */ |
2688 | static void free_block(kmem_cache_t *cachep, void **objpp, int nr_objects, | 2778 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
2689 | int node) | 2779 | int node) |
2690 | { | 2780 | { |
2691 | int i; | 2781 | int i; |
@@ -2694,29 +2784,14 @@ static void free_block(kmem_cache_t *cachep, void **objpp, int nr_objects, | |||
2694 | for (i = 0; i < nr_objects; i++) { | 2784 | for (i = 0; i < nr_objects; i++) { |
2695 | void *objp = objpp[i]; | 2785 | void *objp = objpp[i]; |
2696 | struct slab *slabp; | 2786 | struct slab *slabp; |
2697 | unsigned int objnr; | ||
2698 | 2787 | ||
2699 | slabp = page_get_slab(virt_to_page(objp)); | 2788 | slabp = virt_to_slab(objp); |
2700 | l3 = cachep->nodelists[node]; | 2789 | l3 = cachep->nodelists[node]; |
2701 | list_del(&slabp->list); | 2790 | list_del(&slabp->list); |
2702 | objnr = (objp - slabp->s_mem) / cachep->objsize; | ||
2703 | check_spinlock_acquired_node(cachep, node); | 2791 | check_spinlock_acquired_node(cachep, node); |
2704 | check_slabp(cachep, slabp); | 2792 | check_slabp(cachep, slabp); |
2705 | 2793 | slab_put_obj(cachep, slabp, objp, node); | |
2706 | #if DEBUG | ||
2707 | /* Verify that the slab belongs to the intended node */ | ||
2708 | WARN_ON(slabp->nodeid != node); | ||
2709 | |||
2710 | if (slab_bufctl(slabp)[objnr] != BUFCTL_FREE) { | ||
2711 | printk(KERN_ERR "slab: double free detected in cache " | ||
2712 | "'%s', objp %p\n", cachep->name, objp); | ||
2713 | BUG(); | ||
2714 | } | ||
2715 | #endif | ||
2716 | slab_bufctl(slabp)[objnr] = slabp->free; | ||
2717 | slabp->free = objnr; | ||
2718 | STATS_DEC_ACTIVE(cachep); | 2794 | STATS_DEC_ACTIVE(cachep); |
2719 | slabp->inuse--; | ||
2720 | l3->free_objects++; | 2795 | l3->free_objects++; |
2721 | check_slabp(cachep, slabp); | 2796 | check_slabp(cachep, slabp); |
2722 | 2797 | ||
@@ -2738,7 +2813,7 @@ static void free_block(kmem_cache_t *cachep, void **objpp, int nr_objects, | |||
2738 | } | 2813 | } |
2739 | } | 2814 | } |
2740 | 2815 | ||
2741 | static void cache_flusharray(kmem_cache_t *cachep, struct array_cache *ac) | 2816 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
2742 | { | 2817 | { |
2743 | int batchcount; | 2818 | int batchcount; |
2744 | struct kmem_list3 *l3; | 2819 | struct kmem_list3 *l3; |
@@ -2797,9 +2872,9 @@ static void cache_flusharray(kmem_cache_t *cachep, struct array_cache *ac) | |||
2797 | * | 2872 | * |
2798 | * Called with disabled ints. | 2873 | * Called with disabled ints. |
2799 | */ | 2874 | */ |
2800 | static inline void __cache_free(kmem_cache_t *cachep, void *objp) | 2875 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
2801 | { | 2876 | { |
2802 | struct array_cache *ac = ac_data(cachep); | 2877 | struct array_cache *ac = cpu_cache_get(cachep); |
2803 | 2878 | ||
2804 | check_irq_off(); | 2879 | check_irq_off(); |
2805 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); | 2880 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
@@ -2810,7 +2885,7 @@ static inline void __cache_free(kmem_cache_t *cachep, void *objp) | |||
2810 | #ifdef CONFIG_NUMA | 2885 | #ifdef CONFIG_NUMA |
2811 | { | 2886 | { |
2812 | struct slab *slabp; | 2887 | struct slab *slabp; |
2813 | slabp = page_get_slab(virt_to_page(objp)); | 2888 | slabp = virt_to_slab(objp); |
2814 | if (unlikely(slabp->nodeid != numa_node_id())) { | 2889 | if (unlikely(slabp->nodeid != numa_node_id())) { |
2815 | struct array_cache *alien = NULL; | 2890 | struct array_cache *alien = NULL; |
2816 | int nodeid = slabp->nodeid; | 2891 | int nodeid = slabp->nodeid; |
@@ -2856,9 +2931,9 @@ static inline void __cache_free(kmem_cache_t *cachep, void *objp) | |||
2856 | * Allocate an object from this cache. The flags are only relevant | 2931 | * Allocate an object from this cache. The flags are only relevant |
2857 | * if the cache has no available objects. | 2932 | * if the cache has no available objects. |
2858 | */ | 2933 | */ |
2859 | void *kmem_cache_alloc(kmem_cache_t *cachep, gfp_t flags) | 2934 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
2860 | { | 2935 | { |
2861 | return __cache_alloc(cachep, flags); | 2936 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); |
2862 | } | 2937 | } |
2863 | EXPORT_SYMBOL(kmem_cache_alloc); | 2938 | EXPORT_SYMBOL(kmem_cache_alloc); |
2864 | 2939 | ||
@@ -2876,12 +2951,12 @@ EXPORT_SYMBOL(kmem_cache_alloc); | |||
2876 | * | 2951 | * |
2877 | * Currently only used for dentry validation. | 2952 | * Currently only used for dentry validation. |
2878 | */ | 2953 | */ |
2879 | int fastcall kmem_ptr_validate(kmem_cache_t *cachep, void *ptr) | 2954 | int fastcall kmem_ptr_validate(struct kmem_cache *cachep, void *ptr) |
2880 | { | 2955 | { |
2881 | unsigned long addr = (unsigned long)ptr; | 2956 | unsigned long addr = (unsigned long)ptr; |
2882 | unsigned long min_addr = PAGE_OFFSET; | 2957 | unsigned long min_addr = PAGE_OFFSET; |
2883 | unsigned long align_mask = BYTES_PER_WORD - 1; | 2958 | unsigned long align_mask = BYTES_PER_WORD - 1; |
2884 | unsigned long size = cachep->objsize; | 2959 | unsigned long size = cachep->buffer_size; |
2885 | struct page *page; | 2960 | struct page *page; |
2886 | 2961 | ||
2887 | if (unlikely(addr < min_addr)) | 2962 | if (unlikely(addr < min_addr)) |
@@ -2917,32 +2992,23 @@ int fastcall kmem_ptr_validate(kmem_cache_t *cachep, void *ptr) | |||
2917 | * New and improved: it will now make sure that the object gets | 2992 | * New and improved: it will now make sure that the object gets |
2918 | * put on the correct node list so that there is no false sharing. | 2993 | * put on the correct node list so that there is no false sharing. |
2919 | */ | 2994 | */ |
2920 | void *kmem_cache_alloc_node(kmem_cache_t *cachep, gfp_t flags, int nodeid) | 2995 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
2921 | { | 2996 | { |
2922 | unsigned long save_flags; | 2997 | unsigned long save_flags; |
2923 | void *ptr; | 2998 | void *ptr; |
2924 | 2999 | ||
2925 | if (nodeid == -1) | ||
2926 | return __cache_alloc(cachep, flags); | ||
2927 | |||
2928 | if (unlikely(!cachep->nodelists[nodeid])) { | ||
2929 | /* Fall back to __cache_alloc if we run into trouble */ | ||
2930 | printk(KERN_WARNING | ||
2931 | "slab: not allocating in inactive node %d for cache %s\n", | ||
2932 | nodeid, cachep->name); | ||
2933 | return __cache_alloc(cachep, flags); | ||
2934 | } | ||
2935 | |||
2936 | cache_alloc_debugcheck_before(cachep, flags); | 3000 | cache_alloc_debugcheck_before(cachep, flags); |
2937 | local_irq_save(save_flags); | 3001 | local_irq_save(save_flags); |
2938 | if (nodeid == numa_node_id()) | 3002 | |
3003 | if (nodeid == -1 || nodeid == numa_node_id() || | ||
3004 | !cachep->nodelists[nodeid]) | ||
2939 | ptr = ____cache_alloc(cachep, flags); | 3005 | ptr = ____cache_alloc(cachep, flags); |
2940 | else | 3006 | else |
2941 | ptr = __cache_alloc_node(cachep, flags, nodeid); | 3007 | ptr = __cache_alloc_node(cachep, flags, nodeid); |
2942 | local_irq_restore(save_flags); | 3008 | local_irq_restore(save_flags); |
2943 | ptr = | 3009 | |
2944 | cache_alloc_debugcheck_after(cachep, flags, ptr, | 3010 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, |
2945 | __builtin_return_address(0)); | 3011 | __builtin_return_address(0)); |
2946 | 3012 | ||
2947 | return ptr; | 3013 | return ptr; |
2948 | } | 3014 | } |
@@ -2950,7 +3016,7 @@ EXPORT_SYMBOL(kmem_cache_alloc_node); | |||
2950 | 3016 | ||
2951 | void *kmalloc_node(size_t size, gfp_t flags, int node) | 3017 | void *kmalloc_node(size_t size, gfp_t flags, int node) |
2952 | { | 3018 | { |
2953 | kmem_cache_t *cachep; | 3019 | struct kmem_cache *cachep; |
2954 | 3020 | ||
2955 | cachep = kmem_find_general_cachep(size, flags); | 3021 | cachep = kmem_find_general_cachep(size, flags); |
2956 | if (unlikely(cachep == NULL)) | 3022 | if (unlikely(cachep == NULL)) |
@@ -2981,9 +3047,10 @@ EXPORT_SYMBOL(kmalloc_node); | |||
2981 | * platforms. For example, on i386, it means that the memory must come | 3047 | * platforms. For example, on i386, it means that the memory must come |
2982 | * from the first 16MB. | 3048 | * from the first 16MB. |
2983 | */ | 3049 | */ |
2984 | void *__kmalloc(size_t size, gfp_t flags) | 3050 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
3051 | void *caller) | ||
2985 | { | 3052 | { |
2986 | kmem_cache_t *cachep; | 3053 | struct kmem_cache *cachep; |
2987 | 3054 | ||
2988 | /* If you want to save a few bytes .text space: replace | 3055 | /* If you want to save a few bytes .text space: replace |
2989 | * __ with kmem_. | 3056 | * __ with kmem_. |
@@ -2993,10 +3060,27 @@ void *__kmalloc(size_t size, gfp_t flags) | |||
2993 | cachep = __find_general_cachep(size, flags); | 3060 | cachep = __find_general_cachep(size, flags); |
2994 | if (unlikely(cachep == NULL)) | 3061 | if (unlikely(cachep == NULL)) |
2995 | return NULL; | 3062 | return NULL; |
2996 | return __cache_alloc(cachep, flags); | 3063 | return __cache_alloc(cachep, flags, caller); |
3064 | } | ||
3065 | |||
3066 | #ifndef CONFIG_DEBUG_SLAB | ||
3067 | |||
3068 | void *__kmalloc(size_t size, gfp_t flags) | ||
3069 | { | ||
3070 | return __do_kmalloc(size, flags, NULL); | ||
2997 | } | 3071 | } |
2998 | EXPORT_SYMBOL(__kmalloc); | 3072 | EXPORT_SYMBOL(__kmalloc); |
2999 | 3073 | ||
3074 | #else | ||
3075 | |||
3076 | void *__kmalloc_track_caller(size_t size, gfp_t flags, void *caller) | ||
3077 | { | ||
3078 | return __do_kmalloc(size, flags, caller); | ||
3079 | } | ||
3080 | EXPORT_SYMBOL(__kmalloc_track_caller); | ||
3081 | |||
3082 | #endif | ||
3083 | |||
3000 | #ifdef CONFIG_SMP | 3084 | #ifdef CONFIG_SMP |
3001 | /** | 3085 | /** |
3002 | * __alloc_percpu - allocate one copy of the object for every present | 3086 | * __alloc_percpu - allocate one copy of the object for every present |
@@ -3054,7 +3138,7 @@ EXPORT_SYMBOL(__alloc_percpu); | |||
3054 | * Free an object which was previously allocated from this | 3138 | * Free an object which was previously allocated from this |
3055 | * cache. | 3139 | * cache. |
3056 | */ | 3140 | */ |
3057 | void kmem_cache_free(kmem_cache_t *cachep, void *objp) | 3141 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
3058 | { | 3142 | { |
3059 | unsigned long flags; | 3143 | unsigned long flags; |
3060 | 3144 | ||
@@ -3075,15 +3159,15 @@ EXPORT_SYMBOL(kmem_cache_free); | |||
3075 | */ | 3159 | */ |
3076 | void kfree(const void *objp) | 3160 | void kfree(const void *objp) |
3077 | { | 3161 | { |
3078 | kmem_cache_t *c; | 3162 | struct kmem_cache *c; |
3079 | unsigned long flags; | 3163 | unsigned long flags; |
3080 | 3164 | ||
3081 | if (unlikely(!objp)) | 3165 | if (unlikely(!objp)) |
3082 | return; | 3166 | return; |
3083 | local_irq_save(flags); | 3167 | local_irq_save(flags); |
3084 | kfree_debugcheck(objp); | 3168 | kfree_debugcheck(objp); |
3085 | c = page_get_cache(virt_to_page(objp)); | 3169 | c = virt_to_cache(objp); |
3086 | mutex_debug_check_no_locks_freed(objp, obj_reallen(c)); | 3170 | mutex_debug_check_no_locks_freed(objp, obj_size(c)); |
3087 | __cache_free(c, (void *)objp); | 3171 | __cache_free(c, (void *)objp); |
3088 | local_irq_restore(flags); | 3172 | local_irq_restore(flags); |
3089 | } | 3173 | } |
@@ -3112,13 +3196,13 @@ void free_percpu(const void *objp) | |||
3112 | EXPORT_SYMBOL(free_percpu); | 3196 | EXPORT_SYMBOL(free_percpu); |
3113 | #endif | 3197 | #endif |
3114 | 3198 | ||
3115 | unsigned int kmem_cache_size(kmem_cache_t *cachep) | 3199 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
3116 | { | 3200 | { |
3117 | return obj_reallen(cachep); | 3201 | return obj_size(cachep); |
3118 | } | 3202 | } |
3119 | EXPORT_SYMBOL(kmem_cache_size); | 3203 | EXPORT_SYMBOL(kmem_cache_size); |
3120 | 3204 | ||
3121 | const char *kmem_cache_name(kmem_cache_t *cachep) | 3205 | const char *kmem_cache_name(struct kmem_cache *cachep) |
3122 | { | 3206 | { |
3123 | return cachep->name; | 3207 | return cachep->name; |
3124 | } | 3208 | } |
@@ -3127,7 +3211,7 @@ EXPORT_SYMBOL_GPL(kmem_cache_name); | |||
3127 | /* | 3211 | /* |
3128 | * This initializes kmem_list3 for all nodes. | 3212 | * This initializes kmem_list3 for all nodes. |
3129 | */ | 3213 | */ |
3130 | static int alloc_kmemlist(kmem_cache_t *cachep) | 3214 | static int alloc_kmemlist(struct kmem_cache *cachep) |
3131 | { | 3215 | { |
3132 | int node; | 3216 | int node; |
3133 | struct kmem_list3 *l3; | 3217 | struct kmem_list3 *l3; |
@@ -3183,7 +3267,7 @@ static int alloc_kmemlist(kmem_cache_t *cachep) | |||
3183 | } | 3267 | } |
3184 | 3268 | ||
3185 | struct ccupdate_struct { | 3269 | struct ccupdate_struct { |
3186 | kmem_cache_t *cachep; | 3270 | struct kmem_cache *cachep; |
3187 | struct array_cache *new[NR_CPUS]; | 3271 | struct array_cache *new[NR_CPUS]; |
3188 | }; | 3272 | }; |
3189 | 3273 | ||
@@ -3193,13 +3277,13 @@ static void do_ccupdate_local(void *info) | |||
3193 | struct array_cache *old; | 3277 | struct array_cache *old; |
3194 | 3278 | ||
3195 | check_irq_off(); | 3279 | check_irq_off(); |
3196 | old = ac_data(new->cachep); | 3280 | old = cpu_cache_get(new->cachep); |
3197 | 3281 | ||
3198 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; | 3282 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3199 | new->new[smp_processor_id()] = old; | 3283 | new->new[smp_processor_id()] = old; |
3200 | } | 3284 | } |
3201 | 3285 | ||
3202 | static int do_tune_cpucache(kmem_cache_t *cachep, int limit, int batchcount, | 3286 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, int batchcount, |
3203 | int shared) | 3287 | int shared) |
3204 | { | 3288 | { |
3205 | struct ccupdate_struct new; | 3289 | struct ccupdate_struct new; |
@@ -3245,7 +3329,7 @@ static int do_tune_cpucache(kmem_cache_t *cachep, int limit, int batchcount, | |||
3245 | return 0; | 3329 | return 0; |
3246 | } | 3330 | } |
3247 | 3331 | ||
3248 | static void enable_cpucache(kmem_cache_t *cachep) | 3332 | static void enable_cpucache(struct kmem_cache *cachep) |
3249 | { | 3333 | { |
3250 | int err; | 3334 | int err; |
3251 | int limit, shared; | 3335 | int limit, shared; |
@@ -3258,13 +3342,13 @@ static void enable_cpucache(kmem_cache_t *cachep) | |||
3258 | * The numbers are guessed, we should auto-tune as described by | 3342 | * The numbers are guessed, we should auto-tune as described by |
3259 | * Bonwick. | 3343 | * Bonwick. |
3260 | */ | 3344 | */ |
3261 | if (cachep->objsize > 131072) | 3345 | if (cachep->buffer_size > 131072) |
3262 | limit = 1; | 3346 | limit = 1; |
3263 | else if (cachep->objsize > PAGE_SIZE) | 3347 | else if (cachep->buffer_size > PAGE_SIZE) |
3264 | limit = 8; | 3348 | limit = 8; |
3265 | else if (cachep->objsize > 1024) | 3349 | else if (cachep->buffer_size > 1024) |
3266 | limit = 24; | 3350 | limit = 24; |
3267 | else if (cachep->objsize > 256) | 3351 | else if (cachep->buffer_size > 256) |
3268 | limit = 54; | 3352 | limit = 54; |
3269 | else | 3353 | else |
3270 | limit = 120; | 3354 | limit = 120; |
@@ -3279,7 +3363,7 @@ static void enable_cpucache(kmem_cache_t *cachep) | |||
3279 | */ | 3363 | */ |
3280 | shared = 0; | 3364 | shared = 0; |
3281 | #ifdef CONFIG_SMP | 3365 | #ifdef CONFIG_SMP |
3282 | if (cachep->objsize <= PAGE_SIZE) | 3366 | if (cachep->buffer_size <= PAGE_SIZE) |
3283 | shared = 8; | 3367 | shared = 8; |
3284 | #endif | 3368 | #endif |
3285 | 3369 | ||
@@ -3297,7 +3381,7 @@ static void enable_cpucache(kmem_cache_t *cachep) | |||
3297 | cachep->name, -err); | 3381 | cachep->name, -err); |
3298 | } | 3382 | } |
3299 | 3383 | ||
3300 | static void drain_array_locked(kmem_cache_t *cachep, struct array_cache *ac, | 3384 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
3301 | int force, int node) | 3385 | int force, int node) |
3302 | { | 3386 | { |
3303 | int tofree; | 3387 | int tofree; |
@@ -3342,12 +3426,12 @@ static void cache_reap(void *unused) | |||
3342 | } | 3426 | } |
3343 | 3427 | ||
3344 | list_for_each(walk, &cache_chain) { | 3428 | list_for_each(walk, &cache_chain) { |
3345 | kmem_cache_t *searchp; | 3429 | struct kmem_cache *searchp; |
3346 | struct list_head *p; | 3430 | struct list_head *p; |
3347 | int tofree; | 3431 | int tofree; |
3348 | struct slab *slabp; | 3432 | struct slab *slabp; |
3349 | 3433 | ||
3350 | searchp = list_entry(walk, kmem_cache_t, next); | 3434 | searchp = list_entry(walk, struct kmem_cache, next); |
3351 | 3435 | ||
3352 | if (searchp->flags & SLAB_NO_REAP) | 3436 | if (searchp->flags & SLAB_NO_REAP) |
3353 | goto next; | 3437 | goto next; |
@@ -3359,7 +3443,7 @@ static void cache_reap(void *unused) | |||
3359 | drain_alien_cache(searchp, l3); | 3443 | drain_alien_cache(searchp, l3); |
3360 | spin_lock_irq(&l3->list_lock); | 3444 | spin_lock_irq(&l3->list_lock); |
3361 | 3445 | ||
3362 | drain_array_locked(searchp, ac_data(searchp), 0, | 3446 | drain_array_locked(searchp, cpu_cache_get(searchp), 0, |
3363 | numa_node_id()); | 3447 | numa_node_id()); |
3364 | 3448 | ||
3365 | if (time_after(l3->next_reap, jiffies)) | 3449 | if (time_after(l3->next_reap, jiffies)) |
@@ -3450,15 +3534,15 @@ static void *s_start(struct seq_file *m, loff_t *pos) | |||
3450 | if (p == &cache_chain) | 3534 | if (p == &cache_chain) |
3451 | return NULL; | 3535 | return NULL; |
3452 | } | 3536 | } |
3453 | return list_entry(p, kmem_cache_t, next); | 3537 | return list_entry(p, struct kmem_cache, next); |
3454 | } | 3538 | } |
3455 | 3539 | ||
3456 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | 3540 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) |
3457 | { | 3541 | { |
3458 | kmem_cache_t *cachep = p; | 3542 | struct kmem_cache *cachep = p; |
3459 | ++*pos; | 3543 | ++*pos; |
3460 | return cachep->next.next == &cache_chain ? NULL | 3544 | return cachep->next.next == &cache_chain ? NULL |
3461 | : list_entry(cachep->next.next, kmem_cache_t, next); | 3545 | : list_entry(cachep->next.next, struct kmem_cache, next); |
3462 | } | 3546 | } |
3463 | 3547 | ||
3464 | static void s_stop(struct seq_file *m, void *p) | 3548 | static void s_stop(struct seq_file *m, void *p) |
@@ -3468,7 +3552,7 @@ static void s_stop(struct seq_file *m, void *p) | |||
3468 | 3552 | ||
3469 | static int s_show(struct seq_file *m, void *p) | 3553 | static int s_show(struct seq_file *m, void *p) |
3470 | { | 3554 | { |
3471 | kmem_cache_t *cachep = p; | 3555 | struct kmem_cache *cachep = p; |
3472 | struct list_head *q; | 3556 | struct list_head *q; |
3473 | struct slab *slabp; | 3557 | struct slab *slabp; |
3474 | unsigned long active_objs; | 3558 | unsigned long active_objs; |
@@ -3528,7 +3612,7 @@ static int s_show(struct seq_file *m, void *p) | |||
3528 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | 3612 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); |
3529 | 3613 | ||
3530 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | 3614 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
3531 | name, active_objs, num_objs, cachep->objsize, | 3615 | name, active_objs, num_objs, cachep->buffer_size, |
3532 | cachep->num, (1 << cachep->gfporder)); | 3616 | cachep->num, (1 << cachep->gfporder)); |
3533 | seq_printf(m, " : tunables %4u %4u %4u", | 3617 | seq_printf(m, " : tunables %4u %4u %4u", |
3534 | cachep->limit, cachep->batchcount, cachep->shared); | 3618 | cachep->limit, cachep->batchcount, cachep->shared); |
@@ -3618,7 +3702,8 @@ ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |||
3618 | mutex_lock(&cache_chain_mutex); | 3702 | mutex_lock(&cache_chain_mutex); |
3619 | res = -EINVAL; | 3703 | res = -EINVAL; |
3620 | list_for_each(p, &cache_chain) { | 3704 | list_for_each(p, &cache_chain) { |
3621 | kmem_cache_t *cachep = list_entry(p, kmem_cache_t, next); | 3705 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, |
3706 | next); | ||
3622 | 3707 | ||
3623 | if (!strcmp(cachep->name, kbuf)) { | 3708 | if (!strcmp(cachep->name, kbuf)) { |
3624 | if (limit < 1 || | 3709 | if (limit < 1 || |
@@ -3656,5 +3741,5 @@ unsigned int ksize(const void *objp) | |||
3656 | if (unlikely(objp == NULL)) | 3741 | if (unlikely(objp == NULL)) |
3657 | return 0; | 3742 | return 0; |
3658 | 3743 | ||
3659 | return obj_reallen(page_get_cache(virt_to_page(objp))); | 3744 | return obj_size(virt_to_cache(objp)); |
3660 | } | 3745 | } |
diff --git a/mm/swap_state.c b/mm/swap_state.c index 7b09ac503fec..db8a3d3e1636 100644 --- a/mm/swap_state.c +++ b/mm/swap_state.c | |||
@@ -27,6 +27,7 @@ static struct address_space_operations swap_aops = { | |||
27 | .writepage = swap_writepage, | 27 | .writepage = swap_writepage, |
28 | .sync_page = block_sync_page, | 28 | .sync_page = block_sync_page, |
29 | .set_page_dirty = __set_page_dirty_nobuffers, | 29 | .set_page_dirty = __set_page_dirty_nobuffers, |
30 | .migratepage = migrate_page, | ||
30 | }; | 31 | }; |
31 | 32 | ||
32 | static struct backing_dev_info swap_backing_dev_info = { | 33 | static struct backing_dev_info swap_backing_dev_info = { |
diff --git a/mm/swapfile.c b/mm/swapfile.c index f1e69c30d203..1f9cf0d073b8 100644 --- a/mm/swapfile.c +++ b/mm/swapfile.c | |||
@@ -554,6 +554,15 @@ static int unuse_mm(struct mm_struct *mm, | |||
554 | return 0; | 554 | return 0; |
555 | } | 555 | } |
556 | 556 | ||
557 | #ifdef CONFIG_MIGRATION | ||
558 | int remove_vma_swap(struct vm_area_struct *vma, struct page *page) | ||
559 | { | ||
560 | swp_entry_t entry = { .val = page_private(page) }; | ||
561 | |||
562 | return unuse_vma(vma, entry, page); | ||
563 | } | ||
564 | #endif | ||
565 | |||
557 | /* | 566 | /* |
558 | * Scan swap_map from current position to next entry still in use. | 567 | * Scan swap_map from current position to next entry still in use. |
559 | * Recycle to start on reaching the end, returning 0 when empty. | 568 | * Recycle to start on reaching the end, returning 0 when empty. |
@@ -646,6 +655,7 @@ static int try_to_unuse(unsigned int type) | |||
646 | */ | 655 | */ |
647 | swap_map = &si->swap_map[i]; | 656 | swap_map = &si->swap_map[i]; |
648 | entry = swp_entry(type, i); | 657 | entry = swp_entry(type, i); |
658 | again: | ||
649 | page = read_swap_cache_async(entry, NULL, 0); | 659 | page = read_swap_cache_async(entry, NULL, 0); |
650 | if (!page) { | 660 | if (!page) { |
651 | /* | 661 | /* |
@@ -680,6 +690,12 @@ static int try_to_unuse(unsigned int type) | |||
680 | wait_on_page_locked(page); | 690 | wait_on_page_locked(page); |
681 | wait_on_page_writeback(page); | 691 | wait_on_page_writeback(page); |
682 | lock_page(page); | 692 | lock_page(page); |
693 | if (!PageSwapCache(page)) { | ||
694 | /* Page migration has occured */ | ||
695 | unlock_page(page); | ||
696 | page_cache_release(page); | ||
697 | goto again; | ||
698 | } | ||
683 | wait_on_page_writeback(page); | 699 | wait_on_page_writeback(page); |
684 | 700 | ||
685 | /* | 701 | /* |
diff --git a/mm/vmscan.c b/mm/vmscan.c index 2e34b61a70c7..5a610804cd06 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c | |||
@@ -477,7 +477,13 @@ static int shrink_list(struct list_head *page_list, struct scan_control *sc) | |||
477 | * processes. Try to unmap it here. | 477 | * processes. Try to unmap it here. |
478 | */ | 478 | */ |
479 | if (page_mapped(page) && mapping) { | 479 | if (page_mapped(page) && mapping) { |
480 | switch (try_to_unmap(page)) { | 480 | /* |
481 | * No unmapping if we do not swap | ||
482 | */ | ||
483 | if (!sc->may_swap) | ||
484 | goto keep_locked; | ||
485 | |||
486 | switch (try_to_unmap(page, 0)) { | ||
481 | case SWAP_FAIL: | 487 | case SWAP_FAIL: |
482 | goto activate_locked; | 488 | goto activate_locked; |
483 | case SWAP_AGAIN: | 489 | case SWAP_AGAIN: |
@@ -492,7 +498,7 @@ static int shrink_list(struct list_head *page_list, struct scan_control *sc) | |||
492 | goto keep_locked; | 498 | goto keep_locked; |
493 | if (!may_enter_fs) | 499 | if (!may_enter_fs) |
494 | goto keep_locked; | 500 | goto keep_locked; |
495 | if (laptop_mode && !sc->may_writepage) | 501 | if (!sc->may_writepage) |
496 | goto keep_locked; | 502 | goto keep_locked; |
497 | 503 | ||
498 | /* Page is dirty, try to write it out here */ | 504 | /* Page is dirty, try to write it out here */ |
@@ -609,6 +615,15 @@ int putback_lru_pages(struct list_head *l) | |||
609 | } | 615 | } |
610 | 616 | ||
611 | /* | 617 | /* |
618 | * Non migratable page | ||
619 | */ | ||
620 | int fail_migrate_page(struct page *newpage, struct page *page) | ||
621 | { | ||
622 | return -EIO; | ||
623 | } | ||
624 | EXPORT_SYMBOL(fail_migrate_page); | ||
625 | |||
626 | /* | ||
612 | * swapout a single page | 627 | * swapout a single page |
613 | * page is locked upon entry, unlocked on exit | 628 | * page is locked upon entry, unlocked on exit |
614 | */ | 629 | */ |
@@ -617,7 +632,7 @@ static int swap_page(struct page *page) | |||
617 | struct address_space *mapping = page_mapping(page); | 632 | struct address_space *mapping = page_mapping(page); |
618 | 633 | ||
619 | if (page_mapped(page) && mapping) | 634 | if (page_mapped(page) && mapping) |
620 | if (try_to_unmap(page) != SWAP_SUCCESS) | 635 | if (try_to_unmap(page, 0) != SWAP_SUCCESS) |
621 | goto unlock_retry; | 636 | goto unlock_retry; |
622 | 637 | ||
623 | if (PageDirty(page)) { | 638 | if (PageDirty(page)) { |
@@ -653,6 +668,167 @@ unlock_retry: | |||
653 | retry: | 668 | retry: |
654 | return -EAGAIN; | 669 | return -EAGAIN; |
655 | } | 670 | } |
671 | EXPORT_SYMBOL(swap_page); | ||
672 | |||
673 | /* | ||
674 | * Page migration was first developed in the context of the memory hotplug | ||
675 | * project. The main authors of the migration code are: | ||
676 | * | ||
677 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | ||
678 | * Hirokazu Takahashi <taka@valinux.co.jp> | ||
679 | * Dave Hansen <haveblue@us.ibm.com> | ||
680 | * Christoph Lameter <clameter@sgi.com> | ||
681 | */ | ||
682 | |||
683 | /* | ||
684 | * Remove references for a page and establish the new page with the correct | ||
685 | * basic settings to be able to stop accesses to the page. | ||
686 | */ | ||
687 | int migrate_page_remove_references(struct page *newpage, | ||
688 | struct page *page, int nr_refs) | ||
689 | { | ||
690 | struct address_space *mapping = page_mapping(page); | ||
691 | struct page **radix_pointer; | ||
692 | |||
693 | /* | ||
694 | * Avoid doing any of the following work if the page count | ||
695 | * indicates that the page is in use or truncate has removed | ||
696 | * the page. | ||
697 | */ | ||
698 | if (!mapping || page_mapcount(page) + nr_refs != page_count(page)) | ||
699 | return 1; | ||
700 | |||
701 | /* | ||
702 | * Establish swap ptes for anonymous pages or destroy pte | ||
703 | * maps for files. | ||
704 | * | ||
705 | * In order to reestablish file backed mappings the fault handlers | ||
706 | * will take the radix tree_lock which may then be used to stop | ||
707 | * processses from accessing this page until the new page is ready. | ||
708 | * | ||
709 | * A process accessing via a swap pte (an anonymous page) will take a | ||
710 | * page_lock on the old page which will block the process until the | ||
711 | * migration attempt is complete. At that time the PageSwapCache bit | ||
712 | * will be examined. If the page was migrated then the PageSwapCache | ||
713 | * bit will be clear and the operation to retrieve the page will be | ||
714 | * retried which will find the new page in the radix tree. Then a new | ||
715 | * direct mapping may be generated based on the radix tree contents. | ||
716 | * | ||
717 | * If the page was not migrated then the PageSwapCache bit | ||
718 | * is still set and the operation may continue. | ||
719 | */ | ||
720 | try_to_unmap(page, 1); | ||
721 | |||
722 | /* | ||
723 | * Give up if we were unable to remove all mappings. | ||
724 | */ | ||
725 | if (page_mapcount(page)) | ||
726 | return 1; | ||
727 | |||
728 | write_lock_irq(&mapping->tree_lock); | ||
729 | |||
730 | radix_pointer = (struct page **)radix_tree_lookup_slot( | ||
731 | &mapping->page_tree, | ||
732 | page_index(page)); | ||
733 | |||
734 | if (!page_mapping(page) || page_count(page) != nr_refs || | ||
735 | *radix_pointer != page) { | ||
736 | write_unlock_irq(&mapping->tree_lock); | ||
737 | return 1; | ||
738 | } | ||
739 | |||
740 | /* | ||
741 | * Now we know that no one else is looking at the page. | ||
742 | * | ||
743 | * Certain minimal information about a page must be available | ||
744 | * in order for other subsystems to properly handle the page if they | ||
745 | * find it through the radix tree update before we are finished | ||
746 | * copying the page. | ||
747 | */ | ||
748 | get_page(newpage); | ||
749 | newpage->index = page->index; | ||
750 | newpage->mapping = page->mapping; | ||
751 | if (PageSwapCache(page)) { | ||
752 | SetPageSwapCache(newpage); | ||
753 | set_page_private(newpage, page_private(page)); | ||
754 | } | ||
755 | |||
756 | *radix_pointer = newpage; | ||
757 | __put_page(page); | ||
758 | write_unlock_irq(&mapping->tree_lock); | ||
759 | |||
760 | return 0; | ||
761 | } | ||
762 | EXPORT_SYMBOL(migrate_page_remove_references); | ||
763 | |||
764 | /* | ||
765 | * Copy the page to its new location | ||
766 | */ | ||
767 | void migrate_page_copy(struct page *newpage, struct page *page) | ||
768 | { | ||
769 | copy_highpage(newpage, page); | ||
770 | |||
771 | if (PageError(page)) | ||
772 | SetPageError(newpage); | ||
773 | if (PageReferenced(page)) | ||
774 | SetPageReferenced(newpage); | ||
775 | if (PageUptodate(page)) | ||
776 | SetPageUptodate(newpage); | ||
777 | if (PageActive(page)) | ||
778 | SetPageActive(newpage); | ||
779 | if (PageChecked(page)) | ||
780 | SetPageChecked(newpage); | ||
781 | if (PageMappedToDisk(page)) | ||
782 | SetPageMappedToDisk(newpage); | ||
783 | |||
784 | if (PageDirty(page)) { | ||
785 | clear_page_dirty_for_io(page); | ||
786 | set_page_dirty(newpage); | ||
787 | } | ||
788 | |||
789 | ClearPageSwapCache(page); | ||
790 | ClearPageActive(page); | ||
791 | ClearPagePrivate(page); | ||
792 | set_page_private(page, 0); | ||
793 | page->mapping = NULL; | ||
794 | |||
795 | /* | ||
796 | * If any waiters have accumulated on the new page then | ||
797 | * wake them up. | ||
798 | */ | ||
799 | if (PageWriteback(newpage)) | ||
800 | end_page_writeback(newpage); | ||
801 | } | ||
802 | EXPORT_SYMBOL(migrate_page_copy); | ||
803 | |||
804 | /* | ||
805 | * Common logic to directly migrate a single page suitable for | ||
806 | * pages that do not use PagePrivate. | ||
807 | * | ||
808 | * Pages are locked upon entry and exit. | ||
809 | */ | ||
810 | int migrate_page(struct page *newpage, struct page *page) | ||
811 | { | ||
812 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | ||
813 | |||
814 | if (migrate_page_remove_references(newpage, page, 2)) | ||
815 | return -EAGAIN; | ||
816 | |||
817 | migrate_page_copy(newpage, page); | ||
818 | |||
819 | /* | ||
820 | * Remove auxiliary swap entries and replace | ||
821 | * them with real ptes. | ||
822 | * | ||
823 | * Note that a real pte entry will allow processes that are not | ||
824 | * waiting on the page lock to use the new page via the page tables | ||
825 | * before the new page is unlocked. | ||
826 | */ | ||
827 | remove_from_swap(newpage); | ||
828 | return 0; | ||
829 | } | ||
830 | EXPORT_SYMBOL(migrate_page); | ||
831 | |||
656 | /* | 832 | /* |
657 | * migrate_pages | 833 | * migrate_pages |
658 | * | 834 | * |
@@ -666,11 +842,6 @@ retry: | |||
666 | * are movable anymore because t has become empty | 842 | * are movable anymore because t has become empty |
667 | * or no retryable pages exist anymore. | 843 | * or no retryable pages exist anymore. |
668 | * | 844 | * |
669 | * SIMPLIFIED VERSION: This implementation of migrate_pages | ||
670 | * is only swapping out pages and never touches the second | ||
671 | * list. The direct migration patchset | ||
672 | * extends this function to avoid the use of swap. | ||
673 | * | ||
674 | * Return: Number of pages not migrated when "to" ran empty. | 845 | * Return: Number of pages not migrated when "to" ran empty. |
675 | */ | 846 | */ |
676 | int migrate_pages(struct list_head *from, struct list_head *to, | 847 | int migrate_pages(struct list_head *from, struct list_head *to, |
@@ -691,6 +862,9 @@ redo: | |||
691 | retry = 0; | 862 | retry = 0; |
692 | 863 | ||
693 | list_for_each_entry_safe(page, page2, from, lru) { | 864 | list_for_each_entry_safe(page, page2, from, lru) { |
865 | struct page *newpage = NULL; | ||
866 | struct address_space *mapping; | ||
867 | |||
694 | cond_resched(); | 868 | cond_resched(); |
695 | 869 | ||
696 | rc = 0; | 870 | rc = 0; |
@@ -698,6 +872,9 @@ redo: | |||
698 | /* page was freed from under us. So we are done. */ | 872 | /* page was freed from under us. So we are done. */ |
699 | goto next; | 873 | goto next; |
700 | 874 | ||
875 | if (to && list_empty(to)) | ||
876 | break; | ||
877 | |||
701 | /* | 878 | /* |
702 | * Skip locked pages during the first two passes to give the | 879 | * Skip locked pages during the first two passes to give the |
703 | * functions holding the lock time to release the page. Later we | 880 | * functions holding the lock time to release the page. Later we |
@@ -734,12 +911,69 @@ redo: | |||
734 | } | 911 | } |
735 | } | 912 | } |
736 | 913 | ||
914 | if (!to) { | ||
915 | rc = swap_page(page); | ||
916 | goto next; | ||
917 | } | ||
918 | |||
919 | newpage = lru_to_page(to); | ||
920 | lock_page(newpage); | ||
921 | |||
737 | /* | 922 | /* |
738 | * Page is properly locked and writeback is complete. | 923 | * Pages are properly locked and writeback is complete. |
739 | * Try to migrate the page. | 924 | * Try to migrate the page. |
740 | */ | 925 | */ |
741 | rc = swap_page(page); | 926 | mapping = page_mapping(page); |
742 | goto next; | 927 | if (!mapping) |
928 | goto unlock_both; | ||
929 | |||
930 | if (mapping->a_ops->migratepage) { | ||
931 | rc = mapping->a_ops->migratepage(newpage, page); | ||
932 | goto unlock_both; | ||
933 | } | ||
934 | |||
935 | /* | ||
936 | * Trigger writeout if page is dirty | ||
937 | */ | ||
938 | if (PageDirty(page)) { | ||
939 | switch (pageout(page, mapping)) { | ||
940 | case PAGE_KEEP: | ||
941 | case PAGE_ACTIVATE: | ||
942 | goto unlock_both; | ||
943 | |||
944 | case PAGE_SUCCESS: | ||
945 | unlock_page(newpage); | ||
946 | goto next; | ||
947 | |||
948 | case PAGE_CLEAN: | ||
949 | ; /* try to migrate the page below */ | ||
950 | } | ||
951 | } | ||
952 | /* | ||
953 | * If we have no buffer or can release the buffer | ||
954 | * then do a simple migration. | ||
955 | */ | ||
956 | if (!page_has_buffers(page) || | ||
957 | try_to_release_page(page, GFP_KERNEL)) { | ||
958 | rc = migrate_page(newpage, page); | ||
959 | goto unlock_both; | ||
960 | } | ||
961 | |||
962 | /* | ||
963 | * On early passes with mapped pages simply | ||
964 | * retry. There may be a lock held for some | ||
965 | * buffers that may go away. Later | ||
966 | * swap them out. | ||
967 | */ | ||
968 | if (pass > 4) { | ||
969 | unlock_page(newpage); | ||
970 | newpage = NULL; | ||
971 | rc = swap_page(page); | ||
972 | goto next; | ||
973 | } | ||
974 | |||
975 | unlock_both: | ||
976 | unlock_page(newpage); | ||
743 | 977 | ||
744 | unlock_page: | 978 | unlock_page: |
745 | unlock_page(page); | 979 | unlock_page(page); |
@@ -752,7 +986,10 @@ next: | |||
752 | list_move(&page->lru, failed); | 986 | list_move(&page->lru, failed); |
753 | nr_failed++; | 987 | nr_failed++; |
754 | } else { | 988 | } else { |
755 | /* Success */ | 989 | if (newpage) { |
990 | /* Successful migration. Return page to LRU */ | ||
991 | move_to_lru(newpage); | ||
992 | } | ||
756 | list_move(&page->lru, moved); | 993 | list_move(&page->lru, moved); |
757 | } | 994 | } |
758 | } | 995 | } |
@@ -1170,7 +1407,7 @@ int try_to_free_pages(struct zone **zones, gfp_t gfp_mask) | |||
1170 | int i; | 1407 | int i; |
1171 | 1408 | ||
1172 | sc.gfp_mask = gfp_mask; | 1409 | sc.gfp_mask = gfp_mask; |
1173 | sc.may_writepage = 0; | 1410 | sc.may_writepage = !laptop_mode; |
1174 | sc.may_swap = 1; | 1411 | sc.may_swap = 1; |
1175 | 1412 | ||
1176 | inc_page_state(allocstall); | 1413 | inc_page_state(allocstall); |
@@ -1273,7 +1510,7 @@ loop_again: | |||
1273 | total_scanned = 0; | 1510 | total_scanned = 0; |
1274 | total_reclaimed = 0; | 1511 | total_reclaimed = 0; |
1275 | sc.gfp_mask = GFP_KERNEL; | 1512 | sc.gfp_mask = GFP_KERNEL; |
1276 | sc.may_writepage = 0; | 1513 | sc.may_writepage = !laptop_mode; |
1277 | sc.may_swap = 1; | 1514 | sc.may_swap = 1; |
1278 | sc.nr_mapped = read_page_state(nr_mapped); | 1515 | sc.nr_mapped = read_page_state(nr_mapped); |
1279 | 1516 | ||
@@ -1586,40 +1823,61 @@ module_init(kswapd_init) | |||
1586 | */ | 1823 | */ |
1587 | int zone_reclaim_mode __read_mostly; | 1824 | int zone_reclaim_mode __read_mostly; |
1588 | 1825 | ||
1826 | #define RECLAIM_OFF 0 | ||
1827 | #define RECLAIM_ZONE (1<<0) /* Run shrink_cache on the zone */ | ||
1828 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ | ||
1829 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ | ||
1830 | #define RECLAIM_SLAB (1<<3) /* Do a global slab shrink if the zone is out of memory */ | ||
1831 | |||
1589 | /* | 1832 | /* |
1590 | * Mininum time between zone reclaim scans | 1833 | * Mininum time between zone reclaim scans |
1591 | */ | 1834 | */ |
1592 | #define ZONE_RECLAIM_INTERVAL HZ/2 | 1835 | int zone_reclaim_interval __read_mostly = 30*HZ; |
1836 | |||
1837 | /* | ||
1838 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | ||
1839 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | ||
1840 | * a zone. | ||
1841 | */ | ||
1842 | #define ZONE_RECLAIM_PRIORITY 4 | ||
1843 | |||
1593 | /* | 1844 | /* |
1594 | * Try to free up some pages from this zone through reclaim. | 1845 | * Try to free up some pages from this zone through reclaim. |
1595 | */ | 1846 | */ |
1596 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | 1847 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
1597 | { | 1848 | { |
1598 | int nr_pages = 1 << order; | 1849 | int nr_pages; |
1599 | struct task_struct *p = current; | 1850 | struct task_struct *p = current; |
1600 | struct reclaim_state reclaim_state; | 1851 | struct reclaim_state reclaim_state; |
1601 | struct scan_control sc = { | 1852 | struct scan_control sc; |
1602 | .gfp_mask = gfp_mask, | 1853 | cpumask_t mask; |
1603 | .may_writepage = 0, | 1854 | int node_id; |
1604 | .may_swap = 0, | 1855 | |
1605 | .nr_mapped = read_page_state(nr_mapped), | 1856 | if (time_before(jiffies, |
1606 | .nr_scanned = 0, | 1857 | zone->last_unsuccessful_zone_reclaim + zone_reclaim_interval)) |
1607 | .nr_reclaimed = 0, | 1858 | return 0; |
1608 | .priority = 0 | ||
1609 | }; | ||
1610 | 1859 | ||
1611 | if (!(gfp_mask & __GFP_WAIT) || | 1860 | if (!(gfp_mask & __GFP_WAIT) || |
1612 | zone->zone_pgdat->node_id != numa_node_id() || | ||
1613 | zone->all_unreclaimable || | 1861 | zone->all_unreclaimable || |
1614 | atomic_read(&zone->reclaim_in_progress) > 0) | 1862 | atomic_read(&zone->reclaim_in_progress) > 0) |
1615 | return 0; | 1863 | return 0; |
1616 | 1864 | ||
1617 | if (time_before(jiffies, | 1865 | node_id = zone->zone_pgdat->node_id; |
1618 | zone->last_unsuccessful_zone_reclaim + ZONE_RECLAIM_INTERVAL)) | 1866 | mask = node_to_cpumask(node_id); |
1619 | return 0; | 1867 | if (!cpus_empty(mask) && node_id != numa_node_id()) |
1868 | return 0; | ||
1869 | |||
1870 | sc.may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE); | ||
1871 | sc.may_swap = !!(zone_reclaim_mode & RECLAIM_SWAP); | ||
1872 | sc.nr_scanned = 0; | ||
1873 | sc.nr_reclaimed = 0; | ||
1874 | sc.priority = ZONE_RECLAIM_PRIORITY + 1; | ||
1875 | sc.nr_mapped = read_page_state(nr_mapped); | ||
1876 | sc.gfp_mask = gfp_mask; | ||
1620 | 1877 | ||
1621 | disable_swap_token(); | 1878 | disable_swap_token(); |
1622 | 1879 | ||
1880 | nr_pages = 1 << order; | ||
1623 | if (nr_pages > SWAP_CLUSTER_MAX) | 1881 | if (nr_pages > SWAP_CLUSTER_MAX) |
1624 | sc.swap_cluster_max = nr_pages; | 1882 | sc.swap_cluster_max = nr_pages; |
1625 | else | 1883 | else |
@@ -1629,14 +1887,37 @@ int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |||
1629 | p->flags |= PF_MEMALLOC; | 1887 | p->flags |= PF_MEMALLOC; |
1630 | reclaim_state.reclaimed_slab = 0; | 1888 | reclaim_state.reclaimed_slab = 0; |
1631 | p->reclaim_state = &reclaim_state; | 1889 | p->reclaim_state = &reclaim_state; |
1632 | shrink_zone(zone, &sc); | 1890 | |
1891 | /* | ||
1892 | * Free memory by calling shrink zone with increasing priorities | ||
1893 | * until we have enough memory freed. | ||
1894 | */ | ||
1895 | do { | ||
1896 | sc.priority--; | ||
1897 | shrink_zone(zone, &sc); | ||
1898 | |||
1899 | } while (sc.nr_reclaimed < nr_pages && sc.priority > 0); | ||
1900 | |||
1901 | if (sc.nr_reclaimed < nr_pages && (zone_reclaim_mode & RECLAIM_SLAB)) { | ||
1902 | /* | ||
1903 | * shrink_slab does not currently allow us to determine | ||
1904 | * how many pages were freed in the zone. So we just | ||
1905 | * shake the slab and then go offnode for a single allocation. | ||
1906 | * | ||
1907 | * shrink_slab will free memory on all zones and may take | ||
1908 | * a long time. | ||
1909 | */ | ||
1910 | shrink_slab(sc.nr_scanned, gfp_mask, order); | ||
1911 | sc.nr_reclaimed = 1; /* Avoid getting the off node timeout */ | ||
1912 | } | ||
1913 | |||
1633 | p->reclaim_state = NULL; | 1914 | p->reclaim_state = NULL; |
1634 | current->flags &= ~PF_MEMALLOC; | 1915 | current->flags &= ~PF_MEMALLOC; |
1635 | 1916 | ||
1636 | if (sc.nr_reclaimed == 0) | 1917 | if (sc.nr_reclaimed == 0) |
1637 | zone->last_unsuccessful_zone_reclaim = jiffies; | 1918 | zone->last_unsuccessful_zone_reclaim = jiffies; |
1638 | 1919 | ||
1639 | return sc.nr_reclaimed > nr_pages; | 1920 | return sc.nr_reclaimed >= nr_pages; |
1640 | } | 1921 | } |
1641 | #endif | 1922 | #endif |
1642 | 1923 | ||