Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux

rmap: add exclusively owned pages to the newest anon_vma

The recent anon_vma fixes cause many anonymous pages to end up
in the parent process anon_vma, even when the page is exclusively
owned by the current process.

Adding exclusively owned anonymous pages to the top anon_vma
reduces rmap scanning overhead, especially in workloads with
forking servers.

This patch adds a parameter to __page_set_anon_rmap that can
be used to indicate whether or not the added page is exclusively
owned by the current process.

Pages added through page_add_new_anon_rmap are exclusively
owned by the current process, and can be added to the top
anon_vma.

Pages added through page_add_anon_rmap can be either shared
or exclusively owned, so we do the conservative thing and
add it to the oldest anon_vma.

A next step would be to add the exclusive parameter to
page_add_anon_rmap, to be used from functions where we do
know for sure whether a page is exclusively owned.

Signed-off-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Lightly-tested-by: Borislav Petkov <bp@alien8.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
[ Edited to look nicer - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

authored by

Rik van Riel and committed by
Linus Torvalds
e8a03feb 9b030e20

+16 -11
+16 -11
mm/rmap.c
··· 730 730 * @page: the page to add the mapping to 731 731 * @vma: the vm area in which the mapping is added 732 732 * @address: the user virtual address mapped 733 + * @exclusive: the page is exclusively owned by the current process 733 734 */ 734 735 static void __page_set_anon_rmap(struct page *page, 735 - struct vm_area_struct *vma, unsigned long address) 736 + struct vm_area_struct *vma, unsigned long address, int exclusive) 736 737 { 737 - struct anon_vma_chain *avc; 738 - struct anon_vma *anon_vma; 738 + struct anon_vma *anon_vma = vma->anon_vma; 739 739 740 - BUG_ON(!vma->anon_vma); 740 + BUG_ON(!anon_vma); 741 741 742 742 /* 743 - * We must use the _oldest_ possible anon_vma for the page mapping! 743 + * If the page isn't exclusively mapped into this vma, 744 + * we must use the _oldest_ possible anon_vma for the 745 + * page mapping! 744 746 * 745 - * So take the last AVC chain entry in the vma, which is the deepest 746 - * ancestor, and use the anon_vma from that. 747 + * So take the last AVC chain entry in the vma, which is 748 + * the deepest ancestor, and use the anon_vma from that. 747 749 */ 748 - avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma); 749 - anon_vma = avc->anon_vma; 750 + if (!exclusive) { 751 + struct anon_vma_chain *avc; 752 + avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma); 753 + anon_vma = avc->anon_vma; 754 + } 750 755 751 756 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 752 757 page->mapping = (struct address_space *) anon_vma; ··· 807 802 VM_BUG_ON(!PageLocked(page)); 808 803 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); 809 804 if (first) 810 - __page_set_anon_rmap(page, vma, address); 805 + __page_set_anon_rmap(page, vma, address, 0); 811 806 else 812 807 __page_check_anon_rmap(page, vma, address); 813 808 } ··· 829 824 SetPageSwapBacked(page); 830 825 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ 831 826 __inc_zone_page_state(page, NR_ANON_PAGES); 832 - __page_set_anon_rmap(page, vma, address); 827 + __page_set_anon_rmap(page, vma, address, 1); 833 828 if (page_evictable(page, vma)) 834 829 lru_cache_add_lru(page, LRU_ACTIVE_ANON); 835 830 else