aboutsummaryrefslogtreecommitdiffstats
path: root/mm
diff options
context:
space:
mode:
authorLee Schermerhorn <lee.schermerhorn@hp.com>2008-10-18 23:26:52 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2008-10-20 11:52:31 -0400
commit64d6519dda3905dfb94d3f93c07c5f263f41813f (patch)
tree62cae88edcb858c42a5b4b568eb77801299250bb /mm
parent5344b7e648980cc2ca613ec03a56a8222ff48820 (diff)
swap: cull unevictable pages in fault path
In the fault paths that install new anonymous pages, check whether the page is evictable or not using lru_cache_add_active_or_unevictable(). If the page is evictable, just add it to the active lru list [via the pagevec cache], else add it to the unevictable list. This "proactive" culling in the fault path mimics the handling of mlocked pages in Nick Piggin's series to keep mlocked pages off the lru lists. Notes: 1) This patch is optional--e.g., if one is concerned about the additional test in the fault path. We can defer the moving of nonreclaimable pages until when vmscan [shrink_*_list()] encounters them. Vmscan will only need to handle such pages once, but if there are a lot of them it could impact system performance. 2) The 'vma' argument to page_evictable() is require to notice that we're faulting a page into an mlock()ed vma w/o having to scan the page's rmap in the fault path. Culling mlock()ed anon pages is currently the only reason for this patch. 3) We can't cull swap pages in read_swap_cache_async() because the vma argument doesn't necessarily correspond to the swap cache offset passed in by swapin_readahead(). This could [did!] result in mlocking pages in non-VM_LOCKED vmas if [when] we tried to cull in this path. 4) Move set_pte_at() to after where we add page to lru to keep it hidden from other tasks that might walk the page table. We already do it in this order in do_anonymous() page. And, these are COW'd anon pages. Is this safe? [riel@redhat.com: undo an overzealous code cleanup] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'mm')
-rw-r--r--mm/memory.c18
-rw-r--r--mm/swap.c21
2 files changed, 31 insertions, 8 deletions
diff --git a/mm/memory.c b/mm/memory.c
index 9fef7272fb9e..450127f4c582 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1922,12 +1922,13 @@ gotten:
1922 * thread doing COW. 1922 * thread doing COW.
1923 */ 1923 */
1924 ptep_clear_flush_notify(vma, address, page_table); 1924 ptep_clear_flush_notify(vma, address, page_table);
1925 set_pte_at(mm, address, page_table, entry);
1926 update_mmu_cache(vma, address, entry);
1927 SetPageSwapBacked(new_page); 1925 SetPageSwapBacked(new_page);
1928 lru_cache_add_active_anon(new_page); 1926 lru_cache_add_active_or_unevictable(new_page, vma);
1929 page_add_new_anon_rmap(new_page, vma, address); 1927 page_add_new_anon_rmap(new_page, vma, address);
1930 1928
1929//TODO: is this safe? do_anonymous_page() does it this way.
1930 set_pte_at(mm, address, page_table, entry);
1931 update_mmu_cache(vma, address, entry);
1931 if (old_page) { 1932 if (old_page) {
1932 /* 1933 /*
1933 * Only after switching the pte to the new page may 1934 * Only after switching the pte to the new page may
@@ -2420,7 +2421,7 @@ static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2420 goto release; 2421 goto release;
2421 inc_mm_counter(mm, anon_rss); 2422 inc_mm_counter(mm, anon_rss);
2422 SetPageSwapBacked(page); 2423 SetPageSwapBacked(page);
2423 lru_cache_add_active_anon(page); 2424 lru_cache_add_active_or_unevictable(page, vma);
2424 page_add_new_anon_rmap(page, vma, address); 2425 page_add_new_anon_rmap(page, vma, address);
2425 set_pte_at(mm, address, page_table, entry); 2426 set_pte_at(mm, address, page_table, entry);
2426 2427
@@ -2564,12 +2565,11 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2564 entry = mk_pte(page, vma->vm_page_prot); 2565 entry = mk_pte(page, vma->vm_page_prot);
2565 if (flags & FAULT_FLAG_WRITE) 2566 if (flags & FAULT_FLAG_WRITE)
2566 entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2567 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2567 set_pte_at(mm, address, page_table, entry);
2568 if (anon) { 2568 if (anon) {
2569 inc_mm_counter(mm, anon_rss); 2569 inc_mm_counter(mm, anon_rss);
2570 SetPageSwapBacked(page); 2570 SetPageSwapBacked(page);
2571 lru_cache_add_active_anon(page); 2571 lru_cache_add_active_or_unevictable(page, vma);
2572 page_add_new_anon_rmap(page, vma, address); 2572 page_add_new_anon_rmap(page, vma, address);
2573 } else { 2573 } else {
2574 inc_mm_counter(mm, file_rss); 2574 inc_mm_counter(mm, file_rss);
2575 page_add_file_rmap(page); 2575 page_add_file_rmap(page);
@@ -2578,6 +2578,8 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2578 get_page(dirty_page); 2578 get_page(dirty_page);
2579 } 2579 }
2580 } 2580 }
2581//TODO: is this safe? do_anonymous_page() does it this way.
2582 set_pte_at(mm, address, page_table, entry);
2581 2583
2582 /* no need to invalidate: a not-present page won't be cached */ 2584 /* no need to invalidate: a not-present page won't be cached */
2583 update_mmu_cache(vma, address, entry); 2585 update_mmu_cache(vma, address, entry);
diff --git a/mm/swap.c b/mm/swap.c
index bc58c1369dd6..2152e48a7b8f 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -31,6 +31,8 @@
31#include <linux/backing-dev.h> 31#include <linux/backing-dev.h>
32#include <linux/memcontrol.h> 32#include <linux/memcontrol.h>
33 33
34#include "internal.h"
35
34/* How many pages do we try to swap or page in/out together? */ 36/* How many pages do we try to swap or page in/out together? */
35int page_cluster; 37int page_cluster;
36 38
@@ -244,6 +246,25 @@ void add_page_to_unevictable_list(struct page *page)
244 spin_unlock_irq(&zone->lru_lock); 246 spin_unlock_irq(&zone->lru_lock);
245} 247}
246 248
249/**
250 * lru_cache_add_active_or_unevictable
251 * @page: the page to be added to LRU
252 * @vma: vma in which page is mapped for determining reclaimability
253 *
254 * place @page on active or unevictable LRU list, depending on
255 * page_evictable(). Note that if the page is not evictable,
256 * it goes directly back onto it's zone's unevictable list. It does
257 * NOT use a per cpu pagevec.
258 */
259void lru_cache_add_active_or_unevictable(struct page *page,
260 struct vm_area_struct *vma)
261{
262 if (page_evictable(page, vma))
263 lru_cache_add_lru(page, LRU_ACTIVE + page_is_file_cache(page));
264 else
265 add_page_to_unevictable_list(page);
266}
267
247/* 268/*
248 * Drain pages out of the cpu's pagevecs. 269 * Drain pages out of the cpu's pagevecs.
249 * Either "cpu" is the current CPU, and preemption has already been 270 * Either "cpu" is the current CPU, and preemption has already been