From: Hugh Dickins <hugh@veritas.com>
Date: Tue, 5 Feb 2008 06:28:46 +0000 (-0800)
Subject: swapoff: scan ptes preemptibly
X-Git-Tag: v2.6.25-rc1~971
X-Git-Url: https://err.no/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=2e441889c38fe1b6ef6b963e6993076aa120176c;p=linux-2.6

swapoff: scan ptes preemptibly

Provided that CONFIG_HIGHPTE is not set, unuse_pte_range can reduce latency
in swapoff by scanning the page table preemptibly: so long as unuse_pte is
careful to recheck that entry under pte lock.

(To tell the truth, this patch was not inspired by any cries for lower
latency here: rather, this restructuring permits a future memory controller
patch to allocate with GFP_KERNEL in unuse_pte, where before it could not.
But it would be wrong to tuck this change away inside a memcgroup patch.)

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Tested-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---

diff --git a/mm/swapfile.c b/mm/swapfile.c
index f5ba723faf..14bc4f28a8 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -506,9 +506,19 @@ unsigned int count_swap_pages(int type, int free)
  * just let do_wp_page work it out if a write is requested later - to
  * force COW, vm_page_prot omits write permission from any private vma.
  */
-static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
+static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
 		unsigned long addr, swp_entry_t entry, struct page *page)
 {
+	spinlock_t *ptl;
+	pte_t *pte;
+	int found = 1;
+
+	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+	if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
+		found = 0;
+		goto out;
+	}
+
 	inc_mm_counter(vma->vm_mm, anon_rss);
 	get_page(page);
 	set_pte_at(vma->vm_mm, addr, pte,
@@ -520,6 +530,9 @@ static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
 	 * immediately swapped out again after swapon.
 	 */
 	activate_page(page);
+out:
+	pte_unmap_unlock(pte, ptl);
+	return found;
 }
 
 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
@@ -528,22 +541,33 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 {
 	pte_t swp_pte = swp_entry_to_pte(entry);
 	pte_t *pte;
-	spinlock_t *ptl;
 	int found = 0;
 
-	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
+	/*
+	 * We don't actually need pte lock while scanning for swp_pte: since
+	 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
+	 * page table while we're scanning; though it could get zapped, and on
+	 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
+	 * of unmatched parts which look like swp_pte, so unuse_pte must
+	 * recheck under pte lock.  Scanning without pte lock lets it be
+	 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
+	 */
+	pte = pte_offset_map(pmd, addr);
 	do {
 		/*
 		 * swapoff spends a _lot_ of time in this loop!
 		 * Test inline before going to call unuse_pte.
 		 */
 		if (unlikely(pte_same(*pte, swp_pte))) {
-			unuse_pte(vma, pte++, addr, entry, page);
-			found = 1;
-			break;
+			pte_unmap(pte);
+			found = unuse_pte(vma, pmd, addr, entry, page);
+			if (found)
+				goto out;
+			pte = pte_offset_map(pmd, addr);
 		}
 	} while (pte++, addr += PAGE_SIZE, addr != end);
-	pte_unmap_unlock(pte - 1, ptl);
+	pte_unmap(pte - 1);
+out:
 	return found;
 }