The patch titled
swapoff: scan ptes preemptibly
has been added to the -mm tree. Its filename is
swapoff-scan-ptes-preemptibly.patch
*** Remember to use Documentation/SubmitChecklist when testing your code ***
See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
out what to do about this
------------------------------------------------------
Subject: swapoff: scan ptes preemptibly
From: Hugh Dickins <hugh@xxxxxxxxxxx>
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@xxxxxxxxxxx>
Cc: Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/swapfile.c | 38 +++++++++++++++++++++++++++++++-------
1 file changed, 31 insertions(+), 7 deletions(-)
diff -puN mm/swapfile.c~swapoff-scan-ptes-preemptibly mm/swapfile.c
--- a/mm/swapfile.c~swapoff-scan-ptes-preemptibly
+++ a/mm/swapfile.c
@@ -506,9 +506,19 @@ unsigned int count_swap_pages(int type,
* 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_str
* 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_are
{
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;
}
_
Patches currently in -mm which might be from hugh@xxxxxxxxxxx are
git-unionfs.patch
i386-and-x86_64-randomize-brk-fix-2.patch
swapin_readahead-excise-numa-bogosity.patch
swapin_readahead-move-and-rearrange-args.patch
swapin-needs-gfp_mask-for-loop-on-tmpfs.patch
shmem-sgp_quick-and-sgp_fault-redundant.patch
shmem_getpage-return-page-locked.patch
shmem_file_write-is-redundant.patch
swapin-fix-valid_swaphandles-defect.patch
swapoff-scan-ptes-preemptibly.patch
maps4-add-proportional-set-size-accounting-in-smaps.patch
tmpfs-fix-mounts-when-size-is-less-than-the-page-size.patch
r-o-bind-mounts-track-number-of-mount-writer-fix-buggy-loop.patch
r-o-bind-mounts-track-number-of-mount-writer-fix-buggy-loop-checkpatch-fixes.patch
memory-controller-memory-accounting-v7.patch
prio_tree-debugging-patch.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
