lkml.org 
[lkml]   [1999]   [Dec]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject2.3.29 patch for releasing swap cache pages on exit and fixing race in do_swap_page.Larry
We have run into two problems in 2.3 related to the swap cache: 1.) When
a process
exits, some of its private pages can remain in the swap cache and tie up
swap space
unless shrink_mmap runs. 2.) What looks like a race between
do_swap_page
and shrink_mmap.

Here is my attempt at a patch that fixes both problems.

Larry Woodman
http://www/missioncriticallinux.com
--- mm/memory.c.sav Tue Dec 7 10:04:33 1999
+++ mm/memory.c Tue Dec 7 10:06:14 1999
@@ -1004,12 +1004,15 @@
pte = mk_pte(page, vma->vm_page_prot);

set_bit(PG_swap_entry, &page->flags);
+ lock_page(page);
if (write_access && !is_page_shared(page)) {
- delete_from_swap_cache(page);
+ delete_from_swap_cache_nolock(page);
+ UnlockPage(page);
page = replace_with_highmem(page);
pte = mk_pte(page, vma->vm_page_prot);
pte = pte_mkwrite(pte_mkdirty(pte));
- }
+ } else
+ UnlockPage(page);
set_pte(page_table, pte);
/* No need to invalidate - it was non-present before */
update_mmu_cache(vma, address, pte);
--- mm/swap_state.c.sav Wed Dec 1 18:20:23 1999
+++ mm/swap_state.c Tue Dec 7 10:07:17 1999
@@ -171,7 +171,7 @@
unlock_kernel();
}

-static void delete_from_swap_cache_nolock(struct page *page)
+void delete_from_swap_cache_nolock(struct page *page)
{
if (block_flushpage(NULL, page, 0))
lru_cache_del(page);
--- include/linux/swap.h.sav Wed Dec 1 12:28:42 1999
+++ include/linux/swap.h Tue Dec 7 10:01:16 1999
@@ -147,8 +147,9 @@
if (PageReserved(page))
return 1;
count = page_count(page);
- if (PageSwapCache(page))
- count += swap_count(page) - 2;
+ if (PageSwapCache(page))
+ /* page_count is 3 if shrink_mmap didnt free buffer yet */
+ count += swap_count(page) - 2 - !!page->buffers;
return count > 1;
}
\
 
 \ /
  Last update: 2005-03-22 13:55    [W:0.457 / U:0.060 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site