The problem is that there is no good reason to map in the process a
read-only swap cache page if the swap cache page was not shared.
Mapping the swap cache page even if it wasn't shared not only cause the
kerenl to go low of swap entry, but it will also cause an additional page
fault and cow if after the read access there will be a write access.
With this my patch everything is normal while running `./bw_mem 120m rd'.
I have not yet benchmarked the thing but I think it will be an improvement
while using swap cache pages (for the reasons above).
Patch against 2.2.7:
Index: linux/mm//page_alloc.c
===================================================================
RCS file: /var/cvs/linux/mm/page_alloc.c,v
retrieving revision 1.1.1.4
diff -u -r1.1.1.4 page_alloc.c
--- page_alloc.c 1999/04/28 20:46:47 1.1.1.4
+++ linux/mm/page_alloc.c 1999/05/04 01:20:30
@@ -414,17 +414,16 @@
tsk->min_flt++;
swap_free(entry);
- if (!write_access || is_page_shared(page_map)) {
+ if (is_page_shared(page_map)) {
set_pte(page_table, mk_pte(page, vma->vm_page_prot));
return;
}
- /* The page is unshared, and we want write access. In this
- case, it is safe to tear down the swap cache and give the
- page over entirely to this process. */
+ /* The page is unshared. In this case, it is safe to tear down the
+ * swap cache and give the page over entirely to this process.
+ */
- if (PageSwapCache(page_map))
- delete_from_swap_cache(page_map);
+ delete_from_swap_cache(page_map);
set_pte(page_table, pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot))));
return;
}
The check for PageSwapCache() was useless.
Andrea Arcangeli
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/