Here is the use-once patch, updated to 2.4.7.
For my test load, I noticed that the stock 2.4.7 is a little faster
and 2.4.7+use.once is a little bit slower than 2.4.5+use.once. But
the patch still turns in a performance improvement of about 6% over
the stock kernel.
I investigated this a little bit and what I saw was some very erratic
behavior of the inactive lists. In particular, both free counts and
inactive_dirty length would gyrate wildly while inactive_dirty stayed
unchanged for long periods. This is for both stock and patched
kernels. Quite different from the 2.4.5 behaviour, and not very nice
looking.
I think what this means is, we need to take a good hard look at the
way the queues are balanced. Hmmm, I mean *yet another* good hard
look.
To apply:
cd /usr/src/your-2.4.7-tree
patch <this.patch -p0
--- ../2.4.7.clean/mm/filemap.c Wed Jul 11 23:52:58 2001
+++ ./mm/filemap.c Thu Jul 26 16:40:53 2001
@@ -778,51 +778,6 @@
#endif
/*
- * We combine this with read-ahead to deactivate pages when we
- * think there's sequential IO going on. Note that this is
- * harmless since we don't actually evict the pages from memory
- * but just move them to the inactive list.
- *
- * TODO:
- * - make the readahead code smarter
- * - move readahead to the VMA level so we can do the same
- * trick with mmap()
- *
- * Rik van Riel, 2000
- */
-static void drop_behind(struct file * file, unsigned long index)
-{
- struct inode *inode = file->f_dentry->d_inode;
- struct address_space *mapping = inode->i_mapping;
- struct page *page;
- unsigned long start;
-
- /* Nothing to drop-behind if we're on the first page. */
- if (!index)
- return;
-
- if (index > file->f_rawin)
- start = index - file->f_rawin;
- else
- start = 0;
-
- /*
- * Go backwards from index-1 and drop all pages in the
- * readahead window. Since the readahead window may have
- * been increased since the last time we were called, we
- * stop when the page isn't there.
- */
- spin_lock(&pagecache_lock);
- while (--index >= start) {
- page = __find_page_simple(mapping, index);
- if (!page)
- break;
- deactivate_page(page);
- }
- spin_unlock(&pagecache_lock);
-}
-
-/*
* Read-ahead profiling information
* --------------------------------
* Every PROFILE_MAXREADCOUNT, the following information is written
@@ -1041,12 +996,6 @@
if (filp->f_ramax > max_readahead)
filp->f_ramax = max_readahead;
- /*
- * Move the pages that have already been passed
- * to the inactive list.
- */
- drop_behind(filp, index);
-
#ifdef PROFILE_READAHEAD
profile_readahead((reada_ok == 2), filp);
#endif
@@ -1055,6 +1004,21 @@
return;
}
+/*
+ * Optimization for use-once file pages: queue immediately for eviction,
+ * but check later for additional uses before actually evicting.
+ *
+ * Daniel Phillips, 7/2001
+ */
+
+static inline void check_used_once (struct page *page)
+{
+ if (!page->age)
+ {
+ page->age = PAGE_AGE_START;
+ ClearPageReferenced(page);
+ }
+}
/*
* This is a generic file read routine, and uses the
@@ -1171,7 +1135,8 @@
offset += ret;
index += offset >> PAGE_CACHE_SHIFT;
offset &= ~PAGE_CACHE_MASK;
-
+
+ check_used_once (page);
page_cache_release(page);
if (ret == nr && desc->count)
continue;
@@ -2608,7 +2573,6 @@
while (count) {
unsigned long index, offset;
char *kaddr;
- int deactivate = 1;
/*
* Try to find the page in the cache. If it isn't there,
@@ -2617,10 +2581,8 @@
offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
index = pos >> PAGE_CACHE_SHIFT;
bytes = PAGE_CACHE_SIZE - offset;
- if (bytes > count) {
+ if (bytes > count)
bytes = count;
- deactivate = 0;
- }
/*
* Bring in the user page that we will copy from _first_.
@@ -2664,8 +2626,7 @@
unlock:
/* Mark it unlocked again and drop the page.. */
UnlockPage(page);
- if (deactivate)
- deactivate_page(page);
+ check_used_once(page);
page_cache_release(page);
if (status < 0)
--- ../2.4.7.clean/mm/swap.c Mon Jan 22 22:30:21 2001
+++ ./mm/swap.c Thu Jul 26 15:59:30 2001
@@ -231,11 +231,8 @@
spin_lock(&pagemap_lru_lock);
if (!PageLocked(page))
BUG();
- DEBUG_ADD_PAGE
- add_page_to_active_list(page);
- /* This should be relatively rare */
- if (!page->age)
- deactivate_page_nolock(page);
+ add_page_to_inactive_dirty_list(page);
+ page->age = 0;
spin_unlock(&pagemap_lru_lock);
}
--- ../2.4.7.clean/mm/vmscan.c Mon Jul 9 19:18:50 2001
+++ ./mm/vmscan.c Thu Jul 26 15:59:30 2001
@@ -355,10 +355,10 @@
}
/* Page is or was in use? Move it to the active list. */
- if (PageReferenced(page) || page->age > 0 ||
- (!page->buffers && page_count(page) > 1)) {
+ if (PageReferenced(page) || (!page->buffers && page_count(page) > 1)) {
del_page_from_inactive_clean_list(page);
add_page_to_active_list(page);
+ page->age = PAGE_AGE_START;
continue;
}
@@ -453,11 +453,11 @@
}
/* Page is or was in use? Move it to the active list. */
- if (PageReferenced(page) || page->age > 0 ||
- (!page->buffers && page_count(page) > 1) ||
+ if (PageReferenced(page) || (!page->buffers && page_count(page) > 1) ||
page_ramdisk(page)) {
del_page_from_inactive_dirty_list(page);
add_page_to_active_list(page);
+ page->age = PAGE_AGE_START;
continue;
}