Re: [PATCH][8/8] mm: lru interface change

From: Nick Piggin
Date: Tue Mar 21 2006 - 07:50:21 EST


Stone Wang wrote:
Add Wired list to LRU.
Add PG_Wired bit to page.


I may have missed something very trivial, but... why are they on a
list at all if they don't get scanned?


diff -urN linux-2.6.15.orig/mm/swap.c linux-2.6.15/mm/swap.c
--- linux-2.6.15.orig/mm/swap.c 2006-01-02 22:21:10.000000000 -0500
+++ linux-2.6.15/mm/swap.c 2006-03-07 11:45:37.000000000 -0500
@@ -110,6 +110,44 @@
spin_unlock_irq(&zone->lru_lock);
}

+/* Wire the page; if the page is in LRU,
+ * try move it to Wired list.
+ */
+void fastcall wire_page(struct page *page)

Ahh, here's the elusive beast.

+{
+ struct zone *zone = page_zone(page);
+
+ spin_lock_irq(&zone->lru_lock);

This is a fairly heavy lock for something which is going into page fault
fastpaths and such. You might be better off making wired_count atomic and
not using a lock in the common case if possible (even if it is only for
mlocked pages, I'm sure someone will complain).

+ page->wired_count ++;

Oh dear, I missed this change you made to struct page, tucked away in 5/8.
This alone pretty much makes it a showstopper, I'm afraid. You'll have to
work out some other way to do it so as not to penalise 99.999% of machines
which don't need this.

(Oh, and making the field a short usually won't help either, because of
alignment constraints).

+ if(!PageWired(page)){
+ if(PageLRU(page)){
+ del_page_from_lru(zone, page);
+ add_page_to_wired_list(zone,page);
+ SetPageWired(page);
+ }
+ }
+ spin_unlock_irq(&zone->lru_lock);
+}
+
+/* Unwire the page.
+ * If it isnt wired by any process, try move it to active list.
+ */
+void fastcall unwire_page(struct page *page)
+{
+ struct zone *zone = page_zone(page);
+
+ spin_lock_irq(&zone->lru_lock);
+ page->wired_count --;
+ if(!page->wired_count){
+ if(PageLRU(page) && TestClearPageWired(page)){
+ del_page_from_wired_list(zone,page);
+ add_page_to_active_list(zone,page);
+ SetPageActive(page);
+ }
+ }
+ spin_unlock_irq(&zone->lru_lock);
+}
+
/*
* Mark a page as having seen activity.
*
@@ -119,11 +157,13 @@
*/
void fastcall mark_page_accessed(struct page *page)
{
- if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
- activate_page(page);
- ClearPageReferenced(page);
- } else if (!PageReferenced(page)) {
- SetPageReferenced(page);
+ if(!PageWired(page)) {
+ if (!PageActive(page) && PageReferenced(page) && PageLRU(page)) {
+ activate_page(page);
+ ClearPageReferenced(page);
+ } else if (!PageReferenced(page)) {
+ SetPageReferenced(page);
+ }
}
}


So, umm... what happens when the page gets wired before activate_page()?

@@ -178,13 +218,15 @@
struct zone *zone = page_zone(page);

spin_lock_irqsave(&zone->lru_lock, flags);
- if (TestClearPageLRU(page))
- del_page_from_lru(zone, page);
- if (page_count(page) != 0)
- page = NULL;
+ if(!PageWired(page)) {
+ if (TestClearPageLRU(page))
+ del_page_from_lru(zone, page);
+ if (page_count(page) != 0)
+ page = NULL;
+ if (page)
+ free_hot_page(page);
+ }
spin_unlock_irqrestore(&zone->lru_lock, flags);
- if (page)
- free_hot_page(page);
}


Hmm... how do PageWired pages get freed, then?

EXPORT_SYMBOL(__page_cache_release);
@@ -214,7 +256,8 @@

if (!put_page_testzero(page))
continue;
-
+ if(PageWired(page))
+ continue;
pagezone = page_zone(page);
if (pagezone != zone) {
if (zone)

Ditto.

--
SUSE Labs, Novell Inc.


Send instant messages to your online friends http://au.messenger.yahoo.com -
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/