[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH][RFC] Serialize scrubbing pages



Hi,

I noticed that parallel scrubbing pages is not efficient
due to spinlock contention. (also memory bandwidth?)
Actually heap_lock becomes a bottleneck.

In my investigation, it often takes 1 millisec just to acquire
the lock on ia64 with 8cpus. It's very wasteful.

For example, creating a domain is too slow while scrubbing pages.

# xm create vm memory=4000
# xm destroy vm; time xm create vm memory=4000
real    0m6.083s
user    0m0.024s
sys     0m0.008s

After the attached patch is applied,

# xm destroy vm; time xm create vm memory=4000
real    0m1.463s
user    0m0.132s
sys     0m0.028s

This workaround is an easy solution but ugly.
There must be another good way. 

Thanks,
Kouya

Signed-off-by: Kouya Shimura <kouya@xxxxxxxxxxxxxx>

diff -r 4e3316ed1af5 xen/common/page_alloc.c
--- a/xen/common/page_alloc.c   Fri Aug 08 15:02:19 2008 +0100
+++ b/xen/common/page_alloc.c   Mon Aug 11 14:19:21 2008 +0900
@@ -70,6 +70,7 @@ integer_param("dma_bits", dma_bitsize);
 #define scrub_page(p) clear_page(p)
 #endif
 
+static DEFINE_SPINLOCK(page_scrub_mutex);
 static DEFINE_SPINLOCK(page_scrub_lock);
 LIST_HEAD(page_scrub_list);
 static unsigned long scrub_pages;
@@ -951,6 +952,11 @@ static void page_scrub_softirq(void)
     int               i;
     s_time_t          start = NOW();
 
+    if (!spin_trylock(&page_scrub_mutex)) {
+        set_timer(&this_cpu(page_scrub_timer), NOW() + MILLISECS(1));
+        return;
+    }
+
     /* Aim to do 1ms of work every 10ms. */
     do {
         spin_lock(&page_scrub_lock);
@@ -958,6 +964,7 @@ static void page_scrub_softirq(void)
         if ( unlikely((ent = page_scrub_list.next) == &page_scrub_list) )
         {
             spin_unlock(&page_scrub_lock);
+            spin_unlock(&page_scrub_mutex);
             return;
         }
         
@@ -987,6 +994,8 @@ static void page_scrub_softirq(void)
             free_heap_pages(pfn_dom_zone_type(page_to_mfn(pg)), pg, 0);
         }
     } while ( (NOW() - start) < MILLISECS(1) );
+
+    spin_unlock(&page_scrub_mutex);
 
     set_timer(&this_cpu(page_scrub_timer), NOW() + MILLISECS(10));
 }
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.