Skip to contentSkip to conversation

RHEL3 2.4 domU kernel has load average of 2

PATCH Applied John Byrne 2 participants 2 messages Feb 23, 2007 – Feb 26, 2007 Xen hypervisor interface

lore
Share & tools
lore

John Byrne

author
#0 lore raw

PATCH 1/1 RHEL3 2.4 domU kernel has load average of 2

The backport of the xenbus code for the 2.4 kernel changed wait_event_interruptible() into wait_event() into a couple of places it looks to deal with the fact that signals weren't being blocked. This results in an idle domU showing a load-average of 2, which was messing up some work-management tools.

The attached patch blocks all signals for the xenbus and xenwatch daemons, instead.

drivers/xen/xenbus/xenbus_comms.c +1 -1
diff -Nar -U4 linux-2.4.21/drivers/xen/xenbus/xenbus_comms.c linux-2.4.21/drivers/xen/xenbus/xenbus_comms.c
@@ -149,9 +149,9 @@
 	while (len != 0) {
 		unsigned int avail;
 		const char *src;
 
-		wait_event/*_interruptible*/(xb_waitq,
+		wait_event_interruptible(xb_waitq,
 					 intf->rsp_cons != intf->rsp_prod);
 
 		/* Read indexes, then verify. */
 		cons = intf->rsp_cons;
drivers/xen/xenbus/xenbus_xs.c +17 -3
diff -Nar -U4 linux-2.4.21/drivers/xen/xenbus/xenbus_xs.c linux-2.4.21/drivers/xen/xenbus/xenbus_xs.c
@@ -45,8 +45,21 @@
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
 #include <linux/module.h>
 #include <linux/uio.h>
 #define kvec iovec
+
+static void block_signals(void)
+{
+	sigset_t blocked;
+
+	sigfillset(&blocked);
+	spin_lock_irq(&current->sighand->siglock);
+	current->blocked = blocked;
+	recalc_sigpending();
+	spin_unlock_irq(&current->sighand->siglock);
+	flush_signals(current);
+}
+
 #endif
 
 /* xenbus_probe.c */
 extern char *kasprintf(const char *fmt, ...);
@@ -712,8 +722,9 @@
 	struct xs_stored_msg *msg;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
 	daemonize();
+	block_signals();
 	memcpy(current->comm, "xenwatch", 10);
 #endif
 
 
@@ -724,11 +735,10 @@
 		schedule();
 #endif
 
 	for (;;) {
-		/* XXXAP _interruptable spins here after killall */
-		wait_event(watch_events_waitq,
-			   !list_empty(&watch_events));
+		wait_event_interruptible(watch_events_waitq,
+					 !list_empty(&watch_events));
 
 #if 0				/* XXXAP 2.4 */
 		if (kthread_should_stop())
 			break;
@@ -827,8 +837,9 @@
 	int err;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)
 	daemonize();
+	block_signals();
 	memcpy(current->comm, "xenbus", 8);
 #endif
 
 #ifdef CONFIG_HOTPLUG_CPU

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

Applied
#1 lore raw
Quoted 8 lines · John Byrne

The backport of the xenbus code for the 2.4 kernel changed wait_event_interruptible() into wait_event() into a couple of places it looks to deal with the fact that signals weren't being blocked. This results in an idle domU showing a load-average of 2, which was messing up some work-management tools.

The attached patch blocks all signals for the xenbus and xenwatch daemons, instead.

Applied, thanks.

Ian.