From Gerd Knorr <kraxel@suse.de>:

Watch events may get lost, the xenstored response races against the new
watch being added to the watch list in register_xenbus_watch().  Fixed
by adding the watch to the list unconditionally, and remove it in the
error case.

That one fixes the network issues I'm seeing.  Due to the race the
"online" hotplug event for the vif backend was lost sometimes.

Signed-off-by: Ewan Mellor <ewan@xensource.com>

diff --git a/linux-2.6-xen-sparse/drivers/xen/xenbus/xenbus_xs.c b/linux-2.6-xen-sparse/drivers/xen/xenbus/xenbus_xs.c
index d55037177ba9ffccc0ebcccb5a9bf79b5256955c..55ba4b41ad51d68a4eaf5c18d1344f9ab07867ef 100644 (file)
--- a/linux-2.6-xen-sparse/drivers/xen/xenbus/xenbus_xs.c
+++ b/linux-2.6-xen-sparse/drivers/xen/xenbus/xenbus_xs.c
@@ -682,14 +682,15 @@ int register_xenbus_watch(struct xenbus_watch *watch)
 
        spin_lock(&watches_lock);
        BUG_ON(find_watch(token));
+       list_add(&watch->list, &watches);
        spin_unlock(&watches_lock);
 
        err = xs_watch(watch->node, token);
 
        /* Ignore errors due to multiple registration. */
-       if ((err == 0) || (err == -EEXIST)) {
+       if ((err != 0) && (err != -EEXIST)) {
                spin_lock(&watches_lock);
-               list_add(&watch->list, &watches);
+               list_del(&watch->list);
                spin_unlock(&watches_lock);
        }