Ian Jackson
2010-Oct-12 14:35 UTC
[Xen-devel] [BROKEN PATCH] Work around pvops lost event channel bug
Our automated tests are failing to push because, with some
probability, a lost event channel event makes xenconsoled, and the
domU get stuck.
We think this is going to be fixed by
http://xenbits.xen.org/gitweb?p=people/ianc/linux-2.6.git;a=commit;h=5d30cb2a85912ffb5f6556d55472c26801eef2ea
http://lists.xensource.com/archives/html/xen-devel/2010-10/msg00370.html
But here is a workaround (against xen-4.0-testing) that in my tests
seemed to make dom0 xenconsoled unwedge.
Ian.
diff -r c32e9163328d tools/console/daemon/io.c
--- a/tools/console/daemon/io.c Mon Oct 04 12:53:03 2010 +0100
+++ b/tools/console/daemon/io.c Tue Oct 12 15:28:36 2010 +0100
@@ -891,8 +891,9 @@ static void handle_ring_read(struct doma
if (dom->is_dead)
return;
- if ((port = xc_evtchn_pending(dom->xce_handle)) == -1)
- return;
+// if ((port = xc_evtchn_pending(dom->xce_handle)) == -1)
+// return;
+ port = dom->local_port;
dom->event_count++;
@@ -975,6 +976,7 @@ void handle_io(void)
{
fd_set readfds, writefds;
int ret;
+ static struct timeval abs_max_timeout = { 10, 0 };
if (log_hv) {
xc_handle = xc_interface_open();
@@ -1076,7 +1078,7 @@ void handle_io(void)
}
ret = select(max_fd + 1, &readfds, &writefds, 0,
- next_timeout ? &timeout : NULL);
+ next_timeout ? &timeout : &abs_max_timeout);
if (log_reload) {
handle_log_reload();
@@ -1105,9 +1107,7 @@ void handle_io(void)
for (d = dom_head; d; d = n) {
n = d->next;
if (d->event_count < RATE_LIMIT_ALLOWANCE) {
- if (d->xce_handle != -1 &&
- FD_ISSET(xc_evtchn_fd(d->xce_handle),
- &readfds))
+ if (d->xce_handle != -1)
handle_ring_read(d);
}
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel