Xen devel - Mar 2012 - [PATCH] Mem event: don't leave zombie domains if there are wait-queued vcpus

xen/arch/x86/mm/mem_event.c |  19 ++++++++++++++++---
 1 files changed, 16 insertions(+), 3 deletions(-)


Vcpus in wait queues retain a domain reference. Upon domain destruction, we
were not taking care of draining the wait queues.

Signed-off-by: Andres Lagar-Cavilla <andres@lagarcavilla.org>

diff -r 9e44b357d988 -r 06a927b592df xen/arch/x86/mm/mem_event.c
--- a/xen/arch/x86/mm/mem_event.c
+++ b/xen/arch/x86/mm/mem_event.c
@@ -490,12 +490,25 @@ int do_mem_event_op(int op, uint32_t dom
 /* Clean up on domain destruction */
 void mem_event_cleanup(struct domain *d)
 {
-    if ( d->mem_event->paging.ring_page )
+    if ( d->mem_event->paging.ring_page ) {
+        /* Destroying the wait queue head means waking up all
+         * queued vcpus. This will drain the list, allowing
+         * the disable routine to complete. It will also drop
+         * all domain refs the wait-queued vcpus are holding.
+         * Finally, because this code path involves previously
+         * pausing the domain (domain_kill), unpausing the 
+         * vcpus causes no harm. */
+        destroy_waitqueue_head(&d->mem_event->paging.wq);
         (void)mem_event_disable(d, &d->mem_event->paging);
-    if ( d->mem_event->access.ring_page )
+    }
+    if ( d->mem_event->access.ring_page ) {
+        destroy_waitqueue_head(&d->mem_event->access.wq);
         (void)mem_event_disable(d, &d->mem_event->access);
-    if ( d->mem_event->share.ring_page )
+    }
+    if ( d->mem_event->share.ring_page ) {
+        destroy_waitqueue_head(&d->mem_event->share.wq);
         (void)mem_event_disable(d, &d->mem_event->share);
+    }
 }
 
 int mem_event_domctl(struct domain *d, xen_domctl_mem_event_op_t *mec,

At 11:17 -0400 on 12 Mar (1331551026), Andres Lagar-Cavilla
wrote:
>  xen/arch/x86/mm/mem_event.c |  19 ++++++++++++++++---
>  1 files changed, 16 insertions(+), 3 deletions(-)
> 
> 
> Vcpus in wait queues retain a domain reference. Upon domain destruction, we
> were not taking care of draining the wait queues.
Applied, thanks.

Tim.