Adjust the critical section protecting _xen_flush_queue to cover the

entire range where the page mapping request queue needs to be atomically
examined and modified.

Oddly, while this doesn't seem to affect the overall rate of panics
(running 'make index' on EC2 t1.micro instances, there are 0.6 +/- 0.1
panics per hour, both before and after this change), it eliminates
vm_fault from panic backtraces, leaving only backtraces going through
vmspace_fork.

1 files changed, 11 insertions, 4 deletions

diff --git a/sys/i386/xen/xen_machdep.c b/sys/i386/xen/xen_machdep.c
index a9180fc..2a3eade 100644
--- a/sys/i386/xen/xen_machdep.c
+++ b/sys/i386/xen/xen_machdep.c
@@ -42,6 +42,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/malloc.h>
 #include <sys/mutex.h>
 #include <sys/kernel.h>
+#include <sys/proc.h>
 #include <sys/reboot.h>
 #include <sys/sysproto.h>
 
@@ -249,10 +250,13 @@ _xen_flush_queue(void)
 	SET_VCPU();
 	int _xpq_idx = XPQ_IDX;
 	int error, i;
-	/* window of vulnerability here? */
 
+#ifdef INVARIANTS
 	if (__predict_true(gdtset))
-		critical_enter();
+		KASSERT(curthread->td_critnest > 0,
+		    ("xen queue flush should be in a critical section"));
+#endif
+
 	XPQ_IDX = 0;
 	/* Make sure index is cleared first to avoid double updates. */
 	error = HYPERVISOR_mmu_update((mmu_update_t *)&XPQ_QUEUE,
@@ -286,8 +290,6 @@ _xen_flush_queue(void)
 		}
 	}
 #endif	
-	if (__predict_true(gdtset))
-		critical_exit();
 	if (__predict_false(error < 0)) {
 		for (i = 0; i < _xpq_idx; i++)
 			printf("val: %llx ptr: %llx\n",
@@ -301,7 +303,12 @@ void
 xen_flush_queue(void)
 {
 	SET_VCPU();
+
+	if (__predict_true(gdtset))
+		critical_enter();
 	if (XPQ_IDX != 0) _xen_flush_queue();
+	if (__predict_true(gdtset))
+		critical_exit();
 }
 
 static __inline void