Skip to contentSkip to conversation

ppc64: Fix iseries_veth module unload race and memory leak

PATCH New Michael Ellerman 1 participant 1 message Jan 11, 2005 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 2/2 ppc64: Fix iseries_veth module unload race and memory leak

Hi All,

When the iseries_veth driver module is unloaded there is the potential for an oops and also some memory leakage.

Because the HvLpEvent_unregisterHandler() function did no synchronisation, it was possible for the handler that was being unregistered to be running on another CPU *after* HvLpEvent_unregisterHandler() had returned. This could cause the iseries_veth driver to leave work in the events work queue after the module had been unloaded. When that work was eventually executed we got an oops.

In addition some of the data structures in the iseries_veth driver were not being correctly freed when the module was unloaded.

This is the second patch, we make iseries_veth call flush_scheduled_work() after we are sure the handler is no longer running, and also fix the memory leaks.

 iseries_veth.c |   26 ++++++++++++++++++++++----
 1 files changed, 22 insertions(+), 4 deletions(-)
drivers/net/iseries_veth.c +22 -4
diff -urN 2.6.10-ppc64-stock/drivers/net/iseries_veth.c 2.6.10-ppc64-work/drivers/net/iseries_veth.c
@@ -642,7 +642,7 @@
 	return 0;
 }
 
-static void veth_destroy_connection(u8 rlp)
+static void veth_stop_connection(u8 rlp)
 {
 	struct veth_lpar_connection *cnx = veth_cnx[rlp];
 
@@ -671,9 +671,18 @@
 				      HvLpEvent_Type_VirtualLan,
 				      cnx->num_ack_events,
 				      NULL, NULL);
+}
+
+static void veth_destroy_connection(u8 rlp)
+{
+	struct veth_lpar_connection *cnx = veth_cnx[rlp];
 
-	if (cnx->msgs)
-		kfree(cnx->msgs);
+	if (! cnx)
+		return;
+
+	kfree(cnx->msgs);
+	kfree(cnx);
+	veth_cnx[rlp] = NULL;
 }
 
 /*
@@ -1375,9 +1384,18 @@
 	vio_unregister_driver(&veth_driver);
 
 	for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
-		veth_destroy_connection(i);
+		veth_stop_connection(i);
 
 	HvLpEvent_unregisterHandler(HvLpEvent_Type_VirtualLan);
+
+	/* Hypervisor callbacks may have scheduled more work while we
+	 * were destroying connections. Now that we've disconnected from
+	 * the hypervisor make sure everything's finished. */
+	flush_scheduled_work();
+
+	for (i = 0; i < HVMAXARCHITECTEDLPS; ++i)
+		veth_destroy_connection(i);
+	
 }
 module_exit(veth_module_cleanup);