IB/multicast: Report errors on multicast groups if P_key changes

P_key changes can invalidate multicast groups.  Report errors on all
multicast groups affected by a pkey change.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 45 insertions, 10 deletions

diff --git a/drivers/infiniband/core/multicast.c b/drivers/infiniband/core/multicast.c
index 1bc1fe60528..107f170c57c 100644
--- a/drivers/infiniband/core/multicast.c
+++ b/drivers/infiniband/core/multicast.c
@@ -73,11 +73,20 @@ struct mcast_device {
 };
 
 enum mcast_state {
-        MCAST_IDLE,
         MCAST_JOINING,
         MCAST_MEMBER,
+        MCAST_ERROR,
+};
+
+enum mcast_group_state {
+        MCAST_IDLE,
         MCAST_BUSY,
-        MCAST_ERROR
+        MCAST_GROUP_ERROR,
+        MCAST_PKEY_EVENT
+};
+
+enum {
+        MCAST_INVALID_PKEY_INDEX = 0xFFFF
 };
 
 struct mcast_member;
@@ -93,9 +102,10 @@ struct mcast_group {
         struct mcast_member     *last_join;
         int                     members[3];
         atomic_t                refcount;
-        enum mcast_state        state;
+        enum mcast_group_state  state;
         struct ib_sa_query      *query;
         int                     query_id;
+        u16                     pkey_index;
 };
 
 struct mcast_member {
@@ -378,9 +388,19 @@ static int fail_join(struct mcast_group *group, struct mcast_member *member,
 static void process_group_error(struct mcast_group *group)
 {
         struct mcast_member *member;
-        int ret;
+        int ret = 0;
+        u16 pkey_index;
+
+        if (group->state == MCAST_PKEY_EVENT)
+                ret = ib_find_pkey(group->port->dev->device,
+                                   group->port->port_num,
+                                   be16_to_cpu(group->rec.pkey), &pkey_index);
 
         spin_lock_irq(&group->lock);
+        if (group->state == MCAST_PKEY_EVENT && !ret &&
+            group->pkey_index == pkey_index)
+                goto out;
+
         while (!list_empty(&group->active_list)) {
                 member = list_entry(group->active_list.next,
                                     struct mcast_member, list);
@@ -399,6 +419,7 @@ static void process_group_error(struct mcast_group *group)
         }
 
         group->rec.join_state = 0;
+out:
         group->state = MCAST_BUSY;
         spin_unlock_irq(&group->lock);
 }
@@ -415,9 +436,9 @@ static void mcast_work_handler(struct work_struct *work)
 retest:
         spin_lock_irq(&group->lock);
         while (!list_empty(&group->pending_list) ||
-               (group->state == MCAST_ERROR)) {
+               (group->state != MCAST_BUSY)) {
 
-                if (group->state == MCAST_ERROR) {
+                if (group->state != MCAST_BUSY) {
                         spin_unlock_irq(&group->lock);
                         process_group_error(group);
                         goto retest;
@@ -494,12 +515,19 @@ static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
                          void *context)
 {
         struct mcast_group *group = context;
+        u16 pkey_index = MCAST_INVALID_PKEY_INDEX;
 
         if (status)
                 process_join_error(group, status);
         else {
+                ib_find_pkey(group->port->dev->device, group->port->port_num,
+                             be16_to_cpu(rec->pkey), &pkey_index);
+
                 spin_lock_irq(&group->port->lock);
                 group->rec = *rec;
+                if (group->state == MCAST_BUSY &&
+                    group->pkey_index == MCAST_INVALID_PKEY_INDEX)
+                        group->pkey_index = pkey_index;
                 if (!memcmp(&mgid0, &group->rec.mgid, sizeof mgid0)) {
                         rb_erase(&group->node, &group->port->table);
                         mcast_insert(group->port, group, 1);
@@ -539,6 +567,7 @@ static struct mcast_group *acquire_group(struct mcast_port *port,
 
         group->port = port;
         group->rec.mgid = *mgid;
+        group->pkey_index = MCAST_INVALID_PKEY_INDEX;
         INIT_LIST_HEAD(&group->pending_list);
         INIT_LIST_HEAD(&group->active_list);
         INIT_WORK(&group->work, mcast_work_handler);
@@ -707,7 +736,8 @@ int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num,
 }
 EXPORT_SYMBOL(ib_init_ah_from_mcmember);
 
-static void mcast_groups_lost(struct mcast_port *port)
+static void mcast_groups_event(struct mcast_port *port,
+                               enum mcast_group_state state)
 {
         struct mcast_group *group;
         struct rb_node *node;
@@ -721,7 +751,8 @@ static void mcast_groups_lost(struct mcast_port *port)
                         atomic_inc(&group->refcount);
                         queue_work(mcast_wq, &group->work);
                 }
-                group->state = MCAST_ERROR;
+                if (group->state != MCAST_GROUP_ERROR)
+                        group->state = state;
                 spin_unlock(&group->lock);
         }
         spin_unlock_irqrestore(&port->lock, flags);
@@ -731,16 +762,20 @@ static void mcast_event_handler(struct ib_event_handler *handler,
                                 struct ib_event *event)
 {
         struct mcast_device *dev;
+        int index;
 
         dev = container_of(handler, struct mcast_device, event_handler);
+        index = event->element.port_num - dev->start_port;
 
         switch (event->event) {
         case IB_EVENT_PORT_ERR:
         case IB_EVENT_LID_CHANGE:
         case IB_EVENT_SM_CHANGE:
         case IB_EVENT_CLIENT_REREGISTER:
-                mcast_groups_lost(&dev->port[event->element.port_num -
-                                             dev->start_port]);
+                mcast_groups_event(&dev->port[index], MCAST_GROUP_ERROR);
+                break;
+        case IB_EVENT_PKEY_CHANGE:
+                mcast_groups_event(&dev->port[index], MCAST_PKEY_EVENT);
                 break;
         default:
                 break;