RDMA/cxgb4: Don't change QP state outside EP lock

Concurrent ingress CLOSE and ULP ABORT operations causes a crash due
to a race condition where the close path releases the EP lock and then
tries to move the QP state to CLOSED.  This must be done inside the EP
lock to avoid the race.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>

3 files changed, 9 insertions, 12 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index 9d8dcfab2b38..d235810e52df 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -1466,7 +1466,7 @@ static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
         struct c4iw_qp_attributes attrs;
         int disconnect = 1;
         int release = 0;
-        int closing = 0;
+        int abort = 0;
         struct tid_info *t = dev->rdev.lldi.tids;
         unsigned int tid = GET_TID(hdr);
 
@@ -1507,8 +1507,11 @@ static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
         case FPDU_MODE:
                 start_ep_timer(ep);
                 __state_set(&ep->com, CLOSING);
-                closing = 1;
+                attrs.next_state = C4IW_QP_STATE_CLOSING;
+                abort = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+                                       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
                 peer_close_upcall(ep);
+                disconnect = 1;
                 break;
         case ABORTING:
                 disconnect = 0;
@@ -1536,11 +1539,6 @@ static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
                 BUG_ON(1);
         }
         mutex_unlock(&ep->com.mutex);
-        if (closing) {
-                attrs.next_state = C4IW_QP_STATE_CLOSING;
-                c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
-                               C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
-        }
         if (disconnect)
                 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
         if (release)
@@ -1710,14 +1708,14 @@ static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
         ep = lookup_tid(t, tid);
         BUG_ON(!ep);
 
-        if (ep->com.qp) {
+        if (ep && ep->com.qp) {
                 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
                        ep->com.qp->wq.sq.qid);
                 attrs.next_state = C4IW_QP_STATE_TERMINATE;
                 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
                                C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
         } else
-                printk(KERN_WARNING MOD "TERM received tid %u no qp\n", tid);
+                printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
 
         return 0;
 }
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index 9f6166f59268..8e16eb2de91f 100644
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -161,8 +161,8 @@ static inline int c4iw_wait_for_reply(struct c4iw_rdev *rdev,
                 }
         } while (!wr_waitp->done);
         if (wr_waitp->ret)
-                printk(KERN_WARNING MOD "%s: FW reply %d tid %u qpid %u\n",
-                       pci_name(rdev->lldi.pdev), wr_waitp->ret, hwtid, qpid);
+                PDBG("%s: FW reply %d tid %u qpid %u\n",
+                     pci_name(rdev->lldi.pdev), wr_waitp->ret, hwtid, qpid);
         return wr_waitp->ret;
 }
 
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index 70a5a3c646da..a1824a5f3d76 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -1210,7 +1210,6 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
                         if (ret) {
                                 if (internal)
                                         c4iw_get_ep(&qhp->ep->com);
-                                disconnect = abort = 1;
                                 goto err;
                         }
                         break;