b=23575 O2iblnd credit deadlock regression

This fixed a regression of bug 14425.

i=liang

diff --git a/lnet/klnds/o2iblnd/o2iblnd.c b/lnet/klnds/o2iblnd/o2iblnd.c
index 5a9c93a..ab2803e 100644 (file)
--- a/lnet/klnds/o2iblnd/o2iblnd.c
+++ b/lnet/klnds/o2iblnd/o2iblnd.c
@@ -621,6 +621,10 @@ kiblnd_debug_conn (kib_conn_t *conn)
         list_for_each(tmp, &conn->ibc_early_rxs)
                 kiblnd_debug_rx(list_entry(tmp, kib_rx_t, rx_list));
 
+        CDEBUG(D_CONSOLE, "   tx_noops:\n");
+        list_for_each(tmp, &conn->ibc_tx_noops)
+                kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list));
+
         CDEBUG(D_CONSOLE, "   tx_queue_nocred:\n");
         list_for_each(tmp, &conn->ibc_tx_queue_nocred)
                 kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list));
@@ -730,6 +734,7 @@ kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
         conn->ibc_cmid = cmid;
 
         INIT_LIST_HEAD(&conn->ibc_early_rxs);
+        INIT_LIST_HEAD(&conn->ibc_tx_noops);
         INIT_LIST_HEAD(&conn->ibc_tx_queue);
         INIT_LIST_HEAD(&conn->ibc_tx_queue_rsrvd);
         INIT_LIST_HEAD(&conn->ibc_tx_queue_nocred);
@@ -866,6 +871,7 @@ kiblnd_destroy_conn (kib_conn_t *conn)
         LASSERT (!in_interrupt());
         LASSERT (atomic_read(&conn->ibc_refcount) == 0);
         LASSERT (list_empty(&conn->ibc_early_rxs));
+        LASSERT (list_empty(&conn->ibc_tx_noops));
         LASSERT (list_empty(&conn->ibc_tx_queue));
         LASSERT (list_empty(&conn->ibc_tx_queue_rsrvd));
         LASSERT (list_empty(&conn->ibc_tx_queue_nocred));

diff --git a/lnet/klnds/o2iblnd/o2iblnd.h b/lnet/klnds/o2iblnd/o2iblnd.h
index 5989a74..a9b0d7e 100644 (file)
--- a/lnet/klnds/o2iblnd/o2iblnd.h
+++ b/lnet/klnds/o2iblnd/o2iblnd.h
@@ -548,6 +548,7 @@ typedef struct kib_conn
         int                 ibc_ready:1;        /* CQ callback fired */
         unsigned long       ibc_last_send;      /* time of last send */
         struct list_head    ibc_early_rxs;      /* rxs completed before ESTABLISHED */
+        struct list_head    ibc_tx_noops;       /* IBLND_MSG_NOOPs for IBLND_MSG_VERSION_1 */
         struct list_head    ibc_tx_queue;       /* sends that need a credit */
         struct list_head    ibc_tx_queue_nocred;/* sends that don't need a credit */
         struct list_head    ibc_tx_queue_rsrvd; /* sends that need to reserve an ACK/DONE msg */
@@ -672,14 +673,25 @@ kiblnd_send_noop(kib_conn_t *conn)
             !kiblnd_send_keepalive(conn))
                 return 0; /* No need to send NOOP */
 
-        if (!list_empty(&conn->ibc_tx_queue_nocred))
-                return 0; /* NOOP can be piggybacked */
+        if (IBLND_OOB_CAPABLE(conn->ibc_version)) {
+                if (!list_empty(&conn->ibc_tx_queue_nocred))
+                        return 0; /* NOOP can be piggybacked */
 
-        if (!IBLND_OOB_CAPABLE(conn->ibc_version))
-                return list_empty(&conn->ibc_tx_queue); /* can't piggyback? */
+                /* No tx to piggyback NOOP onto or no credit to send a tx */
+                return (list_empty(&conn->ibc_tx_queue) || conn->ibc_credits == 0);
+        }
+
+        if (!list_empty(&conn->ibc_tx_noops) ||       /* NOOP already queued */
+            !list_empty(&conn->ibc_tx_queue_nocred) || /* can be piggybacked */
+            conn->ibc_credits == 0)                    /* no credit */
+                return 0;
+
+        if (conn->ibc_credits == 1 &&      /* last credit reserved for */
+            conn->ibc_outstanding_credits == 0) /* giving back credits */
+                return 0;
 
         /* No tx to piggyback NOOP onto or no credit to send a tx */
-        return (list_empty(&conn->ibc_tx_queue) || conn->ibc_credits == 0);
+        return (list_empty(&conn->ibc_tx_queue) || conn->ibc_credits == 1);
 }
 
 static inline void

diff --git a/lnet/klnds/o2iblnd/o2iblnd_cb.c b/lnet/klnds/o2iblnd/o2iblnd_cb.c
index f4a02dd..ec5ce4f 100644 (file)
--- a/lnet/klnds/o2iblnd/o2iblnd_cb.c
+++ b/lnet/klnds/o2iblnd/o2iblnd_cb.c
@@ -326,6 +326,11 @@ kiblnd_handle_rx (kib_rx_t *rx)
 
                 conn->ibc_credits += credits;
 
+                /* This ensures the credit taken by NOOP can be returned */
+                if (msg->ibm_type == IBLND_MSG_NOOP &&
+                    !IBLND_OOB_CAPABLE(conn->ibc_version)) /* v1 only */
+                        conn->ibc_outstanding_credits++;
+
                 spin_unlock(&conn->ibc_lock);
                 kiblnd_check_sends(conn);
         }
@@ -339,9 +344,14 @@ kiblnd_handle_rx (kib_rx_t *rx)
                 break;
 
         case IBLND_MSG_NOOP:
-                if (IBLND_OOB_CAPABLE(conn->ibc_version))
+                if (IBLND_OOB_CAPABLE(conn->ibc_version)) {
                         post_credit = IBLND_POSTRX_NO_CREDIT;
-                else
+                        break;
+                }
+
+                if (credits != 0) /* credit already posted */
+                        post_credit = IBLND_POSTRX_NO_CREDIT;
+                else              /* a keepalive NOOP */
                         post_credit = IBLND_POSTRX_PEER_CREDIT;
                 break;
 
@@ -786,8 +796,8 @@ kiblnd_post_tx_locked (kib_conn_t *conn, kib_tx_t *tx, int credit)
         }
 
         if (credit != 0 && !IBLND_OOB_CAPABLE(ver) &&
-            conn->ibc_credits == 1 &&   /* last credit reserved for */
-            conn->ibc_outstanding_credits == 0) { /* giving back credits */
+            conn->ibc_credits == 1 &&   /* last credit reserved */
+            msg->ibm_type != IBLND_MSG_NOOP) {      /* for NOOP */
                 CDEBUG(D_NET, "%s: not using last credit\n",
                        libcfs_nid2str(peer->ibp_nid));
                 return -EAGAIN;
@@ -928,6 +938,11 @@ kiblnd_check_sends (kib_conn_t *conn)
                         credit = 0;
                         tx = list_entry(conn->ibc_tx_queue_nocred.next,
                                         kib_tx_t, tx_list);
+                } else if (!list_empty(&conn->ibc_tx_noops)) {
+                        LASSERT (!IBLND_OOB_CAPABLE(ver));
+                        credit = 1;
+                        tx = list_entry(conn->ibc_tx_noops.next,
+                                        kib_tx_t, tx_list);
                 } else if (!list_empty(&conn->ibc_tx_queue)) {
                         credit = 1;
                         tx = list_entry(conn->ibc_tx_queue.next,
@@ -1161,7 +1176,7 @@ kiblnd_queue_tx_locked (kib_tx_t *tx, kib_conn_t *conn)
                 if (IBLND_OOB_CAPABLE(conn->ibc_version))
                         q = &conn->ibc_tx_queue_nocred;
                 else
-                        q = &conn->ibc_tx_queue;
+                        q = &conn->ibc_tx_noops;
                 break;
 
         case IBLND_MSG_IMMEDIATE:
@@ -1775,6 +1790,7 @@ kiblnd_close_conn_locked (kib_conn_t *conn, int error)
                 return; /* already being handled  */
 
         if (error == 0 &&
+            list_empty(&conn->ibc_tx_noops) &&
             list_empty(&conn->ibc_tx_queue) &&
             list_empty(&conn->ibc_tx_queue_rsrvd) &&
             list_empty(&conn->ibc_tx_queue_nocred) &&
@@ -1782,9 +1798,10 @@ kiblnd_close_conn_locked (kib_conn_t *conn, int error)
                 CDEBUG(D_NET, "closing conn to %s\n",
                        libcfs_nid2str(peer->ibp_nid));
         } else {
-                CNETERR("Closing conn to %s: error %d%s%s%s%s\n",
+                CNETERR("Closing conn to %s: error %d%s%s%s%s%s\n",
                         libcfs_nid2str(peer->ibp_nid), error,
                         list_empty(&conn->ibc_tx_queue) ? "" : "(sending)",
+                        list_empty(&conn->ibc_tx_noops) ? "" : "(sending_noops)",
                         list_empty(&conn->ibc_tx_queue_rsrvd) ?
                                 "" : "(sending_rsrvd)",
                         list_empty(&conn->ibc_tx_queue_nocred) ?
@@ -1901,6 +1918,7 @@ kiblnd_finalise_conn (kib_conn_t *conn)
         /* Complete all tx descs not waiting for sends to complete.
          * NB we should be safe from RDMA now that the QP has changed state */
 
+        kiblnd_abort_txs(conn, &conn->ibc_tx_noops);
         kiblnd_abort_txs(conn, &conn->ibc_tx_queue);
         kiblnd_abort_txs(conn, &conn->ibc_tx_queue_rsrvd);
         kiblnd_abort_txs(conn, &conn->ibc_tx_queue_nocred);
@@ -2904,6 +2922,7 @@ int
 kiblnd_conn_timed_out (kib_conn_t *conn)
 {
         return  kiblnd_check_txs(conn, &conn->ibc_tx_queue) ||
+                kiblnd_check_txs(conn, &conn->ibc_tx_noops) ||
                 kiblnd_check_txs(conn, &conn->ibc_tx_queue_rsrvd) ||
                 kiblnd_check_txs(conn, &conn->ibc_tx_queue_nocred) ||
                 kiblnd_check_txs(conn, &conn->ibc_active_txs);