* vibnal fixes

    5807 - incorrect outstanding SEND accounting
    changed refcounting inlines to macros to improve logging
    cleanup unused per-work-item debug on send

*   lctl debug_file wasn't passing the mode argument to open(O_CREAT)
    fixed with restrictive (owner only) permissions

*   5803 ranal initialisation fix

diff --git a/lnet/klnds/ralnd/ralnd.c b/lnet/klnds/ralnd/ralnd.c
index 1aa3556..0da7af4 100644 (file)
--- a/lnet/klnds/ralnd/ralnd.c
+++ b/lnet/klnds/ralnd/ralnd.c
@@ -2057,6 +2057,7 @@ kranal_api_startup (nal_t *nal, ptl_pid_t requested_pid,
         }
         
         for (i = 0; i < kranal_data.kra_ndevs; i++) {
+                dev = &kranal_data.kra_devices[i];
                 rc = kranal_thread_start(kranal_scheduler, dev);
                 if (rc != 0) {
                         CERROR("Can't spawn ranal scheduler[%d]: %d\n",

diff --git a/lnet/klnds/viblnd/viblnd.h b/lnet/klnds/viblnd/viblnd.h
index 7a5e733..1026995 100644 (file)
--- a/lnet/klnds/viblnd/viblnd.h
+++ b/lnet/klnds/viblnd/viblnd.h
@@ -549,50 +549,48 @@ wrq_signals_completion (vv_wr_t *wrq)
         return wrq->completion_notification != 0;
 }
 
-static inline void
-kibnal_conn_addref (kib_conn_t *conn)
-{
-        CDEBUG(D_NET, "++conn[%p] (%d)\n",
-               conn, atomic_read(&conn->ibc_refcount));
-        LASSERT(atomic_read(&conn->ibc_refcount) > 0);
-        atomic_inc(&conn->ibc_refcount);
-}
-
-static inline void
-kibnal_conn_decref (kib_conn_t *conn)
-{
-        unsigned long   flags;
-
-        CDEBUG(D_NET, "--conn[%p] (%d)\n",
-               conn, atomic_read(&conn->ibc_refcount));
-        LASSERT(atomic_read(&conn->ibc_refcount) > 0);
-        if (atomic_dec_and_test(&conn->ibc_refcount)) {
-                spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
-                list_add_tail(&conn->ibc_list, &kibnal_data.kib_connd_zombies);
-                wake_up(&kibnal_data.kib_connd_waitq);
-                spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
-        }
-}
-
-static inline void
-kibnal_peer_addref (kib_peer_t *peer)
-{
-        CDEBUG(D_NET, "++peer[%p] -> "LPX64" (%d)\n",
-               peer, peer->ibp_nid, atomic_read (&peer->ibp_refcount));
-        LASSERT(atomic_read(&peer->ibp_refcount) > 0);
-        atomic_inc(&peer->ibp_refcount);
-}
-
-static inline void
-kibnal_peer_decref (kib_peer_t *peer)
-{
-        CDEBUG(D_NET, "--peer[%p] -> "LPX64" (%d)\n",
-               peer, peer->ibp_nid, atomic_read (&peer->ibp_refcount));
-
-        LASSERT(atomic_read(&peer->ibp_refcount) > 0);
-        if (atomic_dec_and_test (&peer->ibp_refcount))
-                kibnal_destroy_peer (peer);
-}
+#define kibnal_conn_addref(conn)                                \
+do {                                                            \
+        CDEBUG(D_NET, "conn[%p] (%d)++\n",                      \
+               (conn), atomic_read(&(conn)->ibc_refcount));     \
+        LASSERT(atomic_read(&(conn)->ibc_refcount) > 0);        \
+        atomic_inc(&(conn)->ibc_refcount);                      \
+} while (0)
+
+#define kibnal_conn_decref(conn)                                              \
+do {                                                                          \
+        unsigned long   flags;                                                \
+                                                                              \
+        CDEBUG(D_NET, "conn[%p] (%d)--\n",                                    \
+               (conn), atomic_read(&(conn)->ibc_refcount));                   \
+        LASSERT(atomic_read(&(conn)->ibc_refcount) > 0);                      \
+        if (atomic_dec_and_test(&(conn)->ibc_refcount)) {                     \
+                spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);        \
+                list_add_tail(&(conn)->ibc_list,                              \
+                              &kibnal_data.kib_connd_zombies);                \
+                wake_up(&kibnal_data.kib_connd_waitq);                        \
+                spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);   \
+        }                                                                     \
+} while (0)
+
+#define kibnal_peer_addref(peer)                                \
+do {                                                            \
+        CDEBUG(D_NET, "peer[%p] -> "LPX64" (%d)++\n",           \
+               (peer), (peer)->ibp_nid,                         \
+               atomic_read (&(peer)->ibp_refcount));            \
+        LASSERT(atomic_read(&(peer)->ibp_refcount) > 0);        \
+        atomic_inc(&(peer)->ibp_refcount);                      \
+} while (0)
+
+#define kibnal_peer_decref(peer)                                \
+do {                                                            \
+        CDEBUG(D_NET, "peer[%p] -> "LPX64" (%d)--\n",           \
+               (peer), (peer)->ibp_nid,                         \
+               atomic_read (&(peer)->ibp_refcount));            \
+        LASSERT(atomic_read(&(peer)->ibp_refcount) > 0);        \
+        if (atomic_dec_and_test(&(peer)->ibp_refcount))         \
+                kibnal_destroy_peer(peer);                      \
+} while (0)
 
 static inline struct list_head *
 kibnal_nid2peerlist (ptl_nid_t nid)

diff --git a/lnet/klnds/viblnd/viblnd_cb.c b/lnet/klnds/viblnd/viblnd_cb.c
index 3bb4a1d..477d190 100644 (file)
--- a/lnet/klnds/viblnd/viblnd_cb.c
+++ b/lnet/klnds/viblnd/viblnd_cb.c
@@ -950,7 +950,7 @@ kibnal_check_sends (kib_conn_t *conn)
                  * stashed on ibc_active_txs, matched by an incoming PUT_ACK,
                  * and then re-queued here.  It's (just) possible that
                  * tx_sending is non-zero if we've not done the tx_complete() from
-                 * the first send; hence the += rather than = below. */
+                 * the first send; hence the ++ rather than = below. */
                 tx->tx_sending++;
 
                 list_add (&tx->tx_list, &conn->ibc_active_txs);
@@ -965,61 +965,12 @@ kibnal_check_sends (kib_conn_t *conn)
                 vvrc = vv_return_ok;
                 if (conn->ibc_state == IBNAL_CONN_ESTABLISHED) {
                         tx->tx_status = 0;
-#if 1
                         vvrc = vv_post_send_list(kibnal_data.kib_hca,
                                                  conn->ibc_qp,
                                                  tx->tx_nwrq,
                                                  tx->tx_wrq,
                                                  vv_operation_type_send_rc);
                         rc = (vvrc == vv_return_ok) ? 0 : -EIO;
-#else
-                        /* Only post 1 item at a time for now (so we know
-                         * exactly how many got posted successfully) */
-                        for (i = 0; i < tx->tx_nwrq; i++) {
-                                switch (tx->tx_wrq[i].wr_type) {
-                                case vv_wr_send:
-                                        CDEBUG(D_NET, "[%d]posting send [%d %x %p]%s: %x\n", 
-                                               i,
-                                               tx->tx_wrq[i].scatgat_list->length,
-                                               tx->tx_wrq[i].scatgat_list->l_key,
-                                               tx->tx_wrq[i].scatgat_list->v_address,
-                                               tx->tx_wrq[i].type.send.send_qp_type.rc_type.fance_indicator ?
-                                               "(fence)":"",
-                                               tx->tx_msg->ibm_type);
-                                        break;
-                                case vv_wr_rdma_write:
-                                        CDEBUG(D_NET, "[%d]posting PUT  [%d %x %p]->[%x "LPX64"]\n", 
-                                               i,
-                                               tx->tx_wrq[i].scatgat_list->length,
-                                               tx->tx_wrq[i].scatgat_list->l_key,
-                                               tx->tx_wrq[i].scatgat_list->v_address,
-                                               tx->tx_wrq[i].type.send.send_qp_type.rc_type.r_r_key,
-                                               tx->tx_wrq[i].type.send.send_qp_type.rc_type.r_addr);
-                                        break;
-                                case vv_wr_rdma_read:
-                                        CDEBUG(D_NET, "[%d]posting GET  [%d %x %p]->[%x "LPX64"]\n", 
-                                               i,
-                                               tx->tx_wrq[i].scatgat_list->length,
-                                               tx->tx_wrq[i].scatgat_list->l_key,
-                                               tx->tx_wrq[i].scatgat_list->v_address,
-                                               tx->tx_wrq[i].type.send.send_qp_type.rc_type.r_r_key,
-                                               tx->tx_wrq[i].type.send.send_qp_type.rc_type.r_addr);
-                                        break;
-                                default:
-                                        LBUG();
-                                }
-                                vvrc = vv_post_send(kibnal_data.kib_hca,
-                                                    conn->ibc_qp, 
-                                                    &tx->tx_wrq[i], 
-                                                    vv_operation_type_send_rc);
-                                CDEBUG(D_NET, LPX64": post %d/%d\n",
-                                       conn->ibc_peer->ibp_nid, i, tx->tx_nwrq);
-                                if (vvrc != vv_return_ok) {
-                                        rc = -EIO;
-                                        break;
-                                }
-                        }
-#endif
                 }
 
                 if (rc != 0) {
@@ -1081,6 +1032,7 @@ kibnal_tx_complete (kib_tx_t *tx, vv_comp_status_t vvrc)
          * gets to free it, which also drops its ref on 'conn'. */
 
         tx->tx_sending--;
+        conn->ibc_nsends_posted--;
 
         if (failed) {
                 tx->tx_waiting = 0;
@@ -1094,9 +1046,6 @@ kibnal_tx_complete (kib_tx_t *tx, vv_comp_status_t vvrc)
 
         kibnal_conn_addref(conn);               /* 1 ref for me.... */
 
-        if (tx->tx_sending == 0)
-                conn->ibc_nsends_posted--;
-
         spin_unlock(&conn->ibc_lock);
 
         if (idle)

diff --git a/lnet/utils/debug.c b/lnet/utils/debug.c
index 7b0ca54..f8d94ed 100644 (file)
--- a/lnet/utils/debug.c
+++ b/lnet/utils/debug.c
@@ -467,7 +467,7 @@ int jt_dbg_debug_file(int argc, char **argv)
                 return 1;
         }
         if (argc > 2) {
-                fdout = open(argv[2], O_CREAT | O_WRONLY | O_LARGEFILE);
+                fdout = open(argv[2], O_CREAT | O_WRONLY | O_LARGEFILE, 0600);
                 if (fdout == -1) {
                         fprintf(stderr, "open(%s) failed: %s\n", argv[2],
                                 strerror(errno));