Added an almost obscene amount of debugging and assertion code while

tracking down a bug that turned out to be a free_iob() used where I
needed a netdev_tx_complete().  This left the freed I/O buffer on the
net device's TX list, with bad, bad consequences later.

Also fixed the bug in question.

src/drivers/net/ipoib.c

@@ -47,22 +47,22 @@ extern struct ib_address_vector hack_ipoib_bcast_av;
 #define IPOIB_MTU 2048
 
 /** Number of IPoIB data send work queue entries */
-#define IPOIB_DATA_NUM_SEND_WQES 4
+#define IPOIB_DATA_NUM_SEND_WQES 2
 
 /** Number of IPoIB data receive work queue entries */
-#define IPOIB_DATA_NUM_RECV_WQES 4
+#define IPOIB_DATA_NUM_RECV_WQES 2
 
 /** Number of IPoIB data completion entries */
-#define IPOIB_DATA_NUM_CQES 8
+#define IPOIB_DATA_NUM_CQES 32
 
 /** Number of IPoIB metadata send work queue entries */
-#define IPOIB_META_NUM_SEND_WQES 4
+#define IPOIB_META_NUM_SEND_WQES 2
 
 /** Number of IPoIB metadata receive work queue entries */
-#define IPOIB_META_NUM_RECV_WQES 4
+#define IPOIB_META_NUM_RECV_WQES 2
 
 /** Number of IPoIB metadata completion entries */
-#define IPOIB_META_NUM_CQES 8
+#define IPOIB_META_NUM_CQES 32
 
 /** An IPoIB queue set */
 struct ipoib_queue_set {
@@ -205,6 +205,14 @@ static int ipoib_rx ( struct io_buffer *iobuf, struct net_device *netdev ) {
 	}
 
 	/* Strip off IPoIB header */
+	int len = iob_len ( iobuf );
+	DBG ( "WTF iob_len = %zd\n", len );
+	if ( len < 0 ) {
+		DBG_HD ( iobuf, sizeof ( *iobuf ) );
+		DBG ( "locking\n" );
+		while ( 1 ) {}
+	}
+
 	iob_pull ( iobuf, sizeof ( *ipoib_hdr ) );
 
 	/* Hand off to network-layer protocol */
@@ -492,7 +500,7 @@ static int ipoib_transmit ( struct net_device *netdev,
 			/* No path entry - get path record */
 			rc = ipoib_get_path_record ( ipoib,
 						     &ipoib_pshdr->peer.gid );
-			free_iob ( iobuf );
+			netdev_tx_complete ( netdev, iobuf );
 			return rc;
 		}
 		av.dest_qp = ntohl ( ipoib_pshdr->peer.qpn );

src/drivers/net/mlx_ipoib/mt25218.c

@@ -398,6 +398,9 @@ static int arbel_create_cq ( struct ib_device *ibdev,
 		goto err_sw2hw_cq;
 	}
 
+	DBGC ( arbel, "Arbel %p CQN %#lx ring at [%p,%p)\n",
+	       arbel, cq->cqn, arbel_cq->cqe,
+	       ( ( ( void * ) arbel_cq->cqe ) + arbel_cq->cqe_size ) );
 	cq->dev_priv = arbel_cq;
 	return 0;
 
@@ -650,6 +653,12 @@ static int arbel_create_qp ( struct ib_device *ibdev,
 		goto err_rtr2rts_qpee;
 	}
 
+	DBGC ( arbel, "Arbel %p QPN %#lx send ring at [%p,%p)\n",
+	       arbel, qp->qpn, arbel_qp->send.wqe,
+	       ( ( (void *) arbel_qp->send.wqe ) + arbel_qp->send.wqe_size ) );
+	DBGC ( arbel, "Arbel %p QPN %#lx receive ring at [%p,%p)\n",
+	       arbel, qp->qpn, arbel_qp->recv.wqe,
+	       ( ( (void *) arbel_qp->recv.wqe ) + arbel_qp->recv.wqe_size ) );
 	qp->dev_priv = arbel_qp;
 	return 0;
 
@@ -904,6 +913,7 @@ static int arbel_complete ( struct ib_device *ibdev,
 	struct arbel_queue_pair *arbel_qp;
 	struct arbel_send_work_queue *arbel_send_wq;
 	struct arbel_recv_work_queue *arbel_recv_wq;
+	struct arbelprm_recv_wqe *recv_wqe;
 	struct io_buffer *iobuf;
 	ib_completer_t complete;
 	unsigned int opcode;
@@ -915,7 +925,6 @@ static int arbel_complete ( struct ib_device *ibdev,
 
 	/* Parse completion */
 	memset ( &completion, 0, sizeof ( completion ) );
-	completion.len = MLX_GET ( &cqe->normal, byte_cnt );
 	qpn = MLX_GET ( &cqe->normal, my_qpn );
 	is_send = MLX_GET ( &cqe->normal, s );
 	wqe_adr = ( MLX_GET ( &cqe->normal, wqe_adr ) << 6 );
@@ -946,10 +955,12 @@ static int arbel_complete ( struct ib_device *ibdev,
 		arbel_send_wq = &arbel_qp->send;
 		wqe_idx = ( ( wqe_adr - virt_to_bus ( arbel_send_wq->wqe ) ) /
 			    sizeof ( arbel_send_wq->wqe[0] ) );
+		assert ( wqe_idx < qp->send.num_wqes );
 	} else {
 		arbel_recv_wq = &arbel_qp->recv;
 		wqe_idx = ( ( wqe_adr - virt_to_bus ( arbel_recv_wq->wqe ) ) /
 			    sizeof ( arbel_recv_wq->wqe[0] ) );
+		assert ( wqe_idx < qp->recv.num_wqes );
 	}
 
 	/* Identify I/O buffer */
@@ -961,6 +972,27 @@ static int arbel_complete ( struct ib_device *ibdev,
 	}
 	wq->iobufs[wqe_idx] = NULL;
 
+	/* Fill in length for received packets */
+	if ( ! is_send ) {
+		completion.len = MLX_GET ( &cqe->normal, byte_cnt );
+		recv_wqe = &arbel_recv_wq->wqe[wqe_idx].recv;
+		assert ( MLX_GET ( &recv_wqe->data[0], local_address_l ) ==
+			 virt_to_bus ( iobuf->data ) );
+		assert ( MLX_GET ( &recv_wqe->data[0], byte_count ) ==
+			 iob_tailroom ( iobuf ) );
+		DBG ( "CPQ %lx QPN %lx WQE %x\n", cq->cqn, qp->qpn, wqe_idx );
+		//		DBG_HD ( iobuf, sizeof ( *iobuf ) );
+		MLX_FILL_1 ( &recv_wqe->data[0], 0, byte_count, 0 );
+		MLX_FILL_1 ( &recv_wqe->data[0], 1,
+			     l_key, ARBEL_INVALID_LKEY );
+		if ( completion.len > iob_tailroom ( iobuf ) ) {
+			DBGC ( arbel, "Arbel %p CQN %lx QPN %lx IDX %x "
+			       "overlength received packet length %zd\n",
+			       arbel, cq->cqn, qpn, wqe_idx, completion.len );
+			return -EIO;
+		}
+	}
+
 	/* Pass off to caller's completion handler */
 	complete = ( is_send ? complete_send : complete_recv );
 	complete ( ibdev, qp, &completion, iobuf );
@@ -1252,7 +1284,7 @@ static int arbel_get_sm_lid ( struct arbel *arbel,
 	return 0;
 }
 
-static int arbel_get_pkey ( struct arbel *arbel, unsigned long *pkey ) {
+static int arbel_get_pkey ( struct arbel *arbel, unsigned int *pkey ) {
 	struct ib_mad_pkey_table pkey_table;
 	int rc;
 

src/net/infiniband.c

@@ -64,8 +64,8 @@ struct ib_completion_queue * ib_create_cq ( struct ib_device *ibdev,
 		return NULL;
 	}
 
-	DBGC ( ibdev, "IBDEV %p created completion queue %#lx\n",
-	       ibdev, cq->cqn );
+	DBGC ( ibdev, "IBDEV %p created %d-entry completion queue %p (%p) "
+	       "with CQN %#lx\n", ibdev, num_cqes, cq, cq->dev_priv, cq->cqn );
 	return cq;
 }
 
@@ -102,14 +102,16 @@ struct ib_queue_pair * ib_create_qp ( struct ib_device *ibdev,
 				      struct ib_completion_queue *recv_cq,
 				      unsigned long qkey ) {
 	struct ib_queue_pair *qp;
+	size_t total_size;
 	int rc;
 
 	DBGC ( ibdev, "IBDEV %p creating queue pair\n", ibdev );
 
 	/* Allocate and initialise data structure */
-	qp = zalloc ( sizeof ( *qp ) +
-		      ( num_send_wqes * sizeof ( qp->send.iobufs[0] ) ) +
-		      ( num_recv_wqes * sizeof ( qp->recv.iobufs[0] ) ) );
+	total_size = ( sizeof ( *qp ) +
+		       ( num_send_wqes * sizeof ( qp->send.iobufs[0] ) ) +
+		       ( num_recv_wqes * sizeof ( qp->recv.iobufs[0] ) ) );
+	qp = zalloc ( total_size );
 	if ( ! qp )
 		return NULL;
 	qp->qkey = qkey;
@@ -134,8 +136,14 @@ struct ib_queue_pair * ib_create_qp ( struct ib_device *ibdev,
 		return NULL;
 	}
 
-	DBGC ( ibdev, "IBDEV %p created queue pair %#lx\n",
-	       ibdev, qp->qpn );
+	DBGC ( ibdev, "IBDEV %p created queue pair %p (%p) with QPN %#lx\n",
+	       ibdev, qp, qp->dev_priv, qp->qpn );
+	DBGC ( ibdev, "IBDEV %p QPN %#lx has %d send entries at [%p,%p)\n",
+	       ibdev, qp->qpn, num_send_wqes, qp->send.iobufs,
+	       qp->recv.iobufs );
+	DBGC ( ibdev, "IBDEV %p QPN %#lx has %d receive entries at [%p,%p)\n",
+	       ibdev, qp->qpn, num_send_wqes, qp->recv.iobufs,
+	       ( ( ( void * ) qp ) + total_size ) );
 	return qp;
 }