Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  RDMA/cxgb3: Wrap the software send queue pointer as needed on flush
  IB/ipath: Change ipath_devdata.ipath_sdma_status to be unsigned long
  IB/ipath: Make ipath_portdata work with struct pid * not pid_t
  IB/ipath: Fix RDMA read response sequence checking
  IB/ipath: Fix many locking issues when switching to error state
  IB/ipath: Fix RC and UC error handling
  RDMA/nes: Fix up nes_lro_max_aggr module parameter

15 files changed, 725 insertions, 555 deletions

diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.c b/drivers/infiniband/hw/cxgb3/cxio_hal.c
index ebf9d3043f8..3f441fc57c1 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_hal.c
+++ b/drivers/infiniband/hw/cxgb3/cxio_hal.c
@@ -405,11 +405,11 @@ int cxio_flush_sq(struct t3_wq *wq, struct t3_cq *cq, int count)
 	struct t3_swsq *sqp = wq->sq + Q_PTR2IDX(wq->sq_rptr, wq->sq_size_log2);
 
 	ptr = wq->sq_rptr + count;
-	sqp += count;
+	sqp = wq->sq + Q_PTR2IDX(ptr, wq->sq_size_log2);
 	while (ptr != wq->sq_wptr) {
 		insert_sq_cqe(wq, cq, sqp);
-		sqp++;
 		ptr++;
+		sqp = wq->sq + Q_PTR2IDX(ptr, wq->sq_size_log2);
 		flushed++;
 	}
 	return flushed;
diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index ce7b7c34360..daad09a4591 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -1894,7 +1894,7 @@ void ipath_cancel_sends(struct ipath_devdata *dd, int restore_sendctrl)
 	 */
 	if (dd->ipath_flags & IPATH_HAS_SEND_DMA) {
 		int skip_cancel;
-		u64 *statp = &dd->ipath_sdma_status;
+		unsigned long *statp = &dd->ipath_sdma_status;
 
 		spin_lock_irqsave(&dd->ipath_sdma_lock, flags);
 		skip_cancel =
@@ -2616,7 +2616,7 @@ int ipath_reset_device(int unit)
 				ipath_dbg("unit %u port %d is in use "
 					  "(PID %u cmd %s), can't reset\n",
 					  unit, i,
-					  dd->ipath_pd[i]->port_pid,
+					  pid_nr(dd->ipath_pd[i]->port_pid),
 					  dd->ipath_pd[i]->port_comm);
 				ret = -EBUSY;
 				goto bail;
@@ -2654,19 +2654,21 @@ bail:
 static int ipath_signal_procs(struct ipath_devdata *dd, int sig)
 {
 	int i, sub, any = 0;
-	pid_t pid;
+	struct pid *pid;
 
 	if (!dd->ipath_pd)
 		return 0;
 	for (i = 1; i < dd->ipath_cfgports; i++) {
-		if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt ||
-		    !dd->ipath_pd[i]->port_pid)
+		if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt)
 			continue;
 		pid = dd->ipath_pd[i]->port_pid;
+		if (!pid)
+			continue;
+
 		dev_info(&dd->pcidev->dev, "context %d in use "
 			  "(PID %u), sending signal %d\n",
-			  i, pid, sig);
-		kill_proc(pid, sig, 1);
+			  i, pid_nr(pid), sig);
+		kill_pid(pid, sig, 1);
 		any++;
 		for (sub = 0; sub < INFINIPATH_MAX_SUBPORT; sub++) {
 			pid = dd->ipath_pd[i]->port_subpid[sub];
@@ -2674,8 +2676,8 @@ static int ipath_signal_procs(struct ipath_devdata *dd, int sig)
 				continue;
 			dev_info(&dd->pcidev->dev, "sub-context "
 				"%d:%d in use (PID %u), sending "
-				"signal %d\n", i, sub, pid, sig);
-			kill_proc(pid, sig, 1);
+				"signal %d\n", i, sub, pid_nr(pid), sig);
+			kill_pid(pid, sig, 1);
 			any++;
 		}
 	}
diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c
index 3295177c937..b472b15637f 100644
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c
@@ -555,7 +555,7 @@ static int ipath_tid_free(struct ipath_portdata *pd, unsigned subport,
 			p = dd->ipath_pageshadow[porttid + tid];
 			dd->ipath_pageshadow[porttid + tid] = NULL;
 			ipath_cdbg(VERBOSE, "PID %u freeing TID %u\n",
-				   pd->port_pid, tid);
+				   pid_nr(pd->port_pid), tid);
 			dd->ipath_f_put_tid(dd, &tidbase[tid],
 					    RCVHQ_RCV_TYPE_EXPECTED,
 					    dd->ipath_tidinvalid);
@@ -1609,7 +1609,7 @@ static int try_alloc_port(struct ipath_devdata *dd, int port,
 			   port);
 		pd->port_cnt = 1;
 		port_fp(fp) = pd;
-		pd->port_pid = current->pid;
+		pd->port_pid = get_pid(task_pid(current));
 		strncpy(pd->port_comm, current->comm, sizeof(pd->port_comm));
 		ipath_stats.sps_ports++;
 		ret = 0;
@@ -1793,14 +1793,15 @@ static int find_shared_port(struct file *fp,
 			}
 			port_fp(fp) = pd;
 			subport_fp(fp) = pd->port_cnt++;
-			pd->port_subpid[subport_fp(fp)] = current->pid;
+			pd->port_subpid[subport_fp(fp)] =
+				get_pid(task_pid(current));
 			tidcursor_fp(fp) = 0;
 			pd->active_slaves |= 1 << subport_fp(fp);
 			ipath_cdbg(PROC,
 				   "%s[%u] %u sharing %s[%u] unit:port %u:%u\n",
 				   current->comm, current->pid,
 				   subport_fp(fp),
-				   pd->port_comm, pd->port_pid,
+				   pd->port_comm, pid_nr(pd->port_pid),
 				   dd->ipath_unit, pd->port_port);
 			ret = 1;
 			goto done;
@@ -2066,7 +2067,8 @@ static int ipath_close(struct inode *in, struct file *fp)
 		 * the slave(s) don't wait for receive data forever.
 		 */
 		pd->active_slaves &= ~(1 << fd->subport);
-		pd->port_subpid[fd->subport] = 0;
+		put_pid(pd->port_subpid[fd->subport]);
+		pd->port_subpid[fd->subport] = NULL;
 		mutex_unlock(&ipath_mutex);
 		goto bail;
 	}
@@ -2074,7 +2076,7 @@ static int ipath_close(struct inode *in, struct file *fp)
 
 	if (pd->port_hdrqfull) {
 		ipath_cdbg(PROC, "%s[%u] had %u rcvhdrqfull errors "
-			   "during run\n", pd->port_comm, pd->port_pid,
+			   "during run\n", pd->port_comm, pid_nr(pd->port_pid),
 			   pd->port_hdrqfull);
 		pd->port_hdrqfull = 0;
 	}
@@ -2134,11 +2136,12 @@ static int ipath_close(struct inode *in, struct file *fp)
 			unlock_expected_tids(pd);
 		ipath_stats.sps_ports--;
 		ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
-			   pd->port_comm, pd->port_pid,
+			   pd->port_comm, pid_nr(pd->port_pid),
 			   dd->ipath_unit, port);
 	}
 
-	pd->port_pid = 0;
+	put_pid(pd->port_pid);
+	pd->port_pid = NULL;
 	dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
 	mutex_unlock(&ipath_mutex);
 	ipath_free_pddata(dd, pd); /* after releasing the mutex */
diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h
index 02b24a34059..59a8b254b97 100644
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h
@@ -159,8 +159,8 @@ struct ipath_portdata {
 	/* saved total number of polled urgent packets for poll edge trigger */
 	u32 port_urgent_poll;
 	/* pid of process using this port */
-	pid_t port_pid;
-	pid_t port_subpid[INFINIPATH_MAX_SUBPORT];
+	struct pid *port_pid;
+	struct pid *port_subpid[INFINIPATH_MAX_SUBPORT];
 	/* same size as task_struct .comm[] */
 	char port_comm[16];
 	/* pkeys set by this use of this port */
@@ -483,7 +483,7 @@ struct ipath_devdata {
 
 	/* SendDMA related entries */
 	spinlock_t            ipath_sdma_lock;
-	u64                   ipath_sdma_status;
+	unsigned long         ipath_sdma_status;
 	unsigned long         ipath_sdma_abort_jiffies;
 	unsigned long         ipath_sdma_abort_intr_timeout;
 	unsigned long         ipath_sdma_buf_jiffies;
@@ -822,8 +822,8 @@ struct ipath_devdata {
 #define IPATH_SDMA_DISARMED  1
 #define IPATH_SDMA_DISABLED  2
 #define IPATH_SDMA_LAYERBUF  3
-#define IPATH_SDMA_RUNNING  62
-#define IPATH_SDMA_SHUTDOWN 63
+#define IPATH_SDMA_RUNNING  30
+#define IPATH_SDMA_SHUTDOWN 31
 
 /* bit combinations that correspond to abort states */
 #define IPATH_SDMA_ABORT_NONE 0
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index dd5b6e9d57c..4715911101e 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -242,7 +242,6 @@ static void ipath_free_qp(struct ipath_qp_table *qpt, struct ipath_qp *qp)
 {
 	struct ipath_qp *q, **qpp;
 	unsigned long flags;
-	int fnd = 0;
 
 	spin_lock_irqsave(&qpt->lock, flags);
 
@@ -253,51 +252,40 @@ static void ipath_free_qp(struct ipath_qp_table *qpt, struct ipath_qp *qp)
 			*qpp = qp->next;
 			qp->next = NULL;
 			atomic_dec(&qp->refcount);
-			fnd = 1;
 			break;
 		}
 	}
 
 	spin_unlock_irqrestore(&qpt->lock, flags);
-
-	if (!fnd)
-		return;
-
-	free_qpn(qpt, qp->ibqp.qp_num);
-
-	wait_event(qp->wait, !atomic_read(&qp->refcount));
 }
 
 /**
- * ipath_free_all_qps - remove all QPs from the table
+ * ipath_free_all_qps - check for QPs still in use
  * @qpt: the QP table to empty
+ *
+ * There should not be any QPs still in use.
+ * Free memory for table.
  */
-void ipath_free_all_qps(struct ipath_qp_table *qpt)
+unsigned ipath_free_all_qps(struct ipath_qp_table *qpt)
 {
 	unsigned long flags;
-	struct ipath_qp *qp, *nqp;
-	u32 n;
+	struct ipath_qp *qp;
+	u32 n, qp_inuse = 0;
 
+	spin_lock_irqsave(&qpt->lock, flags);
 	for (n = 0; n < qpt->max; n++) {
-		spin_lock_irqsave(&qpt->lock, flags);
 		qp = qpt->table[n];
 		qpt->table[n] = NULL;
-		spin_unlock_irqrestore(&qpt->lock, flags);
-
-		while (qp) {
-			nqp = qp->next;
-			free_qpn(qpt, qp->ibqp.qp_num);
-			if (!atomic_dec_and_test(&qp->refcount) ||
-			    !ipath_destroy_qp(&qp->ibqp))
-				ipath_dbg("QP memory leak!\n");
-			qp = nqp;
-		}
+
+		for (; qp; qp = qp->next)
+			qp_inuse++;
 	}
+	spin_unlock_irqrestore(&qpt->lock, flags);
 
-	for (n = 0; n < ARRAY_SIZE(qpt->map); n++) {
+	for (n = 0; n < ARRAY_SIZE(qpt->map); n++)
 		if (qpt->map[n].page)
-			free_page((unsigned long)qpt->map[n].page);
-	}
+			free_page((unsigned long) qpt->map[n].page);
+	return qp_inuse;
 }
 
 /**
@@ -336,11 +324,12 @@ static void ipath_reset_qp(struct ipath_qp *qp, enum ib_qp_type type)
 	qp->remote_qpn = 0;
 	qp->qkey = 0;
 	qp->qp_access_flags = 0;
-	qp->s_busy = 0;
+	atomic_set(&qp->s_dma_busy, 0);
 	qp->s_flags &= IPATH_S_SIGNAL_REQ_WR;
 	qp->s_hdrwords = 0;
 	qp->s_wqe = NULL;
 	qp->s_pkt_delay = 0;
+	qp->s_draining = 0;
 	qp->s_psn = 0;
 	qp->r_psn = 0;
 	qp->r_msn = 0;
@@ -353,7 +342,8 @@ static void ipath_reset_qp(struct ipath_qp *qp, enum ib_qp_type type)
 	}
 	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
 	qp->r_nak_state = 0;
-	qp->r_wrid_valid = 0;
+	qp->r_aflags = 0;
+	qp->r_flags = 0;
 	qp->s_rnr_timeout = 0;
 	qp->s_head = 0;
 	qp->s_tail = 0;
@@ -361,7 +351,6 @@ static void ipath_reset_qp(struct ipath_qp *qp, enum ib_qp_type type)
 	qp->s_last = 0;
 	qp->s_ssn = 1;
 	qp->s_lsn = 0;
-	qp->s_wait_credit = 0;
 	memset(qp->s_ack_queue, 0, sizeof(qp->s_ack_queue));
 	qp->r_head_ack_queue = 0;
 	qp->s_tail_ack_queue = 0;
@@ -370,17 +359,17 @@ static void ipath_reset_qp(struct ipath_qp *qp, enum ib_qp_type type)
 		qp->r_rq.wq->head = 0;
 		qp->r_rq.wq->tail = 0;
 	}
-	qp->r_reuse_sge = 0;
 }
 
 /**
- * ipath_error_qp - put a QP into an error state
- * @qp: the QP to put into an error state
+ * ipath_error_qp - put a QP into the error state
+ * @qp: the QP to put into the error state
  * @err: the receive completion error to signal if a RWQE is active
  *
  * Flushes both send and receive work queues.
  * Returns true if last WQE event should be generated.
  * The QP s_lock should be held and interrupts disabled.
+ * If we are already in error state, just return.
  */
 
 int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
@@ -389,8 +378,10 @@ int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
 	struct ib_wc wc;
 	int ret = 0;
 
-	ipath_dbg("QP%d/%d in error state (%d)\n",
-		  qp->ibqp.qp_num, qp->remote_qpn, err);
+	if (qp->state == IB_QPS_ERR)
+		goto bail;
+
+	qp->state = IB_QPS_ERR;
 
 	spin_lock(&dev->pending_lock);
 	if (!list_empty(&qp->timerwait))
@@ -399,39 +390,21 @@ int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
 		list_del_init(&qp->piowait);
 	spin_unlock(&dev->pending_lock);
 
-	wc.vendor_err = 0;
-	wc.byte_len = 0;
-	wc.imm_data = 0;
+	/* Schedule the sending tasklet to drain the send work queue. */
+	if (qp->s_last != qp->s_head)
+		ipath_schedule_send(qp);
+
+	memset(&wc, 0, sizeof(wc));
 	wc.qp = &qp->ibqp;
-	wc.src_qp = 0;
-	wc.wc_flags = 0;
-	wc.pkey_index = 0;
-	wc.slid = 0;
-	wc.sl = 0;
-	wc.dlid_path_bits = 0;
-	wc.port_num = 0;
-	if (qp->r_wrid_valid) {
-		qp->r_wrid_valid = 0;
+	wc.opcode = IB_WC_RECV;
+
+	if (test_and_clear_bit(IPATH_R_WRID_VALID, &qp->r_aflags)) {
 		wc.wr_id = qp->r_wr_id;
-		wc.opcode = IB_WC_RECV;
 		wc.status = err;
 		ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
 	}
 	wc.status = IB_WC_WR_FLUSH_ERR;
 
-	while (qp->s_last != qp->s_head) {
-		struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
-
-		wc.wr_id = wqe->wr.wr_id;
-		wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
-		if (++qp->s_last >= qp->s_size)
-			qp->s_last = 0;
-		ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 1);
-	}
-	qp->s_cur = qp->s_tail = qp->s_head;
-	qp->s_hdrwords = 0;
-	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
-
 	if (qp->r_rq.wq) {
 		struct ipath_rwq *wq;
 		u32 head;
@@ -447,7 +420,6 @@ int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
 		tail = wq->tail;
 		if (tail >= qp->r_rq.size)
 			tail = 0;
-		wc.opcode = IB_WC_RECV;
 		while (tail != head) {
 			wc.wr_id = get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
 			if (++tail >= qp->r_rq.size)
@@ -460,6 +432,7 @@ int ipath_error_qp(struct ipath_qp *qp, enum ib_wc_status err)
 	} else if (qp->ibqp.event_handler)
 		ret = 1;
 
+bail:
 	return ret;
 }
 
@@ -478,11 +451,10 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	struct ipath_ibdev *dev = to_idev(ibqp->device);
 	struct ipath_qp *qp = to_iqp(ibqp);
 	enum ib_qp_state cur_state, new_state;
-	unsigned long flags;
 	int lastwqe = 0;
 	int ret;
 
-	spin_lock_irqsave(&qp->s_lock, flags);
+	spin_lock_irq(&qp->s_lock);
 
 	cur_state = attr_mask & IB_QP_CUR_STATE ?
 		attr->cur_qp_state : qp->state;
@@ -535,16 +507,42 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 
 	switch (new_state) {
 	case IB_QPS_RESET:
+		if (qp->state != IB_QPS_RESET) {
+			qp->state = IB_QPS_RESET;
+			spin_lock(&dev->pending_lock);
+			if (!list_empty(&qp->timerwait))
+				list_del_init(&qp->timerwait);
+			if (!list_empty(&qp->piowait))
+				list_del_init(&qp->piowait);
+			spin_unlock(&dev->pending_lock);
+			qp->s_flags &= ~IPATH_S_ANY_WAIT;
+			spin_unlock_irq(&qp->s_lock);
+			/* Stop the sending tasklet */
+			tasklet_kill(&qp->s_task);
+			wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+			spin_lock_irq(&qp->s_lock);
+		}
 		ipath_reset_qp(qp, ibqp->qp_type);
 		break;
 
+	case IB_QPS_SQD:
+		qp->s_draining = qp->s_last != qp->s_cur;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_SQE:
+		if (qp->ibqp.qp_type == IB_QPT_RC)
+			goto inval;
+		qp->state = new_state;
+		break;
+
 	case IB_QPS_ERR:
 		lastwqe = ipath_error_qp(qp, IB_WC_WR_FLUSH_ERR);
 		break;
 
 	default:
+		qp->state = new_state;
 		break;
-
 	}
 
 	if (attr_mask & IB_QP_PKEY_INDEX)
@@ -597,8 +595,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
 		qp->s_max_rd_atomic = attr->max_rd_atomic;
 
-	qp->state = new_state;
-	spin_unlock_irqrestore(&qp->s_lock, flags);
+	spin_unlock_irq(&qp->s_lock);
 
 	if (lastwqe) {
 		struct ib_event ev;
@@ -612,7 +609,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	goto bail;
 
 inval:
-	spin_unlock_irqrestore(&qp->s_lock, flags);
+	spin_unlock_irq(&qp->s_lock);
 	ret = -EINVAL;
 
 bail:
@@ -643,7 +640,7 @@ int ipath_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	attr->pkey_index = qp->s_pkey_index;
 	attr->alt_pkey_index = 0;
 	attr->en_sqd_async_notify = 0;
-	attr->sq_draining = 0;
+	attr->sq_draining = qp->s_draining;
 	attr->max_rd_atomic = qp->s_max_rd_atomic;
 	attr->max_dest_rd_atomic = qp->r_max_rd_atomic;
 	attr->min_rnr_timer = qp->r_min_rnr_timer;
@@ -833,6 +830,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 		spin_lock_init(&qp->r_rq.lock);
 		atomic_set(&qp->refcount, 0);
 		init_waitqueue_head(&qp->wait);
+		init_waitqueue_head(&qp->wait_dma);
 		tasklet_init(&qp->s_task, ipath_do_send, (unsigned long)qp);
 		INIT_LIST_HEAD(&qp->piowait);
 		INIT_LIST_HEAD(&qp->timerwait);
@@ -926,6 +924,7 @@ bail_ip:
 	else
 		vfree(qp->r_rq.wq);
 	ipath_free_qp(&dev->qp_table, qp);
+	free_qpn(&dev->qp_table, qp->ibqp.qp_num);
 bail_qp:
 	kfree(qp);
 bail_swq:
@@ -947,41 +946,44 @@ int ipath_destroy_qp(struct ib_qp *ibqp)
 {
 	struct ipath_qp *qp = to_iqp(ibqp);
 	struct ipath_ibdev *dev = to_idev(ibqp->device);
-	unsigned long flags;
 
-	spin_lock_irqsave(&qp->s_lock, flags);
-	qp->state = IB_QPS_ERR;
-	spin_unlock_irqrestore(&qp->s_lock, flags);
-	spin_lock(&dev->n_qps_lock);
-	dev->n_qps_allocated--;
-	spin_unlock(&dev->n_qps_lock);
+	/* Make sure HW and driver activity is stopped. */
+	spin_lock_irq(&qp->s_lock);
+	if (qp->state != IB_QPS_RESET) {
+		qp->state = IB_QPS_RESET;
+		spin_lock(&dev->pending_lock);
+		if (!list_empty(&qp->timerwait))
+			list_del_init(&qp->timerwait);
+		if (!list_empty(&qp->piowait))
+			list_del_init(&qp->piowait);
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~IPATH_S_ANY_WAIT;
+		spin_unlock_irq(&qp->s_lock);
+		/* Stop the sending tasklet */
+		tasklet_kill(&qp->s_task);
+		wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+	} else
+		spin_unlock_irq(&qp->s_lock);
 
-	/* Stop the sending tasklet. */
-	tasklet_kill(&qp->s_task);
+	ipath_free_qp(&dev->qp_table, qp);
 
 	if (qp->s_tx) {
 		atomic_dec(&qp->refcount);
 		if (qp->s_tx->txreq.flags & IPATH_SDMA_TXREQ_F_FREEBUF)
 			kfree(qp->s_tx->txreq.map_addr);
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&qp->s_tx->txreq.list, &dev->txreq_free);
+		spin_unlock_irq(&dev->pending_lock);
+		qp->s_tx = NULL;
 	}
 
-	/* Make sure the QP isn't on the timeout list. */
-	spin_lock_irqsave(&dev->pending_lock, flags);
-	if (!list_empty(&qp->timerwait))
-		list_del_init(&qp->timerwait);
-	if (!list_empty(&qp->piowait))
-		list_del_init(&qp->piowait);
-	if (qp->s_tx)
-		list_add(&qp->s_tx->txreq.list, &dev->txreq_free);
-	spin_unlock_irqrestore(&dev->pending_lock, flags);
+	wait_event(qp->wait, !atomic_read(&qp->refcount));
 
-	/*
-	 * Make sure that the QP is not in the QPN table so receive
-	 * interrupts will discard packets for this QP.  XXX Also remove QP
-	 * from multicast table.
-	 */
-	if (atomic_read(&qp->refcount) != 0)
-		ipath_free_qp(&dev->qp_table, qp);
+	/* all user's cleaned up, mark it available */
+	free_qpn(&dev->qp_table, qp->ibqp.qp_num);
+	spin_lock(&dev->n_qps_lock);
+	dev->n_qps_allocated--;
+	spin_unlock(&dev->n_qps_lock);
 
 	if (qp->ip)
 		kref_put(&qp->ip->ref, ipath_release_mmap_info);
@@ -1026,48 +1028,6 @@ bail:
 }
 
 /**
- * ipath_sqerror_qp - put a QP's send queue into an error state
- * @qp: QP who's send queue will be put into an error state
- * @wc: the WC responsible for putting the QP in this state
- *
- * Flushes the send work queue.
- * The QP s_lock should be held and interrupts disabled.
- */
-
-void ipath_sqerror_qp(struct ipath_qp *qp, struct ib_wc *wc)
-{
-	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
-	struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
-
-	ipath_dbg("Send queue error on QP%d/%d: err: %d\n",
-		  qp->ibqp.qp_num, qp->remote_qpn, wc->status);
-
-	spin_lock(&dev->pending_lock);
-	if (!list_empty(&qp->timerwait))
-		list_del_init(&qp->timerwait);
-	if (!list_empty(&qp->piowait))
-		list_del_init(&qp->piowait);
-	spin_unlock(&dev->pending_lock);
-
-	ipath_cq_enter(to_icq(qp->ibqp.send_cq), wc, 1);
-	if (++qp->s_last >= qp->s_size)
-		qp->s_last = 0;
-
-	wc->status = IB_WC_WR_FLUSH_ERR;
-
-	while (qp->s_last != qp->s_head) {
-		wqe = get_swqe_ptr(qp, qp->s_last);
-		wc->wr_id = wqe->wr.wr_id;
-		wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
-		ipath_cq_enter(to_icq(qp->ibqp.send_cq), wc, 1);
-		if (++qp->s_last >= qp->s_size)
-			qp->s_last = 0;
-	}
-	qp->s_cur = qp->s_tail = qp->s_head;
-	qp->state = IB_QPS_SQE;
-}
-
-/**
  * ipath_get_credit - flush the send work queue of a QP
  * @qp: the qp who's send work queue to flush
  * @aeth: the Acknowledge Extended Transport Header
@@ -1093,9 +1053,10 @@ void ipath_get_credit(struct ipath_qp *qp, u32 aeth)
 	}
 
 	/* Restart sending if it was blocked due to lack of credits. */
-	if (qp->s_cur != qp->s_head &&
+	if ((qp->s_flags & IPATH_S_WAIT_SSN_CREDIT) &&
+	    qp->s_cur != qp->s_head &&
 	    (qp->s_lsn == (u32) -1 ||
 	     ipath_cmp24(get_swqe_ptr(qp, qp->s_cur)->ssn,
 			 qp->s_lsn + 1) <= 0))
-		tasklet_hi_schedule(&qp->s_task);
+		ipath_schedule_send(qp);
 }
diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c
index 08b11b56761..108df667d2e 100644
--- a/drivers/infiniband/hw/ipath/ipath_rc.c
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c
@@ -92,6 +92,10 @@ static int ipath_make_rc_ack(struct ipath_ibdev *dev, struct ipath_qp *qp,
 	u32 bth0;
 	u32 bth2;
 
+	/* Don't send an ACK if we aren't supposed to. */
+	if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK))
+		goto bail;
+
 	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
 	hwords = 5;
 
@@ -238,14 +242,25 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 	    ipath_make_rc_ack(dev, qp, ohdr, pmtu))
 		goto done;
 
-	if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK) ||
-	    qp->s_rnr_timeout || qp->s_wait_credit)
-		goto bail;
+	if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK)) {
+		if (!(ib_ipath_state_ops[qp->state] & IPATH_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= IPATH_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		ipath_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
 
-	/* Limit the number of packets sent without an ACK. */
-	if (ipath_cmp24(qp->s_psn, qp->s_last_psn + IPATH_PSN_CREDIT) > 0) {
-		qp->s_wait_credit = 1;
-		dev->n_rc_stalls++;
+	/* Leave BUSY set until RNR timeout. */
+	if (qp->s_rnr_timeout) {
+		qp->s_flags |= IPATH_S_WAITING;
 		goto bail;
 	}
 
@@ -257,6 +272,9 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 	wqe = get_swqe_ptr(qp, qp->s_cur);
 	switch (qp->s_state) {
 	default:
+		if (!(ib_ipath_state_ops[qp->state] &
+		    IPATH_PROCESS_NEXT_SEND_OK))
+			goto bail;
 		/*
 		 * Resend an old request or start a new one.
 		 *
@@ -294,8 +312,10 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		case IB_WR_SEND_WITH_IMM:
 			/* If no credit, return. */
 			if (qp->s_lsn != (u32) -1 &&
-			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0)
+			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= IPATH_S_WAIT_SSN_CREDIT;
 				goto bail;
+			}
 			wqe->lpsn = wqe->psn;
 			if (len > pmtu) {
 				wqe->lpsn += (len - 1) / pmtu;
@@ -325,8 +345,10 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		case IB_WR_RDMA_WRITE_WITH_IMM:
 			/* If no credit, return. */
 			if (qp->s_lsn != (u32) -1 &&
-			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0)
+			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= IPATH_S_WAIT_SSN_CREDIT;
 				goto bail;
+			}
 			ohdr->u.rc.reth.vaddr =
 				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
 			ohdr->u.rc.reth.rkey =
@@ -570,7 +592,11 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 	ipath_make_ruc_header(dev, qp, ohdr, bth0 | (qp->s_state << 24), bth2);
 done:
 	ret = 1;
+	goto unlock;
+
 bail:
+	qp->s_flags &= ~IPATH_S_BUSY;
+unlock:
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 	return ret;
 }
@@ -606,7 +632,11 @@ static void send_rc_ack(struct ipath_qp *qp)
 
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 
+	/* Don't try to send ACKs if the link isn't ACTIVE */
 	dd = dev->dd;
+	if (!(dd->ipath_flags & IPATH_LINKACTIVE))
+		goto done;
+
 	piobuf = ipath_getpiobuf(dd, 0, NULL);
 	if (!piobuf) {
 		/*
@@ -668,15 +698,16 @@ static void send_rc_ack(struct ipath_qp *qp)
 	goto done;
 
 queue_ack:
-	dev->n_rc_qacks++;
-	qp->s_flags |= IPATH_S_ACK_PENDING;
-	qp->s_nak_state = qp->r_nak_state;
-	qp->s_ack_psn = qp->r_ack_psn;
+	if (ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK) {
+		dev->n_rc_qacks++;
+		qp->s_flags |= IPATH_S_ACK_PENDING;
+		qp->s_nak_state = qp->r_nak_state;
+		qp->s_ack_psn = qp->r_ack_psn;
+
+		/* Schedule the send tasklet. */
+		ipath_schedule_send(qp);
+	}
 	spin_unlock_irqrestore(&qp->s_lock, flags);
-
-	/* Call ipath_do_rc_send() in another thread. */
-	tasklet_hi_schedule(&qp->s_task);
-
 done:
 	return;
 }
@@ -735,7 +766,7 @@ static void reset_psn(struct ipath_qp *qp, u32 psn)
 	/*
 	 * Set the state to restart in the middle of a request.
 	 * Don't change the s_sge, s_cur_sge, or s_cur_size.
-	 * See ipath_do_rc_send().
+	 * See ipath_make_rc_req().
 	 */
 	switch (opcode) {
 	case IB_WR_SEND:
@@ -771,27 +802,14 @@ done:
  *
  * The QP s_lock should be held and interrupts disabled.
  */
-void ipath_restart_rc(struct ipath_qp *qp, u32 psn, struct ib_wc *wc)
+void ipath_restart_rc(struct ipath_qp *qp, u32 psn)
 {
 	struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
 	struct ipath_ibdev *dev;
 
 	if (qp->s_retry == 0) {
-		wc->wr_id = wqe->wr.wr_id;
-		wc->status = IB_WC_RETRY_EXC_ERR;
-		wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
-		wc->vendor_err = 0;
-		wc->byte_len = 0;
-		wc->qp = &qp->ibqp;
-		wc->imm_data = 0;
-		wc->src_qp = qp->remote_qpn;
-		wc->wc_flags = 0;
-		wc->pkey_index = 0;
-		wc->slid = qp->remote_ah_attr.dlid;
-		wc->sl = qp->remote_ah_attr.sl;
-		wc->dlid_path_bits = 0;
-		wc->port_num = 0;
-		ipath_sqerror_qp(qp, wc);
+		ipath_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR);
+		ipath_error_qp(qp, IB_WC_WR_FLUSH_ERR);
 		goto bail;
 	}
 	qp->s_retry--;
@@ -804,6 +822,8 @@ void ipath_restart_rc(struct ipath_qp *qp, u32 psn, struct ib_wc *wc)
 	spin_lock(&dev->pending_lock);
 	if (!list_empty(&qp->timerwait))
 		list_del_init(&qp->timerwait);
+	if (!list_empty(&qp->piowait))
+		list_del_init(&qp->piowait);
 	spin_unlock(&dev->pending_lock);
 
 	if (wqe->wr.opcode == IB_WR_RDMA_READ)
@@ -812,7 +832,7 @@ void ipath_restart_rc(struct ipath_qp *qp, u32 psn, struct ib_wc *wc)
 		dev->n_rc_resends += (qp->s_psn - psn) & IPATH_PSN_MASK;
 
 	reset_psn(qp, psn);
-	tasklet_hi_schedule(&qp->s_task);
+	ipath_schedule_send(qp);
 
 bail:
 	return;
@@ -820,13 +840,7 @@ bail:
 
 static inline void update_last_psn(struct ipath_qp *qp, u32 psn)
 {
-	if (qp->s_last_psn != psn) {
-		qp->s_last_psn = psn;
-		if (qp->s_wait_credit) {
-			qp->s_wait_credit = 0;
-			tasklet_hi_schedule(&qp->s_task);
-		}
-	}
+	qp->s_last_psn = psn;
 }
 
 /**
@@ -845,6 +859,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 {
 	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
 	struct ib_wc wc;
+	enum ib_wc_status status;
 	struct ipath_swqe *wqe;
 	int ret = 0;
 	u32 ack_psn;
@@ -909,7 +924,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 			 */
 			update_last_psn(qp, wqe->psn - 1);
 			/* Retry this request. */
-			ipath_restart_rc(qp, wqe->psn, &wc);
+			ipath_restart_rc(qp, wqe->psn);
 			/*
 			 * No need to process the ACK/NAK since we are
 			 * restarting an earlier request.
@@ -925,32 +940,23 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 		     wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)) {
 			qp->s_num_rd_atomic--;
 			/* Restart sending task if fence is complete */
-			if ((qp->s_flags & IPATH_S_FENCE_PENDING) &&
-			    !qp->s_num_rd_atomic) {
-				qp->s_flags &= ~IPATH_S_FENCE_PENDING;
-				tasklet_hi_schedule(&qp->s_task);
-			} else if (qp->s_flags & IPATH_S_RDMAR_PENDING) {
-				qp->s_flags &= ~IPATH_S_RDMAR_PENDING;
-				tasklet_hi_schedule(&qp->s_task);
-			}
+			if (((qp->s_flags & IPATH_S_FENCE_PENDING) &&
+			     !qp->s_num_rd_atomic) ||
+			    qp->s_flags & IPATH_S_RDMAR_PENDING)
+				ipath_schedule_send(qp);
 		}
 		/* Post a send completion queue entry if requested. */
 		if (!(qp->s_flags & IPATH_S_SIGNAL_REQ_WR) ||
 		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
+			memset(&wc, 0, sizeof wc);
 			wc.wr_id = wqe->wr.wr_id;
 			wc.status = IB_WC_SUCCESS;
 			wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
-			wc.vendor_err = 0;
 			wc.byte_len = wqe->length;
-			wc.imm_data = 0;
 			wc.qp = &qp->ibqp;
 			wc.src_qp = qp->remote_qpn;
-			wc.wc_flags = 0;
-			wc.pkey_index = 0;
 			wc.slid = qp->remote_ah_attr.dlid;
 			wc.sl = qp->remote_ah_attr.sl;
-			wc.dlid_path_bits = 0;
-			wc.port_num = 0;
 			ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
 		}
 		qp->s_retry = qp->s_retry_cnt;
@@ -971,6 +977,8 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 		} else {
 			if (++qp->s_last >= qp->s_size)
 				qp->s_last = 0;
+			if (qp->state == IB_QPS_SQD && qp->s_last == qp->s_cur)
+				qp->s_draining = 0;
 			if (qp->s_last == qp->s_tail)
 				break;
 			wqe = get_swqe_ptr(qp, qp->s_last);
@@ -994,7 +1002,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 			 */
 			if (ipath_cmp24(qp->s_psn, psn) <= 0) {
 				reset_psn(qp, psn + 1);
-				tasklet_hi_schedule(&qp->s_task);
+				ipath_schedule_send(qp);
 			}
 		} else if (ipath_cmp24(qp->s_psn, psn) <= 0) {
 			qp->s_state = OP(SEND_LAST);
@@ -1012,7 +1020,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 		if (qp->s_last == qp->s_tail)
 			goto bail;
 		if (qp->s_rnr_retry == 0) {
-			wc.status = IB_WC_RNR_RETRY_EXC_ERR;
+			status = IB_WC_RNR_RETRY_EXC_ERR;
 			goto class_b;
 		}
 		if (qp->s_rnr_retry_cnt < 7)
@@ -1033,6 +1041,7 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
 			ib_ipath_rnr_table[(aeth >> IPATH_AETH_CREDIT_SHIFT) &