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: (76 commits)
  IB: Update MAINTAINERS with Hal's new email address
  IB/mlx4: Implement query SRQ
  IB/mlx4: Implement query QP
  IB/cm: Send no match if a SIDR REQ does not match a listen
  IB/cm: Fix handling of duplicate SIDR REQs
  IB/cm: cm_msgs.h should include ib_cm.h
  IB/cm: Include HCA ACK delay in local ACK timeout
  IB/cm: Use spin_lock_irq() instead of spin_lock_irqsave() when possible
  IB/sa: Make sure SA queries use default P_Key
  IPoIB: Recycle loopback skbs instead of freeing and reallocating
  IB/mthca: Replace memset(<addr>, 0, PAGE_SIZE) with clear_page(<addr>)
  IPoIB/cm: Fix warning if IPV6 is not enabled
  IB/core: Take sizeof the correct pointer when calling kmalloc()
  IB/ehca: Improve latency by unlocking after triggering the hardware
  IB/ehca: Notify consumers of LID/PKEY/SM changes after nondisruptive events
  IB/ehca: Return QP pointer in poll_cq()
  IB/ehca: Change idr spinlocks into rwlocks
  IB/ehca: Refactor sync between completions and destroy_cq using atomic_t
  IB/ehca: Lock renaming, static initializers
  IB/ehca: Report RDMA atomic attributes in query_qp()
  ...

MAINTAINERS

@@ -370,7 +370,7 @@ P:	Tom Tucker
 M:	tom@opengridcomputing.com
 P:	Steve Wise
 M:	swise@opengridcomputing.com
-L:	openib-general@openib.org
+L:	general@lists.openfabrics.org
 S:	Maintained
 
 AOA (Apple Onboard Audio) ALSA DRIVER
@@ -1395,7 +1395,7 @@ P:	Hoang-Nam Nguyen
 M:	hnguyen@de.ibm.com
 P:	Christoph Raisch
 M:	raisch@de.ibm.com
-L:	openib-general@openib.org
+L:	general@lists.openfabrics.org
 S:	Supported
 
 EMU10K1 SOUND DRIVER
@@ -1849,8 +1849,8 @@ M:	rolandd@cisco.com
 P:	Sean Hefty
 M:	mshefty@ichips.intel.com
 P:	Hal Rosenstock
-M:	halr@voltaire.com
-L:	openib-general@openib.org
+M:	hal.rosenstock@gmail.com 
+L:	general@lists.openfabrics.org
 W:	http://www.openib.org/
 T:	git kernel.org:/pub/scm/linux/kernel/git/roland/infiniband.git
 S:	Supported
@@ -1988,9 +1988,10 @@ M:	jjciarla@raiz.uncu.edu.ar
 S:	Maintained
 
 IPATH DRIVER:
-P:	Bryan O'Sullivan
-M:	support@pathscale.com
-L:	openib-general@openib.org
+P:	Arthur Jones
+M:	infinipath@qlogic.com
+L:	general@lists.openfabrics.org
+T:	git git://git.qlogic.com/ipath-linux-2.6
 S:	Supported
 
 IPMI SUBSYSTEM

drivers/infiniband/Kconfig

@@ -1,14 +1,14 @@
-menu "InfiniBand support"
-	depends on HAS_IOMEM
-
-config INFINIBAND
-	depends on PCI || BROKEN
+menuconfig INFINIBAND
 	tristate "InfiniBand support"
+	depends on PCI || BROKEN
+	depends on HAS_IOMEM
 	---help---
 	  Core support for InfiniBand (IB).  Make sure to also select
 	  any protocols you wish to use as well as drivers for your
 	  InfiniBand hardware.
 
+if INFINIBAND
+
 config INFINIBAND_USER_MAD
 	tristate "InfiniBand userspace MAD support"
 	depends on INFINIBAND
@@ -20,7 +20,6 @@ config INFINIBAND_USER_MAD
 
 config INFINIBAND_USER_ACCESS
 	tristate "InfiniBand userspace access (verbs and CM)"
-	depends on INFINIBAND
 	---help---
 	  Userspace InfiniBand access support.  This enables the
 	  kernel side of userspace verbs and the userspace
@@ -37,7 +36,7 @@ config INFINIBAND_USER_MEM
 
 config INFINIBAND_ADDR_TRANS
 	bool
-	depends on INFINIBAND && INET
+	depends on INET
 	default y
 
 source "drivers/infiniband/hw/mthca/Kconfig"
@@ -54,4 +53,4 @@ source "drivers/infiniband/ulp/srp/Kconfig"
 
 source "drivers/infiniband/ulp/iser/Kconfig"
 
-endmenu
+endif # INFINIBAND

drivers/infiniband/core/agent.c

@@ -3,7 +3,7 @@
  * Copyright (c) 2004, 2005 Infinicon Corporation.  All rights reserved.
  * Copyright (c) 2004, 2005 Intel Corporation.  All rights reserved.
  * Copyright (c) 2004, 2005 Topspin Corporation.  All rights reserved.
- * Copyright (c) 2004, 2005 Voltaire Corporation.  All rights reserved.
+ * Copyright (c) 2004-2007 Voltaire Corporation.  All rights reserved.
  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -34,7 +34,6 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: agent.c 1389 2004-12-27 22:56:47Z roland $
  */
 
 #include <linux/slab.h>
@@ -42,6 +41,7 @@
 
 #include "agent.h"
 #include "smi.h"
+#include "mad_priv.h"
 
 #define SPFX "ib_agent: "
 
@@ -87,8 +87,13 @@ int agent_send_response(struct ib_mad *mad, struct ib_grh *grh,
 	struct ib_mad_send_buf *send_buf;
 	struct ib_ah *ah;
 	int ret;
+	struct ib_mad_send_wr_private *mad_send_wr;
+
+	if (device->node_type == RDMA_NODE_IB_SWITCH)
+		port_priv = ib_get_agent_port(device, 0);
+	else
+		port_priv = ib_get_agent_port(device, port_num);
 
-	port_priv = ib_get_agent_port(device, port_num);
 	if (!port_priv) {
 		printk(KERN_ERR SPFX "Unable to find port agent\n");
 		return -ENODEV;
@@ -113,6 +118,14 @@ int agent_send_response(struct ib_mad *mad, struct ib_grh *grh,
 
 	memcpy(send_buf->mad, mad, sizeof *mad);
 	send_buf->ah = ah;
+
+	if (device->node_type == RDMA_NODE_IB_SWITCH) {
+		mad_send_wr = container_of(send_buf,
+					   struct ib_mad_send_wr_private,
+					   send_buf);
+		mad_send_wr->send_wr.wr.ud.port_num = port_num;
+	}
+
 	if ((ret = ib_post_send_mad(send_buf, NULL))) {
 		printk(KERN_ERR SPFX "ib_post_send_mad error:%d\n", ret);
 		goto err2;

drivers/infiniband/core/cm.c

@@ -87,6 +87,7 @@ struct cm_port {
 struct cm_device {
 	struct list_head list;
 	struct ib_device *device;
+	u8 ack_delay;
 	struct cm_port port[0];
 };
 
@@ -95,7 +96,7 @@ struct cm_av {
 	union ib_gid dgid;
 	struct ib_ah_attr ah_attr;
 	u16 pkey_index;
-	u8 packet_life_time;
+	u8 timeout;
 };
 
 struct cm_work {
@@ -154,6 +155,7 @@ struct cm_id_private {
 	u8 retry_count;
 	u8 rnr_retry_count;
 	u8 service_timeout;
+	u8 target_ack_delay;
 
 	struct list_head work_list;
 	atomic_t work_count;
@@ -293,7 +295,7 @@ static int cm_init_av_by_path(struct ib_sa_path_rec *path, struct cm_av *av)
 	av->port = port;
 	ib_init_ah_from_path(cm_dev->device, port->port_num, path,
 			     &av->ah_attr);
-	av->packet_life_time = path->packet_life_time;
+	av->timeout = path->packet_life_time + 1;
 	return 0;
 }
 
@@ -318,12 +320,10 @@ static int cm_alloc_id(struct cm_id_private *cm_id_priv)
 
 static void cm_free_id(__be32 local_id)
 {
-	unsigned long flags;
-
-	spin_lock_irqsave(&cm.lock, flags);
+	spin_lock_irq(&cm.lock);
 	idr_remove(&cm.local_id_table,
 		   (__force int) (local_id ^ cm.random_id_operand));
-	spin_unlock_irqrestore(&cm.lock, flags);
+	spin_unlock_irq(&cm.lock);
 }
 
 static struct cm_id_private * cm_get_id(__be32 local_id, __be32 remote_id)
@@ -345,11 +345,10 @@ static struct cm_id_private * cm_get_id(__be32 local_id, __be32 remote_id)
 static struct cm_id_private * cm_acquire_id(__be32 local_id, __be32 remote_id)
 {
 	struct cm_id_private *cm_id_priv;
-	unsigned long flags;
 
-	spin_lock_irqsave(&cm.lock, flags);
+	spin_lock_irq(&cm.lock);
 	cm_id_priv = cm_get_id(local_id, remote_id);
-	spin_unlock_irqrestore(&cm.lock, flags);
+	spin_unlock_irq(&cm.lock);
 
 	return cm_id_priv;
 }
@@ -646,6 +645,25 @@ static inline int cm_convert_to_ms(int iba_time)
 	return 1 << max(iba_time - 8, 0);
 }
 
+/*
+ * calculate: 4.096x2^ack_timeout = 4.096x2^ack_delay + 2x4.096x2^life_time
+ * Because of how ack_timeout is stored, adding one doubles the timeout.
+ * To avoid large timeouts, select the max(ack_delay, life_time + 1), and
+ * increment it (round up) only if the other is within 50%.
+ */
+static u8 cm_ack_timeout(u8 ca_ack_delay, u8 packet_life_time)
+{
+	int ack_timeout = packet_life_time + 1;
+
+	if (ack_timeout >= ca_ack_delay)
+		ack_timeout += (ca_ack_delay >= (ack_timeout - 1));
+	else
+		ack_timeout = ca_ack_delay +
+			      (ack_timeout >= (ca_ack_delay - 1));
+
+	return min(31, ack_timeout);
+}
+
 static void cm_cleanup_timewait(struct cm_timewait_info *timewait_info)
 {
 	if (timewait_info->inserted_remote_id) {
@@ -689,7 +707,7 @@ static void cm_enter_timewait(struct cm_id_private *cm_id_priv)
 	 * timewait before notifying the user that we've exited timewait.
 	 */
 	cm_id_priv->id.state = IB_CM_TIMEWAIT;
-	wait_time = cm_convert_to_ms(cm_id_priv->av.packet_life_time + 1);
+	wait_time = cm_convert_to_ms(cm_id_priv->av.timeout);
 	queue_delayed_work(cm.wq, &cm_id_priv->timewait_info->work.work,
 			   msecs_to_jiffies(wait_time));
 	cm_id_priv->timewait_info = NULL;
@@ -713,31 +731,30 @@ static void cm_destroy_id(struct ib_cm_id *cm_id, int err)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_work *work;
-	unsigned long flags;
 
 	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
 retest:
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id->state) {
 	case IB_CM_LISTEN:
 		cm_id->state = IB_CM_IDLE;
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
-		spin_lock_irqsave(&cm.lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
+		spin_lock_irq(&cm.lock);
 		rb_erase(&cm_id_priv->service_node, &cm.listen_service_table);
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 		break;
 	case IB_CM_SIDR_REQ_SENT:
 		cm_id->state = IB_CM_IDLE;
 		ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		break;
 	case IB_CM_SIDR_REQ_RCVD:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		cm_reject_sidr_req(cm_id_priv, IB_SIDR_REJECT);
 		break;
 	case IB_CM_REQ_SENT:
 		ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ib_send_cm_rej(cm_id, IB_CM_REJ_TIMEOUT,
 			       &cm_id_priv->id.device->node_guid,
 			       sizeof cm_id_priv->id.device->node_guid,
@@ -747,9 +764,9 @@ retest:
 		if (err == -ENOMEM) {
 			/* Do not reject to allow future retries. */
 			cm_reset_to_idle(cm_id_priv);
-			spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+			spin_unlock_irq(&cm_id_priv->lock);
 		} else {
-			spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+			spin_unlock_irq(&cm_id_priv->lock);
 			ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
 				       NULL, 0, NULL, 0);
 		}
@@ -762,25 +779,25 @@ retest:
 	case IB_CM_MRA_REQ_SENT:
 	case IB_CM_REP_RCVD:
 	case IB_CM_MRA_REP_SENT:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
 			       NULL, 0, NULL, 0);
 		break;
 	case IB_CM_ESTABLISHED:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ib_send_cm_dreq(cm_id, NULL, 0);
 		goto retest;
 	case IB_CM_DREQ_SENT:
 		ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
 		cm_enter_timewait(cm_id_priv);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		break;
 	case IB_CM_DREQ_RCVD:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ib_send_cm_drep(cm_id, NULL, 0);
 		break;
 	default:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		break;
 	}
 
@@ -912,7 +929,8 @@ static void cm_format_req(struct cm_req_msg *req_msg,
 	cm_req_set_primary_sl(req_msg, param->primary_path->sl);
 	cm_req_set_primary_subnet_local(req_msg, 1); /* local only... */
 	cm_req_set_primary_local_ack_timeout(req_msg,
-		min(31, param->primary_path->packet_life_time + 1));
+		cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay,
+			       param->primary_path->packet_life_time));
 
 	if (param->alternate_path) {
 		req_msg->alt_local_lid = param->alternate_path->slid;
@@ -927,7 +945,8 @@ static void cm_format_req(struct cm_req_msg *req_msg,
 		cm_req_set_alt_sl(req_msg, param->alternate_path->sl);
 		cm_req_set_alt_subnet_local(req_msg, 1); /* local only... */
 		cm_req_set_alt_local_ack_timeout(req_msg,
-			min(31, param->alternate_path->packet_life_time + 1));
+			cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay,
+				       param->alternate_path->packet_life_time));
 	}
 
 	if (param->private_data && param->private_data_len)
@@ -1169,7 +1188,6 @@ static void cm_format_req_event(struct cm_work *work,
 static void cm_process_work(struct cm_id_private *cm_id_priv,
 			    struct cm_work *work)
 {
-	unsigned long flags;
 	int ret;
 
 	/* We will typically only have the current event to report. */
@@ -1177,9 +1195,9 @@ static void cm_process_work(struct cm_id_private *cm_id_priv,
 	cm_free_work(work);
 
 	while (!ret && !atomic_add_negative(-1, &cm_id_priv->work_count)) {
-		spin_lock_irqsave(&cm_id_priv->lock, flags);
+		spin_lock_irq(&cm_id_priv->lock);
 		work = cm_dequeue_work(cm_id_priv);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		BUG_ON(!work);
 		ret = cm_id_priv->id.cm_handler(&cm_id_priv->id,
 						&work->cm_event);
@@ -1250,7 +1268,6 @@ static void cm_dup_req_handler(struct cm_work *work,
 			       struct cm_id_private *cm_id_priv)
 {
 	struct ib_mad_send_buf *msg = NULL;
-	unsigned long flags;
 	int ret;
 
 	/* Quick state check to discard duplicate REQs. */
@@ -1261,7 +1278,7 @@ static void cm_dup_req_handler(struct cm_work *work,
 	if (ret)
 		return;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id_priv->id.state) {
 	case IB_CM_MRA_REQ_SENT:
 		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
@@ -1276,14 +1293,14 @@ static void cm_dup_req_handler(struct cm_work *work,
 	default:
 		goto unlock;
 	}
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	ret = ib_post_send_mad(msg, NULL);
 	if (ret)
 		goto free;
 	return;
 
-unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+unlock:	spin_unlock_irq(&cm_id_priv->lock);
 free:	cm_free_msg(msg);
 }
 
@@ -1293,17 +1310,16 @@ static struct cm_id_private * cm_match_req(struct cm_work *work,
 	struct cm_id_private *listen_cm_id_priv, *cur_cm_id_priv;
 	struct cm_timewait_info *timewait_info;
 	struct cm_req_msg *req_msg;
-	unsigned long flags;
 
 	req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
 
 	/* Check for possible duplicate REQ. */
-	spin_lock_irqsave(&cm.lock, flags);
+	spin_lock_irq(&cm.lock);
 	timewait_info = cm_insert_remote_id(cm_id_priv->timewait_info);
 	if (timewait_info) {
 		cur_cm_id_priv = cm_get_id(timewait_info->work.local_id,
 					   timewait_info->work.remote_id);
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 		if (cur_cm_id_priv) {
 			cm_dup_req_handler(work, cur_cm_id_priv);
 			cm_deref_id(cur_cm_id_priv);
@@ -1315,7 +1331,7 @@ static struct cm_id_private * cm_match_req(struct cm_work *work,
 	timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info);
 	if (timewait_info) {
 		cm_cleanup_timewait(cm_id_priv->timewait_info);
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 		cm_issue_rej(work->port, work->mad_recv_wc,
 			     IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REQ,
 			     NULL, 0);
@@ -1328,7 +1344,7 @@ static struct cm_id_private * cm_match_req(struct cm_work *work,
 					   req_msg->private_data);
 	if (!listen_cm_id_priv) {
 		cm_cleanup_timewait(cm_id_priv->timewait_info);
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 		cm_issue_rej(work->port, work->mad_recv_wc,
 			     IB_CM_REJ_INVALID_SERVICE_ID, CM_MSG_RESPONSE_REQ,
 			     NULL, 0);
@@ -1338,7 +1354,7 @@ static struct cm_id_private * cm_match_req(struct cm_work *work,
 	atomic_inc(&cm_id_priv->refcount);
 	cm_id_priv->id.state = IB_CM_REQ_RCVD;
 	atomic_inc(&cm_id_priv->work_count);
-	spin_unlock_irqrestore(&cm.lock, flags);
+	spin_unlock_irq(&cm.lock);
 out:
 	return listen_cm_id_priv;
 }
@@ -1440,7 +1456,8 @@ static void cm_format_rep(struct cm_rep_msg *rep_msg,
 	cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn));
 	rep_msg->resp_resources = param->responder_resources;
 	rep_msg->initiator_depth = param->initiator_depth;
-	cm_rep_set_target_ack_delay(rep_msg, param->target_ack_delay);
+	cm_rep_set_target_ack_delay(rep_msg,
+				    cm_id_priv->av.port->cm_dev->ack_delay);
 	cm_rep_set_failover(rep_msg, param->failover_accepted);
 	cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
 	cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count);
@@ -1591,7 +1608,6 @@ static void cm_dup_rep_handler(struct cm_work *work)
 	struct cm_id_private *cm_id_priv;
 	struct cm_rep_msg *rep_msg;
 	struct ib_mad_send_buf *msg = NULL;
-	unsigned long flags;
 	int ret;
 
 	rep_msg = (struct cm_rep_msg *) work->mad_recv_wc->recv_buf.mad;
@@ -1604,7 +1620,7 @@ static void cm_dup_rep_handler(struct cm_work *work)
 	if (ret)
 		goto deref;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state == IB_CM_ESTABLISHED)
 		cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
 			      cm_id_priv->private_data,
@@ -1616,14 +1632,14 @@ static void cm_dup_rep_handler(struct cm_work *work)
 			      cm_id_priv->private_data_len);
 	else
 		goto unlock;
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	ret = ib_post_send_mad(msg, NULL);
 	if (ret)
 		goto free;
 	goto deref;
 
-unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+unlock:	spin_unlock_irq(&cm_id_priv->lock);
 free:	cm_free_msg(msg);
 deref:	cm_deref_id(cm_id_priv);
 }
@@ -1632,7 +1648,6 @@ static int cm_rep_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_rep_msg *rep_msg;
-	unsigned long flags;
 	int ret;
 
 	rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -1644,13 +1659,13 @@ static int cm_rep_handler(struct cm_work *work)
 
 	cm_format_rep_event(work);
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id_priv->id.state) {
 	case IB_CM_REQ_SENT:
 	case IB_CM_MRA_REQ_RCVD:
 		break;
 	default:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ret = -EINVAL;
 		goto error;
 	}
@@ -1663,7 +1678,7 @@ static int cm_rep_handler(struct cm_work *work)
 	/* Check for duplicate REP. */
 	if (cm_insert_remote_id(cm_id_priv->timewait_info)) {
 		spin_unlock(&cm.lock);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ret = -EINVAL;
 		goto error;
 	}
@@ -1673,7 +1688,7 @@ static int cm_rep_handler(struct cm_work *work)
 			 &cm.remote_id_table);
 		cm_id_priv->timewait_info->inserted_remote_id = 0;
 		spin_unlock(&cm.lock);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		cm_issue_rej(work->port, work->mad_recv_wc,
 			     IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REP,
 			     NULL, 0);
@@ -1689,6 +1704,13 @@ static int cm_rep_handler(struct cm_work *work)
 	cm_id_priv->responder_resources = rep_msg->initiator_depth;
 	cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg);
 	cm_id_priv->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
+	cm_id_priv->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg);
+	cm_id_priv->av.timeout =
+			cm_ack_timeout(cm_id_priv->target_ack_delay,
+				       cm_id_priv->av.timeout - 1);
+	cm_id_priv->alt_av.timeout =
+			cm_ack_timeout(cm_id_priv->target_ack_delay,
+				       cm_id_priv->alt_av.timeout - 1);
 
 	/* todo: handle peer_to_peer */
 
@@ -1696,7 +1718,7 @@ static int cm_rep_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -1712,7 +1734,6 @@ error:
 static int cm_establish_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
-	unsigned long flags;
 	int ret;
 
 	/* See comment in cm_establish about lookup. */
@@ -1720,9 +1741,9 @@ static int cm_establish_handler(struct cm_work *work)
 	if (!cm_id_priv)
 		return -EINVAL;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_ESTABLISHED) {
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		goto out;
 	}
 
@@ -1730,7 +1751,7 @@ static int cm_establish_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -1746,7 +1767,6 @@ static int cm_rtu_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_rtu_msg *rtu_msg;
-	unsigned long flags;
 	int ret;
 
 	rtu_msg = (struct cm_rtu_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -1757,10 +1777,10 @@ static int cm_rtu_handler(struct cm_work *work)
 
 	work->cm_event.private_data = &rtu_msg->private_data;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_REP_SENT &&
 	    cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) {
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		goto out;
 	}
 	cm_id_priv->id.state = IB_CM_ESTABLISHED;
@@ -1769,7 +1789,7 @@ static int cm_rtu_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -1932,7 +1952,6 @@ static int cm_dreq_handler(struct cm_work *work)
 	struct cm_id_private *cm_id_priv;
 	struct cm_dreq_msg *dreq_msg;
 	struct ib_mad_send_buf *msg = NULL;
-	unsigned long flags;
 	int ret;
 
 	dreq_msg = (struct cm_dreq_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -1945,7 +1964,7 @@ static int cm_dreq_handler(struct cm_work *work)
 
 	work->cm_event.private_data = &dreq_msg->private_data;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->local_qpn != cm_dreq_get_remote_qpn(dreq_msg))
 		goto unlock;
 
@@ -1964,7 +1983,7 @@ static int cm_dreq_handler(struct cm_work *work)
 		cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
 			       cm_id_priv->private_data,
 			       cm_id_priv->private_data_len);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 
 		if (ib_post_send_mad(msg, NULL))
 			cm_free_msg(msg);
@@ -1977,7 +1996,7 @@ static int cm_dreq_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -1985,7 +2004,7 @@ static int cm_dreq_handler(struct cm_work *work)
 		cm_deref_id(cm_id_priv);
 	return 0;
 
-unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+unlock:	spin_unlock_irq(&cm_id_priv->lock);
 deref:	cm_deref_id(cm_id_priv);
 	return -EINVAL;
 }
@@ -1994,7 +2013,6 @@ static int cm_drep_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_drep_msg *drep_msg;
-	unsigned long flags;
 	int ret;
 
 	drep_msg = (struct cm_drep_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -2005,10 +2023,10 @@ static int cm_drep_handler(struct cm_work *work)
 
 	work->cm_event.private_data = &drep_msg->private_data;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_DREQ_SENT &&
 	    cm_id_priv->id.state != IB_CM_DREQ_RCVD) {
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		goto out;
 	}
 	cm_enter_timewait(cm_id_priv);
@@ -2017,7 +2035,7 @@ static int cm_drep_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -2107,17 +2125,16 @@ static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg)
 {
 	struct cm_timewait_info *timewait_info;
 	struct cm_id_private *cm_id_priv;
-	unsigned long flags;
 	__be32 remote_id;
 
 	remote_id = rej_msg->local_comm_id;
 
 	if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_TIMEOUT) {
-		spin_lock_irqsave(&cm.lock, flags);
+		spin_lock_irq(&cm.lock);
 		timewait_info = cm_find_remote_id( *((__be64 *) rej_msg->ari),
 						  remote_id);
 		if (!timewait_info) {
-			spin_unlock_irqrestore(&cm.lock, flags);
+			spin_unlock_irq(&cm.lock);
 			return NULL;
 		}
 		cm_id_priv = idr_find(&cm.local_id_table, (__force int)
@@ -2129,7 +2146,7 @@ static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg)
 			else
 				cm_id_priv = NULL;
 		}
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 	} else if (cm_rej_get_msg_rejected(rej_msg) == CM_MSG_RESPONSE_REQ)
 		cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, 0);
 	else
@@ -2142,7 +2159,6 @@ static int cm_rej_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_rej_msg *rej_msg;
-	unsigned long flags;
 	int ret;
 
 	rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -2152,7 +2168,7 @@ static int cm_rej_handler(struct cm_work *work)
 
 	cm_format_rej_event(work);
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id_priv->id.state) {
 	case IB_CM_REQ_SENT:
 	case IB_CM_MRA_REQ_RCVD:
@@ -2176,7 +2192,7 @@ static int cm_rej_handler(struct cm_work *work)
 		cm_enter_timewait(cm_id_priv);
 		break;
 	default:
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		ret = -EINVAL;
 		goto out;
 	}
@@ -2184,7 +2200,7 @@ static int cm_rej_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -2295,7 +2311,6 @@ static int cm_mra_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_mra_msg *mra_msg;
-	unsigned long flags;
 	int timeout, ret;
 
 	mra_msg = (struct cm_mra_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -2307,9 +2322,9 @@ static int cm_mra_handler(struct cm_work *work)
 	work->cm_event.param.mra_rcvd.service_timeout =
 					cm_mra_get_service_timeout(mra_msg);
 	timeout = cm_convert_to_ms(cm_mra_get_service_timeout(mra_msg)) +
-		  cm_convert_to_ms(cm_id_priv->av.packet_life_time);
+		  cm_convert_to_ms(cm_id_priv->av.timeout);
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id_priv->id.state) {
 	case IB_CM_REQ_SENT:
 		if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REQ ||
@@ -2342,7 +2357,7 @@ static int cm_mra_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -2350,7 +2365,7 @@ static int cm_mra_handler(struct cm_work *work)
 		cm_deref_id(cm_id_priv);
 	return 0;
 out:
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 	cm_deref_id(cm_id_priv);
 	return -EINVAL;
 }
@@ -2379,7 +2394,8 @@ static void cm_format_lap(struct cm_lap_msg *lap_msg,
 	cm_lap_set_sl(lap_msg, alternate_path->sl);
 	cm_lap_set_subnet_local(lap_msg, 1); /* local only... */
 	cm_lap_set_local_ack_timeout(lap_msg,
-		min(31, alternate_path->packet_life_time + 1));
+		cm_ack_timeout(cm_id_priv->av.port->cm_dev->ack_delay,
+			       alternate_path->packet_life_time));
 
 	if (private_data && private_data_len)
 		memcpy(lap_msg->private_data, private_data, private_data_len);
@@ -2410,6 +2426,9 @@ int ib_send_cm_lap(struct ib_cm_id *cm_id,
 	ret = cm_init_av_by_path(alternate_path, &cm_id_priv->alt_av);
 	if (ret)
 		goto out;
+	cm_id_priv->alt_av.timeout =
+			cm_ack_timeout(cm_id_priv->target_ack_delay,
+				       cm_id_priv->alt_av.timeout - 1);
 
 	ret = cm_alloc_msg(cm_id_priv, &msg);
 	if (ret)
@@ -2465,7 +2484,6 @@ static int cm_lap_handler(struct cm_work *work)
 	struct cm_lap_msg *lap_msg;
 	struct ib_cm_lap_event_param *param;
 	struct ib_mad_send_buf *msg = NULL;
-	unsigned long flags;
 	int ret;
 
 	/* todo: verify LAP request and send reject APR if invalid. */
@@ -2480,7 +2498,7 @@ static int cm_lap_handler(struct cm_work *work)
 	cm_format_path_from_lap(cm_id_priv, param->alternate_path, lap_msg);
 	work->cm_event.private_data = &lap_msg->private_data;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_ESTABLISHED)
 		goto unlock;
 
@@ -2497,7 +2515,7 @@ static int cm_lap_handler(struct cm_work *work)
 			      cm_id_priv->service_timeout,
 			      cm_id_priv->private_data,
 			      cm_id_priv->private_data_len);
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 
 		if (ib_post_send_mad(msg, NULL))
 			cm_free_msg(msg);
@@ -2515,7 +2533,7 @@ static int cm_lap_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -2523,7 +2541,7 @@ static int cm_lap_handler(struct cm_work *work)
 		cm_deref_id(cm_id_priv);
 	return 0;
 
-unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+unlock:	spin_unlock_irq(&cm_id_priv->lock);
 deref:	cm_deref_id(cm_id_priv);
 	return -EINVAL;
 }
@@ -2598,7 +2616,6 @@ static int cm_apr_handler(struct cm_work *work)
 {
 	struct cm_id_private *cm_id_priv;
 	struct cm_apr_msg *apr_msg;
-	unsigned long flags;
 	int ret;
 
 	apr_msg = (struct cm_apr_msg *)work->mad_recv_wc->recv_buf.mad;
@@ -2612,11 +2629,11 @@ static int cm_apr_handler(struct cm_work *work)
 	work->cm_event.param.apr_rcvd.info_len = apr_msg->info_length;
 	work->cm_event.private_data = &apr_msg->private_data;
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_ESTABLISHED ||
 	    (cm_id_priv->id.lap_state != IB_CM_LAP_SENT &&
 	     cm_id_priv->id.lap_state != IB_CM_MRA_LAP_RCVD)) {
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		goto out;
 	}
 	cm_id_priv->id.lap_state = IB_CM_LAP_IDLE;
@@ -2626,7 +2643,7 @@ static int cm_apr_handler(struct cm_work *work)
 	ret = atomic_inc_and_test(&cm_id_priv->work_count);
 	if (!ret)
 		list_add_tail(&work->list, &cm_id_priv->work_list);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	if (ret)
 		cm_process_work(cm_id_priv, work);
@@ -2761,7 +2778,6 @@ static int cm_sidr_req_handler(struct cm_work *work)
 	struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
 	struct cm_sidr_req_msg *sidr_req_msg;
 	struct ib_wc *wc;
-	unsigned long flags;
 
 	cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
 	if (IS_ERR(cm_id))
@@ -2778,27 +2794,26 @@ static int cm_sidr_req_handler(struct cm_work *work)
 				work->mad_recv_wc->recv_buf.grh,
 				&cm_id_priv->av);
 	cm_id_priv->id.remote_id = sidr_req_msg->request_id;
-	cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
 	cm_id_priv->tid = sidr_req_msg->hdr.tid;
 	atomic_inc(&cm_id_priv->work_count);
 
-	spin_lock_irqsave(&cm.lock, flags);
+	spin_lock_irq(&cm.lock);
 	cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv);
 	if (cur_cm_id_priv) {
-		spin_unlock_irqrestore(&cm.lock, flags);
+		spin_unlock_irq(&cm.lock);
 		goto out; /* Duplicate message. */
 	}
+	cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
 	cur_cm_id_priv = cm_find_listen(cm_id->device,
 					sidr_req_msg->service_id,
 					sidr_req_msg->private_data);
 	if (!cur_cm_id_priv) {
-		rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
-		spin_unlock_irqrestore(&cm.lock, flags);
-		/* todo: reply with no match */
+		spin_unlock_irq(&cm.lock);
+		cm_reject_sidr_req(cm_id_priv, IB_SIDR_UNSUPPORTED);
 		goto out; /* No match. */
 	}
 	atomic_inc(&cur_cm_id_priv->refcount);
-	spin_unlock_irqrestore(&cm.lock, flags);
+	spin_unlock_irq(&cm.lock);
 
 	cm_id_priv->id.cm_handler = cur_cm_id_priv->id.cm_handler;
 	cm_id_priv->id.context = cur_cm_id_priv->id.context;
@@ -2899,7 +2914,6 @@ static int cm_sidr_rep_handler(struct cm_work *work)
 {
 	struct cm_sidr_rep_msg *sidr_rep_msg;
 	struct cm_id_private *cm_id_priv;
-	unsigned long flags;
 
 	sidr_rep_msg = (struct cm_sidr_rep_msg *)
 				work->mad_recv_wc->recv_buf.mad;
@@ -2907,14 +2921,14 @@ static int cm_sidr_rep_handler(struct cm_work *work)
 	if (!cm_id_priv)
 		return -EINVAL; /* Unmatched reply. */
 
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	if (cm_id_priv->id.state != IB_CM_SIDR_REQ_SENT) {
-		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_unlock_irq(&cm_id_priv->lock);
 		goto out;
 	}
 	cm_id_priv->id.state = IB_CM_IDLE;
 	ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 
 	cm_format_sidr_rep_event(work);
 	cm_process_work(cm_id_priv, work);
@@ -2930,14 +2944,13 @@ static void cm_process_send_error(struct ib_mad_send_buf *msg,
 	struct cm_id_private *cm_id_priv;
 	struct ib_cm_event cm_event;
 	enum ib_cm_state state;
-	unsigned long flags;
 	int ret;
 
 	memset(&cm_event, 0, sizeof cm_event);
 	cm_id_priv = msg->context[0];
 
 	/* Discard old sends or ones without a response. */
-	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	spin_lock_irq(&cm_id_priv->lock);
 	state = (enum ib_cm_state) (unsigned long) msg->context[1];
 	if (msg != cm_id_priv->msg || state != cm_id_priv->id.state)
 		goto discard;
@@ -2964,7 +2977,7 @@ static void cm_process_send_error(struct ib_mad_send_buf *msg,
 	default:
 		goto discard;
 	}
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 	cm_event.param.send_status = wc_status;
 
 	/* No other events can occur on the cm_id at this point. */
@@ -2974,7 +2987,7 @@ static void cm_process_send_error(struct ib_mad_send_buf *msg,
 		ib_destroy_cm_id(&cm_id_priv->id);
 	return;
 discard:
-	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	spin_unlock_irq(&cm_id_priv->lock);
 	cm_free_msg(msg);
 }
 
@@ -3269,8 +3282,7 @@ static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
 			*qp_attr_mask |= IB_QP_ALT_PATH;
 			qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num;
 			qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index;
-			qp_attr->alt_timeout =
-					cm_id_priv->alt_av.packet_life_time + 1;
+			qp_attr->alt_timeout = cm_id_priv->alt_av.timeout;
 			qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
 		}
 		ret = 0;
@@ -3308,8 +3320,7 @@ static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
 				*qp_attr_mask |= IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
 						 IB_QP_RNR_RETRY |
 						 IB_QP_MAX_QP_RD_ATOMIC;
-				qp_attr->timeout =
-					cm_id_priv->av.packet_life_time + 1;
+				qp_attr->timeout = cm_id_priv->av.timeout;
 				qp_attr->retry_cnt = cm_id_priv->retry_count;
 				qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
 				qp_attr->max_rd_atomic =
@@ -3323,8 +3334,7 @@ static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
 			*qp_attr_mask = IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE;
 			qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num;
 			qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index;
-			qp_attr->alt_timeout =
-				cm_id_priv->alt_av.packet_life_time + 1;
+			qp_attr->alt_timeout = cm_id_priv->alt_av.timeout;
 			qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
 			qp_attr->path_mig_state = IB_MIG_REARM;
 		}
@@ -3364,6 +3374,16 @@ int ib_cm_init_qp_attr(struct ib_cm_id *cm_id,
 }
 EXPORT_SYMBOL(ib_cm_init_qp_attr);
 
+void cm_get_ack_delay(struct cm_device *cm_dev)
+{
+	struct ib_device_attr attr;
+
+	if (ib_query_device(cm_dev->device, &attr))
+		cm_dev->ack_delay = 0; /* acks will rely on packet life time */
+	else
+		cm_dev->ack_delay = attr.local_ca_ack_delay;
+}
+
 static void cm_add_one(struct ib_device *device)
 {
 	struct cm_device *cm_dev;
@@ -3388,6 +3408,7 @@ static void cm_add_one(struct ib_device *device)
 		return;
 
 	cm_dev->device = device;
+	cm_get_ack_delay(cm_dev);
 
 	set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask);
 	for (i = 1; i <= device->phys_port_cnt; i++) {

drivers/infiniband/core/cm_msgs.h

@@ -35,6 +35,7 @@
 #define CM_MSGS_H
 
 #include <rdma/ib_mad.h>
+#include <rdma/ib_cm.h>
 
 /*
  * Parameters to routines below should be in network-byte order, and values

drivers/infiniband/core/cma.c

@@ -2326,7 +2326,6 @@ static int cma_accept_ib(struct rdma_id_private *id_priv,
 	rep.private_data_len = conn_param->private_data_len;
 	rep.responder_resources = conn_param->responder_resources;
 	rep.initiator_depth = conn_param->initiator_depth;
-	rep.target_ack_delay = CMA_CM_RESPONSE_TIMEOUT;
 	rep.failover_accepted = 0;
 	rep.flow_control = conn_param->flow_control;
 	rep.rnr_retry_count = conn_param->rnr_retry_count;

drivers/infiniband/core/mad.c

@@ -675,10 +675,16 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 	struct ib_mad_port_private *port_priv;
 	struct ib_mad_agent_private *recv_mad_agent = NULL;
 	struct ib_device *device = mad_agent_priv->agent.device;
-	u8 port_num = mad_agent_priv->agent.port_num;
+	u8 port_num;
 	struct ib_wc mad_wc;
 	struct ib_send_wr *send_wr = &mad_send_wr->send_wr;
 
+	if (device->node_type == RDMA_NODE_IB_SWITCH &&
+	    smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		port_num = send_wr->wr.ud.port_num;
+	else
+		port_num = mad_agent_priv->agent.port_num;
+
 	/*
 	 * Directed route handling starts if the initial LID routed part of
 	 * a request or the ending LID routed part of a response is empty.
@@ -1839,6 +1845,7 @@ static void ib_mad_recv_done_handler(struct ib_mad_port_private *port_priv,
 	struct ib_mad_private *recv, *response;
 	struct ib_mad_list_head *mad_list;
 	struct ib_mad_agent_private *mad_agent;
+	int port_num;
 
 	response = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL);
 	if (!response)
@@ -1872,25 +1879,50 @@ static void ib_mad_recv_done_handler(struct ib_mad_port_private *port_priv,
 	if (!validate_mad(&recv->mad.mad, qp_info->qp->qp_num))
 		goto out;
 
+	if (port_priv->device->node_type == RDMA_NODE_IB_SWITCH)
+		port_num = wc->port_num;
+	else
+		port_num = port_priv->port_num;
+
 	if (recv->mad.mad.mad_hdr.mgmt_class ==
 	    IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		enum smi_forward_action retsmi;
+
 		if (smi_handle_dr_smp_recv(&recv->mad.smp,
 					   port_priv->device->node_type,
-					   port_priv->port_num,
+					   port_num,
 					   port_priv->device->phys_port_cnt) ==
 					   IB_SMI_DISCARD)
 			goto out;
 
-		if (smi_check_forward_dr_smp(&recv->mad.smp) == IB_SMI_LOCAL)
+		retsmi = smi_check_forward_dr_smp(&recv->mad.smp);
+		if (retsmi == IB_SMI_LOCAL)
 			goto local;
 
-		if (smi_handle_dr_smp_send(&recv->mad.smp,
-					   port_priv->device->node_type,
-					   port_priv->port_num) == IB_SMI_DISCARD)
-			goto out;
+		if (retsmi == IB_SMI_SEND) { /* don't forward */
+			if (smi_handle_dr_smp_send(&recv->mad.smp,
+						   port_priv->device->node_type,
+						   port_num) == IB_SMI_DISCARD)
+				goto out;
+
+			if (smi_check_local_smp(&recv->mad.smp, port_priv->device) == IB_SMI_DISCARD)
+				goto out;
+		} else if (port_priv->device->node_type == RDMA_NODE_IB_SWITCH) {
+			/* forward case for switches */
+			memcpy(response, recv, sizeof(*response));
+			response->header.recv_wc.wc = &response->header.wc;
+			response->header.recv_wc.recv_buf.mad = &response->mad.mad;
+			response->header.recv_wc.recv_buf.grh = &response->grh;
+
+			if (!agent_send_response(&response->mad.mad,
+						 &response->grh, wc,
+						 port_priv->device,
+						 smi_get_fwd_port(&recv->mad.smp),
+						 qp_info->qp->qp_num))
+				response = NULL;
 
-		if (smi_check_local_smp(&recv->mad.smp, port_priv->device) == IB_SMI_DISCARD)
 			goto out;
+		}
 	}
 
 local:
@@ -1919,7 +1951,7 @@ local:
 				agent_send_response(&response->mad.mad,
 						    &recv->grh, wc,
 						    port_priv->device,
-						    port_priv->port_num,
+						    port_num,
 						    qp_info->qp->qp_num);
 				goto out;
 			}

drivers/infiniband/core/multicast.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU

drivers/infiniband/core/sa.h

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
- * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/core/sa_query.c

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
- * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -56,6 +56,7 @@ MODULE_LICENSE("Dual BSD/GPL");
 struct ib_sa_sm_ah {
 	struct ib_ah        *ah;
 	struct kref          ref;
+	u16		     pkey_index;
 	u8		     src_path_mask;
 };
 
@@ -382,6 +383,13 @@ static void update_sm_ah(struct work_struct *work)
 	kref_init(&new_ah->ref);
 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
 
+	new_ah->pkey_index = 0;
+	if (ib_find_pkey(port->agent->device, port->port_num,
+			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index) &&
+	    ib_find_pkey(port->agent->device, port->port_num,
+			 IB_DEFAULT_PKEY_PARTIAL, &new_ah->pkey_index))
+		printk(KERN_ERR "Couldn't find index for default PKey\n");
+
 	memset(&ah_attr, 0, sizeof ah_attr);
 	ah_attr.dlid     = port_attr.sm_lid;
 	ah_attr.sl       = port_attr.sm_sl;
@@ -512,6 +520,35 @@ int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
 }
 EXPORT_SYMBOL(ib_init_ah_from_path);
 
+static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&query->port->ah_lock, flags);
+	kref_get(&query->port->sm_ah->ref);
+	query->sm_ah = query->port->sm_ah;
+	spin_unlock_irqrestore(&query->port->ah_lock, flags);
+
+	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
+					    query->sm_ah->pkey_index,
+					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
+					    gfp_mask);
+	if (!query->mad_buf) {
+		kref_put(&query->sm_ah->ref, free_sm_ah);
+		return -ENOMEM;
+	}
+
+	query->mad_buf->ah = query->sm_ah->ah;
+
+	return 0;
+}
+
+static void free_mad(struct ib_sa_query *query)
+{
+	ib_free_send_mad(query->mad_buf);
+	kref_put(&query->sm_ah->ref, free_sm_ah);
+}
+
 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
 {
 	unsigned long flags;
@@ -548,20 +585,11 @@ retry:
 	query->mad_buf->context[0] = query;
 	query->id = id;
 
-	spin_lock_irqsave(&query->port->ah_lock, flags);
-	kref_get(&query->port->sm_ah->ref);
-	query->sm_ah = query->port->sm_ah;
-	spin_unlock_irqrestore(&query->port->ah_lock, flags);
-
-	query->mad_buf->ah = query->sm_ah->ah;
-
 	ret = ib_post_send_mad(query->mad_buf, NULL);
 	if (ret) {
 		spin_lock_irqsave(&idr_lock, flags);
 		idr_remove(&query_idr, id);
 		spin_unlock_irqrestore(&idr_lock, flags);
-
-		kref_put(&query->sm_ah->ref, free_sm_ah);
 	}
 
 	/*
@@ -647,13 +675,10 @@ int ib_sa_path_rec_get(struct ib_sa_client *client,
 	if (!query)
 		return -ENOMEM;
 
-	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
-						     0, IB_MGMT_SA_HDR,
-						     IB_MGMT_SA_DATA, gfp_mask);
-	if (!query->sa_query.mad_buf) {
-		ret = -ENOMEM;
+	query->sa_query.port     = port;
+	ret = alloc_mad(&query->sa_query, gfp_mask);
+	if (ret)
 		goto err1;
-	}
 
 	ib_sa_client_get(client);
 	query->sa_query.client = client;
@@ -665,7 +690,6 @@ int ib_sa_path_rec_get(struct ib_sa_client *client,
 
 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
 	query->sa_query.release  = ib_sa_path_rec_release;
-	query->sa_query.port     = port;
 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
 	mad->sa_hdr.comp_mask	 = comp_mask;
@@ -683,7 +707,7 @@ int ib_sa_path_rec_get(struct ib_sa_client *client,
 err2:
 	*sa_query = NULL;
 	ib_sa_client_put(query->sa_query.client);
-	ib_free_send_mad(query->sa_query.mad_buf);
+	free_mad(&query->sa_query);
 
 err1:
 	kfree(query);
@@ -773,13 +797,10 @@ int ib_sa_service_rec_query(struct ib_sa_client *client,
 	if (!query)
 		return -ENOMEM;
 
-	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
-						     0, IB_MGMT_SA_HDR,
-						     IB_MGMT_SA_DATA, gfp_mask);
-	if (!query->sa_query.mad_buf) {
-		ret = -ENOMEM;
+	query->sa_query.port     = port;
+	ret = alloc_mad(&query->sa_query, gfp_mask);
+	if (ret)
 		goto err1;
-	}
 
 	ib_sa_client_get(client);
 	query->sa_query.client = client;
@@ -791,7 +812,6 @@ int ib_sa_service_rec_query(struct ib_sa_client *client,
 
 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
 	query->sa_query.release  = ib_sa_service_rec_release;
-	query->sa_query.port     = port;
 	mad->mad_hdr.method	 = method;
 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
 	mad->sa_hdr.comp_mask	 = comp_mask;
@@ -810,7 +830,7 @@ int ib_sa_service_rec_query(struct ib_sa_client *client,
 err2:
 	*sa_query = NULL;
 	ib_sa_client_put(query->sa_query.client);
-	ib_free_send_mad(query->sa_query.mad_buf);
+	free_mad(&query->sa_query);
 
 err1:
 	kfree(query);
@@ -869,13 +889,10 @@ int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
 	if (!query)
 		return -ENOMEM;
 
-	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
-						     0, IB_MGMT_SA_HDR,
-						     IB_MGMT_SA_DATA, gfp_mask);
-	if (!query->sa_query.mad_buf) {
-		ret = -ENOMEM;
+	query->sa_query.port     = port;
+	ret = alloc_mad(&query->sa_query, gfp_mask);
+	if (ret)
 		goto err1;
-	}
 
 	ib_sa_client_get(client);
 	query->sa_query.client = client;
@@ -887,7 +904,6 @@ int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
 
 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
 	query->sa_query.release  = ib_sa_mcmember_rec_release;
-	query->sa_query.port     = port;
 	mad->mad_hdr.method	 = method;
 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
 	mad->sa_hdr.comp_mask	 = comp_mask;
@@ -906,7 +922,7 @@ int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
 err2:
 	*sa_query = NULL;
 	ib_sa_client_put(query->sa_query.client);
-	ib_free_send_mad(query->sa_query.mad_buf);
+	free_mad(&query->sa_query);
 
 err1:
 	kfree(query);
@@ -939,8 +955,7 @@ static void send_handler(struct ib_mad_agent *agent,
 	idr_remove(&query_idr, query->id);
 	spin_unlock_irqrestore(&idr_lock, flags);
 
-	ib_free_send_mad(mad_send_wc->send_buf);
-	kref_put(&query->sm_ah->ref, free_sm_ah);
+	free_mad(query);
 	ib_sa_client_put(query->client);
 	query->release(query);
 }

drivers/infiniband/core/smi.c

@@ -192,7 +192,7 @@ enum smi_action smi_handle_dr_smp_recv(struct ib_smp *smp, u8 node_type,
 			}
 			/* smp->hop_ptr updated when sending */
 			return (node_type == RDMA_NODE_IB_SWITCH ?
-				IB_SMI_HANDLE: IB_SMI_DISCARD);
+				IB_SMI_HANDLE : IB_SMI_DISCARD);
 		}
 
 		/* C14-13:4 -- hop_ptr = 0 -> give to SM */
@@ -211,7 +211,7 @@ enum smi_forward_action smi_check_forward_dr_smp(struct ib_smp *smp)
 	if (!ib_get_smp_direction(smp)) {
 		/* C14-9:2 -- intermediate hop */
 		if (hop_ptr && hop_ptr < hop_cnt)
-			return IB_SMI_SEND;
+			return IB_SMI_FORWARD;
 
 		/* C14-9:3 -- at the end of the DR segment of path */
 		if (hop_ptr == hop_cnt)
@@ -224,7 +224,7 @@ enum smi_forward_action smi_check_forward_dr_smp(struct ib_smp *smp)
 	} else {
 		/* C14-13:2  -- intermediate hop */
 		if (2 <= hop_ptr && hop_ptr <= hop_cnt)
-			return IB_SMI_SEND;
+			return IB_SMI_FORWARD;
 
 		/* C14-13:3 -- at the end of the DR segment of path */
 		if (hop_ptr == 1)
@@ -233,3 +233,13 @@ enum smi_forward_action smi_check_forward_dr_smp(struct ib_smp *smp)
 	}
 	return IB_SMI_LOCAL;
 }
+
+/*
+ * Return the forwarding port number from initial_path for outgoing SMP and
+ * from return_path for returning SMP
+ */
+int smi_get_fwd_port(struct ib_smp *smp)
+{
+	return (!ib_get_smp_direction(smp) ? smp->initial_path[smp->hop_ptr+1] :
+		smp->return_path[smp->hop_ptr-1]);
+}

drivers/infiniband/core/smi.h

@@ -48,10 +48,12 @@ enum smi_action {
 enum smi_forward_action {
 	IB_SMI_LOCAL,	/* SMP should be completed up the stack */
 	IB_SMI_SEND,	/* received DR SMP should be forwarded to the send queue */
+	IB_SMI_FORWARD	/* SMP should be forwarded (for switches only) */
 };
 
 enum smi_action smi_handle_dr_smp_recv(struct ib_smp *smp, u8 node_type,
 				       int port_num, int phys_port_cnt);
+int smi_get_fwd_port(struct ib_smp *smp);
 extern enum smi_forward_action smi_check_forward_dr_smp(struct ib_smp *smp);
 extern enum smi_action smi_handle_dr_smp_send(struct ib_smp *smp,
 					      u8 node_type, int port_num);

drivers/infiniband/core/sysfs.c

@@ -311,7 +311,7 @@ static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
 		return sprintf(buf, "N/A (no PMA)\n");
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
-	out_mad = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 	if (!in_mad || !out_mad) {
 		ret = -ENOMEM;
 		goto out;

drivers/infiniband/core/ucm.c

@@ -823,7 +823,6 @@ static ssize_t ib_ucm_send_rep(struct ib_ucm_file *file,
 	param.private_data_len    = cmd.len;
 	param.responder_resources = cmd.responder_resources;
 	param.initiator_depth     = cmd.initiator_depth;
-	param.target_ack_delay    = cmd.target_ack_delay;
 	param.failover_accepted   = cmd.failover_accepted;
 	param.flow_control        = cmd.flow_control;
 	param.rnr_retry_count     = cmd.rnr_retry_count;

drivers/infiniband/core/umem.c

@@ -121,6 +121,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 
 	cur_base = addr & PAGE_MASK;
 
+	ret = 0;
 	while (npages) {
 		ret = get_user_pages(current, current->mm, cur_base,
 				     min_t(int, npages,

drivers/infiniband/hw/amso1100/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_AMSO1100
 	tristate "Ammasso 1100 HCA support"
-	depends on PCI && INET && INFINIBAND
+	depends on PCI && INET
 	---help---
 	  This is a low-level driver for the Ammasso 1100 host
 	  channel adapter (HCA).

drivers/infiniband/hw/cxgb3/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_CXGB3
 	tristate "Chelsio RDMA Driver"
-	depends on CHELSIO_T3 && INFINIBAND && INET
+	depends on CHELSIO_T3 && INET
 	select GENERIC_ALLOCATOR
 	---help---
 	  This is an iWARP/RDMA driver for the Chelsio T3 1GbE and

drivers/infiniband/hw/cxgb3/cxio_hal.c

@@ -144,7 +144,7 @@ static int cxio_hal_clear_qp_ctx(struct cxio_rdev *rdev_p, u32 qpid)
 	}
 	wqe = (struct t3_modify_qp_wr *) skb_put(skb, sizeof(*wqe));
 	memset(wqe, 0, sizeof(*wqe));
-	build_fw_riwrh((struct fw_riwrh *) wqe, T3_WR_QP_MOD, 3, 1, qpid, 7);
+	build_fw_riwrh((struct fw_riwrh *) wqe, T3_WR_QP_MOD, 3, 0, qpid, 7);
 	wqe->flags = cpu_to_be32(MODQP_WRITE_EC);
 	sge_cmd = qpid << 8 | 3;
 	wqe->sge_cmd = cpu_to_be64(sge_cmd);
@@ -548,7 +548,7 @@ static int cxio_hal_init_ctrl_qp(struct cxio_rdev *rdev_p)
 			V_EC_UP_TOKEN(T3_CTL_QP_TID) | F_EC_VALID)) << 32;
 	wqe = (struct t3_modify_qp_wr *) skb_put(skb, sizeof(*wqe));
 	memset(wqe, 0, sizeof(*wqe));
-	build_fw_riwrh((struct fw_riwrh *) wqe, T3_WR_QP_MOD, 0, 1,
+	build_fw_riwrh((struct fw_riwrh *) wqe, T3_WR_QP_MOD, 0, 0,
 		       T3_CTL_QP_TID, 7);
 	wqe->flags = cpu_to_be32(MODQP_WRITE_EC);
 	sge_cmd = (3ULL << 56) | FW_RI_SGEEC_START << 8 | 3;
@@ -833,7 +833,7 @@ int cxio_rdma_init(struct cxio_rdev *rdev_p, struct t3_rdma_init_attr *attr)
 	wqe->ird = cpu_to_be32(attr->ird);
 	wqe->qp_dma_addr = cpu_to_be64(attr->qp_dma_addr);
 	wqe->qp_dma_size = cpu_to_be32(attr->qp_dma_size);
-	wqe->rsvd = 0;
+	wqe->irs = cpu_to_be32(attr->irs);
 	skb->priority = 0;	/* 0=>ToeQ; 1=>CtrlQ */
 	return (cxgb3_ofld_send(rdev_p->t3cdev_p, skb));
 }

drivers/infiniband/hw/cxgb3/cxio_wr.h

@@ -294,6 +294,7 @@ struct t3_rdma_init_attr {
 	u64 qp_dma_addr;
 	u32 qp_dma_size;
 	u32 flags;
+	u32 irs;
 };
 
 struct t3_rdma_init_wr {
@@ -314,7 +315,7 @@ struct t3_rdma_init_wr {
 	__be32 ird;
 	__be64 qp_dma_addr;	/* 7 */
 	__be32 qp_dma_size;	/* 8 */
-	u32 rsvd;
+	u32 irs;
 };
 
 struct t3_genbit {

drivers/infiniband/hw/cxgb3/iwch_cm.c

@@ -254,8 +254,6 @@ static void release_ep_resources(struct iwch_ep *ep)
 	cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
 	dst_release(ep->dst);
 	l2t_release(L2DATA(ep->com.tdev), ep->l2t);
-	if (ep->com.tdev->type == T3B)
-		release_tid(ep->com.tdev, ep->hwtid, NULL);
 	put_ep(&ep->com);
 }
 
@@ -515,7 +513,7 @@ static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
 	req->len = htonl(len);
 	req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 			   V_TX_SNDBUF(snd_win>>15));
-	req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
+	req->flags = htonl(F_TX_INIT);
 	req->sndseq = htonl(ep->snd_seq);
 	BUG_ON(ep->mpa_skb);
 	ep->mpa_skb = skb;
@@ -566,7 +564,7 @@ static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
 	req->len = htonl(mpalen);
 	req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 			   V_TX_SNDBUF(snd_win>>15));
-	req->flags = htonl(F_TX_IMM_ACK|F_TX_INIT);
+	req->flags = htonl(F_TX_INIT);
 	req->sndseq = htonl(ep->snd_seq);
 	BUG_ON(ep->mpa_skb);
 	ep->mpa_skb = skb;
@@ -618,7 +616,7 @@ static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
 	req->len = htonl(len);
 	req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 			   V_TX_SNDBUF(snd_win>>15));
-	req->flags = htonl(F_TX_MORE | F_TX_IMM_ACK | F_TX_INIT);
+	req->flags = htonl(F_TX_INIT);
 	req->sndseq = htonl(ep->snd_seq);
 	ep->mpa_skb = skb;
 	state_set(&ep->com, MPA_REP_SENT);
@@ -641,6 +639,7 @@ static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
 
 	ep->snd_seq = ntohl(req->snd_isn);
+	ep->rcv_seq = ntohl(req->rcv_isn);
 
 	set_emss(ep, ntohs(req->tcp_opt));
 
@@ -1023,6 +1022,9 @@ static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	skb_pull(skb, sizeof(*hdr));
 	skb_trim(skb, dlen);
 
+	ep->rcv_seq += dlen;
+	BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
+
 	switch (state_read(&ep->com)) {
 	case MPA_REQ_SENT:
 		process_mpa_reply(ep, skb);
@@ -1060,7 +1062,6 @@ static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	struct iwch_ep *ep = ctx;
 	struct cpl_wr_ack *hdr = cplhdr(skb);
 	unsigned int credits = ntohs(hdr->credits);
-	enum iwch_qp_attr_mask  mask;
 
 	PDBG("%s ep %p credits %u\n", __FUNCTION__, ep, credits);
 
@@ -1072,30 +1073,6 @@ static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	ep->mpa_skb = NULL;
 	dst_confirm(ep->dst);
 	if (state_read(&ep->com) == MPA_REP_SENT) {
-		struct iwch_qp_attributes attrs;
-
-		/* bind QP to EP and move to RTS */
-		attrs.mpa_attr = ep->mpa_attr;
-		attrs.max_ird = ep->ord;
-		attrs.max_ord = ep->ord;
-		attrs.llp_stream_handle = ep;
-		attrs.next_state = IWCH_QP_STATE_RTS;
-
-		/* bind QP and TID with INIT_WR */
-		mask = IWCH_QP_ATTR_NEXT_STATE |
-				     IWCH_QP_ATTR_LLP_STREAM_HANDLE |
-				     IWCH_QP_ATTR_MPA_ATTR |
-				     IWCH_QP_ATTR_MAX_IRD |
-				     IWCH_QP_ATTR_MAX_ORD;
-
-		ep->com.rpl_err = iwch_modify_qp(ep->com.qp->rhp,
-				     ep->com.qp, mask, &attrs, 1);
-
-		if (!ep->com.rpl_err) {
-			state_set(&ep->com, FPDU_MODE);
-			established_upcall(ep);
-		}
-
 		ep->com.rpl_done = 1;
 		PDBG("waking up ep %p\n", ep);
 		wake_up(&ep->com.waitq);
@@ -1124,6 +1101,15 @@ static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	return CPL_RET_BUF_DONE;
 }
 
+/*
+ * Return whether a failed active open has allocated a TID
+ */
+static inline int act_open_has_tid(int status)
+{
+	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
+	       status != CPL_ERR_ARP_MISS;
+}
+
 static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 {
 	struct iwch_ep *ep = ctx;
@@ -1133,7 +1119,7 @@ static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	     status2errno(rpl->status));
 	connect_reply_upcall(ep, status2errno(rpl->status));
 	state_set(&ep->com, DEAD);
-	if (ep->com.tdev->type == T3B)
+	if (ep->com.tdev->type == T3B && act_open_has_tid(rpl->status))
 		release_tid(ep->com.tdev, GET_TID(rpl), NULL);
 	cxgb3_free_atid(ep->com.tdev, ep->atid);
 	dst_release(ep->dst);
@@ -1378,6 +1364,7 @@ static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 
 	PDBG("%s ep %p\n", __FUNCTION__, ep);
 	ep->snd_seq = ntohl(req->snd_isn);
+	ep->rcv_seq = ntohl(req->rcv_isn);
 
 	set_emss(ep, ntohs(req->tcp_opt));
 
@@ -1485,6 +1472,13 @@ static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	int ret;
 	int state;
 
+	if (is_neg_adv_abort(req->status)) {
+		PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
+		     ep->hwtid);
+		t3_l2t_send_event(ep->com.tdev, ep->l2t);
+		return CPL_RET_BUF_DONE;
+	}
+
 	/*
 	 * We get 2 peer aborts from the HW.  The first one must
 	 * be ignored except for scribbling that we need one more.
@@ -1494,13 +1488,6 @@ static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 		return CPL_RET_BUF_DONE;
 	}
 
-	if (is_neg_adv_abort(req->status)) {
-		PDBG("%s neg_adv_abort ep %p tid %d\n", __FUNCTION__, ep,
-		     ep->hwtid);
-		t3_l2t_send_event(ep->com.tdev, ep->l2t);
-		return CPL_RET_BUF_DONE;
-	}
-
 	state = state_read(&ep->com);
 	PDBG("%s ep %p state %u\n", __FUNCTION__, ep, state);
 	switch (state) {
@@ -1732,10 +1719,8 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
 
 	PDBG("%s ep %p tid %u\n", __FUNCTION__, ep, ep->hwtid);
-	if (state_read(&ep->com) == DEAD) {
-		put_ep(&ep->com);
+	if (state_read(&ep->com) == DEAD)
 		return -ECONNRESET;
-	}
 
 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
 	BUG_ON(!qp);
@@ -1755,17 +1740,8 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	ep->ird = conn_param->ird;
 	ep->ord = conn_param->ord;
 	PDBG("%s %d ird %d ord %d\n", __FUNCTION__, __LINE__, ep->ird, ep->ord);
+
 	get_ep(&ep->com);
-	err = send_mpa_reply(ep, conn_param->private_data,
-			     conn_param->private_data_len);
-	if (err) {
-		ep->com.cm_id = NULL;
-		ep->com.qp = NULL;
-		cm_id->rem_ref(cm_id);
-		abort_connection(ep, NULL, GFP_KERNEL);
-		put_ep(&ep->com);
-		return err;
-	}
 
 	/* bind QP to EP and move to RTS */
 	attrs.mpa_attr = ep->mpa_attr;
@@ -1783,16 +1759,28 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 
 	err = iwch_modify_qp(ep->com.qp->rhp,
 			     ep->com.qp, mask, &attrs, 1);
+	if (err)
+		goto err;
 
-	if (err) {
-		ep->com.cm_id = NULL;
-		ep->com.qp = NULL;
-		cm_id->rem_ref(cm_id);
-		abort_connection(ep, NULL, GFP_KERNEL);
-	} else {
-		state_set(&ep->com, FPDU_MODE);
-		established_upcall(ep);
-	}
+	err = send_mpa_reply(ep, conn_param->private_data,
+			     conn_param->private_data_len);
+	if (err)
+		goto err;
+
+	/* wait for wr_ack */
+	wait_event(ep->com.waitq, ep->com.rpl_done);
+	err = ep->com.rpl_err;
+	if (err)
+		goto err;
+
+	state_set(&ep->com, FPDU_MODE);
+	established_upcall(ep);
+	put_ep(&ep->com);
+	return 0;
+err:
+	ep->com.cm_id = NULL;
+	ep->com.qp = NULL;
+	cm_id->rem_ref(cm_id);
 	put_ep(&ep->com);
 	return err;
 }

drivers/infiniband/hw/cxgb3/iwch_cm.h

@@ -175,6 +175,7 @@ struct iwch_ep {
 	unsigned int atid;
 	u32 hwtid;
 	u32 snd_seq;
+	u32 rcv_seq;
 	struct l2t_entry *l2t;
 	struct dst_entry *dst;
 	struct sk_buff *mpa_skb;

drivers/infiniband/hw/cxgb3/iwch_provider.c

@@ -1163,9 +1163,10 @@ int iwch_register_device(struct iwch_dev *dev)
 	dev->ibdev.post_recv = iwch_post_receive;
 
 
-	dev->ibdev.iwcm =
-	    (struct iw_cm_verbs *) kmalloc(sizeof(struct iw_cm_verbs),
-					   GFP_KERNEL);
+	dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
+	if (!dev->ibdev.iwcm)
+		return -ENOMEM;
+
 	dev->ibdev.iwcm->connect = iwch_connect;
 	dev->ibdev.iwcm->accept = iwch_accept_cr;
 	dev->ibdev.iwcm->reject = iwch_reject_cr;

drivers/infiniband/hw/cxgb3/iwch_qp.c

@@ -628,9 +628,9 @@ int iwch_post_terminate(struct iwch_qp *qhp, struct respQ_msg_t *rsp_msg)
 	/* immediate data starts here. */
 	term = (struct terminate_message *)wqe->send.sgl;
 	build_term_codes(rsp_msg, &term->layer_etype, &term->ecode);
-	build_fw_riwrh((void *)wqe, T3_WR_SEND,
-		       T3_COMPLETION_FLAG | T3_NOTIFY_FLAG, 1,
-		       qhp->ep->hwtid, 5);
+	wqe->send.wrh.op_seop_flags = cpu_to_be32(V_FW_RIWR_OP(T3_WR_SEND) |
+			 V_FW_RIWR_FLAGS(T3_COMPLETION_FLAG | T3_NOTIFY_FLAG));
+	wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(qhp->ep->hwtid));
 	skb->priority = CPL_PRIORITY_DATA;
 	return cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
 }
@@ -732,6 +732,7 @@ static int rdma_init(struct iwch_dev *rhp, struct iwch_qp *qhp,
 	init_attr.qp_dma_addr = qhp->wq.dma_addr;
 	init_attr.qp_dma_size = (1UL << qhp->wq.size_log2);
 	init_attr.flags = rqes_posted(qhp) ? RECVS_POSTED : 0;
+	init_attr.irs = qhp->ep->rcv_seq;
 	PDBG("%s init_attr.rq_addr 0x%x init_attr.rq_size = %d "
 	     "flags 0x%x qpcaps 0x%x\n", __FUNCTION__,
 	     init_attr.rq_addr, init_attr.rq_size,

drivers/infiniband/hw/ehca/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_EHCA
 	tristate "eHCA support"
-	depends on IBMEBUS && INFINIBAND
+	depends on IBMEBUS
 	---help---
 	This driver supports the IBM pSeries eHCA InfiniBand adapter.
 

drivers/infiniband/hw/ehca/ehca_av.c

@@ -118,7 +118,7 @@ struct ib_ah *ehca_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
 		}
 		memcpy(&av->av.grh.word_1, &gid, sizeof(gid));
 	}
-	av->av.pmtu = EHCA_MAX_MTU;
+	av->av.pmtu = shca->max_mtu;
 
 	/* dgid comes in grh.word_3 */
 	memcpy(&av->av.grh.word_3, &ah_attr->grh.dgid,
@@ -137,6 +137,8 @@ int ehca_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 	struct ehca_av *av;
 	struct ehca_ud_av new_ehca_av;
 	struct ehca_pd *my_pd = container_of(ah->pd, struct ehca_pd, ib_pd);
+	struct ehca_shca *shca = container_of(ah->pd->device, struct ehca_shca,
+					      ib_device);
 	u32 cur_pid = current->tgid;
 
 	if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context &&
@@ -192,7 +194,7 @@ int ehca_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 		memcpy(&new_ehca_av.grh.word_1, &gid, sizeof(gid));
 	}
 
-	new_ehca_av.pmtu = EHCA_MAX_MTU;
+	new_ehca_av.pmtu = shca->max_mtu;
 
 	memcpy(&new_ehca_av.grh.word_3, &ah_attr->grh.dgid,
 	       sizeof(ah_attr->grh.dgid));

drivers/infiniband/hw/ehca/ehca_classes.h

@@ -5,6 +5,7 @@
  *
  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
  *           Christoph Raisch <raisch@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
  *
@@ -86,11 +87,17 @@ struct ehca_eq {
 	struct ehca_eqe_cache_entry eqe_cache[EHCA_EQE_CACHE_SIZE];
 };
 
+struct ehca_sma_attr {
+	u16 lid, lmc, sm_sl, sm_lid;
+	u16 pkey_tbl_len, pkeys[16];
+};
+
 struct ehca_sport {
 	struct ib_cq *ibcq_aqp1;
 	struct ib_qp *ibqp_aqp1;
 	enum ib_rate  rate;
 	enum ib_port_state port_state;
+	struct ehca_sma_attr saved_attr;
 };
 
 struct ehca_shca {
@@ -107,6 +114,8 @@ struct ehca_shca {
 	struct ehca_pd *pd;
 	struct h_galpas galpas;
 	struct mutex modify_mutex;
+	u64 hca_cap;
+	int max_mtu;
 };
 
 struct ehca_pd {
@@ -115,9 +124,20 @@ struct ehca_pd {
 	u32 ownpid;
 };
 
+enum ehca_ext_qp_type {
+	EQPT_NORMAL    = 0,
+	EQPT_LLQP      = 1,
+	EQPT_SRQBASE   = 2,
+	EQPT_SRQ       = 3,
+};
+
 struct ehca_qp {
-	struct ib_qp ib_qp;
+	union {
+		struct ib_qp ib_qp;
+		struct ib_srq ib_srq;
+	};
 	u32 qp_type;
+	enum ehca_ext_qp_type ext_type;
 	struct ipz_queue ipz_squeue;
 	struct ipz_queue ipz_rqueue;
 	struct h_galpas galpas;
@@ -140,6 +160,10 @@ struct ehca_qp {
 	u32 mm_count_galpa;
 };
 
+#define IS_SRQ(qp) (qp->ext_type == EQPT_SRQ)
+#define HAS_SQ(qp) (qp->ext_type != EQPT_SRQ)
+#define HAS_RQ(qp) (qp->ext_type != EQPT_SRQBASE)
+
 /* must be power of 2 */
 #define QP_HASHTAB_LEN 8
 
@@ -156,8 +180,8 @@ struct ehca_cq {
 	spinlock_t cb_lock;
 	struct hlist_head qp_hashtab[QP_HASHTAB_LEN];
 	struct list_head entry;
-	u32 nr_callbacks; /* #events assigned to cpu by scaling code */
-	u32 nr_events;    /* #events seen */
+	u32 nr_callbacks;   /* #events assigned to cpu by scaling code */
+	atomic_t nr_events; /* #events seen */
 	wait_queue_head_t wait_completion;
 	spinlock_t task_lock;
 	u32 ownpid;
@@ -275,9 +299,8 @@ void ehca_cleanup_av_cache(void);
 int ehca_init_mrmw_cache(void);
 void ehca_cleanup_mrmw_cache(void);
 
-extern spinlock_t ehca_qp_idr_lock;
-extern spinlock_t ehca_cq_idr_lock;
-extern spinlock_t hcall_lock;
+extern rwlock_t ehca_qp_idr_lock;
+extern rwlock_t ehca_cq_idr_lock;
 extern struct idr ehca_qp_idr;
 extern struct idr ehca_cq_idr;
 
@@ -305,6 +328,7 @@ struct ehca_create_qp_resp {
 	u32 qp_num;
 	u32 token;
 	u32 qp_type;
+	u32 ext_type;
 	u32 qkey;
 	/* qp_num assigned by ehca: sqp0/1 may have got different numbers */
 	u32 real_qp_num;
@@ -320,14 +344,42 @@ struct ehca_alloc_cq_parms {
 	struct ipz_eq_handle eq_handle;
 };
 
+enum ehca_service_type {
+	ST_RC  = 0,
+	ST_UC  = 1,
+	ST_RD  = 2,
+	ST_UD  = 3,
+};
+
+enum ehca_ll_comp_flags {
+	LLQP_SEND_COMP = 0x20,
+	LLQP_RECV_COMP = 0x40,
+	LLQP_COMP_MASK = 0x60,
+};
+
 struct ehca_alloc_qp_parms {
-	int servicetype;
+/* input parameters */
+	enum ehca_service_type servicetype;
 	int sigtype;
-	int daqp_ctrl;
-	int max_send_sge;
-	int max_recv_sge;
+	enum ehca_ext_qp_type ext_type;
+	enum ehca_ll_comp_flags ll_comp_flags;
+
+	int max_send_wr, max_recv_wr;
+	int max_send_sge, max_recv_sge;
 	int ud_av_l_key_ctl;
 
+	u32 token;
+	struct ipz_eq_handle eq_handle;
+	struct ipz_pd pd;
+	struct ipz_cq_handle send_cq_handle, recv_cq_handle;
+
+	u32 srq_qpn, srq_token, srq_limit;
+
+/* output parameters */
+	u32 real_qp_num;
+	struct ipz_qp_handle qp_handle;
+	struct h_galpas galpas;
+
 	u16 act_nr_send_wqes;
 	u16 act_nr_recv_wqes;
 	u8  act_nr_recv_sges;
@@ -335,9 +387,6 @@ struct ehca_alloc_qp_parms {
 
 	u32 nr_rq_pages;
 	u32 nr_sq_pages;
-
-	struct ipz_eq_handle ipz_eq_handle;
-	struct ipz_pd pd;
 };
 
 int ehca_cq_assign_qp(struct ehca_cq *cq, struct ehca_qp *qp);

drivers/infiniband/hw/ehca/ehca_classes_pSeries.h

@@ -228,8 +228,8 @@ struct hcp_modify_qp_control_block {
 #define MQPCB_QP_NUMBER                         EHCA_BMASK_IBM(8,31)
 #define MQPCB_MASK_QP_ENABLE                    EHCA_BMASK_IBM(48,48)
 #define MQPCB_QP_ENABLE                         EHCA_BMASK_IBM(31,31)
-#define MQPCB_MASK_CURR_SQR_LIMIT               EHCA_BMASK_IBM(49,49)
-#define MQPCB_CURR_SQR_LIMIT                    EHCA_BMASK_IBM(15,31)
+#define MQPCB_MASK_CURR_SRQ_LIMIT               EHCA_BMASK_IBM(49,49)
+#define MQPCB_CURR_SRQ_LIMIT                    EHCA_BMASK_IBM(16,31)
 #define MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG       EHCA_BMASK_IBM(50,50)
 #define MQPCB_MASK_SHARED_RQ_HNDL               EHCA_BMASK_IBM(51,51)
 

drivers/infiniband/hw/ehca/ehca_cq.c

@@ -56,11 +56,11 @@ int ehca_cq_assign_qp(struct ehca_cq *cq, struct ehca_qp *qp)
 {
 	unsigned int qp_num = qp->real_qp_num;
 	unsigned int key = qp_num & (QP_HASHTAB_LEN-1);
-	unsigned long spl_flags;
+	unsigned long flags;
 
-	spin_lock_irqsave(&cq->spinlock, spl_flags);
+	spin_lock_irqsave(&cq->spinlock, flags);
 	hlist_add_head(&qp->list_entries, &cq->qp_hashtab[key]);
-	spin_unlock_irqrestore(&cq->spinlock, spl_flags);
+	spin_unlock_irqrestore(&cq->spinlock, flags);
 
 	ehca_dbg(cq->ib_cq.device, "cq_num=%x real_qp_num=%x",
 		 cq->cq_number, qp_num);
@@ -74,9 +74,9 @@ int ehca_cq_unassign_qp(struct ehca_cq *cq, unsigned int real_qp_num)
 	unsigned int key = real_qp_num & (QP_HASHTAB_LEN-1);
 	struct hlist_node *iter;
 	struct ehca_qp *qp;
-	unsigned long spl_flags;
+	unsigned long flags;
 
-	spin_lock_irqsave(&cq->spinlock, spl_flags);
+	spin_lock_irqsave(&cq->spinlock, flags);
 	hlist_for_each(iter, &cq->qp_hashtab[key]) {
 		qp = hlist_entry(iter, struct ehca_qp, list_entries);
 		if (qp->real_qp_num == real_qp_num) {
@@ -88,7 +88,7 @@ int ehca_cq_unassign_qp(struct ehca_cq *cq, unsigned int real_qp_num)
 			break;
 		}
 	}
-	spin_unlock_irqrestore(&cq->spinlock, spl_flags);
+	spin_unlock_irqrestore(&cq->spinlock, flags);
 	if (ret)
 		ehca_err(cq->ib_cq.device,
 			 "qp not found cq_num=%x real_qp_num=%x",
@@ -146,6 +146,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 	spin_lock_init(&my_cq->spinlock);
 	spin_lock_init(&my_cq->cb_lock);
 	spin_lock_init(&my_cq->task_lock);
+	atomic_set(&my_cq->nr_events, 0);
 	init_waitqueue_head(&my_cq->wait_completion);
 	my_cq->ownpid = current->tgid;
 
@@ -162,9 +163,9 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 			goto create_cq_exit1;
 		}
 
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+		write_lock_irqsave(&ehca_cq_idr_lock, flags);
 		ret = idr_get_new(&ehca_cq_idr, my_cq, &my_cq->token);
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+		write_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 
 	} while (ret == -EAGAIN);
 
@@ -293,9 +294,9 @@ create_cq_exit3:
 			 "cq_num=%x h_ret=%lx", my_cq, my_cq->cq_number, h_ret);
 
 create_cq_exit2:
-	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+	write_lock_irqsave(&ehca_cq_idr_lock, flags);
 	idr_remove(&ehca_cq_idr, my_cq->token);
-	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+	write_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 
 create_cq_exit1:
 	kmem_cache_free(cq_cache, my_cq);
@@ -303,16 +304,6 @@ create_cq_exit1:
 	return cq;
 }
 
-static int get_cq_nr_events(struct ehca_cq *my_cq)
-{
-	int ret;
-	unsigned long flags;
-	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-	ret = my_cq->nr_events;
-	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
-	return ret;
-}
-
 int ehca_destroy_cq(struct ib_cq *cq)
 {
 	u64 h_ret;
@@ -339,17 +330,18 @@ int ehca_destroy_cq(struct ib_cq *cq)
 		}
 	}
 
-	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-	while (my_cq->nr_events) {
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
-		wait_event(my_cq->wait_completion, !get_cq_nr_events(my_cq));
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-		/* recheck nr_events to assure no cqe has just arrived */
-	}
-
+	/*
+	 * remove the CQ from the idr first to make sure
+	 * no more interrupt tasklets will touch this CQ
+	 */
+	write_lock_irqsave(&ehca_cq_idr_lock, flags);
 	idr_remove(&ehca_cq_idr, my_cq->token);
-	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+	write_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+
+	/* now wait until all pending events have completed */
+	wait_event(my_cq->wait_completion, !atomic_read(&my_cq->nr_events));
 
+	/* nobody's using our CQ any longer -- we can destroy it */
 	h_ret = hipz_h_destroy_cq(adapter_handle, my_cq, 0);
 	if (h_ret == H_R_STATE) {
 		/* cq in err: read err data and destroy it forcibly */

drivers/infiniband/hw/ehca/ehca_hca.c

@@ -45,11 +45,25 @@
 
 int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props)
 {
-	int ret = 0;
+	int i, ret = 0;
 	struct ehca_shca *shca = container_of(ibdev, struct ehca_shca,
 					      ib_device);
 	struct hipz_query_hca *rblock;
 
+	static const u32 cap_mapping[] = {
+		IB_DEVICE_RESIZE_MAX_WR,      HCA_CAP_WQE_RESIZE,
+		IB_DEVICE_BAD_PKEY_CNTR,      HCA_CAP_BAD_P_KEY_CTR,
+		IB_DEVICE_BAD_QKEY_CNTR,      HCA_CAP_Q_KEY_VIOL_CTR,
+		IB_DEVICE_RAW_MULTI,          HCA_CAP_RAW_PACKET_MCAST,
+		IB_DEVICE_AUTO_PATH_MIG,      HCA_CAP_AUTO_PATH_MIG,
+		IB_DEVICE_CHANGE_PHY_PORT,    HCA_CAP_SQD_RTS_PORT_CHANGE,
+		IB_DEVICE_UD_AV_PORT_ENFORCE, HCA_CAP_AH_PORT_NR_CHECK,
+		IB_DEVICE_CURR_QP_STATE_MOD,  HCA_CAP_CUR_QP_STATE_MOD,
+		IB_DEVICE_SHUTDOWN_PORT,      HCA_CAP_SHUTDOWN_PORT,
+		IB_DEVICE_INIT_TYPE,          HCA_CAP_INIT_TYPE,
+		IB_DEVICE_PORT_ACTIVE_EVENT,  HCA_CAP_PORT_ACTIVE_EVENT,
+	};
+
 	rblock = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
@@ -96,6 +110,13 @@ int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props)
 	props->max_total_mcast_qp_attach
 		= min_t(int, rblock->max_total_mcast_qp_attach, INT_MAX);
 
+	/* translate device capabilities */
+	props->device_cap_flags = IB_DEVICE_SYS_IMAGE_GUID |
+		IB_DEVICE_RC_RNR_NAK_GEN | IB_DEVICE_N_NOTIFY_CQ;
+	for (i = 0; i < ARRAY_SIZE(cap_mapping); i += 2)
+		if (rblock->hca_cap_indicators & cap_mapping[i + 1])
+			props->device_cap_flags |= cap_mapping[i];
+
 query_device1:
 	ehca_free_fw_ctrlblock(rblock);
 
@@ -172,6 +193,40 @@ query_port1:
 	return ret;
 }
 
+int ehca_query_sma_attr(struct ehca_shca *shca,
+			u8 port, struct ehca_sma_attr *attr)
+{
+	int ret = 0;
+	struct hipz_query_port *rblock;
+
+	rblock = ehca_alloc_fw_ctrlblock(GFP_ATOMIC);
+	if (!rblock) {
+		ehca_err(&shca->ib_device, "Can't allocate rblock memory.");
+		return -ENOMEM;
+	}
+
+	if (hipz_h_query_port(shca->ipz_hca_handle, port, rblock) != H_SUCCESS) {
+		ehca_err(&shca->ib_device, "Can't query port properties");
+		ret = -EINVAL;
+		goto query_sma_attr1;
+	}
+
+	memset(attr, 0, sizeof(struct ehca_sma_attr));
+
+	attr->lid    = rblock->lid;
+	attr->lmc    = rblock->lmc;
+	attr->sm_sl  = rblock->sm_sl;
+	attr->sm_lid = rblock->sm_lid;
+
+	attr->pkey_tbl_len = rblock->pkey_tbl_len;
+	memcpy(attr->pkeys, rblock->pkey_entries, sizeof(attr->pkeys));
+
+query_sma_attr1:
+	ehca_free_fw_ctrlblock(rblock);
+
+	return ret;
+}
+
 int ehca_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 {
 	int ret = 0;
@@ -261,7 +316,7 @@ int ehca_modify_port(struct ib_device *ibdev,
 	}
 
 	if (mutex_lock_interruptible(&shca->modify_mutex))
-                return -ERESTARTSYS;
+		return -ERESTARTSYS;
 
 	rblock = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
 	if (!rblock) {
@@ -290,7 +345,7 @@ modify_port2:
 	ehca_free_fw_ctrlblock(rblock);
 
 modify_port1:
-        mutex_unlock(&shca->modify_mutex);
+	mutex_unlock(&shca->modify_mutex);
 
 	return ret;
 }

drivers/infiniband/hw/ehca/ehca_irq.c

@@ -5,6 +5,8 @@
  *
  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
  *           Khadija Souissi <souissi@de.ibm.com>
+ *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
  *
@@ -59,6 +61,7 @@
 #define NEQE_EVENT_CODE        EHCA_BMASK_IBM(2,7)
 #define NEQE_PORT_NUMBER       EHCA_BMASK_IBM(8,15)
 #define NEQE_PORT_AVAILABILITY EHCA_BMASK_IBM(16,16)
+#define NEQE_DISRUPTIVE        EHCA_BMASK_IBM(16,16)
 
 #define ERROR_DATA_LENGTH      EHCA_BMASK_IBM(52,63)
 #define ERROR_DATA_TYPE        EHCA_BMASK_IBM(0,7)
@@ -178,12 +181,11 @@ static void qp_event_callback(struct ehca_shca *shca,
 {
 	struct ib_event event;
 	struct ehca_qp *qp;
-	unsigned long flags;
 	u32 token = EHCA_BMASK_GET(EQE_QP_TOKEN, eqe);
 
-	spin_lock_irqsave(&ehca_qp_idr_lock, flags);
+	read_lock(&ehca_qp_idr_lock);
 	qp = idr_find(&ehca_qp_idr, token);
-	spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
+	read_unlock(&ehca_qp_idr_lock);
 
 
 	if (!qp)
@@ -207,18 +209,22 @@ static void cq_event_callback(struct ehca_shca *shca,
 			      u64 eqe)
 {
 	struct ehca_cq *cq;
-	unsigned long flags;
 	u32 token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe);
 
-	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+	read_lock(&ehca_cq_idr_lock);
 	cq = idr_find(&ehca_cq_idr, token);
-	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+	if (cq)
+		atomic_inc(&cq->nr_events);
+	read_unlock(&ehca_cq_idr_lock);
 
 	if (!cq)
 		return;
 
 	ehca_error_data(shca, cq, cq->ipz_cq_handle.handle);
 
+	if (atomic_dec_and_test(&cq->nr_events))
+		wake_up(&cq->wait_completion);
+
 	return;
 }
 
@@ -281,30 +287,61 @@ static void parse_identifier(struct ehca_shca *shca, u64 eqe)
 	return;
 }
 
-static void parse_ec(struct ehca_shca *shca, u64 eqe)
+static void dispatch_port_event(struct ehca_shca *shca, int port_num,
+				enum ib_event_type type, const char *msg)
 {
 	struct ib_event event;
+
+	ehca_info(&shca->ib_device, "port %d %s.", port_num, msg);
+	event.device = &shca->ib_device;
+	event.event = type;
+	event.element.port_num = port_num;
+	ib_dispatch_event(&event);
+}
+
+static void notify_port_conf_change(struct ehca_shca *shca, int port_num)
+{
+	struct ehca_sma_attr  new_attr;
+	struct ehca_sma_attr *old_attr = &shca->sport[port_num - 1].saved_attr;
+
+	ehca_query_sma_attr(shca, port_num, &new_attr);
+
+	if (new_attr.sm_sl  != old_attr->sm_sl ||
+	    new_attr.sm_lid != old_attr->sm_lid)
+		dispatch_port_event(shca, port_num, IB_EVENT_SM_CHANGE,
+				    "SM changed");
+
+	if (new_attr.lid != old_attr->lid ||
+	    new_attr.lmc != old_attr->lmc)
+		dispatch_port_event(shca, port_num, IB_EVENT_LID_CHANGE,
+				    "LID changed");
+
+	if (new_attr.pkey_tbl_len != old_attr->pkey_tbl_len ||
+	    memcmp(new_attr.pkeys, old_attr->pkeys,
+		   sizeof(u16) * new_attr.pkey_tbl_len))
+		dispatch_port_event(shca, port_num, IB_EVENT_PKEY_CHANGE,
+				    "P_Key changed");
+
+	*old_attr = new_attr;
+}
+
+static void parse_ec(struct ehca_shca *shca, u64 eqe)
+{
 	u8 ec   = EHCA_BMASK_GET(NEQE_EVENT_CODE, eqe);
 	u8 port = EHCA_BMASK_GET(NEQE_PORT_NUMBER, eqe);
 
 	switch (ec) {
 	case 0x30: /* port availability change */
 		if (EHCA_BMASK_GET(NEQE_PORT_AVAILABILITY, eqe)) {
-			ehca_info(&shca->ib_device,
-				  "port %x is active.", port);
-			event.device = &shca->ib_device;
-			event.event = IB_EVENT_PORT_ACTIVE;
-			event.element.port_num = port;
 			shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
-			ib_dispatch_event(&event);
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
+					    "is active");
+			ehca_query_sma_attr(shca, port,
+					    &shca->sport[port - 1].saved_attr);
 		} else {
-			ehca_info(&shca->ib_device,
-				  "port %x is inactive.", port);
-			event.device = &shca->ib_device;
-			event.event = IB_EVENT_PORT_ERR;
-			event.element.port_num = port;
 			shca->sport[port - 1].port_state = IB_PORT_DOWN;
-			ib_dispatch_event(&event);
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ERR,
+					    "is inactive");
 		}
 		break;
 	case 0x31:
@@ -312,24 +349,19 @@ static void parse_ec(struct ehca_shca *shca, u64 eqe)
 		 * disruptive change is caused by
 		 * LID, PKEY or SM change
 		 */
-		ehca_warn(&shca->ib_device,
-			  "disruptive port %x configuration change", port);
-
-		ehca_info(&shca->ib_device,
-			  "port %x is inactive.", port);
-		event.device = &shca->ib_device;
-		event.event = IB_EVENT_PORT_ERR;
-		event.element.port_num = port;
-		shca->sport[port - 1].port_state = IB_PORT_DOWN;
-		ib_dispatch_event(&event);
-
-		ehca_info(&shca->ib_device,
-			  "port %x is active.", port);
-		event.device = &shca->ib_device;
-		event.event = IB_EVENT_PORT_ACTIVE;
-		event.element.port_num = port;
-		shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
-		ib_dispatch_event(&event);
+		if (EHCA_BMASK_GET(NEQE_DISRUPTIVE, eqe)) {
+			ehca_warn(&shca->ib_device, "disruptive port "
+				  "%d configuration change", port);
+
+			shca->sport[port - 1].port_state = IB_PORT_DOWN;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ERR,
+					    "is inactive");
+
+			shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
+					    "is active");
+		} else
+			notify_port_conf_change(shca, port);
 		break;
 	case 0x32: /* adapter malfunction */
 		ehca_err(&shca->ib_device, "Adapter malfunction.");
@@ -404,7 +436,6 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 {
 	u64 eqe_value;
 	u32 token;
-	unsigned long flags;
 	struct ehca_cq *cq;
 
 	eqe_value = eqe->entry;
@@ -412,27 +443,24 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 	if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
 		ehca_dbg(&shca->ib_device, "Got completion event");
 		token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+		read_lock(&ehca_cq_idr_lock);
 		cq = idr_find(&ehca_cq_idr, token);
+		if (cq)
+			atomic_inc(&cq->nr_events);
+		read_unlock(&ehca_cq_idr_lock);
 		if (cq == NULL) {
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 			ehca_err(&shca->ib_device,
 				 "Invalid eqe for non-existing cq token=%x",
 				 token);
 			return;
 		}
 		reset_eq_pending(cq);
-		cq->nr_events++;
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 		if (ehca_scaling_code)
 			queue_comp_task(cq);
 		else {
 			comp_event_callback(cq);
-			spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-			cq->nr_events--;
-			if (!cq->nr_events)
+			if (atomic_dec_and_test(&cq->nr_events))
 				wake_up(&cq->wait_completion);
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 		}
 	} else {
 		ehca_dbg(&shca->ib_device, "Got non completion event");
@@ -476,17 +504,17 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 		eqe_value = eqe_cache[eqe_cnt].eqe->entry;
 		if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
 			token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
-			spin_lock(&ehca_cq_idr_lock);
+			read_lock(&ehca_cq_idr_lock);
 			eqe_cache[eqe_cnt].cq = idr_find(&ehca_cq_idr, token);
+			if (eqe_cache[eqe_cnt].cq)
+				atomic_inc(&eqe_cache[eqe_cnt].cq->nr_events);
+			read_unlock(&ehca_cq_idr_lock);
 			if (!eqe_cache[eqe_cnt].cq) {
-				spin_unlock(&ehca_cq_idr_lock);
 				ehca_err(&shca->ib_device,
 					 "Invalid eqe for non-existing cq "
 					 "token=%x", token);
 				continue;
 			}
-			eqe_cache[eqe_cnt].cq->nr_events++;
-			spin_unlock(&ehca_cq_idr_lock);
 		} else
 			eqe_cache[eqe_cnt].cq = NULL;
 		eqe_cnt++;
@@ -517,11 +545,8 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 			else {
 				struct ehca_cq *cq = eq->eqe_cache[i].cq;
 				comp_event_callback(cq);
-				spin_lock(&ehca_cq_idr_lock);
-				cq->nr_events--;
-				if (!cq->nr_events)
+				if (atomic_dec_and_test(&cq->nr_events))
 					wake_up(&cq->wait_completion);
-				spin_unlock(&ehca_cq_idr_lock);
 			}
 		} else {
 			ehca_dbg(&shca->ib_device, "Got non completion event");
@@ -621,13 +646,10 @@ static void run_comp_task(struct ehca_cpu_comp_task* cct)
 	while (!list_empty(&cct->cq_list)) {
 		cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
 		spin_unlock_irqrestore(&cct->task_lock, flags);
-		comp_event_callback(cq);
 
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-		cq->nr_events--;
-		if (!cq->nr_events)
+		comp_event_callback(cq);
+		if (atomic_dec_and_test(&cq->nr_events))
 			wake_up(&cq->wait_completion);
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 
 		spin_lock_irqsave(&cct->task_lock, flags);
 		spin_lock(&cq->task_lock);

drivers/infiniband/hw/ehca/ehca_irq.h

@@ -47,7 +47,6 @@ struct ehca_shca;
 
 #include <linux/interrupt.h>
 #include <linux/types.h>
-#include <asm/atomic.h>
 
 int ehca_error_data(struct ehca_shca *shca, void *data, u64 resource);
 

drivers/infiniband/hw/ehca/ehca_iverbs.h

@@ -49,6 +49,9 @@ int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props);
 int ehca_query_port(struct ib_device *ibdev, u8 port,
 		    struct ib_port_attr *props);
 
+int ehca_query_sma_attr(struct ehca_shca *shca, u8 port,
+			struct ehca_sma_attr *attr);
+
 int ehca_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 * pkey);
 
 int ehca_query_gid(struct ib_device *ibdev, u8 port, int index,
@@ -154,6 +157,21 @@ int ehca_post_send(struct ib_qp *qp, struct ib_send_wr *send_wr,
 int ehca_post_recv(struct ib_qp *qp, struct ib_recv_wr *recv_wr,
 		   struct ib_recv_wr **bad_recv_wr);
 
+int ehca_post_srq_recv(struct ib_srq *srq,
+		       struct ib_recv_wr *recv_wr,
+		       struct ib_recv_wr **bad_recv_wr);
+
+struct ib_srq *ehca_create_srq(struct ib_pd *pd,
+			       struct ib_srq_init_attr *init_attr,
+			       struct ib_udata *udata);
+
+int ehca_modify_srq(struct ib_srq *srq, struct ib_srq_attr *attr,
+		    enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
+
+int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
+
+int ehca_destroy_srq(struct ib_srq *srq);
+
 u64 ehca_define_sqp(struct ehca_shca *shca, struct ehca_qp *ibqp,
 		    struct ib_qp_init_attr *qp_init_attr);
 

drivers/infiniband/hw/ehca/ehca_main.c

@@ -94,17 +94,15 @@ MODULE_PARM_DESC(poll_all_eqs,
 MODULE_PARM_DESC(static_rate,
 		 "set permanent static rate (default: disabled)");
 MODULE_PARM_DESC(scaling_code,
-		 "set scaling code (0: disabled, 1: enabled/default)");
+		 "set scaling code (0: disabled/default, 1: enabled)");
 
-spinlock_t ehca_qp_idr_lock;
-spinlock_t ehca_cq_idr_lock;
-spinlock_t hcall_lock;
+DEFINE_RWLOCK(ehca_qp_idr_lock);
+DEFINE_RWLOCK(ehca_cq_idr_lock);
 DEFINE_IDR(ehca_qp_idr);
 DEFINE_IDR(ehca_cq_idr);
 
-
-static struct list_head shca_list; /* list of all registered ehcas */
-static spinlock_t shca_list_lock;
+static LIST_HEAD(shca_list); /* list of all registered ehcas */
+static DEFINE_SPINLOCK(shca_list_lock);
 
 static struct timer_list poll_eqs_timer;
 
@@ -205,11 +203,35 @@ static void ehca_destroy_slab_caches(void)
 #define EHCA_HCAAVER  EHCA_BMASK_IBM(32,39)
 #define EHCA_REVID    EHCA_BMASK_IBM(40,63)
 
+static struct cap_descr {
+	u64 mask;
+	char *descr;
+} hca_cap_descr[] = {
+	{ HCA_CAP_AH_PORT_NR_CHECK, "HCA_CAP_AH_PORT_NR_CHECK" },
+	{ HCA_CAP_ATOMIC, "HCA_CAP_ATOMIC" },
+	{ HCA_CAP_AUTO_PATH_MIG, "HCA_CAP_AUTO_PATH_MIG" },
+	{ HCA_CAP_BAD_P_KEY_CTR, "HCA_CAP_BAD_P_KEY_CTR" },
+	{ HCA_CAP_SQD_RTS_PORT_CHANGE, "HCA_CAP_SQD_RTS_PORT_CHANGE" },
+	{ HCA_CAP_CUR_QP_STATE_MOD, "HCA_CAP_CUR_QP_STATE_MOD" },
+	{ HCA_CAP_INIT_TYPE, "HCA_CAP_INIT_TYPE" },
+	{ HCA_CAP_PORT_ACTIVE_EVENT, "HCA_CAP_PORT_ACTIVE_EVENT" },
+	{ HCA_CAP_Q_KEY_VIOL_CTR, "HCA_CAP_Q_KEY_VIOL_CTR" },
+	{ HCA_CAP_WQE_RESIZE, "HCA_CAP_WQE_RESIZE" },
+	{ HCA_CAP_RAW_PACKET_MCAST, "HCA_CAP_RAW_PACKET_MCAST" },
+	{ HCA_CAP_SHUTDOWN_PORT, "HCA_CAP_SHUTDOWN_PORT" },
+	{ HCA_CAP_RC_LL_QP, "HCA_CAP_RC_LL_QP" },
+	{ HCA_CAP_SRQ, "HCA_CAP_SRQ" },
+	{ HCA_CAP_UD_LL_QP, "HCA_CAP_UD_LL_QP" },
+	{ HCA_CAP_RESIZE_MR, "HCA_CAP_RESIZE_MR" },
+	{ HCA_CAP_MINI_QP, "HCA_CAP_MINI_QP" },
+};
+
 int ehca_sense_attributes(struct ehca_shca *shca)
 {
-	int ret = 0;
+	int i, ret = 0;
 	u64 h_ret;
 	struct hipz_query_hca *rblock;
+	struct hipz_query_port *port;
 
 	rblock = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
 	if (!rblock) {
@@ -222,7 +244,7 @@ int ehca_sense_attributes(struct ehca_shca *shca)
 		ehca_gen_err("Cannot query device properties. h_ret=%lx",
 			     h_ret);
 		ret = -EPERM;
-		goto num_ports1;
+		goto sense_attributes1;
 	}
 
 	if (ehca_nr_ports == 1)
@@ -242,18 +264,44 @@ int ehca_sense_attributes(struct ehca_shca *shca)
 		ehca_gen_dbg(" ... hardware version=%x:%x", hcaaver, revid);
 
 		if ((hcaaver == 1) && (revid == 0))
-			shca->hw_level = 0;
+			shca->hw_level = 0x11;
 		else if ((hcaaver == 1) && (revid == 1))
-			shca->hw_level = 1;
+			shca->hw_level = 0x12;
 		else if ((hcaaver == 1) && (revid == 2))
-			shca->hw_level = 2;
+			shca->hw_level = 0x13;
+		else if ((hcaaver == 2) && (revid == 0))
+			shca->hw_level = 0x21;
+		else if ((hcaaver == 2) && (revid == 0x10))
+			shca->hw_level = 0x22;
+		else {
+			ehca_gen_warn("unknown hardware version"
+				      " - assuming default level");
+			shca->hw_level = 0x22;
+		}
 	}
 	ehca_gen_dbg(" ... hardware level=%x", shca->hw_level);
 
 	shca->sport[0].rate = IB_RATE_30_GBPS;
 	shca->sport[1].rate = IB_RATE_30_GBPS;
 
-num_ports1:
+	shca->hca_cap = rblock->hca_cap_indicators;
+	ehca_gen_dbg(" ... HCA capabilities:");
+	for (i = 0; i < ARRAY_SIZE(hca_cap_descr); i++)
+		if (EHCA_BMASK_GET(hca_cap_descr[i].mask, shca->hca_cap))
+			ehca_gen_dbg("   %s", hca_cap_descr[i].descr);
+
+	port = (struct hipz_query_port *) rblock;
+	h_ret = hipz_h_query_port(shca->ipz_hca_handle, 1, port);
+	if (h_ret != H_SUCCESS) {
+		ehca_gen_err("Cannot query port properties. h_ret=%lx",
+			     h_ret);
+		ret = -EPERM;
+		goto sense_attributes1;
+	}
+
+	shca->max_mtu = port->max_mtu;
+
+sense_attributes1:
 	ehca_free_fw_ctrlblock(rblock);
 	return ret;
 }
@@ -293,7 +341,7 @@ int ehca_init_device(struct ehca_shca *shca)
 	strlcpy(shca->ib_device.name, "ehca%d", IB_DEVICE_NAME_MAX);
 	shca->ib_device.owner               = THIS_MODULE;
 
-	shca->ib_device.uverbs_abi_ver	    = 6;
+	shca->ib_device.uverbs_abi_ver	    = 7;
 	shca->ib_device.uverbs_cmd_mask	    =
 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
@@ -361,6 +409,20 @@ int ehca_init_device(struct ehca_shca *shca)
 	/* shca->ib_device.process_mad	    = ehca_process_mad;	    */
 	shca->ib_device.mmap		    = ehca_mmap;
 
+	if (EHCA_BMASK_GET(HCA_CAP_SRQ, shca->hca_cap)) {
+		shca->ib_device.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
+
+		shca->ib_device.create_srq          = ehca_create_srq;
+		shca->ib_device.modify_srq          = ehca_modify_srq;
+		shca->ib_device.query_srq           = ehca_query_srq;
+		shca->ib_device.destroy_srq         = ehca_destroy_srq;
+		shca->ib_device.post_srq_recv       = ehca_post_srq_recv;
+	}
+
 	return ret;
 }
 
@@ -800,14 +862,6 @@ int __init ehca_module_init(void)
 
 	printk(KERN_INFO "eHCA Infiniband Device Driver "
 	       "(Rel.: SVNEHCA_0023)\n");
-	idr_init(&ehca_qp_idr);
-	idr_init(&ehca_cq_idr);
-	spin_lock_init(&ehca_qp_idr_lock);
-	spin_lock_init(&ehca_cq_idr_lock);
-	spin_lock_init(&hcall_lock);
-
-	INIT_LIST_HEAD(&shca_list);
-	spin_lock_init(&shca_list_lock);
 
 	if ((ret = ehca_create_comp_pool())) {
 		ehca_gen_err("Cannot create comp pool.");

drivers/infiniband/hw/ehca/ehca_reqs.c

@@ -3,8 +3,9 @@
  *
  *  post_send/recv, poll_cq, req_notify
  *
- *  Authors: Waleri Fomin <fomin@de.ibm.com>
- *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *  Authors: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Waleri Fomin <fomin@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *           Reinhard Ernst <rernst@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
@@ -362,10 +363,10 @@ int ehca_post_send(struct ib_qp *qp,
 	struct ehca_wqe *wqe_p;
 	int wqe_cnt = 0;
 	int ret = 0;
-	unsigned long spl_flags;
+	unsigned long flags;
 
 	/* LOCK the QUEUE */
-	spin_lock_irqsave(&my_qp->spinlock_s, spl_flags);
+	spin_lock_irqsave(&my_qp->spinlock_s, flags);
 
 	/* loop processes list of send reqs */
 	for (cur_send_wr = send_wr; cur_send_wr != NULL;
@@ -406,26 +407,31 @@ int ehca_post_send(struct ib_qp *qp,
 	} /* eof for cur_send_wr */
 
 post_send_exit0:
-	/* UNLOCK the QUEUE */
-	spin_unlock_irqrestore(&my_qp->spinlock_s, spl_flags);
 	iosync(); /* serialize GAL register access */
 	hipz_update_sqa(my_qp, wqe_cnt);
+	spin_unlock_irqrestore(&my_qp->spinlock_s, flags);
 	return ret;
 }
 
-int ehca_post_recv(struct ib_qp *qp,
-		   struct ib_recv_wr *recv_wr,
-		   struct ib_recv_wr **bad_recv_wr)
+static int internal_post_recv(struct ehca_qp *my_qp,
+			      struct ib_device *dev,
+			      struct ib_recv_wr *recv_wr,
+			      struct ib_recv_wr **bad_recv_wr)
 {
-	struct ehca_qp *my_qp = container_of(qp, struct ehca_qp, ib_qp);
 	struct ib_recv_wr *cur_recv_wr;
 	struct ehca_wqe *wqe_p;
 	int wqe_cnt = 0;
 	int ret = 0;
-	unsigned long spl_flags;
+	unsigned long flags;
+
+	if (unlikely(!HAS_RQ(my_qp))) {
+		ehca_err(dev, "QP has no RQ  ehca_qp=%p qp_num=%x ext_type=%d",
+			 my_qp, my_qp->real_qp_num, my_qp->ext_type);
+		return -ENODEV;
+	}
 
 	/* LOCK the QUEUE */
-	spin_lock_irqsave(&my_qp->spinlock_r, spl_flags);
+	spin_lock_irqsave(&my_qp->spinlock_r, flags);
 
 	/* loop processes list of send reqs */
 	for (cur_recv_wr = recv_wr; cur_recv_wr != NULL;
@@ -439,8 +445,8 @@ int ehca_post_recv(struct ib_qp *qp,
 				*bad_recv_wr = cur_recv_wr;
 			if (wqe_cnt == 0) {
 				ret = -ENOMEM;
-				ehca_err(qp->device, "Too many posted WQEs "
-					 "qp_num=%x", qp->qp_num);
+				ehca_err(dev, "Too many posted WQEs "
+					 "qp_num=%x", my_qp->real_qp_num);
 			}
 			goto post_recv_exit0;
 		}
@@ -455,23 +461,39 @@ int ehca_post_recv(struct ib_qp *qp,
 			*bad_recv_wr = cur_recv_wr;
 			if (wqe_cnt == 0) {
 				ret = -EINVAL;
-				ehca_err(qp->device, "Could not write WQE "
-					 "qp_num=%x", qp->qp_num);
+				ehca_err(dev, "Could not write WQE "
+					 "qp_num=%x", my_qp->real_qp_num);
 			}
 			goto post_recv_exit0;
 		}
 		wqe_cnt++;
-		ehca_gen_dbg("ehca_qp=%p qp_num=%x wqe_cnt=%d",
-		     my_qp, qp->qp_num, wqe_cnt);
+		ehca_dbg(dev, "ehca_qp=%p qp_num=%x wqe_cnt=%d",
+			 my_qp, my_qp->real_qp_num, wqe_cnt);
 	} /* eof for cur_recv_wr */
 
 post_recv_exit0:
-	spin_unlock_irqrestore(&my_qp->spinlock_r, spl_flags);
 	iosync(); /* serialize GAL register access */
 	hipz_update_rqa(my_qp, wqe_cnt);
+	spin_unlock_irqrestore(&my_qp->spinlock_r, flags);
 	return ret;
 }
 
+int ehca_post_recv(struct ib_qp *qp,
+		   struct ib_recv_wr *recv_wr,
+		   struct ib_recv_wr **bad_recv_wr)
+{
+	return internal_post_recv(container_of(qp, struct ehca_qp, ib_qp),
+				  qp->device, recv_wr, bad_recv_wr);
+}
+
+int ehca_post_srq_recv(struct ib_srq *srq,
+		       struct ib_recv_wr *recv_wr,
+		       struct ib_recv_wr **bad_recv_wr)
+{
+	return internal_post_recv(container_of(srq, struct ehca_qp, ib_srq),
+				  srq->device, recv_wr, bad_recv_wr);
+}
+
 /*
  * ib_wc_opcode table converts ehca wc opcode to ib
  * Since we use zero to indicate invalid opcode, the actual ib opcode must
@@ -494,6 +516,7 @@ static inline int ehca_poll_cq_one(struct ib_cq *cq, struct ib_wc *wc)
 	int ret = 0;
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
 	struct ehca_cqe *cqe;
+	struct ehca_qp *my_qp;
 	int cqe_count = 0;
 
 poll_cq_one_read_cqe:
@@ -513,7 +536,7 @@ poll_cq_one_read_cqe:
 	if (unlikely(cqe->status & WC_STATUS_PURGE_BIT)) {
 		struct ehca_qp *qp=ehca_cq_get_qp(my_cq, cqe->local_qp_number);
 		int purgeflag;
-		unsigned long spl_flags;
+		unsigned long flags;
 		if (!qp) {
 			ehca_err(cq->device, "cq_num=%x qp_num=%x "
 				 "could not find qp -> ignore cqe",
@@ -523,9 +546,9 @@ poll_cq_one_read_cqe:
 			/* ignore this purged cqe */
 			goto poll_cq_one_read_cqe;
 		}
-		spin_lock_irqsave(&qp->spinlock_s, spl_flags);
+		spin_lock_irqsave(&qp->spinlock_s, flags);
 		purgeflag = qp->sqerr_purgeflag;
-		spin_unlock_irqrestore(&qp->spinlock_s, spl_flags);
+		spin_unlock_irqrestore(&qp->spinlock_s, flags);
 
 		if (purgeflag) {
 			ehca_dbg(cq->device, "Got CQE with purged bit qp_num=%x "
@@ -545,7 +568,7 @@ poll_cq_one_read_cqe:
 	}
 
 	/* tracing cqe */
-	if (ehca_debug_level) {
+	if (unlikely(ehca_debug_level)) {
 		ehca_dbg(cq->device,
 			 "Received COMPLETION ehca_cq=%p cq_num=%x -----",
 			 my_cq, my_cq->cq_number);
@@ -579,7 +602,11 @@ poll_cq_one_read_cqe:
 	} else
 		wc->status = IB_WC_SUCCESS;
 
-	wc->qp = NULL;
+	read_lock(&ehca_qp_idr_lock);
+	my_qp = idr_find(&ehca_qp_idr, cqe->qp_token);
+	wc->qp = &my_qp->ib_qp;
+	read_unlock(&ehca_qp_idr_lock);
+
 	wc->byte_len = cqe->nr_bytes_transferred;
 	wc->pkey_index = cqe->pkey_index;
 	wc->slid = cqe->rlid;
@@ -589,7 +616,7 @@ poll_cq_one_read_cqe:
 	wc->imm_data = cpu_to_be32(cqe->immediate_data);
 	wc->sl = cqe->service_level;
 
-	if (wc->status != IB_WC_SUCCESS)
+	if (unlikely(wc->status != IB_WC_SUCCESS))
 		ehca_dbg(cq->device,
 			 "ehca_cq=%p cq_num=%x WARNING unsuccessful cqe "
 			 "OPType=%x status=%x qp_num=%x src_qp=%x wr_id=%lx "
@@ -610,7 +637,7 @@ int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 	int nr;
 	struct ib_wc *current_wc = wc;
 	int ret = 0;
-	unsigned long spl_flags;
+	unsigned long flags;
 
 	if (num_entries < 1) {
 		ehca_err(cq->device, "Invalid num_entries=%d ehca_cq=%p "
@@ -619,14 +646,14 @@ int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 		goto poll_cq_exit0;
 	}
 
-	spin_lock_irqsave(&my_cq->spinlock, spl_flags);
+	spin_lock_irqsave(&my_cq->spinlock, flags);
 	for (nr = 0; nr < num_entries; nr++) {
 		ret = ehca_poll_cq_one(cq, current_wc);
 		if (ret)
 			break;
 		current_wc++;
 	} /* eof for nr */
-	spin_unlock_irqrestore(&my_cq->spinlock, spl_flags);
+	spin_unlock_irqrestore(&my_cq->spinlock, flags);
 	if (ret == -EAGAIN  || !ret)
 		ret = nr;
 
@@ -637,7 +664,6 @@ poll_cq_exit0:
 int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify_flags notify_flags)
 {
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
-	unsigned long spl_flags;
 	int ret = 0;
 
 	switch (notify_flags & IB_CQ_SOLICITED_MASK) {
@@ -652,6 +678,7 @@ int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify_flags notify_flags)
 	}
 
 	if (notify_flags & IB_CQ_REPORT_MISSED_EVENTS) {
+		unsigned long spl_flags;
 		spin_lock_irqsave(&my_cq->spinlock, spl_flags);
 		ret = ipz_qeit_is_valid(&my_cq->ipz_queue);
 		spin_unlock_irqrestore(&my_cq->spinlock, spl_flags);

drivers/infiniband/hw/ehca/ehca_tools.h

@@ -59,6 +59,7 @@
 #include <linux/cpu.h>
 #include <linux/device.h>
 
+#include <asm/atomic.h>
 #include <asm/abs_addr.h>
 #include <asm/ibmebus.h>
 #include <asm/io.h>

drivers/infiniband/hw/ehca/ehca_uverbs.c

@@ -253,16 +253,16 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 	u32 rsrc_type = (fileoffset >> 24) & 0xF; /* sq,rq,cmnd_window */
 	u32 cur_pid = current->tgid;
 	u32 ret;
-	unsigned long flags;
 	struct ehca_cq *cq;
 	struct ehca_qp *qp;
 	struct ehca_pd *pd;
+	struct ib_uobject *uobject;
 
 	switch (q_type) {
 	case  1: /* CQ */
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+		read_lock(&ehca_cq_idr_lock);
 		cq = idr_find(&ehca_cq_idr, idr_handle);
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+		read_unlock(&ehca_cq_idr_lock);
 
 		/* make sure this mmap really belongs to the authorized user */
 		if (!cq)
@@ -288,9 +288,9 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 		break;
 
 	case 2: /* QP */
-		spin_lock_irqsave(&ehca_qp_idr_lock, flags);
+		read_lock(&ehca_qp_idr_lock);
 		qp = idr_find(&ehca_qp_idr, idr_handle);
-		spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
+		read_unlock(&ehca_qp_idr_lock);
 
 		/* make sure this mmap really belongs to the authorized user */
 		if (!qp)
@@ -304,7 +304,8 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 			return -ENOMEM;
 		}
 
-		if (!qp->ib_qp.uobject || qp->ib_qp.uobject->context != context)
+		uobject = IS_SRQ(qp) ? qp->ib_srq.uobject : qp->ib_qp.uobject;
+		if (!uobject || uobject->context != context)
 			return -EINVAL;
 
 		ret = ehca_mmap_qp(vma, qp, rsrc_type);

drivers/infiniband/hw/ehca/hcp_if.c

@@ -5,6 +5,7 @@
  *
  *  Authors: Christoph Raisch <raisch@de.ibm.com>
  *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *           Gerd Bayer <gerd.bayer@de.ibm.com>
  *           Waleri Fomin <fomin@de.ibm.com>
  *
@@ -62,6 +63,12 @@
 #define H_ALL_RES_QP_MAX_SEND_SGE       EHCA_BMASK_IBM(32, 39)
 #define H_ALL_RES_QP_MAX_RECV_SGE       EHCA_BMASK_IBM(40, 47)
 
+#define H_ALL_RES_QP_UD_AV_LKEY         EHCA_BMASK_IBM(32, 63)
+#define H_ALL_RES_QP_SRQ_QP_TOKEN       EHCA_BMASK_IBM(0, 31)
+#define H_ALL_RES_QP_SRQ_QP_HANDLE      EHCA_BMASK_IBM(0, 64)
+#define H_ALL_RES_QP_SRQ_LIMIT          EHCA_BMASK_IBM(48, 63)
+#define H_ALL_RES_QP_SRQ_QPN            EHCA_BMASK_IBM(40, 63)
+
 #define H_ALL_RES_QP_ACT_OUTST_SEND_WR  EHCA_BMASK_IBM(16, 31)
 #define H_ALL_RES_QP_ACT_OUTST_RECV_WR  EHCA_BMASK_IBM(48, 63)
 #define H_ALL_RES_QP_ACT_SEND_SGE       EHCA_BMASK_IBM(8, 15)
@@ -74,10 +81,7 @@
 #define H_MP_SHUTDOWN                   EHCA_BMASK_IBM(48, 48)
 #define H_MP_RESET_QKEY_CTR             EHCA_BMASK_IBM(49, 49)
 
-/* direct access qp controls */
-#define DAQP_CTRL_ENABLE    0x01
-#define DAQP_CTRL_SEND_COMP 0x20
-#define DAQP_CTRL_RECV_COMP 0x40
+static DEFINE_SPINLOCK(hcall_lock);
 
 static u32 get_longbusy_msecs(int longbusy_rc)
 {
@@ -155,7 +159,7 @@ static long ehca_plpar_hcall9(unsigned long opcode,
 {
 	long ret;
 	int i, sleep_msecs, lock_is_set = 0;
-	unsigned long flags;
+	unsigned long flags = 0;
 
 	ehca_gen_dbg("opcode=%lx arg1=%lx arg2=%lx arg3=%lx arg4=%lx "
 		     "arg5=%lx arg6=%lx arg7=%lx arg8=%lx arg9=%lx",
@@ -284,53 +288,53 @@ u64 hipz_h_alloc_resource_cq(const struct ipz_adapter_handle adapter_handle,
 }
 
 u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
-			     struct ehca_qp *qp,
 			     struct ehca_alloc_qp_parms *parms)
 {
 	u64 ret;
-	u64 allocate_controls;
-	u64 max_r10_reg;
+	u64 allocate_controls, max_r10_reg, r11, r12;
 	u64 outs[PLPAR_HCALL9_BUFSIZE];
-	u16 max_nr_receive_wqes = qp->init_attr.cap.max_recv_wr + 1;
-	u16 max_nr_send_wqes = qp->init_attr.cap.max_send_wr + 1;
-	int daqp_ctrl = parms->daqp_ctrl;
 
 	allocate_controls =
-		EHCA_BMASK_SET(H_ALL_RES_QP_ENHANCED_OPS,
-			       (daqp_ctrl & DAQP_CTRL_ENABLE) ? 1 : 0)
+		EHCA_BMASK_SET(H_ALL_RES_QP_ENHANCED_OPS, parms->ext_type)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_PTE_PIN, 0)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_SERVICE_TYPE, parms->servicetype)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_SIGNALING_TYPE, parms->sigtype)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_LL_RQ_CQE_POSTING,
-				 (daqp_ctrl & DAQP_CTRL_RECV_COMP) ? 1 : 0)
+				 !!(parms->ll_comp_flags & LLQP_RECV_COMP))
 		| EHCA_BMASK_SET(H_ALL_RES_QP_LL_SQ_CQE_POSTING,
-				 (daqp_ctrl & DAQP_CTRL_SEND_COMP) ? 1 : 0)
+				 !!(parms->ll_comp_flags & LLQP_SEND_COMP))
 		| EHCA_BMASK_SET(H_ALL_RES_QP_UD_AV_LKEY_CTRL,
 				 parms->ud_av_l_key_ctl)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_RESOURCE_TYPE, 1);
 
 	max_r10_reg =
 		EHCA_BMASK_SET(H_ALL_RES_QP_MAX_OUTST_SEND_WR,
-			       max_nr_send_wqes)
+			       parms->max_send_wr + 1)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_MAX_OUTST_RECV_WR,
-				 max_nr_receive_wqes)
+				 parms->max_recv_wr + 1)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_MAX_SEND_SGE,
 				 parms->max_send_sge)
 		| EHCA_BMASK_SET(H_ALL_RES_QP_MAX_RECV_SGE,
 				 parms->max_recv_sge);
 
+	r11 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_QP_TOKEN, parms->srq_token);
+
+	if (parms->ext_type == EQPT_SRQ)
+		r12 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_LIMIT, parms->srq_limit);
+	else
+		r12 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_QPN, parms->srq_qpn);
+
 	ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
 				adapter_handle.handle,	           /* r4  */
 				allocate_controls,	           /* r5  */
-				qp->send_cq->ipz_cq_handle.handle,
-				qp->recv_cq->ipz_cq_handle.handle,
-				parms->ipz_eq_handle.handle,
-				((u64)qp->token << 32) | parms->pd.value,
-				max_r10_reg,	                   /* r10 */
-				parms->ud_av_l_key_ctl,            /* r11 */
-				0);
-	qp->ipz_qp_handle.handle = outs[0];
-	qp->real_qp_num = (u32)outs[1];
+				parms->send_cq_handle.handle,
+				parms->recv_cq_handle.handle,
+				parms->eq_handle.handle,
+				((u64)parms->token << 32) | parms->pd.value,
+				max_r10_reg, r11, r12);
+
+	parms->qp_handle.handle = outs[0];
+	parms->real_qp_num = (u32)outs[1];
 	parms->act_nr_send_wqes =
 		(u16)EHCA_BMASK_GET(H_ALL_RES_QP_ACT_OUTST_SEND_WR, outs[2]);
 	parms->act_nr_recv_wqes =
@@ -345,7 +349,7 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
 		(u32)EHCA_BMASK_GET(H_ALL_RES_QP_RQUEUE_SIZE_PAGES, outs[4]);
 
 	if (ret == H_SUCCESS)
-		hcp_galpas_ctor(&qp->galpas, outs[6], outs[6]);
+		hcp_galpas_ctor(&parms->galpas, outs[6], outs[6]);
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
 		ehca_gen_err("Not enough resources. ret=%lx", ret);

drivers/infiniband/hw/ehca/hcp_if.h

@@ -78,7 +78,6 @@ u64 hipz_h_alloc_resource_cq(const struct ipz_adapter_handle adapter_handle,
  * initialize resources, create empty QPPTs (2 rings).
  */
 u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
-			     struct ehca_qp *qp,
 			     struct ehca_alloc_qp_parms *parms);
 
 u64 hipz_h_query_port(const struct ipz_adapter_handle adapter_handle,

drivers/infiniband/hw/ehca/hipz_hw.h

@@ -163,6 +163,7 @@ struct hipz_qptemm {
 
 #define QPX_SQADDER EHCA_BMASK_IBM(48,63)
 #define QPX_RQADDER EHCA_BMASK_IBM(48,63)
+#define QPX_AAELOG_RESET_SRQ_LIMIT EHCA_BMASK_IBM(3,3)
 
 #define QPTEMM_OFFSET(x) offsetof(struct hipz_qptemm,x)
 
@@ -360,6 +361,24 @@ struct hipz_query_hca {
 	u32 max_neq;
 } __attribute__ ((packed));
 
+#define HCA_CAP_AH_PORT_NR_CHECK      EHCA_BMASK_IBM( 0,  0)
+#define HCA_CAP_ATOMIC                EHCA_BMASK_IBM( 1,  1)
+#define HCA_CAP_AUTO_PATH_MIG         EHCA_BMASK_IBM( 2,  2)
+#define HCA_CAP_BAD_P_KEY_CTR         EHCA_BMASK_IBM( 3,  3)
+#define HCA_CAP_SQD_RTS_PORT_CHANGE   EHCA_BMASK_IBM( 4,  4)
+#define HCA_CAP_CUR_QP_STATE_MOD      EHCA_BMASK_IBM( 5,  5)
+#define HCA_CAP_INIT_TYPE             EHCA_BMASK_IBM( 6,  6)
+#define HCA_CAP_PORT_ACTIVE_EVENT     EHCA_BMASK_IBM( 7,  7)
+#define HCA_CAP_Q_KEY_VIOL_CTR        EHCA_BMASK_IBM( 8,  8)
+#define HCA_CAP_WQE_RESIZE            EHCA_BMASK_IBM( 9,  9)
+#define HCA_CAP_RAW_PACKET_MCAST      EHCA_BMASK_IBM(10, 10)
+#define HCA_CAP_SHUTDOWN_PORT         EHCA_BMASK_IBM(11, 11)
+#define HCA_CAP_RC_LL_QP              EHCA_BMASK_IBM(12, 12)
+#define HCA_CAP_SRQ                   EHCA_BMASK_IBM(13, 13)
+#define HCA_CAP_UD_LL_QP              EHCA_BMASK_IBM(16, 16)
+#define HCA_CAP_RESIZE_MR             EHCA_BMASK_IBM(17, 17)
+#define HCA_CAP_MINI_QP               EHCA_BMASK_IBM(18, 18)
+
 /* query port response block */
 struct hipz_query_port {
 	u32 state;

drivers/infiniband/hw/ehca/ipz_pt_fn.h

@@ -105,7 +105,6 @@ void *ipz_qpageit_get_inc(struct ipz_queue *queue);
  * step in struct ipz_queue, will wrap in ringbuffer
  * returns address (kv) of Queue Entry BEFORE increment
  * warning don't use in parallel with ipz_qpageit_get_inc()
- * warning unpredictable results may occur if steps>act_nr_of_queue_entries
  */
 static inline void *ipz_qeit_get_inc(struct ipz_queue *queue)
 {
@@ -120,32 +119,25 @@ static inline void *ipz_qeit_get_inc(struct ipz_queue *queue)
 	return ret;
 }
 
+/*
+ * return a bool indicating whether current Queue Entry is valid
+ */
+static inline int ipz_qeit_is_valid(struct ipz_queue *queue)
+{
+	struct ehca_cqe *cqe = ipz_qeit_get(queue);
+	return ((cqe->cqe_flags >> 7) == (queue->toggle_state & 1));
+}
+
 /*
  * return current Queue Entry, increment Queue Entry iterator by one
  * step in struct ipz_queue, will wrap in ringbuffer
  * returns address (kv) of Queue Entry BEFORE increment
  * returns 0 and does not increment, if wrong valid state
  * warning don't use in parallel with ipz_qpageit_get_inc()
- * warning unpredictable results may occur if steps>act_nr_of_queue_entries
  */
 static inline void *ipz_qeit_get_inc_valid(struct ipz_queue *queue)
 {
-	struct ehca_cqe *cqe = ipz_qeit_get(queue);
-	u32 cqe_flags = cqe->cqe_flags;
-
-	if ((cqe_flags >> 7) != (queue->toggle_state & 1))
-		return NULL;
-
-	ipz_qeit_get_inc(queue);
-	return cqe;
-}
-
-static inline int ipz_qeit_is_valid(struct ipz_queue *queue)
-{
-	struct ehca_cqe *cqe = ipz_qeit_get(queue);
-	u32 cqe_flags = cqe->cqe_flags;
-
-	return cqe_flags >> 7 == (queue->toggle_state & 1);
+	return ipz_qeit_is_valid(queue) ? ipz_qeit_get_inc(queue) : NULL;
 }
 
 /*

drivers/infiniband/hw/ipath/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_IPATH
 	tristate "QLogic InfiniPath Driver"
-	depends on (PCI_MSI || HT_IRQ) && 64BIT && INFINIBAND && NET
+	depends on (PCI_MSI || HT_IRQ) && 64BIT && NET
 	---help---
 	This is a driver for QLogic InfiniPath host channel adapters,
 	including InfiniBand verbs support.  This driver allows these

drivers/infiniband/hw/ipath/ipath_common.h

@@ -189,8 +189,7 @@ typedef enum _ipath_ureg {
 #define IPATH_RUNTIME_FORCE_WC_ORDER	0x4
 #define IPATH_RUNTIME_RCVHDR_COPY	0x8
 #define IPATH_RUNTIME_MASTER	0x10
-#define IPATH_RUNTIME_PBC_REWRITE 0x20
-#define IPATH_RUNTIME_LOOSE_DMA_ALIGN 0x40
+/* 0x20 and 0x40 are no longer used, but are reserved for ABI compatibility */
 
 /*
  * This structure is returned by ipath_userinit() immediately after
@@ -432,8 +431,15 @@ struct ipath_user_info {
 #define IPATH_CMD_UNUSED_1	25
 #define IPATH_CMD_UNUSED_2	26
 #define IPATH_CMD_PIOAVAILUPD	27	/* force an update of PIOAvail reg */
+#define IPATH_CMD_POLL_TYPE	28	/* set the kind of polling we want */
 
-#define IPATH_CMD_MAX		27
+#define IPATH_CMD_MAX		28
+
+/*
+ * Poll types
+ */
+#define IPATH_POLL_TYPE_URGENT	 0x01
+#define IPATH_POLL_TYPE_OVERFLOW 0x02
 
 struct ipath_port_info {
 	__u32 num_active;	/* number of active units */
@@ -474,6 +480,8 @@ struct ipath_cmd {
 		__u16 part_key;
 		/* user address of __u32 bitmask of active slaves */
 		__u64 slave_mask_addr;
+		/* type of polling we want */
+		__u16 poll_type;
 	} cmd;
 };
 
@@ -502,13 +510,30 @@ struct __ipath_sendpkt {
 	struct ipath_iovec sps_iov[4];
 };
 
-/* Passed into diag data special file's ->write method. */
+/*
+ * diagnostics can send a packet by "writing" one of the following
+ * two structs to diag data special file
+ * The first is the legacy version for backward compatibility
+ */
 struct ipath_diag_pkt {
 	__u32 unit;
 	__u64 data;
 	__u32 len;
 };
 
+/* The second diag_pkt struct is the expanded version that allows
+ * more control over the packet, specifically, by allowing a custom
+ * pbc (+ extra) qword, so that special modes and deliberate
+ * changes to CRCs can be used. The elements were also re-ordered
+ * for better alignment and to avoid padding issues.
+ */
+struct ipath_diag_xpkt {
+	__u64 data;
+	__u64 pbc_wd;
+	__u32 unit;
+	__u32 len;
+};
+
 /*
  * Data layout in I2C flash (for GUID, etc.)
  * All fields are little-endian binary unless otherwise stated

drivers/infiniband/hw/ipath/ipath_cq.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -90,6 +90,8 @@ void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited)
 	wc->queue[head].sl = entry->sl;
 	wc->queue[head].dlid_path_bits = entry->dlid_path_bits;
 	wc->queue[head].port_num = entry->port_num;
+	/* Make sure queue entry is written before the head index. */
+	smp_wmb();
 	wc->head = next;
 
 	if (cq->notify == IB_CQ_NEXT_COMP ||
@@ -139,7 +141,8 @@ int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 
 		if (tail == wc->head)
 			break;
-
+		/* Make sure entry is read after head index is read. */
+		smp_rmb();
 		qp = ipath_lookup_qpn(&to_idev(cq->ibcq.device)->qp_table,
 				      wc->queue[tail].qp_num);
 		entry->qp = &qp->ibqp;

drivers/infiniband/hw/ipath/ipath_debug.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_diag.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -323,13 +323,14 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 {
 	u32 __iomem *piobuf;
 	u32 plen, clen, pbufn;
-	struct ipath_diag_pkt dp;
+	struct ipath_diag_pkt odp;
+	struct ipath_diag_xpkt dp;
 	u32 *tmpbuf = NULL;
 	struct ipath_devdata *dd;
 	ssize_t ret = 0;
 	u64 val;
 
-	if (count < sizeof(dp)) {
+	if (count != sizeof(dp)) {
 		ret = -EINVAL;
 		goto bail;
 	}
@@ -339,6 +340,29 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 		goto bail;
 	}
 
+	/*
+	 * Due to padding/alignment issues (lessened with new struct)
+	 * the old and new structs are the same length. We need to
+	 * disambiguate them, which we can do because odp.len has never
+	 * been less than the total of LRH+BTH+DETH so far, while
+	 * dp.unit (same offset) unit is unlikely to get that high.
+	 * Similarly, dp.data, the pointer to user at the same offset
+	 * as odp.unit, is almost certainly at least one (512byte)page
+	 * "above" NULL. The if-block below can be omitted if compatibility
+	 * between a new driver and older diagnostic code is unimportant.
+	 * compatibility the other direction (new diags, old driver) is
+	 * handled in the diagnostic code, with a warning.
+	 */
+	if (dp.unit >= 20 && dp.data < 512) {
+		/* very probable version mismatch. Fix it up */
+		memcpy(&odp, &dp, sizeof(odp));
+		/* We got a legacy dp, copy elements to dp */
+		dp.unit = odp.unit;
+		dp.data = odp.data;
+		dp.len = odp.len;
+		dp.pbc_wd = 0; /* Indicate we need to compute PBC wd */
+	}
+
 	/* send count must be an exact number of dwords */
 	if (dp.len & 3) {
 		ret = -EINVAL;
@@ -371,9 +395,10 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 		ret = -ENODEV;
 		goto bail;
 	}
+	/* Check link state, but not if we have custom PBC */
 	val = dd->ipath_lastibcstat & IPATH_IBSTATE_MASK;
-	if (val != IPATH_IBSTATE_INIT && val != IPATH_IBSTATE_ARM &&
-	    val != IPATH_IBSTATE_ACTIVE) {
+	if (!dp.pbc_wd && val != IPATH_IBSTATE_INIT &&
+		val != IPATH_IBSTATE_ARM && val != IPATH_IBSTATE_ACTIVE) {
 		ipath_cdbg(VERBOSE, "unit %u not ready (state %llx)\n",
 			   dd->ipath_unit, (unsigned long long) val);
 		ret = -EINVAL;
@@ -419,9 +444,13 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 		ipath_cdbg(VERBOSE, "unit %u 0x%x+1w pio%d\n",
 			   dd->ipath_unit, plen - 1, pbufn);
 
+	if (dp.pbc_wd == 0)
+		/* Legacy operation, use computed pbc_wd */
+		dp.pbc_wd = plen;
+
 	/* we have to flush after the PBC for correctness on some cpus
 	 * or WC buffer can be written out of order */
-	writeq(plen, piobuf);
+	writeq(dp.pbc_wd, piobuf);
 	ipath_flush_wc();
 	/* copy all by the trigger word, then flush, so it's written
 	 * to chip before trigger word, then write trigger word, then

drivers/infiniband/hw/ipath/ipath_driver.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -104,6 +104,9 @@ static int __devinit ipath_init_one(struct pci_dev *,
 #define PCI_DEVICE_ID_INFINIPATH_HT 0xd
 #define PCI_DEVICE_ID_INFINIPATH_PE800 0x10
 
+/* Number of seconds before our card status check...  */
+#define STATUS_TIMEOUT 60
+
 static const struct pci_device_id ipath_pci_tbl[] = {
 	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_HT) },
 	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_PE800) },
@@ -119,6 +122,18 @@ static struct pci_driver ipath_driver = {
 	.id_table = ipath_pci_tbl,
 };
 
+static void ipath_check_status(struct work_struct *work)
+{
+	struct ipath_devdata *dd = container_of(work, struct ipath_devdata,
+						status_work.work);
+
+	/*
+	 * If we don't have any interrupts, let the user know and
+	 * don't bother checking again.
+	 */
+	if (dd->ipath_int_counter == 0)
+		dev_err(&dd->pcidev->dev, "No interrupts detected.\n");
+}
 
 static inline void read_bars(struct ipath_devdata *dd, struct pci_dev *dev,
 			     u32 *bar0, u32 *bar1)
@@ -187,6 +202,8 @@ static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
 	dd->pcidev = pdev;
 	pci_set_drvdata(pdev, dd);
 
+	INIT_DELAYED_WORK(&dd->status_work, ipath_check_status);
+
 	list_add(&dd->ipath_list, &ipath_dev_list);
 
 bail_unlock:
@@ -504,6 +521,9 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 	ipath_diag_add(dd);
 	ipath_register_ib_device(dd);
 
+	/* Check that card status in STATUS_TIMEOUT seconds. */
+	schedule_delayed_work(&dd->status_work, HZ * STATUS_TIMEOUT);
+
 	goto bail;
 
 bail_irqsetup:
@@ -631,6 +651,9 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev)
 	 */
 	ipath_shutdown_device(dd);
 
+	cancel_delayed_work(&dd->status_work);
+	flush_scheduled_work();
+
 	if (dd->verbs_dev)
 		ipath_unregister_ib_device(dd->verbs_dev);
 
@@ -699,9 +722,9 @@ void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first,
 	u64 sendctrl, sendorig;
 
 	ipath_cdbg(PKT, "disarm %u PIObufs first=%u\n", cnt, first);
-	sendorig = dd->ipath_sendctrl | INFINIPATH_S_DISARM;
+	sendorig = dd->ipath_sendctrl;
 	for (i = first; i < last; i++) {
-		sendctrl = sendorig |
+		sendctrl = sendorig  | INFINIPATH_S_DISARM |
 			(i << INFINIPATH_S_DISARMPIOBUF_SHIFT);
 		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
 				 sendctrl);
@@ -712,12 +735,12 @@ void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first,
 	 * while we were looping; no critical bits that would require
 	 * locking.
 	 *
-	 * Write a 0, and then the original value, reading scratch in
+	 * disable PIOAVAILUPD, then re-enable, reading scratch in
 	 * between.  This seems to avoid a chip timing race that causes
 	 * pioavail updates to memory to stop.
 	 */
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
-			 0);
+			 sendorig & ~IPATH_S_PIOBUFAVAILUPD);
 	sendorig = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
 			 dd->ipath_sendctrl);
@@ -1014,14 +1037,10 @@ void ipath_kreceive(struct ipath_devdata *dd)
 		goto bail;
 	}
 
-	/* There is already a thread processing this queue. */
-	if (test_and_set_bit(0, &dd->ipath_rcv_pending))
-		goto bail;
-
 	l = dd->ipath_port0head;
 	hdrqtail = (u32) le64_to_cpu(*dd->ipath_hdrqtailptr);
 	if (l == hdrqtail)
-		goto done;
+		goto bail;
 
 reloop:
 	for (i = 0; l != hdrqtail; i++) {
@@ -1156,10 +1175,6 @@ reloop:
 	ipath_stats.sps_avgpkts_call =
 		ipath_stats.sps_port0pkts / ++totcalls;
 
-done:
-	clear_bit(0, &dd->ipath_rcv_pending);
-	smp_mb__after_clear_bit();
-
 bail:;
 }
 
@@ -1589,6 +1604,35 @@ int ipath_waitfor_mdio_cmdready(struct ipath_devdata *dd)
 	return ret;
 }
 
+
+/*
+ * Flush all sends that might be in the ready to send state, as well as any
+ * that are in the process of being sent.   Used whenever we need to be
+ * sure the send side is idle.  Cleans up all buffer state by canceling
+ * all pio buffers, and issuing an abort, which cleans up anything in the
+ * launch fifo.  The cancel is superfluous on some chip versions, but
+ * it's safer to always do it.
+ * PIOAvail bits are updated by the chip as if normal send had happened.
+ */
+void ipath_cancel_sends(struct ipath_devdata *dd)
+{
+	ipath_dbg("Cancelling all in-progress send buffers\n");
+	dd->ipath_lastcancel = jiffies+HZ/2; /* skip armlaunch errs a bit */
+	/*
+	 * the abort bit is auto-clearing.  We read scratch to be sure
+	 * that cancels and the abort have taken effect in the chip.
+	 */
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+		INFINIPATH_S_ABORT);
+	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
+	ipath_disarm_piobufs(dd, 0,
+		(unsigned)(dd->ipath_piobcnt2k + dd->ipath_piobcnt4k));
+
+	/* and again, be sure all have hit the chip */
+	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
+}
+
+
 static void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
 {
 	static const char *what[4] = {
@@ -1610,14 +1654,8 @@ static void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
 			   INFINIPATH_IBCS_LINKTRAININGSTATE_MASK]);
 	/* flush all queued sends when going to DOWN or INIT, to be sure that
 	 * they don't block MAD packets */
-	if (!linkcmd || linkcmd == INFINIPATH_IBCC_LINKCMD_INIT) {
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
-				 INFINIPATH_S_ABORT);
-		ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
-		                    (unsigned)(dd->ipath_piobcnt2k +
-				    dd->ipath_piobcnt4k) -
-				    dd->ipath_lastport_piobuf);
-	}
+	if (!linkcmd || linkcmd == INFINIPATH_IBCC_LINKCMD_INIT)
+		ipath_cancel_sends(dd);
 
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 			 dd->ipath_ibcctrl | which);
@@ -1839,6 +1877,87 @@ void ipath_write_kreg_port(const struct ipath_devdata *dd, ipath_kreg regno,
 	ipath_write_kreg(dd, where, value);
 }
 
+/*
+ * Following deal with the "obviously simple" task of overriding the state
+ * of the LEDS, which normally indicate link physical and logical status.
+ * The complications arise in dealing with different hardware mappings
+ * and the board-dependent routine being called from interrupts.
+ * and then there's the requirement to _flash_ them.
+ */
+#define LED_OVER_FREQ_SHIFT 8
+#define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT)
+/* Below is "non-zero" to force override, but both actual LEDs are off */
+#define LED_OVER_BOTH_OFF (8)
+
+void ipath_run_led_override(unsigned long opaque)
+{
+	struct ipath_devdata *dd = (struct ipath_devdata *)opaque;
+	int timeoff;
+	int pidx;
+	u64 lstate, ltstate, val;
+
+	if (!(dd->ipath_flags & IPATH_INITTED))
+		return;
+
+	pidx = dd->ipath_led_override_phase++ & 1;
+	dd->ipath_led_override = dd->ipath_led_override_vals[pidx];
+	timeoff = dd->ipath_led_override_timeoff;
+
+	/*
+	 * below potentially restores the LED values per current status,
+	 * should also possibly setup the traffic-blink register,
+	 * but leave that to per-chip functions.
+	 */
+	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
+	ltstate = (val >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
+		  INFINIPATH_IBCS_LINKTRAININGSTATE_MASK;
+	lstate = (val >> INFINIPATH_IBCS_LINKSTATE_SHIFT) &
+		 INFINIPATH_IBCS_LINKSTATE_MASK;
+
+	dd->ipath_f_setextled(dd, lstate, ltstate);
+	mod_timer(&dd->ipath_led_override_timer, jiffies + timeoff);
+}
+
+void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val)
+{
+	int timeoff, freq;
+
+	if (!(dd->ipath_flags & IPATH_INITTED))
+		return;
+
+	/* First check if we are blinking. If not, use 1HZ polling */
+	timeoff = HZ;
+	freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT;
+
+	if (freq) {
+		/* For blink, set each phase from one nybble of val */
+		dd->ipath_led_override_vals[0] = val & 0xF;
+		dd->ipath_led_override_vals[1] = (val >> 4) & 0xF;
+		timeoff = (HZ << 4)/freq;
+	} else {
+		/* Non-blink set both phases the same. */
+		dd->ipath_led_override_vals[0] = val & 0xF;
+		dd->ipath_led_override_vals[1] = val & 0xF;
+	}
+	dd->ipath_led_override_timeoff = timeoff;
+
+	/*
+	 * If the timer has not already been started, do so. Use a "quick"
+	 * timeout so the function will be called soon, to look at our request.
+	 */
+	if (atomic_inc_return(&dd->ipath_led_override_timer_active) == 1) {
+		/* Need to start timer */
+		init_timer(&dd->ipath_led_override_timer);
+		dd->ipath_led_override_timer.function =
+						 ipath_run_led_override;
+		dd->ipath_led_override_timer.data = (unsigned long) dd;
+		dd->ipath_led_override_timer.expires = jiffies + 1;
+		add_timer(&dd->ipath_led_override_timer);
+	} else {
+		atomic_dec(&dd->ipath_led_override_timer_active);
+	}
+}
+
 /**
  * ipath_shutdown_device - shut down a device
  * @dd: the infinipath device
@@ -1879,17 +1998,9 @@ void ipath_shutdown_device(struct ipath_devdata *dd)
 	 */
 	udelay(5);
 
-	/*
-	 * abort any armed or launched PIO buffers that didn't go. (self
-	 * clearing).  Will cause any packet currently being transmitted to
-	 * go out with an EBP, and may also cause a short packet error on
-	 * the receiver.
-	 */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
-			 INFINIPATH_S_ABORT);
-
 	ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
 			    INFINIPATH_IBCC_LINKINITCMD_SHIFT);
+	ipath_cancel_sends(dd);
 
 	/* disable IBC */
 	dd->ipath_control &= ~INFINIPATH_C_LINKENABLE;
@@ -1902,7 +2013,6 @@ void ipath_shutdown_device(struct ipath_devdata *dd)
 	 * Turn the LEDs off explictly for the same reason.
 	 */
 	dd->ipath_f_quiet_serdes(dd);
-	dd->ipath_f_setextled(dd, 0, 0);
 
 	if (dd->ipath_stats_timer_active) {
 		del_timer_sync(&dd->ipath_stats_timer);
@@ -1918,6 +2028,9 @@ void ipath_shutdown_device(struct ipath_devdata *dd)
 			 ~0ULL & ~INFINIPATH_HWE_MEMBISTFAILED);
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, -1LL);
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, -1LL);
+
+	ipath_cdbg(VERBOSE, "Flush time and errors to EEPROM\n");
+	ipath_update_eeprom_log(dd);
 }
 
 /**
@@ -2078,6 +2191,16 @@ int ipath_reset_device(int unit)
 		goto bail;
 	}
 
+	if (atomic_read(&dd->ipath_led_override_timer_active)) {
+		/* Need to stop LED timer, _then_ shut off LEDs */
+		del_timer_sync(&dd->ipath_led_override_timer);
+		atomic_set(&dd->ipath_led_override_timer_active, 0);
+	}
+
+	/* Shut off LEDs after we are sure timer is not running */
+	dd->ipath_led_override = LED_OVER_BOTH_OFF;
+	dd->ipath_f_setextled(dd, 0, 0);
+
 	dev_info(&dd->pcidev->dev, "Reset on unit %u requested\n", unit);
 
 	if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT)) {

drivers/infiniband/hw/ipath/ipath_eeprom.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -95,39 +95,37 @@ static int i2c_gpio_set(struct ipath_devdata *dd,
 			enum i2c_type line,
 			enum i2c_state new_line_state)
 {
-	u64 read_val, write_val, mask, *gpioval;
+	u64 out_mask, dir_mask, *gpioval;
+	unsigned long flags = 0;
 
 	gpioval = &dd->ipath_gpio_out;
-	read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl);
-	if (line == i2c_line_scl)
-		mask = dd->ipath_gpio_scl;
-	else
-		mask = dd->ipath_gpio_sda;
 
-	if (new_line_state == i2c_line_high)
+	if (line == i2c_line_scl) {
+		dir_mask = dd->ipath_gpio_scl;
+		out_mask = (1UL << dd->ipath_gpio_scl_num);
+	} else {
+		dir_mask = dd->ipath_gpio_sda;
+		out_mask = (1UL << dd->ipath_gpio_sda_num);
+	}
+
+	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
+	if (new_line_state == i2c_line_high) {
 		/* tri-state the output rather than force high */
-		write_val = read_val & ~mask;
-	else
+		dd->ipath_extctrl &= ~dir_mask;
+	} else {
 		/* config line to be an output */
-		write_val = read_val | mask;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, write_val);
+		dd->ipath_extctrl |= dir_mask;
+	}
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, dd->ipath_extctrl);
 
-	/* set high and verify */
+	/* set output as well (no real verify) */
 	if (new_line_state == i2c_line_high)
-		write_val = 0x1UL;
+		*gpioval |= out_mask;
 	else
-		write_val = 0x0UL;
+		*gpioval &= ~out_mask;
 
-	if (line == i2c_line_scl) {
-		write_val <<= dd->ipath_gpio_scl_num;
-		*gpioval = *gpioval & ~(1UL << dd->ipath_gpio_scl_num);
-		*gpioval |= write_val;
-	} else {
-		write_val <<= dd->ipath_gpio_sda_num;
-		*gpioval = *gpioval & ~(1UL << dd->ipath_gpio_sda_num);
-		*gpioval |= write_val;
-	}
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_out, *gpioval);
+	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
 
 	return 0;
 }
@@ -145,8 +143,9 @@ static int i2c_gpio_get(struct ipath_devdata *dd,
 			enum i2c_type line,
 			enum i2c_state *curr_statep)
 {
-	u64 read_val, write_val, mask;
+	u64 read_val, mask;
 	int ret;
+	unsigned long flags = 0;
 
 	/* check args */
 	if (curr_statep == NULL) {
@@ -154,15 +153,21 @@ static int i2c_gpio_get(struct ipath_devdata *dd,
 		goto bail;
 	}
 
-	read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl);
 	/* config line to be an input */
 	if (line == i2c_line_scl)
 		mask = dd->ipath_gpio_scl;
 	else
 		mask = dd->ipath_gpio_sda;
-	write_val = read_val & ~mask;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, write_val);
+
+	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
+	dd->ipath_extctrl &= ~mask;
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, dd->ipath_extctrl);
+	/*
+	 * Below is very unlikely to reflect true input state if Output
+	 * Enable actually changed.
+	 */
 	read_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
+	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
 
 	if (read_val & mask)
 		*curr_statep = i2c_line_high;
@@ -192,6 +197,7 @@ static void i2c_wait_for_writes(struct ipath_devdata *dd)
 
 static void scl_out(struct ipath_devdata *dd, u8 bit)
 {
+	udelay(1);
 	i2c_gpio_set(dd, i2c_line_scl, bit ? i2c_line_high : i2c_line_low);
 
 	i2c_wait_for_writes(dd);
@@ -314,12 +320,18 @@ static int eeprom_reset(struct ipath_devdata *dd)
 	int clock_cycles_left = 9;
 	u64 *gpioval = &dd->ipath_gpio_out;
 	int ret;
+	unsigned long flags;
 
-	eeprom_init = 1;
+	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
+	/* Make sure shadows are consistent */
+	dd->ipath_extctrl = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extctrl);
 	*gpioval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_gpio_out);
+	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
+
 	ipath_cdbg(VERBOSE, "Resetting i2c eeprom; initial gpioout reg "
 		   "is %llx\n", (unsigned long long) *gpioval);
 
+	eeprom_init = 1;
 	/*
 	 * This is to get the i2c into a known state, by first going low,
 	 * then tristate sda (and then tristate scl as first thing
@@ -355,8 +367,8 @@ bail:
  * @len: number of bytes to receive
  */
 
-int ipath_eeprom_read(struct ipath_devdata *dd, u8 eeprom_offset,
-		      void *buffer, int len)
+static int ipath_eeprom_internal_read(struct ipath_devdata *dd,
+					u8 eeprom_offset, void *buffer, int len)
 {
 	/* compiler complains unless initialized */
 	u8 single_byte = 0;
@@ -406,6 +418,7 @@ bail:
 	return ret;
 }
 
+
 /**
  * ipath_eeprom_write - writes data to the eeprom via I2C
  * @dd: the infinipath device
@@ -413,8 +426,8 @@ bail:
  * @buffer: data to write
  * @len: number of bytes to write
  */
-int ipath_eeprom_write(struct ipath_devdata *dd, u8 eeprom_offset,
-		       const void *buffer, int len)
+int ipath_eeprom_internal_write(struct ipath_devdata *dd, u8 eeprom_offset,
+				const void *buffer, int len)
 {
 	u8 single_byte;
 	int sub_len;
@@ -488,6 +501,38 @@ bail:
 	return ret;
 }
 
+/*
+ * The public entry-points ipath_eeprom_read() and ipath_eeprom_write()
+ * are now just wrappers around the internal functions.
+ */
+int ipath_eeprom_read(struct ipath_devdata *dd, u8 eeprom_offset,
+			void *buff, int len)
+{
+	int ret;
+
+	ret = down_interruptible(&dd->ipath_eep_sem);
+	if (!ret) {
+		ret = ipath_eeprom_internal_read(dd, eeprom_offset, buff, len);
+		up(&dd->ipath_eep_sem);
+	}
+
+	return ret;
+}
+
+int ipath_eeprom_write(struct ipath_devdata *dd, u8 eeprom_offset,
+			const void *buff, int len)
+{
+	int ret;
+
+	ret = down_interruptible(&dd->ipath_eep_sem);
+	if (!ret) {
+		ret = ipath_eeprom_internal_write(dd, eeprom_offset, buff, len);
+		up(&dd->ipath_eep_sem);
+	}
+
+	return ret;
+}
+
 static u8 flash_csum(struct ipath_flash *ifp, int adjust)
 {
 	u8 *ip = (u8 *) ifp;
@@ -515,7 +560,7 @@ void ipath_get_eeprom_info(struct ipath_devdata *dd)
 	void *buf;
 	struct ipath_flash *ifp;
 	__be64 guid;
-	int len;
+	int len, eep_stat;
 	u8 csum, *bguid;
 	int t = dd->ipath_unit;
 	struct ipath_devdata *dd0 = ipath_lookup(0);
@@ -559,7 +604,11 @@ void ipath_get_eeprom_info(struct ipath_devdata *dd)
 		goto bail;
 	}
 
-	if (ipath_eeprom_read(dd, 0, buf, len)) {
+	down(&dd->ipath_eep_sem);
+	eep_stat = ipath_eeprom_internal_read(dd, 0, buf, len);
+	up(&dd->ipath_eep_sem);
+
+	if (eep_stat) {
 		ipath_dev_err(dd, "Failed reading GUID from eeprom\n");
 		goto done;
 	}
@@ -634,8 +683,192 @@ void ipath_get_eeprom_info(struct ipath_devdata *dd)
 	ipath_cdbg(VERBOSE, "Initted GUID to %llx from eeprom\n",
 		   (unsigned long long) be64_to_cpu(dd->ipath_guid));
 
+	memcpy(&dd->ipath_eep_st_errs, &ifp->if_errcntp, IPATH_EEP_LOG_CNT);
+	/*
+	 * Power-on (actually "active") hours are kept as little-endian value
+	 * in EEPROM, but as seconds in a (possibly as small as 24-bit)
+	 * atomic_t while running.
+	 */
+	atomic_set(&dd->ipath_active_time, 0);
+	dd->ipath_eep_hrs = ifp->if_powerhour[0] | (ifp->if_powerhour[1] << 8);
+
 done:
 	vfree(buf);
 
 bail:;
 }
+
+/**
+ * ipath_update_eeprom_log - copy active-time and error counters to eeprom
+ * @dd: the infinipath device
+ *
+ * Although the time is kept as seconds in the ipath_devdata struct, it is
+ * rounded to hours for re-write, as we have only 16 bits in EEPROM.
+ * First-cut code reads whole (expected) struct ipath_flash, modifies,
+ * re-writes. Future direction: read/write only what we need, assuming
+ * that the EEPROM had to have been "good enough" for driver init, and
+ * if not, we aren't making it worse.
+ *
+ */
+
+int ipath_update_eeprom_log(struct ipath_devdata *dd)
+{
+	void *buf;
+	struct ipath_flash *ifp;
+	int len, hi_water;
+	uint32_t new_time, new_hrs;
+	u8 csum;
+	int ret, idx;
+	unsigned long flags;
+
+	/* first, check if we actually need to do anything. */
+	ret = 0;
+	for (idx = 0; idx < IPATH_EEP_LOG_CNT; ++idx) {
+		if (dd->ipath_eep_st_new_errs[idx]) {
+			ret = 1;
+			break;
+		}
+	}
+	new_time = atomic_read(&dd->ipath_active_time);
+
+	if (ret == 0 && new_time < 3600)
+		return 0;
+
+	/*
+	 * The quick-check above determined that there is something worthy
+	 * of logging, so get current contents and do a more detailed idea.
+	 */
+	len = offsetof(struct ipath_flash, if_future);
+	buf = vmalloc(len);
+	ret = 1;
+	if (!buf) {
+		ipath_dev_err(dd, "Couldn't allocate memory to read %u "
+				"bytes from eeprom for logging\n", len);
+		goto bail;
+	}
+
+	/* Grab semaphore and read current EEPROM. If we get an
+	 * error, let go, but if not, keep it until we finish write.
+	 */
+	ret = down_interruptible(&dd->ipath_eep_sem);
+	if (ret) {
+		ipath_dev_err(dd, "Unable to acquire EEPROM for logging\n");
+		goto free_bail;
+	}
+	ret = ipath_eeprom_internal_read(dd, 0, buf, len);
+	if (ret) {
+		up(&dd->ipath_eep_sem);
+		ipath_dev_err(dd, "Unable read EEPROM for logging\n");
+		goto free_bail;
+	}
+	ifp = (struct ipath_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		up(&dd->ipath_eep_sem);
+		ipath_dev_err(dd, "EEPROM cks err (0x%02X, S/B 0x%02X)\n",
+				csum, ifp->if_csum);
+		ret = 1;
+		goto free_bail;
+	}
+	hi_water = 0;
+	spin_lock_irqsave(&dd->ipath_eep_st_lock, flags);
+	for (idx = 0; idx < IPATH_EEP_LOG_CNT; ++idx) {
+		int new_val = dd->ipath_eep_st_new_errs[idx];
+		if (new_val) {
+			/*
+			 * If we have seen any errors, add to EEPROM values
+			 * We need to saturate at 0xFF (255) and we also
+			 * would need to adjust the checksum if we were
+			 * trying to minimize EEPROM traffic
+			 * Note that we add to actual current count in EEPROM,
+			 * in case it was altered while we were running.
+			 */
+			new_val += ifp->if_errcntp[idx];
+			if (new_val > 0xFF)
+				new_val = 0xFF;
+			if (ifp->if_errcntp[idx] != new_val) {
+				ifp->if_errcntp[idx] = new_val;
+				hi_water = offsetof(struct ipath_flash,
+						if_errcntp) + idx;
+			}
+			/*
+			 * update our shadow (used to minimize EEPROM
+			 * traffic), to match what we are about to write.
+			 */
+			dd->ipath_eep_st_errs[idx] = new_val;
+			dd->ipath_eep_st_new_errs[idx] = 0;
+		}
+	}
+	/*
+	 * now update active-time. We would like to round to the nearest hour
+	 * but unless atomic_t are sure to be proper signed ints we cannot,
+	 * because we need to account for what we "transfer" to EEPROM and
+	 * if we log an hour at 31 minutes, then we would need to set
+	 * active_time to -29 to accurately count the _next_ hour.
+	 */
+	if (new_time > 3600) {
+		new_hrs = new_time / 3600;
+		atomic_sub((new_hrs * 3600), &dd->ipath_active_time);
+		new_hrs += dd->ipath_eep_hrs;
+		if (new_hrs > 0xFFFF)
+			new_hrs = 0xFFFF;
+		dd->ipath_eep_hrs = new_hrs;
+		if ((new_hrs & 0xFF) != ifp->if_powerhour[0]) {
+			ifp->if_powerhour[0] = new_hrs & 0xFF;
+			hi_water = offsetof(struct ipath_flash, if_powerhour);
+		}
+		if ((new_hrs >> 8) != ifp->if_powerhour[1]) {
+			ifp->if_powerhour[1] = new_hrs >> 8;
+			hi_water = offsetof(struct ipath_flash, if_powerhour)
+					+ 1;
+		}
+	}
+	/*
+	 * There is a tiny possibility that we could somehow fail to write
+	 * the EEPROM after updating our shadows, but problems from holding
+	 * the spinlock too long are a much bigger issue.
+	 */
+	spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags);
+	if (hi_water) {
+		/* we made some change to the data, uopdate cksum and write */
+		csum = flash_csum(ifp, 1);
+		ret = ipath_eeprom_internal_write(dd, 0, buf, hi_water + 1);
+	}
+	up(&dd->ipath_eep_sem);
+	if (ret)
+		ipath_dev_err(dd, "Failed updating EEPROM\n");
+
+free_bail:
+	vfree(buf);
+bail:
+	return ret;
+
+}
+
+/**
+ * ipath_inc_eeprom_err - increment one of the four error counters
+ * that are logged to EEPROM.
+ * @dd: the infinipath device
+ * @eidx: 0..3, the counter to increment
+ * @incr: how much to add
+ *
+ * Each counter is 8-bits, and saturates at 255 (0xFF). They
+ * are copied to the EEPROM (aka flash) whenever ipath_update_eeprom_log()
+ * is called, but it can only be called in a context that allows sleep.
+ * This function can be called even at interrupt level.
+ */
+
+void ipath_inc_eeprom_err(struct ipath_devdata *dd, u32 eidx, u32 incr)
+{
+	uint new_val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->ipath_eep_st_lock, flags);
+	new_val = dd->ipath_eep_st_new_errs[eidx] + incr;
+	if (new_val > 255)
+		new_val = 255;
+	dd->ipath_eep_st_new_errs[eidx] = new_val;
+	spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags);
+	return;
+}

drivers/infiniband/hw/ipath/ipath_file_ops.c

@@ -396,7 +396,8 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
 			   "TID %u, vaddr %lx, physaddr %llx pgp %p\n",
 			   tid, vaddr, (unsigned long long) physaddr,
 			   pagep[i]);
-		dd->ipath_f_put_tid(dd, &tidbase[tid], 1, physaddr);
+		dd->ipath_f_put_tid(dd, &tidbase[tid], RCVHQ_RCV_TYPE_EXPECTED,
+				    physaddr);
 		/*
 		 * don't check this tid in ipath_portshadow, since we
 		 * just filled it in; start with the next one.
@@ -422,7 +423,8 @@ static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
 			if (dd->ipath_pageshadow[porttid + tid]) {
 				ipath_cdbg(VERBOSE, "Freeing TID %u\n",
 					   tid);
-				dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
+				dd->ipath_f_put_tid(dd, &tidbase[tid],
+						    RCVHQ_RCV_TYPE_EXPECTED,
 						    dd->ipath_tidinvalid);
 				pci_unmap_page(dd->pcidev,
 					dd->ipath_physshadow[porttid + tid],
@@ -538,7 +540,8 @@ static int ipath_tid_free(struct ipath_portdata *pd, unsigned subport,
 		if (dd->ipath_pageshadow[porttid + tid]) {
 			ipath_cdbg(VERBOSE, "PID %u freeing TID %u\n",
 				   pd->port_pid, tid);
-			dd->ipath_f_put_tid(dd, &tidbase[tid], 1,
+			dd->ipath_f_put_tid(dd, &tidbase[tid],
+					    RCVHQ_RCV_TYPE_EXPECTED,
 					    dd->ipath_tidinvalid);
 			pci_unmap_page(dd->pcidev,
 				dd->ipath_physshadow[porttid + tid],
@@ -921,7 +924,8 @@ static int ipath_create_user_egr(struct ipath_portdata *pd)
 					    (u64 __iomem *)
 					    ((char __iomem *)
 					     dd->ipath_kregbase +
-					     dd->ipath_rcvegrbase), 0, pa);
+					     dd->ipath_rcvegrbase),
+					    RCVHQ_RCV_TYPE_EAGER, pa);
 			pa += egrsize;
 		}
 		cond_resched();	/* don't hog the cpu */
@@ -1337,68 +1341,133 @@ bail:
 	return ret;
 }
 
-static unsigned int ipath_poll(struct file *fp,
-			       struct poll_table_struct *pt)
+static unsigned int ipath_poll_urgent(struct ipath_portdata *pd,
+				      struct file *fp,
+				      struct poll_table_struct *pt)
 {
-	struct ipath_portdata *pd;
-	u32 head, tail;
-	int bit;
 	unsigned pollflag = 0;
 	struct ipath_devdata *dd;
 
-	pd = port_fp(fp);
-	if (!pd)
-		goto bail;
 	dd = pd->port_dd;
 
-	bit = pd->port_port + INFINIPATH_R_INTRAVAIL_SHIFT;
-	set_bit(bit, &dd->ipath_rcvctrl);
+	if (test_bit(IPATH_PORT_WAITING_OVERFLOW, &pd->int_flag)) {
+		pollflag |= POLLERR;
+		clear_bit(IPATH_PORT_WAITING_OVERFLOW, &pd->int_flag);
+	}
 
-	/*
-	 * Before blocking, make sure that head is still == tail,
-	 * reading from the chip, so we can be sure the interrupt
-	 * enable has made it to the chip.  If not equal, disable
-	 * interrupt again and return immediately.  This avoids races,
-	 * and the overhead of the chip read doesn't matter much at
-	 * this point, since we are waiting for something anyway.
-	 */
+	if (test_bit(IPATH_PORT_WAITING_URG, &pd->int_flag)) {
+		pollflag |= POLLIN | POLLRDNORM;
+		clear_bit(IPATH_PORT_WAITING_URG, &pd->int_flag);
+	}
 
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
-			 dd->ipath_rcvctrl);
+	if (!pollflag) {
+		set_bit(IPATH_PORT_WAITING_URG, &pd->port_flag);
+		if (pd->poll_type & IPATH_POLL_TYPE_OVERFLOW)
+			set_bit(IPATH_PORT_WAITING_OVERFLOW,
+				&pd->port_flag);
+
+		poll_wait(fp, &pd->port_wait, pt);
+	}
+
+	return pollflag;
+}
+
+static unsigned int ipath_poll_next(struct ipath_portdata *pd,
+				    struct file *fp,
+				    struct poll_table_struct *pt)
+{
+	u32 head, tail;
+	unsigned pollflag = 0;
+	struct ipath_devdata *dd;
+
+	dd = pd->port_dd;
 
 	head = ipath_read_ureg32(dd, ur_rcvhdrhead, pd->port_port);
-	tail = ipath_read_ureg32(dd, ur_rcvhdrtail, pd->port_port);
+	tail = *(volatile u64 *)pd->port_rcvhdrtail_kvaddr;
 
-	if (tail == head) {
+	if (test_bit(IPATH_PORT_WAITING_OVERFLOW, &pd->int_flag)) {
+		pollflag |= POLLERR;
+		clear_bit(IPATH_PORT_WAITING_OVERFLOW, &pd->int_flag);
+	}
+
+	if (tail != head ||
+	    test_bit(IPATH_PORT_WAITING_RCV, &pd->int_flag)) {
+		pollflag |= POLLIN | POLLRDNORM;
+		clear_bit(IPATH_PORT_WAITING_RCV, &pd->int_flag);
+	}
+
+	if (!pollflag) {
 		set_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag);
+		if (pd->poll_type & IPATH_POLL_TYPE_OVERFLOW)
+			set_bit(IPATH_PORT_WAITING_OVERFLOW,
+				&pd->port_flag);
+
+		set_bit(pd->port_port + INFINIPATH_R_INTRAVAIL_SHIFT,
+			&dd->ipath_rcvctrl);
+
+		ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
+				 dd->ipath_rcvctrl);
+
 		if (dd->ipath_rhdrhead_intr_off) /* arm rcv interrupt */
-			(void)ipath_write_ureg(dd, ur_rcvhdrhead,
-					       dd->ipath_rhdrhead_intr_off
-					       | head, pd->port_port);
-		poll_wait(fp, &pd->port_wait, pt);
+			ipath_write_ureg(dd, ur_rcvhdrhead,
+					 dd->ipath_rhdrhead_intr_off | head,
+					 pd->port_port);
 
-		if (test_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag)) {
-			/* timed out, no packets received */
-			clear_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag);
-			pd->port_rcvwait_to++;
-		}
-		else
-			pollflag = POLLIN | POLLRDNORM;
-	}
-	else {
-		/* it's already happened; don't do wait_event overhead */
-		pollflag = POLLIN | POLLRDNORM;
-		pd->port_rcvnowait++;
+		poll_wait(fp, &pd->port_wait, pt);
 	}
 
-	clear_bit(bit, &dd->ipath_rcvctrl);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
-			 dd->ipath_rcvctrl);
+	return pollflag;
+}
+
+static unsigned int ipath_poll(struct file *fp,
+			       struct poll_table_struct *pt)
+{
+	struct ipath_portdata *pd;
+	unsigned pollflag;
+
+	pd = port_fp(fp);
+	if (!pd)
+		pollflag = 0;
+	else if (pd->poll_type & IPATH_POLL_TYPE_URGENT)
+		pollflag = ipath_poll_urgent(pd, fp, pt);
+	else
+		pollflag = ipath_poll_next(pd, fp, pt);
 
-bail:
 	return pollflag;
 }
 
+static int ipath_supports_subports(int user_swmajor, int user_swminor)
+{
+	/* no subport implementation prior to software version 1.3 */
+	return (user_swmajor > 1) || (user_swminor >= 3);
+}
+
+static int ipath_compatible_subports(int user_swmajor, int user_swminor)
+{
+	/* this code is written long-hand for clarity */
+	if (IPATH_USER_SWMAJOR != user_swmajor) {
+		/* no promise of compatibility if major mismatch */
+		return 0;
+	}
+	if (IPATH_USER_SWMAJOR == 1) {
+		switch (IPATH_USER_SWMINOR) {
+		case 0:
+		case 1:
+		case 2:
+			/* no subport implementation so cannot be compatible */
+			return 0;
+		case 3:
+			/* 3 is only compatible with itself */
+			return user_swminor == 3;
+		default:
+			/* >= 4 are compatible (or are expected to be) */
+			return user_swminor >= 4;
+		}
+	}
+	/* make no promises yet for future major versions */
+	return 0;
+}
+
 static int init_subports(struct ipath_devdata *dd,
 			 struct ipath_portdata *pd,
 			 const struct ipath_user_info *uinfo)
@@ -1408,20 +1477,32 @@ static int init_subports(struct ipath_devdata *dd,
 	size_t size;
 
 	/*
-	 * If the user is requesting zero or one port,
+	 * If the user is requesting zero subports,
 	 * skip the subport allocation.
 	 */
-	if (uinfo->spu_subport_cnt <= 1)
+	if (uinfo->spu_subport_cnt <= 0)
+		goto bail;
+
+	/* Self-consistency check for ipath_compatible_subports() */
+	if (ipath_supports_subports(IPATH_USER_SWMAJOR, IPATH_USER_SWMINOR) &&
+	    !ipath_compatible_subports(IPATH_USER_SWMAJOR,
+				       IPATH_USER_SWMINOR)) {
+		dev_info(&dd->pcidev->dev,
+			 "Inconsistent ipath_compatible_subports()\n");
 		goto bail;
+	}
 
-	/* Old user binaries don't know about new subport implementation */
-	if ((uinfo->spu_userversion & 0xffff) != IPATH_USER_SWMINOR) {
+	/* Check for subport compatibility */
+	if (!ipath_compatible_subports(uinfo->spu_userversion >> 16,
+				       uinfo->spu_userversion & 0xffff)) {
 		dev_info(&dd->pcidev->dev,
-			 "Mismatched user minor version (%d) and driver "
-                         "minor version (%d) while port sharing. Ensure "
+			 "Mismatched user version (%d.%d) and driver "
+			 "version (%d.%d) while port sharing. Ensure "
                          "that driver and library are from the same "
                          "release.\n",
+			 (int) (uinfo->spu_userversion >> 16),
                          (int) (uinfo->spu_userversion & 0xffff),
+			 IPATH_USER_SWMAJOR,
 	                 IPATH_USER_SWMINOR);
 		goto bail;
 	}
@@ -1725,14 +1806,13 @@ static int ipath_open(struct inode *in, struct file *fp)
 	return fp->private_data ? 0 : -ENOMEM;
 }
 
-
 /* Get port early, so can set affinity prior to memory allocation */
 static int ipath_assign_port(struct file *fp,
 			      const struct ipath_user_info *uinfo)
 {
 	int ret;
 	int i_minor;
-	unsigned swminor;
+	unsigned swmajor, swminor;
 
 	/* Check to be sure we haven't already initialized this file */
 	if (port_fp(fp)) {
@@ -1741,7 +1821,8 @@ static int ipath_assign_port(struct file *fp,
 	}
 
 	/* for now, if major version is different, bail */
-	if ((uinfo->spu_userversion >> 16) != IPATH_USER_SWMAJOR) {
+	swmajor = uinfo->spu_userversion >> 16;
+	if (swmajor != IPATH_USER_SWMAJOR) {
 		ipath_dbg("User major version %d not same as driver "
 			  "major %d\n", uinfo->spu_userversion >> 16,
 			  IPATH_USER_SWMAJOR);
@@ -1756,7 +1837,8 @@ static int ipath_assign_port(struct file *fp,
 
 	mutex_lock(&ipath_mutex);
 
-	if (swminor == IPATH_USER_SWMINOR && uinfo->spu_subport_cnt &&
+	if (ipath_compatible_subports(swmajor, swminor) &&
+	    uinfo->spu_subport_cnt &&
 	    (ret = find_shared_port(fp, uinfo))) {
 		mutex_unlock(&ipath_mutex);
 		if (ret > 0)
@@ -2020,7 +2102,8 @@ static int ipath_port_info(struct ipath_portdata *pd, u16 subport,
 	info.port = pd->port_port;
 	info.subport = subport;
 	/* Don't return new fields if old library opened the port. */
-	if ((pd->userversion & 0xffff) == IPATH_USER_SWMINOR) {
+	if (ipath_supports_subports(pd->userversion >> 16,
+				    pd->userversion & 0xffff)) {
 		/* Number of user ports available for this device. */
 		info.num_ports = pd->port_dd->ipath_cfgports - 1;
 		info.num_subports = pd->port_subport_cnt;
@@ -2123,6 +2206,11 @@ static ssize_t ipath_write(struct file *fp, const char __user *data,
 		src = NULL;
 		dest = NULL;
 		break;
+	case IPATH_CMD_POLL_TYPE:
+		copy = sizeof(cmd.cmd.poll_type);
+		dest = &cmd.cmd.poll_type;
+		src = &ucmd->cmd.poll_type;
+		break;
 	default:
 		ret = -EINVAL;
 		goto bail;
@@ -2195,6 +2283,9 @@ static ssize_t ipath_write(struct file *fp, const char __user *data,
 	case IPATH_CMD_PIOAVAILUPD:
 		ret = ipath_force_pio_avail_update(pd->port_dd);
 		break;
+	case IPATH_CMD_POLL_TYPE:
+		pd->poll_type = cmd.cmd.poll_type;
+		break;
 	}
 
 	if (ret >= 0)

drivers/infiniband/hw/ipath/ipath_fs.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -257,9 +257,14 @@ static ssize_t atomic_port_info_read(struct file *file, char __user *buf,
 		/* Notimpl InitType (actually, an SMA decision) */
 		/* VLHighLimit is 0 (only one VL) */
 		; /* VLArbitrationHighCap is 0 (only one VL) */
+	/*
+	 * Note: the chips support a maximum MTU of 4096, but the driver
+	 * hasn't implemented this feature yet, so set the maximum
+	 * to 2048.
+	 */
 	portinfo[10] = 	/* VLArbitrationLowCap is 0 (only one VL) */
 		/* InitTypeReply is SMA decision */
-		(5 << 16)	/* MTUCap 4096 */
+		(4 << 16)	/* MTUCap 2048 */
 		| (7 << 13)	/* VLStallCount */
 		| (0x1f << 8)	/* HOQLife */
 		| (1 << 4)

drivers/infiniband/hw/ipath/ipath_iba6110.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -36,6 +36,7 @@
  * HT chip.
  */
 
+#include <linux/vmalloc.h>
 #include <linux/pci.h>
 #include <linux/delay.h>
 #include <linux/htirq.h>
@@ -439,6 +440,7 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 	u32 bits, ctrl;
 	int isfatal = 0;
 	char bitsmsg[64];
+	int log_idx;
 
 	hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
 
@@ -467,6 +469,11 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 
 	hwerrs &= dd->ipath_hwerrmask;
 
+	/* We log some errors to EEPROM, check if we have any of those. */
+	for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
+		if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
+			ipath_inc_eeprom_err(dd, log_idx, 1);
+
 	/*
 	 * make sure we get this much out, unless told to be quiet,
 	 * it's a parity error we may recover from,
@@ -502,9 +509,7 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 		if (!hwerrs) {
 			ipath_dbg("Clearing freezemode on ignored or "
 				  "recovered hardware error\n");
-			ctrl &= ~INFINIPATH_C_FREEZEMODE;
-			ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-					 ctrl);
+			ipath_clear_freeze(dd);
 		}
 	}
 
@@ -672,10 +677,16 @@ static int ipath_ht_boardname(struct ipath_devdata *dd, char *name,
 	if (n)
 		snprintf(name, namelen, "%s", n);
 
+	if (dd->ipath_boardrev != 6 && dd->ipath_boardrev != 7 &&
+	    dd->ipath_boardrev != 11) {
+		ipath_dev_err(dd, "Unsupported InfiniPath board %s!\n", name);
+		ret = 1;
+		goto bail;
+	}
 	if (dd->ipath_majrev != 3 || (dd->ipath_minrev < 2 ||
-		dd->ipath_minrev > 3)) {
+		dd->ipath_minrev > 4)) {
 		/*
-		 * This version of the driver only supports Rev 3.2 and 3.3
+		 * This version of the driver only supports Rev 3.2 - 3.4
 		 */
 		ipath_dev_err(dd,
 			      "Unsupported InfiniPath hardware revision %u.%u!\n",
@@ -689,36 +700,11 @@ static int ipath_ht_boardname(struct ipath_devdata *dd, char *name,
 	 * copies
 	 */
 	dd->ipath_flags |= IPATH_32BITCOUNTERS;
+	dd->ipath_flags |= IPATH_GPIO_INTR;
 	if (dd->ipath_htspeed != 800)
 		ipath_dev_err(dd,
 			      "Incorrectly configured for HT @ %uMHz\n",
 			      dd->ipath_htspeed);
-	if (dd->ipath_boardrev == 7 || dd->ipath_boardrev == 11 ||
-	    dd->ipath_boardrev == 6)
-		dd->ipath_flags |= IPATH_GPIO_INTR;
-	else
-		dd->ipath_flags |= IPATH_POLL_RX_INTR;
-	if (dd->ipath_boardrev == 8) {	/* LS/X-1 */
-		u64 val;
-		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
-		if (val & INFINIPATH_EXTS_SERDESSEL) {
-			/*
-			 * hardware disabled
-			 *
-			 * This means that the chip is hardware disabled,
-			 * and will not be able to bring up the link,
-			 * in any case.  We special case this and abort
-			 * early, to avoid later messages.  We also set
-			 * the DISABLED status bit
-			 */
-			ipath_dbg("Unit %u is hardware-disabled\n",
-				  dd->ipath_unit);
-			*dd->ipath_statusp |= IPATH_STATUS_DISABLED;
-			/* this value is handled differently */
-			ret = 2;
-			goto bail;
-		}
-	}
 	ret = 0;
 
 bail:
@@ -1058,12 +1044,24 @@ static void ipath_setup_ht_setextled(struct ipath_devdata *dd,
 				     u64 lst, u64 ltst)
 {
 	u64 extctl;
+	unsigned long flags = 0;
 
 	/* the diags use the LED to indicate diag info, so we leave
 	 * the external LED alone when the diags are running */
 	if (ipath_diag_inuse)
 		return;
 
+	/* Allow override of LED display for, e.g. Locating system in rack */
+	if (dd->ipath_led_override) {
+		ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
+			? INFINIPATH_IBCS_LT_STATE_LINKUP
+			: INFINIPATH_IBCS_LT_STATE_DISABLED;
+		lst = (dd->ipath_led_override & IPATH_LED_LOG)
+			? INFINIPATH_IBCS_L_STATE_ACTIVE
+			: INFINIPATH_IBCS_L_STATE_DOWN;
+	}
+
+	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
 	/*
 	 * start by setting both LED control bits to off, then turn
 	 * on the appropriate bit(s).
@@ -1092,6 +1090,7 @@ static void ipath_setup_ht_setextled(struct ipath_devdata *dd,
 	}
 	dd->ipath_extctrl = extctl;
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
+	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
 }
 
 static void ipath_init_ht_variables(struct ipath_devdata *dd)
@@ -1157,6 +1156,22 @@ static void ipath_init_ht_variables(struct ipath_devdata *dd)
 
 	dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
 	dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
+
+	/*
+	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
+	 * 2 is Some Misc, 3 is reserved for future.
+	 */
+	dd->ipath_eep_st_masks[0].hwerrs_to_log =
+		INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
+		INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
+
+	dd->ipath_eep_st_masks[1].hwerrs_to_log =
+		INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
+		INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
+
+	dd->ipath_eep_st_masks[2].errs_to_log =
+		INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET;
+
 }
 
 /**
@@ -1372,7 +1387,7 @@ static void ipath_ht_quiet_serdes(struct ipath_devdata *dd)
  * ipath_pe_put_tid - write a TID in chip
  * @dd: the infinipath device
  * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: 0 for eager, 1 for expected
+ * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
  *
  * This exists as a separate routine to allow for special locking etc.
@@ -1393,7 +1408,7 @@ static void ipath_ht_put_tid(struct ipath_devdata *dd,
 				 "40 bits, using only 40!!!\n", pa);
 			pa &= INFINIPATH_RT_ADDR_MASK;
 		}
-		if (type == 0)
+		if (type == RCVHQ_RCV_TYPE_EAGER)
 			pa |= dd->ipath_tidtemplate;
 		else {
 			/* in words (fixed, full page).  */
@@ -1433,7 +1448,8 @@ static void ipath_ht_clear_tids(struct ipath_devdata *dd, unsigned port)
 				   port * dd->ipath_rcvtidcnt *
 				   sizeof(*tidbase));
 	for (i = 0; i < dd->ipath_rcvtidcnt; i++)
-		ipath_ht_put_tid(dd, &tidbase[i], 1, dd->ipath_tidinvalid);
+		ipath_ht_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
+				 dd->ipath_tidinvalid);
 
 	tidbase = (u64 __iomem *) ((char __iomem *)(dd->ipath_kregbase) +
 				   dd->ipath_rcvegrbase +
@@ -1441,7 +1457,8 @@ static void ipath_ht_clear_tids(struct ipath_devdata *dd, unsigned port)
 				   sizeof(*tidbase));
 
 	for (i = 0; i < dd->ipath_rcvegrcnt; i++)
-		ipath_ht_put_tid(dd, &tidbase[i], 0, dd->ipath_tidinvalid);
+		ipath_ht_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
+				 dd->ipath_tidinvalid);
 }
 
 /**
@@ -1528,11 +1545,6 @@ static int ipath_ht_early_init(struct ipath_devdata *dd)
 		writel(16, piobuf);
 		piobuf += pioincr;
 	}
-	/*
-	 * self-clearing
-	 */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
-			 INFINIPATH_S_ABORT);
 
 	ipath_get_eeprom_info(dd);
 	if (dd->ipath_boardrev == 5 && dd->ipath_serial[0] == '1' &&
@@ -1543,8 +1555,10 @@ static int ipath_ht_early_init(struct ipath_devdata *dd)
 		 * with 128, rather than 112.
 		 */
 		dd->ipath_flags |= IPATH_GPIO_INTR;
-		dd->ipath_flags &= ~IPATH_POLL_RX_INTR;
-	}
+	} else
+		ipath_dev_err(dd, "Unsupported InfiniPath serial "
+			      "number %.16s!\n", dd->ipath_serial);
+
 	return 0;
 }
 
@@ -1561,7 +1575,6 @@ static int ipath_ht_txe_recover(struct ipath_devdata *dd)
 	}
 	dev_info(&dd->pcidev->dev,
 		"Recovering from TXE PIO parity error\n");
-	ipath_disarm_senderrbufs(dd, 1);
 	return 1;
 }
 

drivers/infiniband/hw/ipath/ipath_iba6120.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -296,13 +296,6 @@ static const struct ipath_cregs ipath_pe_cregs = {
 #define IPATH_GPIO_SCL (1ULL << \
 	(_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
 
-/*
- * Rev2 silicon allows suppressing check for ArmLaunch errors.
- * this can speed up short packet sends on systems that do
- * not guaranteee write-order.
- */
-#define INFINIPATH_XGXS_SUPPRESS_ARMLAUNCH_ERR (1ULL<<63)
-
 /* 6120 specific hardware errors... */
 static const struct ipath_hwerror_msgs ipath_6120_hwerror_msgs[] = {
 	INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"),
@@ -347,6 +340,7 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 	u32 bits, ctrl;
 	int isfatal = 0;
 	char bitsmsg[64];
+	int log_idx;
 
 	hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
 	if (!hwerrs) {
@@ -374,6 +368,11 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 
 	hwerrs &= dd->ipath_hwerrmask;
 
+	/* We log some errors to EEPROM, check if we have any of those. */
+	for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
+		if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
+			ipath_inc_eeprom_err(dd, log_idx, 1);
+
 	/*
 	 * make sure we get this much out, unless told to be quiet,
 	 * or it's occurred within the last 5 seconds
@@ -431,10 +430,12 @@ static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 			*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
 			dd->ipath_flags &= ~IPATH_INITTED;
 		} else {
-			ipath_dbg("Clearing freezemode on ignored hardware "
-				  "error\n");
-			ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-			   		 dd->ipath_control);
+			static u32 freeze_cnt;
+
+			freeze_cnt++;
+			ipath_dbg("Clearing freezemode on ignored or recovered "
+				  "hardware error (%u)\n", freeze_cnt);
+			ipath_clear_freeze(dd);
 		}
 	}
 
@@ -680,17 +681,6 @@ static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
 		val |= dd->ipath_rx_pol_inv <<
 			INFINIPATH_XGXS_RX_POL_SHIFT;
 	}
-	if (dd->ipath_minrev >= 2) {
-		/* Rev 2. can tolerate multiple writes to PBC, and
-		 * allowing them can provide lower latency on some
-		 * CPUs, but this feature is off by default, only
-		 * turned on by setting D63 of XGXSconfig reg.
-		 * May want to make this conditional more
-		 * fine-grained in future. This is not exactly
-		 * related to XGXS, but where the bit ended up.
-		 */
-		val |= INFINIPATH_XGXS_SUPPRESS_ARMLAUNCH_ERR;
-	}
 	if (val != prev_val)
 		ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
 
@@ -791,12 +781,24 @@ static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst,
 				     u64 ltst)
 {
 	u64 extctl;
+	unsigned long flags = 0;
 
 	/* the diags use the LED to indicate diag info, so we leave
 	 * the external LED alone when the diags are running */
 	if (ipath_diag_inuse)
 		return;
 
+	/* Allow override of LED display for, e.g. Locating system in rack */
+	if (dd->ipath_led_override) {
+		ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
+			? INFINIPATH_IBCS_LT_STATE_LINKUP
+			: INFINIPATH_IBCS_LT_STATE_DISABLED;
+		lst = (dd->ipath_led_override & IPATH_LED_LOG)
+			? INFINIPATH_IBCS_L_STATE_ACTIVE
+			: INFINIPATH_IBCS_L_STATE_DOWN;
+	}
+
+	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
 	extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON |
 				       INFINIPATH_EXTC_LED2PRIPORT_ON);
 
@@ -806,6 +808,7 @@ static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst,
 		extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON;
 	dd->ipath_extctrl = extctl;
 	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
+	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
 }
 
 /**
@@ -955,6 +958,27 @@ static void ipath_init_pe_variables(struct ipath_devdata *dd)
 
 	dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
 	dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
+
+	/*
+	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
+	 * 2 is Some Misc, 3 is reserved for future.
+	 */
+	dd->ipath_eep_st_masks[0].hwerrs_to_log =
+		INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
+		INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
+
+	/* Ignore errors in PIO/PBC on systems with unordered write-combining */
+	if (ipath_unordered_wc())
+		dd->ipath_eep_st_masks[0].hwerrs_to_log &= ~TXE_PIO_PARITY;
+
+	dd->ipath_eep_st_masks[1].hwerrs_to_log =
+		INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
+		INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
+
+	dd->ipath_eep_st_masks[2].errs_to_log =
+		INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET;
+
+
 }
 
 /* setup the MSI stuff again after a reset.  I'd like to just call
@@ -1082,7 +1106,7 @@ bail:
  * ipath_pe_put_tid - write a TID in chip
  * @dd: the infinipath device
  * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: 0 for eager, 1 for expected
+ * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
  *
  * This exists as a separate routine to allow for special locking etc.
@@ -1108,7 +1132,7 @@ static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
 				      "BUG: Physical page address 0x%lx "
 				      "has bits set in 31-29\n", pa);
 
-		if (type == 0)
+		if (type == RCVHQ_RCV_TYPE_EAGER)
 			pa |= dd->ipath_tidtemplate;
 		else /* for now, always full 4KB page */
 			pa |= 2 << 29;
@@ -1132,7 +1156,7 @@ static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
  * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher
  * @dd: the infinipath device
  * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: 0 for eager, 1 for expected
+ * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
  *
  * This exists as a separate routine to allow for selection of the
@@ -1157,7 +1181,7 @@ static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
 				      "BUG: Physical page address 0x%lx "
 				      "has bits set in 31-29\n", pa);
 
-		if (type == 0)
+		if (type == RCVHQ_RCV_TYPE_EAGER)
 			pa |= dd->ipath_tidtemplate;
 		else /* for now, always full 4KB page */
 			pa |= 2 << 29;
@@ -1196,7 +1220,8 @@ static void ipath_pe_clear_tids(struct ipath_devdata *dd, unsigned port)
 		 port * dd->ipath_rcvtidcnt * sizeof(*tidbase));
 
 	for (i = 0; i < dd->ipath_rcvtidcnt; i++)
-		ipath_pe_put_tid(dd, &tidbase[i], 0, tidinv);
+		ipath_pe_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
+				 tidinv);
 
 	tidbase = (u64 __iomem *)
 		((char __iomem *)(dd->ipath_kregbase) +
@@ -1204,7 +1229,8 @@ static void ipath_pe_clear_tids(struct ipath_devdata *dd, unsigned port)
 		 port * dd->ipath_rcvegrcnt * sizeof(*tidbase));
 
 	for (i = 0; i < dd->ipath_rcvegrcnt; i++)
-		ipath_pe_put_tid(dd, &tidbase[i], 1, tidinv);
+		ipath_pe_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
+				 tidinv);
 }
 
 /**
@@ -1311,13 +1337,6 @@ static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
 
 	dd = pd->port_dd;
 
-	if (dd != NULL && dd->ipath_minrev >= 2) {
-		ipath_cdbg(PROC, "IBA6120 Rev2, allow multiple PBC write\n");
-		kinfo->spi_runtime_flags |= IPATH_RUNTIME_PBC_REWRITE;
-		ipath_cdbg(PROC, "IBA6120 Rev2, allow loose DMA alignment\n");
-		kinfo->spi_runtime_flags |= IPATH_RUNTIME_LOOSE_DMA_ALIGN;
-	}
-
 done:
 	kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE;
 	return 0;
@@ -1354,7 +1373,6 @@ static int ipath_pe_txe_recover(struct ipath_devdata *dd)
 		dev_info(&dd->pcidev->dev,
 			"Recovering from TXE PIO parity error\n");
 	}
-	ipath_disarm_senderrbufs(dd, 1);
 	return 1;
 }
 

drivers/infiniband/hw/ipath/ipath_init_chip.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -133,7 +133,8 @@ static int create_port0_egr(struct ipath_devdata *dd)
 				   dd->ipath_ibmaxlen, PCI_DMA_FROMDEVICE);
 		dd->ipath_f_put_tid(dd, e + (u64 __iomem *)
 				    ((char __iomem *) dd->ipath_kregbase +
-				     dd->ipath_rcvegrbase), 0,
+				     dd->ipath_rcvegrbase),
+				    RCVHQ_RCV_TYPE_EAGER,
 				    dd->ipath_port0_skbinfo[e].phys);
 	}
 
@@ -310,7 +311,12 @@ static int init_chip_first(struct ipath_devdata *dd,
 	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiosize);
 	dd->ipath_piosize2k = val & ~0U;
 	dd->ipath_piosize4k = val >> 32;
-	dd->ipath_ibmtu = 4096;	/* default to largest legal MTU */
+	/*
+	 * Note: the chips support a maximum MTU of 4096, but the driver
+	 * hasn't implemented this feature yet, so set the initial value
+	 * to 2048.
+	 */
+	dd->ipath_ibmtu = 2048;
 	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_sendpiobufcnt);
 	dd->ipath_piobcnt2k = val & ~0U;
 	dd->ipath_piobcnt4k = val >> 32;
@@ -340,6 +346,10 @@ static int init_chip_first(struct ipath_devdata *dd,
 
 	spin_lock_init(&dd->ipath_tid_lock);
 
+	spin_lock_init(&dd->ipath_gpio_lock);
+	spin_lock_init(&dd->ipath_eep_st_lock);
+	sema_init(&dd->ipath_eep_sem, 1);
+
 done:
 	*pdp = pd;
 	return ret;
@@ -646,7 +656,7 @@ static int init_housekeeping(struct ipath_devdata *dd,
 	ret = dd->ipath_f_get_boardname(dd, boardn, sizeof boardn);
 
 	snprintf(dd->ipath_boardversion, sizeof(dd->ipath_boardversion),
-		 "Driver %u.%u, %s, InfiniPath%u %u.%u, PCI %u, "
+		 "ChipABI %u.%u, %s, InfiniPath%u %u.%u, PCI %u, "
 		 "SW Compat %u\n",
 		 IPATH_CHIP_VERS_MAJ, IPATH_CHIP_VERS_MIN, boardn,
 		 (unsigned)(dd->ipath_revision >> INFINIPATH_R_ARCH_SHIFT) &
@@ -727,7 +737,7 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit)
 	uports = dd->ipath_cfgports ? dd->ipath_cfgports - 1 : 0;
 	if (ipath_kpiobufs == 0) {
 		/* not set by user (this is default) */
-		if (piobufs >= (uports * IPATH_MIN_USER_PORT_BUFCNT) + 32)
+		if (piobufs > 144)
 			kpiobufs = 32;
 		else
 			kpiobufs = 16;
@@ -767,6 +777,12 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit)
 		   piobufs, dd->ipath_pbufsport, uports);
 
 	dd->ipath_f_early_init(dd);
+	/*
+	 * cancel any possible active sends from early driver load.
+	 * Follows early_init because some chips have to initialize
+	 * PIO buffers in early_init to avoid false parity errors.
+	 */
+	ipath_cancel_sends(dd);
 
 	/* early_init sets rcvhdrentsize and rcvhdrsize, so this must be
 	 * done after early_init */

drivers/infiniband/hw/ipath/ipath_intr.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -93,7 +93,8 @@ void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite)
 
 	if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) {
 		int i;
-		if (ipath_debug & (__IPATH_PKTDBG|__IPATH_DBG)) {
+		if (ipath_debug & (__IPATH_PKTDBG|__IPATH_DBG) &&
+			dd->ipath_lastcancel > jiffies) {
 			__IPATH_DBG_WHICH(__IPATH_PKTDBG|__IPATH_DBG,
 					  "SendbufErrs %lx %lx", sbuf[0],
 					  sbuf[1]);
@@ -108,7 +109,8 @@ void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite)
 					ipath_clrpiobuf(dd, i);
 				ipath_disarm_piobufs(dd, i, 1);
 			}
-		dd->ipath_lastcancel = jiffies+3; /* no armlaunch for a bit */
+		/* ignore armlaunch errs for a bit */
+		dd->ipath_lastcancel = jiffies+3;
 	}
 }
 
@@ -130,6 +132,17 @@ void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite)
 	 INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SPKTLEN | \
 	 INFINIPATH_E_INVALIDADDR)
 
+/*
+ * this is similar to E_SUM_ERRS, but can't ignore armlaunch, don't ignore
+ * errors not related to freeze and cancelling buffers.  Can't ignore
+ * armlaunch because could get more while still cleaning up, and need
+ * to cancel those as they happen.
+ */
+#define E_SPKT_ERRS_IGNORE \
+	 (INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
+	 INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SMINPKTLEN | \
+	 INFINIPATH_E_SPKTLEN)
+
 /*
  * these are errors that can occur when the link changes state while
  * a packet is being sent or received.  This doesn't cover things
@@ -290,12 +303,7 @@ static void handle_e_ibstatuschanged(struct ipath_devdata *dd,
 		 * Flush all queued sends when link went to DOWN or INIT,
 		 * to be sure that they don't block SMA and other MAD packets
 		 */
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
-				 INFINIPATH_S_ABORT);
-		ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
-							(unsigned)(dd->ipath_piobcnt2k +
-					dd->ipath_piobcnt4k) -
-					dd->ipath_lastport_piobuf);
+		ipath_cancel_sends(dd);
 	}
 	else if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM ||
 	    lstate == IPATH_IBSTATE_ACTIVE) {
@@ -505,6 +513,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
 	int i, iserr = 0;
 	int chkerrpkts = 0, noprint = 0;
 	unsigned supp_msgs;
+	int log_idx;
 
 	supp_msgs = handle_frequent_errors(dd, errs, msg, &noprint);
 
@@ -518,6 +527,13 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
 	if (errs & INFINIPATH_E_HARDWARE) {
 		/* reuse same msg buf */
 		dd->ipath_f_handle_hwerrors(dd, msg, sizeof msg);
+	} else {
+		u64 mask;
+		for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx) {
+			mask = dd->ipath_eep_st_masks[log_idx].errs_to_log;
+			if (errs & mask)
+				ipath_inc_eeprom_err(dd, log_idx, 1);
+		}
 	}
 
 	if (!noprint && (errs & ~dd->ipath_e_bitsextant))
@@ -675,6 +691,17 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
 					chkerrpkts = 1;
 				dd->ipath_lastrcvhdrqtails[i] = tl;
 				pd->port_hdrqfull++;
+				if (test_bit(IPATH_PORT_WAITING_OVERFLOW,
+					     &pd->port_flag)) {
+					clear_bit(
+					  IPATH_PORT_WAITING_OVERFLOW,
+					  &pd->port_flag);
+					set_bit(
+					  IPATH_PORT_WAITING_OVERFLOW,
+					  &pd->int_flag);
+					wake_up_interruptible(
+					  &pd->port_wait);
+				}
 			}
 		}
 	}
@@ -744,6 +771,72 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
 	return chkerrpkts;
 }
 
+
+/*
+ * try to cleanup as much as possible for anything that might have gone
+ * wrong while in freeze mode, such as pio buffers being written by user
+ * processes (causing armlaunch), send errors due to going into freeze mode,
+ * etc., and try to avoid causing extra interrupts while doing so.
+ * Forcibly update the in-memory pioavail register copies after cleanup
+ * because the chip won't do it for anything changing while in freeze mode
+ * (we don't want to wait for the next pio buffer state change).
+ * Make sure that we don't lose any important interrupts by using the chip
+ * feature that says that writing 0 to a bit in *clear that is set in
+ * *status will cause an interrupt to be generated again (if allowed by
+ * the *mask value).
+ */
+void ipath_clear_freeze(struct ipath_devdata *dd)
+{
+	int i, im;
+	__le64 val;
+
+	/* disable error interrupts, to avoid confusion */
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, 0ULL);
+
+	/*
+	 * clear all sends, because they have may been
+	 * completed by usercode while in freeze mode, and
+	 * therefore would not be sent, and eventually
+	 * might cause the process to run out of bufs
+	 */
+	ipath_cancel_sends(dd);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
+			 dd->ipath_control);
+
+	/* ensure pio avail updates continue */
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+		 dd->ipath_sendctrl & ~IPATH_S_PIOBUFAVAILUPD);
+	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
+		 dd->ipath_sendctrl);
+
+	/*
+	 * We just enabled pioavailupdate, so dma copy is almost certainly
+	 * not yet right, so read the registers directly.  Similar to init
+	 */
+	for (i = 0; i < dd->ipath_pioavregs; i++) {
+		/* deal with 6110 chip bug */
+		im = i > 3 ? ((i&1) ? i-1 : i+1) : i;
+		val = ipath_read_kreg64(dd, 0x1000+(im*sizeof(u64)));
+		dd->ipath_pioavailregs_dma[i] = dd->ipath_pioavailshadow[i]
+			= le64_to_cpu(val);
+	}
+
+	/*
+	 * force new interrupt if any hwerr, error or interrupt bits are
+	 * still set, and clear "safe" send packet errors related to freeze
+	 * and cancelling sends.  Re-enable error interrupts before possible
+	 * force of re-interrupt on pending interrupts.
+	 */
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, 0ULL);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear,
+		E_SPKT_ERRS_IGNORE);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask,
+		~dd->ipath_maskederrs);
+	ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, 0ULL);
+}
+
+
 /* this is separate to allow for better optimization of ipath_intr() */
 
 static void ipath_bad_intr(struct ipath_devdata *dd, u32 * unexpectp)
@@ -872,14 +965,25 @@ static void handle_urcv(struct ipath_devdata *dd, u32 istat)
 		   dd->ipath_i_rcvurg_mask);
 	for (i = 1; i < dd->ipath_cfgports; i++) {
 		struct ipath_portdata *pd = dd->ipath_pd[i];
-		if (portr & (1 << i) && pd && pd->port_cnt &&
-			test_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag)) {
-			clear_bit(IPATH_PORT_WAITING_RCV,
-				  &pd->port_flag);
-			clear_bit(i + INFINIPATH_R_INTRAVAIL_SHIFT,
-				  &dd->ipath_rcvctrl);
-			wake_up_interruptible(&pd->port_wait);
-			rcvdint = 1;
+		if (portr & (1 << i) && pd && pd->port_cnt) {
+			if (test_bit(IPATH_PORT_WAITING_RCV,
+				     &pd->port_flag)) {
+				clear_bit(IPATH_PORT_WAITING_RCV,
+					  &pd->port_flag);
+				set_bit(IPATH_PORT_WAITING_RCV,
+					&pd->int_flag);
+				clear_bit(i + INFINIPATH_R_INTRAVAIL_SHIFT,
+					  &dd->ipath_rcvctrl);
+				wake_up_interruptible(&pd->port_wait);
+				rcvdint = 1;
+			} else if (test_bit(IPATH_PORT_WAITING_URG,
+					    &pd->port_flag)) {
+				clear_bit(IPATH_PORT_WAITING_URG,
+					  &pd->port_flag);
+				set_bit(IPATH_PORT_WAITING_URG,
+					&pd->int_flag);
+				wake_up_interruptible(&pd->port_wait);
+			}
 		}
 	}
 	if (rcvdint) {
@@ -905,6 +1009,9 @@ irqreturn_t ipath_intr(int irq, void *data)
 
 	ipath_stats.sps_ints++;
 
+	if (dd->ipath_int_counter != (u32) -1)
+		dd->ipath_int_counter++;
+
 	if (!(dd->ipath_flags & IPATH_PRESENT)) {
 		/*
 		 * This return value is not great, but we do not want the

drivers/infiniband/hw/ipath/ipath_kernel.h

@@ -1,7 +1,7 @@
 #ifndef _IPATH_KERNEL_H
 #define _IPATH_KERNEL_H
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -57,6 +57,24 @@
 extern struct infinipath_stats ipath_stats;
 
 #define IPATH_CHIP_SWVERSION IPATH_CHIP_VERS_MAJ
+/*
+ * First-cut critierion for "device is active" is
+ * two thousand dwords combined Tx, Rx traffic per
+ * 5-second interval. SMA packets are 64 dwords,
+ * and occur "a few per second", presumably each way.
+ */
+#define IPATH_TRAFFIC_ACTIVE_THRESHOLD (2000)
+/*
+ * Struct used to indicate which errors are logged in each of the
+ * error-counters that are logged to EEPROM. A counter is incremented
+ * _once_ (saturating at 255) for each event with any bits set in
+ * the error or hwerror register masks below.
+ */
+#define IPATH_EEP_LOG_CNT (4)
+struct ipath_eep_log_mask {
+	u64 errs_to_log;
+	u64 hwerrs_to_log;
+};
 
 struct ipath_portdata {
 	void **port_rcvegrbuf;
@@ -109,6 +127,8 @@ struct ipath_portdata {
 	u32 port_tidcursor;
 	/* next expected TID to check */
 	unsigned long port_flag;
+	/* what happened */
+	unsigned long int_flag;
 	/* WAIT_RCV that timed out, no interrupt */
 	u32 port_rcvwait_to;
 	/* WAIT_PIO that timed out, no interrupt */
@@ -137,6 +157,8 @@ struct ipath_portdata {
 	u32 userversion;
 	/* Bitmask of active slaves */
 	u32 active_slaves;
+	/* Type of packets or conditions we want to poll for */
+	u16 poll_type;
 };
 
 struct sk_buff;
@@ -275,6 +297,8 @@ struct ipath_devdata {
 	u32 ipath_lastport_piobuf;
 	/* is a stats timer active */
 	u32 ipath_stats_timer_active;
+	/* number of interrupts for this device -- saturates... */
+	u32 ipath_int_counter;
 	/* dwords sent read from counter */
 	u32 ipath_lastsword;
 	/* dwords received read from counter */
@@ -369,9 +393,6 @@ struct ipath_devdata {
 	struct class_device *diag_class_dev;
 	/* timer used to prevent stats overflow, error throttling, etc. */
 	struct timer_list ipath_stats_timer;
-	/* check for stale messages in rcv queue */
-	/* only allow one intr at a time. */
-	unsigned long ipath_rcv_pending;
 	void *ipath_dummy_hdrq;	/* used after port close */
 	dma_addr_t ipath_dummy_hdrq_phys;
 
@@ -399,6 +420,8 @@ struct ipath_devdata {
 	u64 ipath_gpio_out;
 	/* shadow the gpio mask register */
 	u64 ipath_gpio_mask;
+	/* shadow the gpio output enable, etc... */
+	u64 ipath_extctrl;
 	/* kr_revision shadow */
 	u64 ipath_revision;
 	/*
@@ -473,8 +496,6 @@ struct ipath_devdata {
 	u32 ipath_cregbase;
 	/* shadow the control register contents */
 	u32 ipath_control;
-	/* shadow the gpio output contents */
-	u32 ipath_extctrl;
 	/* PCI revision register (HTC rev on FPGA) */
 	u32 ipath_pcirev;
 
@@ -552,6 +573,9 @@ struct ipath_devdata {
 	u32 ipath_overrun_thresh_errs;
 	u32 ipath_lli_errs;
 
+	/* status check work */
+	struct delayed_work status_work;
+
 	/*
 	 * Not all devices managed by a driver instance are the same
 	 * type, so these fields must be per-device.
@@ -575,6 +599,37 @@ struct ipath_devdata {
 	u16 ipath_gpio_scl_num;
 	u64 ipath_gpio_sda;
 	u64 ipath_gpio_scl;
+
+	/* lock for doing RMW of shadows/regs for ExtCtrl and GPIO */
+	spinlock_t ipath_gpio_lock;
+
+	/* used to override LED behavior */
+	u8 ipath_led_override;  /* Substituted for normal value, if non-zero */
+	u16 ipath_led_override_timeoff; /* delta to next timer event */
+	u8 ipath_led_override_vals[2]; /* Alternates per blink-frame */
+	u8 ipath_led_override_phase; /* Just counts, LSB picks from vals[] */
+	atomic_t ipath_led_override_timer_active;
+	/* Used to flash LEDs in override mode */
+	struct timer_list ipath_led_override_timer;
+
+	/* Support (including locks) for EEPROM logging of errors and time */
+	/* control access to actual counters, timer */
+	spinlock_t ipath_eep_st_lock;
+	/* control high-level access to EEPROM */
+	struct semaphore ipath_eep_sem;
+	/* Below inc'd by ipath_snap_cntrs(), locked by ipath_eep_st_lock */
+	uint64_t ipath_traffic_wds;
+	/* active time is kept in seconds, but logged in hours */
+	atomic_t ipath_active_time;
+	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
+	uint8_t ipath_eep_st_errs[IPATH_EEP_LOG_CNT];
+	uint8_t ipath_eep_st_new_errs[IPATH_EEP_LOG_CNT];
+	uint16_t ipath_eep_hrs;
+	/*
+	 * masks for which bits of errs, hwerrs that cause
+	 * each of the counters to increment.
+	 */
+	struct ipath_eep_log_mask ipath_eep_st_masks[IPATH_EEP_LOG_CNT];
 };
 
 /* Private data for file operations */
@@ -592,6 +647,7 @@ int ipath_enable_wc(struct ipath_devdata *dd);
 void ipath_disable_wc(struct ipath_devdata *dd);
 int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp);
 void ipath_shutdown_device(struct ipath_devdata *);
+void ipath_clear_freeze(struct ipath_devdata *);
 
 struct file_operations;
 int ipath_cdev_init(int minor, char *name, const struct file_operations *fops,
@@ -627,6 +683,7 @@ int ipath_unordered_wc(void);
 
 void ipath_disarm_piobufs(struct ipath_devdata *, unsigned first,
 			  unsigned cnt);
+void ipath_cancel_sends(struct ipath_devdata *);
 
 int ipath_create_rcvhdrq(struct ipath_devdata *, struct ipath_portdata *);
 void ipath_free_pddata(struct ipath_devdata *, struct ipath_portdata *);
@@ -685,7 +742,6 @@ int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv);
 		 * are 64bit */
 #define IPATH_32BITCOUNTERS 0x20000
 		/* can miss port0 rx interrupts */
-#define IPATH_POLL_RX_INTR  0x40000
 #define IPATH_DISABLED      0x80000 /* administratively disabled */
 		/* Use GPIO interrupts for new counters */
 #define IPATH_GPIO_ERRINTRS 0x100000
@@ -704,6 +760,10 @@ int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv);
 #define IPATH_PORT_WAITING_PIO   3
 		/* master has not finished initializing */
 #define IPATH_PORT_MASTER_UNINIT 4
+		/* waiting for an urgent packet to arrive */
+#define IPATH_PORT_WAITING_URG 5
+		/* waiting for a header overflow */
+#define IPATH_PORT_WAITING_OVERFLOW 6
 
 /* free up any allocated data at closes */
 void ipath_free_data(struct ipath_portdata *dd);
@@ -713,9 +773,20 @@ u32 __iomem *ipath_getpiobuf(struct ipath_devdata *, u32 *);
 void ipath_init_iba6120_funcs(struct ipath_devdata *);
 void ipath_init_iba6110_funcs(struct ipath_devdata *);
 void ipath_get_eeprom_info(struct ipath_devdata *);
+int ipath_update_eeprom_log(struct ipath_devdata *dd);
+void ipath_inc_eeprom_err(struct ipath_devdata *dd, u32 eidx, u32 incr);
 u64 ipath_snap_cntr(struct ipath_devdata *, ipath_creg);
 void ipath_disarm_senderrbufs(struct ipath_devdata *, int);
 
+/*
+ * Set LED override, only the two LSBs have "public" meaning, but
+ * any non-zero value substitutes them for the Link and LinkTrain
+ * LED states.
+ */
+#define IPATH_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
+#define IPATH_LED_LOG 2  /* Logical (link) YELLOW LED */
+void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val);
+
 /*
  * number of words used for protocol header if not set by ipath_userinit();
  */

drivers/infiniband/hw/ipath/ipath_keys.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_layer.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_layer.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_mad.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -103,7 +103,7 @@ static int recv_subn_get_nodeinfo(struct ib_smp *smp,
 	/* This is already in network order */
 	nip->sys_guid = to_idev(ibdev)->sys_image_guid;
 	nip->node_guid = dd->ipath_guid;
-	nip->port_guid = nip->sys_guid;
+	nip->port_guid = dd->ipath_guid;
 	nip->partition_cap = cpu_to_be16(ipath_get_npkeys(dd));
 	nip->device_id = cpu_to_be16(dd->ipath_deviceid);
 	majrev = dd->ipath_majrev;
@@ -292,7 +292,12 @@ static int recv_subn_get_portinfo(struct ib_smp *smp,
 	/* pip->vl_arb_high_cap; // only one VL */
 	/* pip->vl_arb_low_cap; // only one VL */
 	/* InitTypeReply = 0 */
-	pip->inittypereply_mtucap = IB_MTU_4096;
+	/*
+	 * Note: the chips support a maximum MTU of 4096, but the driver
+	 * hasn't implemented this feature yet, so set the maximum value
+	 * to 2048.
+	 */
+	pip->inittypereply_mtucap = IB_MTU_2048;
 	// HCAs ignore VLStallCount and HOQLife
 	/* pip->vlstallcnt_hoqlife; */
 	pip->operationalvl_pei_peo_fpi_fpo = 0x10;	/* OVLs = 1 */

drivers/infiniband/hw/ipath/ipath_mmap.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU

drivers/infiniband/hw/ipath/ipath_mr.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_qp.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -336,7 +336,7 @@ static void ipath_reset_qp(struct ipath_qp *qp)
 	qp->qkey = 0;
 	qp->qp_access_flags = 0;
 	qp->s_busy = 0;
-	qp->s_flags &= ~IPATH_S_SIGNAL_REQ_WR;
+	qp->s_flags &= IPATH_S_SIGNAL_REQ_WR;
 	qp->s_hdrwords = 0;
 	qp->s_psn = 0;
 	qp->r_psn = 0;
@@ -507,16 +507,13 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		    attr->port_num > ibqp->device->phys_port_cnt)
 			goto inval;
 
+	/*
+	 * Note: the chips support a maximum MTU of 4096, but the driver
+	 * hasn't implemented this feature yet, so don't allow Path MTU
+	 * values greater than 2048.
+	 */
 	if (attr_mask & IB_QP_PATH_MTU)
-		if (attr->path_mtu > IB_MTU_4096)
-			goto inval;
-
-	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
-		if (attr->max_dest_rd_atomic > 1)
-			goto inval;
-
-	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
-		if (attr->max_rd_atomic > 1)
+		if (attr->path_mtu > IB_MTU_2048)
 			goto inval;
 
 	if (attr_mask & IB_QP_PATH_MIG_STATE)

drivers/infiniband/hw/ipath/ipath_rc.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -125,8 +125,10 @@ static int ipath_make_rc_ack(struct ipath_qp *qp,
 			if (len > pmtu) {
 				len = pmtu;
 				qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
-			} else
+			} else {
 				qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
+				e->sent = 1;
+			}
 			ohdr->u.aeth = ipath_compute_aeth(qp);
 			hwords++;
 			qp->s_ack_rdma_psn = e->psn;
@@ -143,6 +145,7 @@ static int ipath_make_rc_ack(struct ipath_qp *qp,
 				cpu_to_be32(e->atomic_data);
 			hwords += sizeof(ohdr->u.at) / sizeof(u32);
 			bth2 = e->psn;
+			e->sent = 1;
 		}
 		bth0 = qp->s_ack_state << 24;
 		break;
@@ -158,6 +161,7 @@ static int ipath_make_rc_ack(struct ipath_qp *qp,
 			ohdr->u.aeth = ipath_compute_aeth(qp);
 			hwords++;
 			qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+			qp->s_ack_queue[qp->s_tail_ack_queue].sent = 1;
 		}
 		bth0 = qp->s_ack_state << 24;
 		bth2 = qp->s_ack_rdma_psn++ & IPATH_PSN_MASK;
@@ -188,7 +192,7 @@ static int ipath_make_rc_ack(struct ipath_qp *qp,
 	}
 	qp->s_hdrwords = hwords;
 	qp->s_cur_size = len;
-	*bth0p = bth0;
+	*bth0p = bth0 | (1 << 22); /* Set M bit */
 	*bth2p = bth2;
 	return 1;
 
@@ -240,7 +244,7 @@ int ipath_make_rc_req(struct ipath_qp *qp,
 
 	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
 	hwords = 5;
-	bth0 = 0;
+	bth0 = 1 << 22; /* Set M bit */
 
 	/* Send a request. */
 	wqe = get_swqe_ptr(qp, qp->s_cur);
@@ -604,7 +608,7 @@ static void send_rc_ack(struct ipath_qp *qp)
 	}
 	/* read pkey_index w/o lock (its atomic) */
 	bth0 = ipath_get_pkey(dev->dd, qp->s_pkey_index) |
-		OP(ACKNOWLEDGE) << 24;
+		(OP(ACKNOWLEDGE) << 24) | (1 << 22);
 	if (qp->r_nak_state)
 		ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
 					    (qp->r_nak_state <<
@@ -806,13 +810,15 @@ static inline void update_last_psn(struct ipath_qp *qp, u32 psn)
  * Called at interrupt level with the QP s_lock held and interrupts disabled.
  * Returns 1 if OK, 0 if current operation should be aborted (NAK).
  */
-static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
+static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode,
+		     u64 val)
 {
 	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
 	struct ib_wc wc;
 	struct ipath_swqe *wqe;
 	int ret = 0;
 	u32 ack_psn;
+	int diff;
 
 	/*
 	 * Remove the QP from the timeout queue (or RNR timeout queue).
@@ -840,7 +846,19 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
 	 * The MSN might be for a later WQE than the PSN indicates so
 	 * only complete WQEs that the PSN finishes.
 	 */
-	while (ipath_cmp24(ack_psn, wqe->lpsn) >= 0) {
+	while ((diff = ipath_cmp24(ack_psn, wqe->lpsn)) >= 0) {
+		/*
+		 * RDMA_READ_RESPONSE_ONLY is a special case since
+		 * we want to generate completion events for everything
+		 * before the RDMA read, copy the data, then generate
+		 * the completion for the read.
+		 */
+		if (wqe->wr.opcode == IB_WR_RDMA_READ &&
+		    opcode == OP(RDMA_READ_RESPONSE_ONLY) &&
+		    diff == 0) {
+			ret = 1;
+			goto bail;
+		}
 		/*
 		 * If this request is a RDMA read or atomic, and the ACK is
 		 * for a later operation, this ACK NAKs the RDMA read or
@@ -851,12 +869,10 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
 		 * is sent but before the response is received.
 		 */
 		if ((wqe->wr.opcode == IB_WR_RDMA_READ &&
-		     (opcode != OP(RDMA_READ_RESPONSE_LAST) ||
-		      ipath_cmp24(ack_psn, wqe->lpsn) != 0)) ||
+		     (opcode != OP(RDMA_READ_RESPONSE_LAST) || diff != 0)) ||
 		    ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
 		      wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) &&
-		     (opcode != OP(ATOMIC_ACKNOWLEDGE) ||
-		      ipath_cmp24(wqe->psn, psn) != 0))) {
+		     (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0))) {
 			/*
 			 * The last valid PSN seen is the previous
 			 * request's.
@@ -870,6 +886,9 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
 			 */
 			goto bail;
 		}
+		if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+			*(u64 *) wqe->sg_list[0].vaddr = val;
 		if (qp->s_num_rd_atomic &&
 		    (wqe->wr.opcode == IB_WR_RDMA_READ ||
 		     wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
@@ -1079,6 +1098,7 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 	int diff;
 	u32 pad;
 	u32 aeth;
+	u64 val;
 
 	spin_lock_irqsave(&qp->s_lock, flags);
 
@@ -1118,8 +1138,6 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 			data += sizeof(__be32);
 		}
 		if (opcode == OP(ATOMIC_ACKNOWLEDGE)) {
-			u64 val;
-
 			if (!header_in_data) {
 				__be32 *p = ohdr->u.at.atomic_ack_eth;
 
@@ -1127,12 +1145,13 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 					be32_to_cpu(p[1]);
 			} else
 				val = be64_to_cpu(((__be64 *) data)[0]);
-			*(u64 *) wqe->sg_list[0].vaddr = val;
-		}
-		if (!do_rc_ack(qp, aeth, psn, opcode) ||
+		} else
+			val = 0;
+		if (!do_rc_ack(qp, aeth, psn, opcode, val) ||
 		    opcode != OP(RDMA_READ_RESPONSE_FIRST))
 			goto ack_done;
 		hdrsize += 4;
+		wqe = get_swqe_ptr(qp, qp->s_last);
 		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
 			goto ack_op_err;
 		/*
@@ -1176,13 +1195,12 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 		goto bail;
 
 	case OP(RDMA_READ_RESPONSE_ONLY):
-		if (unlikely(ipath_cmp24(psn, qp->s_last_psn + 1))) {
-			dev->n_rdma_seq++;
-			ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
+		if (!header_in_data)
+			aeth = be32_to_cpu(ohdr->u.aeth);
+		else
+			aeth = be32_to_cpu(((__be32 *) data)[0]);
+		if (!do_rc_ack(qp, aeth, psn, opcode, 0))
 			goto ack_done;
-		}
-		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
-			goto ack_op_err;
 		/* Get the number of bytes the message was padded by. */
 		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
 		/*
@@ -1197,6 +1215,7 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 		 * have to be careful to copy the data to the right
 		 * location.
 		 */
+		wqe = get_swqe_ptr(qp, qp->s_last);
 		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
 						  wqe, psn, pmtu);
 		goto read_last;
@@ -1230,7 +1249,8 @@ static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
 			data += sizeof(__be32);
 		}
 		ipath_copy_sge(&qp->s_rdma_read_sge, data, tlen);
-		(void) do_rc_ack(qp, aeth, psn, OP(RDMA_READ_RESPONSE_LAST));
+		(void) do_rc_ack(qp, aeth, psn,
+				 OP(RDMA_READ_RESPONSE_LAST), 0);
 		goto ack_done;
 	}
 
@@ -1344,8 +1364,11 @@ static inline int ipath_rc_rcv_error(struct ipath_ibdev *dev,
 			e = NULL;
 			break;
 		}
-		if (ipath_cmp24(psn, e->psn) >= 0)
+		if (ipath_cmp24(psn, e->psn) >= 0) {
+			if (prev == qp->s_tail_ack_queue)
+				old_req = 0;
 			break;
+		}
 	}
 	switch (opcode) {
 	case OP(RDMA_READ_REQUEST): {
@@ -1460,6 +1483,22 @@ static void ipath_rc_error(struct ipath_qp *qp, enum ib_wc_status err)
 	spin_unlock_irqrestore(&qp->s_lock, flags);
 }
 
+static inline void ipath_update_ack_queue(struct ipath_qp *qp, unsigned n)
+{
+	unsigned long flags;
+	unsigned next;
+
+	next = n + 1;
+	if (next > IPATH_MAX_RDMA_ATOMIC)
+		next = 0;
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (n == qp->s_tail_ack_queue) {
+		qp->s_tail_ack_queue = next;
+		qp->s_ack_state = OP(ACKNOWLEDGE);
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
 /**
  * ipath_rc_rcv - process an incoming RC packet
  * @dev: the device this packet came in on
@@ -1672,6 +1711,9 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 	case OP(RDMA_WRITE_FIRST):
 	case OP(RDMA_WRITE_ONLY):
 	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE)))
+			goto nack_inv;
 		/* consume RWQE */
 		/* RETH comes after BTH */
 		if (!header_in_data)
@@ -1701,9 +1743,6 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 			qp->r_sge.sge.length = 0;
 			qp->r_sge.sge.sge_length = 0;
 		}
-		if (unlikely(!(qp->qp_access_flags &
-			       IB_ACCESS_REMOTE_WRITE)))
-			goto nack_acc;
 		if (opcode == OP(RDMA_WRITE_FIRST))
 			goto send_middle;
 		else if (opcode == OP(RDMA_WRITE_ONLY))
@@ -1717,13 +1756,17 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		u32 len;
 		u8 next;
 
-		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
-			goto nack_acc;
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_READ)))
+			goto nack_inv;
 		next = qp->r_head_ack_queue + 1;
 		if (next > IPATH_MAX_RDMA_ATOMIC)
 			next = 0;
-		if (unlikely(next == qp->s_tail_ack_queue))
-			goto nack_inv;
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv;
+			ipath_update_ack_queue(qp, next);
+		}
 		e = &qp->s_ack_queue[qp->r_head_ack_queue];
 		/* RETH comes after BTH */
 		if (!header_in_data)
@@ -1758,6 +1801,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 			e->rdma_sge.sge.sge_length = 0;
 		}
 		e->opcode = opcode;
+		e->sent = 0;
 		e->psn = psn;
 		/*
 		 * We need to increment the MSN here instead of when we
@@ -1789,12 +1833,15 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 
 		if (unlikely(!(qp->qp_access_flags &
 			       IB_ACCESS_REMOTE_ATOMIC)))
-			goto nack_acc;
+			goto nack_inv;
 		next = qp->r_head_ack_queue + 1;
 		if (next > IPATH_MAX_RDMA_ATOMIC)
 			next = 0;
-		if (unlikely(next == qp->s_tail_ack_queue))
-			goto nack_inv;
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv;
+			ipath_update_ack_queue(qp, next);
+		}
 		if (!header_in_data)
 			ateth = &ohdr->u.atomic_eth;
 		else
@@ -1819,6 +1866,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 				      be64_to_cpu(ateth->compare_data),
 				      sdata);
 		e->opcode = opcode;
+		e->sent = 0;
 		e->psn = psn & IPATH_PSN_MASK;
 		qp->r_msn++;
 		qp->r_psn++;

drivers/infiniband/hw/ipath/ipath_registers.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_ruc.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -194,6 +194,8 @@ int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
 			ret = 0;
 			goto bail;
 		}
+		/* Make sure entry is read after head index is read. */
+		smp_rmb();
 		wqe = get_rwqe_ptr(rq, tail);
 		if (++tail >= rq->size)
 			tail = 0;
@@ -267,7 +269,7 @@ again:
 	spin_lock_irqsave(&sqp->s_lock, flags);
 
 	if (!(ib_ipath_state_ops[sqp->state] & IPATH_PROCESS_SEND_OK) ||
-	    qp->s_rnr_timeout) {
+	    sqp->s_rnr_timeout) {
 		spin_unlock_irqrestore(&sqp->s_lock, flags);
 		goto done;
 	}
@@ -319,12 +321,22 @@ again:
 		break;
 
 	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE))) {
+			wc.status = IB_WC_REM_INV_REQ_ERR;
+			goto err;
+		}
 		wc.wc_flags = IB_WC_WITH_IMM;
 		wc.imm_data = wqe->wr.imm_data;
 		if (!ipath_get_rwqe(qp, 1))
 			goto rnr_nak;
 		/* FALLTHROUGH */
 	case IB_WR_RDMA_WRITE:
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE))) {
+			wc.status = IB_WC_REM_INV_REQ_ERR;
+			goto err;
+		}
 		if (wqe->length == 0)
 			break;
 		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, wqe->length,
@@ -354,8 +366,10 @@ again:
 
 	case IB_WR_RDMA_READ:
 		if (unlikely(!(qp->qp_access_flags &
-			       IB_ACCESS_REMOTE_READ)))
-			goto acc_err;
+			       IB_ACCESS_REMOTE_READ))) {
+			wc.status = IB_WC_REM_INV_REQ_ERR;
+			goto err;
+		}
 		if (unlikely(!ipath_rkey_ok(qp, &sqp->s_sge, wqe->length,
 					    wqe->wr.wr.rdma.remote_addr,
 					    wqe->wr.wr.rdma.rkey,
@@ -369,8 +383,10 @@ again:
 	case IB_WR_ATOMIC_CMP_AND_SWP:
 	case IB_WR_ATOMIC_FETCH_AND_ADD:
 		if (unlikely(!(qp->qp_access_flags &
-			       IB_ACCESS_REMOTE_ATOMIC)))
-			goto acc_err;
+			       IB_ACCESS_REMOTE_ATOMIC))) {
+			wc.status = IB_WC_REM_INV_REQ_ERR;
+			goto err;
+		}
 		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, sizeof(u64),
 					    wqe->wr.wr.atomic.remote_addr,
 					    wqe->wr.wr.atomic.rkey,
@@ -396,6 +412,8 @@ again:
 
 		if (len > sge->length)
 			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
 		BUG_ON(len == 0);
 		ipath_copy_sge(&qp->r_sge, sge->vaddr, len);
 		sge->vaddr += len;
@@ -503,11 +521,9 @@ void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev)
 	 * could be called.  If we are still in the tasklet function,
 	 * tasklet_hi_schedule() will not call us until the next time
 	 * tasklet_hi_schedule() is called.
-	 * We clear the tasklet flag now since we are committing to return
-	 * from the tasklet function.
+	 * We leave the busy flag set so that another post send doesn't
+	 * try to put the same QP on the piowait list again.
 	 */
-	clear_bit(IPATH_S_BUSY, &qp->s_busy);
-	tasklet_unlock(&qp->s_task);
 	want_buffer(dev->dd);
 	dev->n_piowait++;
 }

drivers/infiniband/hw/ipath/ipath_srq.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -80,6 +80,8 @@ int ipath_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
 		wqe->num_sge = wr->num_sge;
 		for (i = 0; i < wr->num_sge; i++)
 			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
 		wq->head = next;
 		spin_unlock_irqrestore(&srq->rq.lock, flags);
 	}

drivers/infiniband/hw/ipath/ipath_stats.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -55,6 +55,7 @@ u64 ipath_snap_cntr(struct ipath_devdata *dd, ipath_creg creg)
 	u64 val64;
 	unsigned long t0, t1;
 	u64 ret;
+	unsigned long flags;
 
 	t0 = jiffies;
 	/* If fast increment counters are only 32 bits, snapshot them,
@@ -91,12 +92,18 @@ u64 ipath_snap_cntr(struct ipath_devdata *dd, ipath_creg creg)
 	if (creg == dd->ipath_cregs->cr_wordsendcnt) {
 		if (val != dd->ipath_lastsword) {
 			dd->ipath_sword += val - dd->ipath_lastsword;
+			spin_lock_irqsave(&dd->ipath_eep_st_lock, flags);
+			dd->ipath_traffic_wds += val - dd->ipath_lastsword;
+			spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags);
 			dd->ipath_lastsword = val;
 		}
 		val64 = dd->ipath_sword;
 	} else if (creg == dd->ipath_cregs->cr_wordrcvcnt) {
 		if (val != dd->ipath_lastrword) {
 			dd->ipath_rword += val - dd->ipath_lastrword;
+			spin_lock_irqsave(&dd->ipath_eep_st_lock, flags);
+			dd->ipath_traffic_wds += val - dd->ipath_lastrword;
+			spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags);
 			dd->ipath_lastrword = val;
 		}
 		val64 = dd->ipath_rword;
@@ -200,6 +207,7 @@ void ipath_get_faststats(unsigned long opaque)
 	struct ipath_devdata *dd = (struct ipath_devdata *) opaque;
 	u32 val;
 	static unsigned cnt;
+	unsigned long flags;
 
 	/*
 	 * don't access the chip while running diags, or memory diags can
@@ -210,9 +218,20 @@ void ipath_get_faststats(unsigned long opaque)
 		/* but re-arm the timer, for diags case; won't hurt other */
 		goto done;
 
+	/*
+	 * We now try to maintain a "active timer", based on traffic
+	 * exceeding a threshold, so we need to check the word-counts
+	 * even if they are 64-bit.
+	 */
+	ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
+	ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
+	spin_lock_irqsave(&dd->ipath_eep_st_lock, flags);
+	if (dd->ipath_traffic_wds  >= IPATH_TRAFFIC_ACTIVE_THRESHOLD)
+		atomic_add(5, &dd->ipath_active_time); /* S/B #define */
+	dd->ipath_traffic_wds = 0;
+	spin_unlock_irqrestore(&dd->ipath_eep_st_lock, flags);
+
 	if (dd->ipath_flags & IPATH_32BITCOUNTERS) {
-		ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
-		ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
 		ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
 		ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
 	}

drivers/infiniband/hw/ipath/ipath_sysfs.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -596,6 +596,43 @@ bail:
 	return ret;
 }
 
+static ssize_t store_led_override(struct device *dev,
+			  struct device_attribute *attr,
+			  const char *buf,
+			  size_t count)
+{
+	struct ipath_devdata *dd = dev_get_drvdata(dev);
+	int ret;
+	u16 val;
+
+	ret = ipath_parse_ushort(buf, &val);
+	if (ret > 0)
+		ipath_set_led_override(dd, val);
+	else
+		ipath_dev_err(dd, "attempt to set invalid LED override\n");
+	return ret;
+}
+
+static ssize_t show_logged_errs(struct device *dev,
+				struct device_attribute *attr,
+				char *buf)
+{
+	struct ipath_devdata *dd = dev_get_drvdata(dev);
+	int idx, count;
+
+	/* force consistency with actual EEPROM */
+	if (ipath_update_eeprom_log(dd) != 0)
+		return -ENXIO;
+
+	count = 0;
+	for (idx = 0; idx < IPATH_EEP_LOG_CNT; ++idx) {
+		count += scnprintf(buf + count, PAGE_SIZE - count, "%d%c",
+			dd->ipath_eep_st_errs[idx],
+			idx == (IPATH_EEP_LOG_CNT - 1) ? '\n' : ' ');
+	}
+
+	return count;
+}
 
 static DRIVER_ATTR(num_units, S_IRUGO, show_num_units, NULL);
 static DRIVER_ATTR(version, S_IRUGO, show_version, NULL);
@@ -625,6 +662,8 @@ static DEVICE_ATTR(status_str, S_IRUGO, show_status_str, NULL);
 static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL);
 static DEVICE_ATTR(unit, S_IRUGO, show_unit, NULL);
 static DEVICE_ATTR(rx_pol_inv, S_IWUSR, NULL, store_rx_pol_inv);
+static DEVICE_ATTR(led_override, S_IWUSR, NULL, store_led_override);
+static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL);
 
 static struct attribute *dev_attributes[] = {
 	&dev_attr_guid.attr,
@@ -641,6 +680,8 @@ static struct attribute *dev_attributes[] = {
 	&dev_attr_unit.attr,
 	&dev_attr_enabled.attr,
 	&dev_attr_rx_pol_inv.attr,
+	&dev_attr_led_override.attr,
+	&dev_attr_logged_errors.attr,
 	NULL
 };
 

drivers/infiniband/hw/ipath/ipath_uc.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -58,7 +58,6 @@ static void complete_last_send(struct ipath_qp *qp, struct ipath_swqe *wqe,
 		wc->port_num = 0;
 		ipath_cq_enter(to_icq(qp->ibqp.send_cq), wc, 0);
 	}
-	wqe = get_swqe_ptr(qp, qp->s_last);
 }
 
 /**
@@ -87,7 +86,7 @@ int ipath_make_uc_req(struct ipath_qp *qp,
 
 	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
 	hwords = 5;
-	bth0 = 0;
+	bth0 = 1 << 22; /* Set M bit */
 
 	/* Get the next send request. */
 	wqe = get_swqe_ptr(qp, qp->s_last);
@@ -97,8 +96,10 @@ int ipath_make_uc_req(struct ipath_qp *qp,
 		 * Signal the completion of the last send
 		 * (if there is one).
 		 */
-		if (qp->s_last != qp->s_tail)
+		if (qp->s_last != qp->s_tail) {
 			complete_last_send(qp, wqe, &wc);
+			wqe = get_swqe_ptr(qp, qp->s_last);
+		}
 
 		/* Check if send work queue is empty. */
 		if (qp->s_tail == qp->s_head)

drivers/infiniband/hw/ipath/ipath_ud.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -176,6 +176,8 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 			dev->n_pkt_drops++;
 			goto bail_sge;
 		}
+		/* Make sure entry is read after head index is read. */
+		smp_rmb();
 		wqe = get_rwqe_ptr(rq, tail);
 		if (++tail >= rq->size)
 			tail = 0;
@@ -231,6 +233,8 @@ static void ipath_ud_loopback(struct ipath_qp *sqp,
 
 		if (len > length)
 			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
 		BUG_ON(len == 0);
 		ipath_copy_sge(&rsge, sge->vaddr, len);
 		sge->vaddr += len;

drivers/infiniband/hw/ipath/ipath_user_pages.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_verbs.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -164,9 +164,11 @@ void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length)
 	while (length) {
 		u32 len = sge->length;
 
-		BUG_ON(len == 0);
 		if (len > length)
 			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
 		memcpy(sge->vaddr, data, len);
 		sge->vaddr += len;
 		sge->length -= len;
@@ -202,9 +204,11 @@ void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
 	while (length) {
 		u32 len = sge->length;
 
-		BUG_ON(len == 0);
 		if (len > length)
 			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
 		sge->vaddr += len;
 		sge->length -= len;
 		sge->sge_length -= len;
@@ -323,6 +327,8 @@ static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 		wqe->num_sge = wr->num_sge;
 		for (i = 0; i < wr->num_sge; i++)
 			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
 		wq->head = next;
 		spin_unlock_irqrestore(&qp->r_rq.lock, flags);
 	}
@@ -948,6 +954,7 @@ int ipath_ib_piobufavail(struct ipath_ibdev *dev)
 		qp = list_entry(dev->piowait.next, struct ipath_qp,
 				piowait);
 		list_del_init(&qp->piowait);
+		clear_bit(IPATH_S_BUSY, &qp->s_busy);
 		tasklet_hi_schedule(&qp->s_task);
 	}
 	spin_unlock_irqrestore(&dev->pending_lock, flags);
@@ -981,6 +988,8 @@ static int ipath_query_device(struct ib_device *ibdev,
 	props->max_ah = ib_ipath_max_ahs;
 	props->max_cqe = ib_ipath_max_cqes;
 	props->max_mr = dev->lk_table.max;
+	props->max_fmr = dev->lk_table.max;
+	props->max_map_per_fmr = 32767;
 	props->max_pd = ib_ipath_max_pds;
 	props->max_qp_rd_atom = IPATH_MAX_RDMA_ATOMIC;
 	props->max_qp_init_rd_atom = 255;
@@ -1051,7 +1060,12 @@ static int ipath_query_port(struct ib_device *ibdev,
 	props->max_vl_num = 1;		/* VLCap = VL0 */
 	props->init_type_reply = 0;
 
-	props->max_mtu = IB_MTU_4096;
+	/*
+	 * Note: the chips support a maximum MTU of 4096, but the driver
+	 * hasn't implemented this feature yet, so set the maximum value
+	 * to 2048.
+	 */
+	props->max_mtu = IB_MTU_2048;
 	switch (dev->dd->ipath_ibmtu) {
 	case 4096:
 		mtu = IB_MTU_4096;
@@ -1361,13 +1375,6 @@ static void __verbs_timer(unsigned long arg)
 {
 	struct ipath_devdata *dd = (struct ipath_devdata *) arg;
 
-	/*
-	 * If port 0 receive packet interrupts are not available, or
-	 * can be missed, poll the receive queue
-	 */
-	if (dd->ipath_flags & IPATH_POLL_RX_INTR)
-		ipath_kreceive(dd);
-
 	/* Handle verbs layer timeouts. */
 	ipath_ib_timer(dd->verbs_dev);
 

drivers/infiniband/hw/ipath/ipath_verbs.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -321,6 +321,7 @@ struct ipath_sge_state {
  */
 struct ipath_ack_entry {
 	u8 opcode;
+	u8 sent;
 	u32 psn;
 	union {
 		struct ipath_sge_state rdma_sge;

drivers/infiniband/hw/ipath/ipath_verbs_mcast.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two

drivers/infiniband/hw/ipath/ipath_wc_ppc64.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU

drivers/infiniband/hw/ipath/ipath_wc_x86_64.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 QLogic, Inc. All rights reserved.
+ * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -63,12 +63,29 @@ int ipath_enable_wc(struct ipath_devdata *dd)
 	 * of 2 address matching the length (which has to be a power of 2).
 	 * For rev1, that means the base address, for rev2, it will be just
 	 * the PIO buffers themselves.
+	 * For chips with two sets of buffers, the calculations are
+	 * somewhat more complicated; we need to sum, and the piobufbase
+	 * register has both offsets, 2K in low 32 bits, 4K in high 32 bits.
+	 * The buffers are still packed, so a single range covers both.
 	 */
-	pioaddr = addr + dd->ipath_piobufbase;
-	piolen = (dd->ipath_piobcnt2k +
-		  dd->ipath_piobcnt4k) *
-		ALIGN(dd->ipath_piobcnt2k +
-		      dd->ipath_piobcnt4k, dd->ipath_palign);
+	if (dd->ipath_piobcnt2k && dd->ipath_piobcnt4k) { /* 2 sizes */
+		unsigned long pio2kbase, pio4kbase;
+		pio2kbase = dd->ipath_piobufbase & 0xffffffffUL;
+		pio4kbase = (dd->ipath_piobufbase >> 32) & 0xffffffffUL;
+		if (pio2kbase < pio4kbase) { /* all, for now */
+			pioaddr = addr + pio2kbase;
+			piolen = pio4kbase - pio2kbase +
+				dd->ipath_piobcnt4k * dd->ipath_4kalign;
+		} else {
+			pioaddr = addr + pio4kbase;
+			piolen = pio2kbase - pio4kbase +
+				dd->ipath_piobcnt2k * dd->ipath_palign;
+		}
+	} else {  /* single buffer size (2K, currently) */
+		pioaddr = addr + dd->ipath_piobufbase;
+		piolen = dd->ipath_piobcnt2k * dd->ipath_palign +
+			dd->ipath_piobcnt4k * dd->ipath_4kalign;
+	}
 
 	for (bits = 0; !(piolen & (1ULL << bits)); bits++)
 		/* do nothing */ ;

drivers/infiniband/hw/mlx4/Kconfig

@@ -1,6 +1,5 @@
 config MLX4_INFINIBAND
 	tristate "Mellanox ConnectX HCA support"
-	depends on INFINIBAND
 	select MLX4_CORE
 	---help---
 	  This driver provides low-level InfiniBand support for

drivers/infiniband/hw/mlx4/main.c

@@ -169,7 +169,7 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 	props->phys_state	= out_mad->data[33] >> 4;
 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
 	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
-	props->max_msg_sz	= 0x80000000;
+	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
@@ -523,11 +523,13 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
 
 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
@@ -546,10 +548,12 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
+	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
+	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;

drivers/infiniband/hw/mlx4/mlx4_ib.h

@@ -35,6 +35,7 @@
 
 #include <linux/compiler.h>
 #include <linux/list.h>
+#include <linux/mutex.h>
 
 #include <rdma/ib_verbs.h>
 #include <rdma/ib_umem.h>
@@ -255,6 +256,7 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 				  struct ib_udata *udata);
 int mlx4_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
+int mlx4_ib_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
 int mlx4_ib_destroy_srq(struct ib_srq *srq);
 void mlx4_ib_free_srq_wqe(struct mlx4_ib_srq *srq, int wqe_index);
 int mlx4_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
@@ -266,6 +268,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 int mlx4_ib_destroy_qp(struct ib_qp *qp);
 int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		      int attr_mask, struct ib_udata *udata);
+int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr);
 int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		      struct ib_send_wr **bad_wr);
 int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,

drivers/infiniband/hw/mlx4/qp.c

@@ -1455,3 +1455,140 @@ out:
 
 	return err;
 }
+
+static inline enum ib_qp_state to_ib_qp_state(enum mlx4_qp_state mlx4_state)
+{
+	switch (mlx4_state) {
+	case MLX4_QP_STATE_RST:      return IB_QPS_RESET;
+	case MLX4_QP_STATE_INIT:     return IB_QPS_INIT;
+	case MLX4_QP_STATE_RTR:      return IB_QPS_RTR;
+	case MLX4_QP_STATE_RTS:      return IB_QPS_RTS;
+	case MLX4_QP_STATE_SQ_DRAINING:
+	case MLX4_QP_STATE_SQD:      return IB_QPS_SQD;
+	case MLX4_QP_STATE_SQER:     return IB_QPS_SQE;
+	case MLX4_QP_STATE_ERR:      return IB_QPS_ERR;
+	default:		     return -1;
+	}
+}
+
+static inline enum ib_mig_state to_ib_mig_state(int mlx4_mig_state)
+{
+	switch (mlx4_mig_state) {
+	case MLX4_QP_PM_ARMED:		return IB_MIG_ARMED;
+	case MLX4_QP_PM_REARM:		return IB_MIG_REARM;
+	case MLX4_QP_PM_MIGRATED:	return IB_MIG_MIGRATED;
+	default: return -1;
+	}
+}
+
+static int to_ib_qp_access_flags(int mlx4_flags)
+{
+	int ib_flags = 0;
+
+	if (mlx4_flags & MLX4_QP_BIT_RRE)
+		ib_flags |= IB_ACCESS_REMOTE_READ;
+	if (mlx4_flags & MLX4_QP_BIT_RWE)
+		ib_flags |= IB_ACCESS_REMOTE_WRITE;
+	if (mlx4_flags & MLX4_QP_BIT_RAE)
+		ib_flags |= IB_ACCESS_REMOTE_ATOMIC;
+
+	return ib_flags;
+}
+
+static void to_ib_ah_attr(struct mlx4_dev *dev, struct ib_ah_attr *ib_ah_attr,
+				struct mlx4_qp_path *path)
+{
+	memset(ib_ah_attr, 0, sizeof *path);
+	ib_ah_attr->port_num	  = path->sched_queue & 0x40 ? 2 : 1;
+
+	if (ib_ah_attr->port_num == 0 || ib_ah_attr->port_num > dev->caps.num_ports)
+		return;
+
+	ib_ah_attr->dlid	  = be16_to_cpu(path->rlid);
+	ib_ah_attr->sl		  = (path->sched_queue >> 2) & 0xf;
+	ib_ah_attr->src_path_bits = path->grh_mylmc & 0x7f;
+	ib_ah_attr->static_rate   = path->static_rate ? path->static_rate - 5 : 0;
+	ib_ah_attr->ah_flags      = (path->grh_mylmc & (1 << 7)) ? IB_AH_GRH : 0;
+	if (ib_ah_attr->ah_flags) {
+		ib_ah_attr->grh.sgid_index = path->mgid_index;
+		ib_ah_attr->grh.hop_limit  = path->hop_limit;
+		ib_ah_attr->grh.traffic_class =
+			(be32_to_cpu(path->tclass_flowlabel) >> 20) & 0xff;
+		ib_ah_attr->grh.flow_label =
+			be32_to_cpu(path->tclass_flowlabel) & 0xffffff;
+		memcpy(ib_ah_attr->grh.dgid.raw,
+			path->rgid, sizeof ib_ah_attr->grh.dgid.raw);
+	}
+}
+
+int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx4_ib_qp *qp = to_mqp(ibqp);
+	struct mlx4_qp_context context;
+	int mlx4_state;
+	int err;
+
+	if (qp->state == IB_QPS_RESET) {
+		qp_attr->qp_state = IB_QPS_RESET;
+		goto done;
+	}
+
+	err = mlx4_qp_query(dev->dev, &qp->mqp, &context);
+	if (err)
+		return -EINVAL;
+
+	mlx4_state = be32_to_cpu(context.flags) >> 28;
+
+	qp_attr->qp_state	     = to_ib_qp_state(mlx4_state);
+	qp_attr->path_mtu	     = context.mtu_msgmax >> 5;
+	qp_attr->path_mig_state	     =
+		to_ib_mig_state((be32_to_cpu(context.flags) >> 11) & 0x3);
+	qp_attr->qkey		     = be32_to_cpu(context.qkey);
+	qp_attr->rq_psn		     = be32_to_cpu(context.rnr_nextrecvpsn) & 0xffffff;
+	qp_attr->sq_psn		     = be32_to_cpu(context.next_send_psn) & 0xffffff;
+	qp_attr->dest_qp_num	     = be32_to_cpu(context.remote_qpn) & 0xffffff;
+	qp_attr->qp_access_flags     =
+		to_ib_qp_access_flags(be32_to_cpu(context.params2));
+
+	if (qp->ibqp.qp_type == IB_QPT_RC || qp->ibqp.qp_type == IB_QPT_UC) {
+		to_ib_ah_attr(dev->dev, &qp_attr->ah_attr, &context.pri_path);
+		to_ib_ah_attr(dev->dev, &qp_attr->alt_ah_attr, &context.alt_path);
+		qp_attr->alt_pkey_index = context.alt_path.pkey_index & 0x7f;
+		qp_attr->alt_port_num	= qp_attr->alt_ah_attr.port_num;
+	}
+
+	qp_attr->pkey_index = context.pri_path.pkey_index & 0x7f;
+	qp_attr->port_num   = context.pri_path.sched_queue & 0x40 ? 2 : 1;
+
+	/* qp_attr->en_sqd_async_notify is only applicable in modify qp */
+	qp_attr->sq_draining = mlx4_state == MLX4_QP_STATE_SQ_DRAINING;
+
+	qp_attr->max_rd_atomic = 1 << ((be32_to_cpu(context.params1) >> 21) & 0x7);
+
+	qp_attr->max_dest_rd_atomic =
+		1 << ((be32_to_cpu(context.params2) >> 21) & 0x7);
+	qp_attr->min_rnr_timer	    =
+		(be32_to_cpu(context.rnr_nextrecvpsn) >> 24) & 0x1f;
+	qp_attr->timeout	    = context.pri_path.ackto >> 3;
+	qp_attr->retry_cnt	    = (be32_to_cpu(context.params1) >> 16) & 0x7;
+	qp_attr->rnr_retry	    = (be32_to_cpu(context.params1) >> 13) & 0x7;
+	qp_attr->alt_timeout	    = context.alt_path.ackto >> 3;
+
+done:
+	qp_attr->cur_qp_state	     = qp_attr->qp_state;
+	if (!ibqp->uobject) {
+		qp_attr->cap.max_send_wr     = qp->sq.wqe_cnt;
+		qp_attr->cap.max_recv_wr     = qp->rq.wqe_cnt;
+		qp_attr->cap.max_send_sge    = qp->sq.max_gs;
+		qp_attr->cap.max_recv_sge    = qp->rq.max_gs;
+		qp_attr->cap.max_inline_data = (1 << qp->sq.wqe_shift) -
+			send_wqe_overhead(qp->ibqp.qp_type) -
+			sizeof (struct mlx4_wqe_inline_seg);
+		qp_init_attr->cap	     = qp_attr->cap;
+	}
+
+	return 0;
+}
+

drivers/infiniband/hw/mlx4/srq.c

@@ -240,6 +240,24 @@ int mlx4_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 	return 0;
 }
 
+int mlx4_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
+	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
+	int ret;
+	int limit_watermark;
+
+	ret = mlx4_srq_query(dev->dev, &srq->msrq, &limit_watermark);
+	if (ret)
+		return ret;
+
+	srq_attr->srq_limit = be16_to_cpu(limit_watermark);
+	srq_attr->max_wr    = srq->msrq.max - 1;
+	srq_attr->max_sge   = srq->msrq.max_gs;
+
+	return 0;
+}
+
 int mlx4_ib_destroy_srq(struct ib_srq *srq)
 {
 	struct mlx4_ib_dev *dev = to_mdev(srq->device);

drivers/infiniband/hw/mthca/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_MTHCA
 	tristate "Mellanox HCA support"
-	depends on PCI && INFINIBAND
+	depends on PCI
 	---help---
 	  This is a low-level driver for Mellanox InfiniHost host
 	  channel adapters (HCAs), including the MT23108 PCI-X HCA

drivers/infiniband/hw/mthca/mthca_allocator.c

@@ -255,7 +255,7 @@ int mthca_buf_alloc(struct mthca_dev *dev, int size, int max_direct,
 			dma_list[i] = t;
 			pci_unmap_addr_set(&buf->page_list[i], mapping, t);
 
-			memset(buf->page_list[i].buf, 0, PAGE_SIZE);
+			clear_page(buf->page_list[i].buf);
 		}
 	}
 

drivers/infiniband/hw/mthca/mthca_eq.c

@@ -522,7 +522,7 @@ static int mthca_create_eq(struct mthca_dev *dev,
 		dma_list[i] = t;
 		pci_unmap_addr_set(&eq->page_list[i], mapping, t);
 
-		memset(eq->page_list[i].buf, 0, PAGE_SIZE);
+		clear_page(eq->page_list[i].buf);
 	}
 
 	for (i = 0; i < eq->nent; ++i)

drivers/infiniband/ulp/ipoib/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_IPOIB
 	tristate "IP-over-InfiniBand"
-	depends on INFINIBAND && NETDEVICES && INET && (IPV6 || IPV6=n)
+	depends on NETDEVICES && INET && (IPV6 || IPV6=n)
 	---help---
 	  Support for the IP-over-InfiniBand protocol (IPoIB). This
 	  transports IP packets over InfiniBand so you can use your IB

drivers/infiniband/ulp/ipoib/ipoib_cm.c

@@ -281,7 +281,6 @@ static int ipoib_cm_send_rep(struct net_device *dev, struct ib_cm_id *cm_id,
 	rep.private_data_len = sizeof data;
 	rep.flow_control = 0;
 	rep.rnr_retry_count = req->rnr_retry_count;
-	rep.target_ack_delay = 20; /* FIXME */
 	rep.srq = 1;
 	rep.qp_num = qp->qp_num;
 	rep.starting_psn = psn;
@@ -1148,7 +1147,6 @@ static void ipoib_cm_skb_reap(struct work_struct *work)
 {
 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
 						   cm.skb_task);
-	struct net_device *dev = priv->dev;
 	struct sk_buff *skb;
 
 	unsigned mtu = priv->mcast_mtu;
@@ -1162,7 +1160,7 @@ static void ipoib_cm_skb_reap(struct work_struct *work)
 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 		else if (skb->protocol == htons(ETH_P_IPV6))
-			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
+			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, priv->dev);
 #endif
 		dev_kfree_skb_any(skb);
 		spin_lock_irq(&priv->tx_lock);

drivers/infiniband/ulp/ipoib/ipoib_ib.c

@@ -196,6 +196,13 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 		return;
 	}
 
+	/*
+	 * Drop packets that this interface sent, ie multicast packets
+	 * that the HCA has replicated.
+	 */
+	if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num)
+		goto repost;
+
 	/*
 	 * If we can't allocate a new RX buffer, dump
 	 * this packet and reuse the old buffer.
@@ -213,24 +220,18 @@ static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 	skb_put(skb, wc->byte_len);
 	skb_pull(skb, IB_GRH_BYTES);
 
-	if (wc->slid != priv->local_lid ||
-	    wc->src_qp != priv->qp->qp_num) {
-		skb->protocol = ((struct ipoib_header *) skb->data)->proto;
-		skb_reset_mac_header(skb);
-		skb_pull(skb, IPOIB_ENCAP_LEN);
+	skb->protocol = ((struct ipoib_header *) skb->data)->proto;
+	skb_reset_mac_header(skb);
+	skb_pull(skb, IPOIB_ENCAP_LEN);
 
-		dev->last_rx = jiffies;
-		++priv->stats.rx_packets;
-		priv->stats.rx_bytes += skb->len;
+	dev->last_rx = jiffies;
+	++priv->stats.rx_packets;
+	priv->stats.rx_bytes += skb->len;
 
-		skb->dev = dev;
-		/* XXX get correct PACKET_ type here */
-		skb->pkt_type = PACKET_HOST;
-		netif_receive_skb(skb);
-	} else {
-		ipoib_dbg_data(priv, "dropping loopback packet\n");
-		dev_kfree_skb_any(skb);
-	}
+	skb->dev = dev;
+	/* XXX get correct PACKET_ type here */
+	skb->pkt_type = PACKET_HOST;
+	netif_receive_skb(skb);
 
 repost:
 	if (unlikely(ipoib_ib_post_receive(dev, wr_id)))

drivers/infiniband/ulp/iser/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_ISER
 	tristate "iSCSI Extensions for RDMA (iSER)"
-	depends on INFINIBAND && SCSI && INET
+	depends on SCSI && INET
 	select SCSI_ISCSI_ATTRS
 	---help---
 	  Support for the iSCSI Extensions for RDMA (iSER) Protocol

drivers/infiniband/ulp/srp/Kconfig

@@ -1,6 +1,6 @@
 config INFINIBAND_SRP
 	tristate "InfiniBand SCSI RDMA Protocol"
-	depends on INFINIBAND && SCSI
+	depends on SCSI
 	---help---
 	  Support for the SCSI RDMA Protocol over InfiniBand.  This
 	  allows you to access storage devices that speak SRP over

drivers/net/cxgb3/version.h

@@ -39,6 +39,6 @@
 
 /* Firmware version */
 #define FW_VERSION_MAJOR 4
-#define FW_VERSION_MINOR 1
+#define FW_VERSION_MINOR 3
 #define FW_VERSION_MICRO 0
 #endif				/* __CHELSIO_VERSION_H */

drivers/net/mlx4/fw.c

@@ -138,6 +138,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 #define QUERY_DEV_CAP_ACK_DELAY_OFFSET		0x35
 #define QUERY_DEV_CAP_MTU_WIDTH_OFFSET		0x36
 #define QUERY_DEV_CAP_VL_PORT_OFFSET		0x37
+#define QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET		0x38
 #define QUERY_DEV_CAP_MAX_GID_OFFSET		0x3b
 #define QUERY_DEV_CAP_RATE_SUPPORT_OFFSET	0x3c
 #define QUERY_DEV_CAP_MAX_PKEY_OFFSET		0x3f
@@ -220,6 +221,8 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev_cap->local_ca_ack_delay = field & 0x1f;
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
 	dev_cap->num_ports = field & 0xf;
+	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET);
+	dev_cap->max_msg_sz = 1 << (field & 0x1f);
 	MLX4_GET(stat_rate, outbox, QUERY_DEV_CAP_RATE_SUPPORT_OFFSET);
 	dev_cap->stat_rate_support = stat_rate;
 	MLX4_GET(dev_cap->flags, outbox, QUERY_DEV_CAP_FLAGS_OFFSET);

drivers/net/mlx4/fw.h

@@ -60,6 +60,7 @@ struct mlx4_dev_cap {
 	int max_rdma_global;
 	int local_ca_ack_delay;
 	int num_ports;
+	u32 max_msg_sz;
 	int max_mtu[MLX4_MAX_PORTS + 1];
 	int max_port_width[MLX4_MAX_PORTS + 1];
 	int max_vl[MLX4_MAX_PORTS + 1];

drivers/net/mlx4/main.c

@@ -154,6 +154,7 @@ static int __devinit mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev
 	dev->caps.reserved_uars	     = dev_cap->reserved_uars;
 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
 	dev->caps.mtt_entry_sz	     = MLX4_MTT_ENTRY_PER_SEG * dev_cap->mtt_entry_sz;
+	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
 	dev->caps.flags		     = dev_cap->flags;
 	dev->caps.stat_rate_support  = dev_cap->stat_rate_support;

drivers/net/mlx4/mlx4.h

@@ -37,6 +37,7 @@
 #ifndef MLX4_H
 #define MLX4_H
 
+#include <linux/mutex.h>
 #include <linux/radix-tree.h>
 
 #include <linux/mlx4/device.h>

drivers/net/mlx4/qp.c

@@ -277,3 +277,24 @@ void mlx4_cleanup_qp_table(struct mlx4_dev *dev)
 	mlx4_CONF_SPECIAL_QP(dev, 0);
 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->qp_table.bitmap);
 }
+
+int mlx4_qp_query(struct mlx4_dev *dev, struct mlx4_qp *qp,
+		  struct mlx4_qp_context *context)
+{
+	struct mlx4_cmd_mailbox *mailbox;
+	int err;
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	err = mlx4_cmd_box(dev, 0, mailbox->dma, qp->qpn, 0,
+			   MLX4_CMD_QUERY_QP, MLX4_CMD_TIME_CLASS_A);
+	if (!err)
+		memcpy(context, mailbox->buf + 8, sizeof *context);
+
+	mlx4_free_cmd_mailbox(dev, mailbox);
+	return err;
+}
+EXPORT_SYMBOL_GPL(mlx4_qp_query);
+

drivers/net/mlx4/srq.c

@@ -102,6 +102,13 @@ static int mlx4_ARM_SRQ(struct mlx4_dev *dev, int srq_num, int limit_watermark)
 			MLX4_CMD_TIME_CLASS_B);
 }
 
+static int mlx4_QUERY_SRQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox,
+			  int srq_num)
+{
+	return mlx4_cmd_box(dev, 0, mailbox->dma, srq_num, 0, MLX4_CMD_QUERY_SRQ,
+			    MLX4_CMD_TIME_CLASS_A);
+}
+
 int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
 		   u64 db_rec, struct mlx4_srq *srq)
 {
@@ -205,6 +212,29 @@ int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark
 }
 EXPORT_SYMBOL_GPL(mlx4_srq_arm);
 
+int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark)
+{
+	struct mlx4_cmd_mailbox *mailbox;
+	struct mlx4_srq_context *srq_context;
+	int err;
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	srq_context = mailbox->buf;
+
+	err = mlx4_QUERY_SRQ(dev, mailbox, srq->srqn);
+	if (err)
+		goto err_out;
+	*limit_watermark = srq_context->limit_watermark;
+
+err_out:
+	mlx4_free_cmd_mailbox(dev, mailbox);
+	return err;
+}
+EXPORT_SYMBOL_GPL(mlx4_srq_query);
+
 int __devinit mlx4_init_srq_table(struct mlx4_dev *dev)
 {
 	struct mlx4_srq_table *srq_table = &mlx4_priv(dev)->srq_table;

include/linux/mlx4/device.h

@@ -172,6 +172,7 @@ struct mlx4_caps {
 	int			num_pds;
 	int			reserved_pds;
 	int			mtt_entry_sz;
+	u32			max_msg_sz;
 	u32			page_size_cap;
 	u32			flags;
 	u16			stat_rate_support;
@@ -322,6 +323,7 @@ int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
 		   u64 db_rec, struct mlx4_srq *srq);
 void mlx4_srq_free(struct mlx4_dev *dev, struct mlx4_srq *srq);
 int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark);
+int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark);
 
 int mlx4_INIT_PORT(struct mlx4_dev *dev, int port);
 int mlx4_CLOSE_PORT(struct mlx4_dev *dev, int port);

include/linux/mlx4/qp.h

@@ -282,6 +282,9 @@ int mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
 		   struct mlx4_qp_context *context, enum mlx4_qp_optpar optpar,
 		   int sqd_event, struct mlx4_qp *qp);
 
+int mlx4_qp_query(struct mlx4_dev *dev, struct mlx4_qp *qp,
+		  struct mlx4_qp_context *context);
+
 static inline struct mlx4_qp *__mlx4_qp_lookup(struct mlx4_dev *dev, u32 qpn)
 {
 	return radix_tree_lookup(&dev->qp_table_tree, qpn & (dev->caps.num_qps - 1));

include/rdma/ib_cm.h

@@ -385,7 +385,6 @@ struct ib_cm_rep_param {
 	u8		private_data_len;
 	u8		responder_resources;
 	u8		initiator_depth;
-	u8		target_ack_delay;
 	u8		failover_accepted;
 	u8		flow_control;
 	u8		rnr_retry_count;

include/rdma/ib_mad.h

@@ -111,6 +111,9 @@
 #define IB_QP1_QKEY	0x80010000
 #define IB_QP_SET_QKEY	0x80000000
 
+#define IB_DEFAULT_PKEY_PARTIAL 0x7FFF
+#define IB_DEFAULT_PKEY_FULL	0xFFFF
+
 enum {
 	IB_MGMT_MAD_HDR = 24,
 	IB_MGMT_MAD_DATA = 232,