Merge branches 'cma', 'cxgb3', 'ehca', 'ipath', 'ipoib', 'mad', 'misc',...

Merge branches 'cma', 'cxgb3', 'ehca', 'ipath', 'ipoib', 'mad', 'misc', 'mlx4', 'mthca' and 'nes' into for-next

drivers/infiniband/core/mad.c

@@ -1697,9 +1697,8 @@ static inline int rcv_has_same_gid(struct ib_mad_agent_private *mad_agent_priv,
 	u8 port_num = mad_agent_priv->agent.port_num;
 	u8 lmc;
 
-	send_resp = ((struct ib_mad *)(wr->send_buf.mad))->
-		     mad_hdr.method & IB_MGMT_METHOD_RESP;
-	rcv_resp = rwc->recv_buf.mad->mad_hdr.method & IB_MGMT_METHOD_RESP;
+	send_resp = ib_response_mad((struct ib_mad *)wr->send_buf.mad);
+	rcv_resp = ib_response_mad(rwc->recv_buf.mad);
 
 	if (send_resp == rcv_resp)
 		/* both requests, or both responses. GIDs different */

drivers/infiniband/hw/amso1100/c2_provider.c

@@ -272,7 +272,6 @@ static struct ib_qp *c2_create_qp(struct ib_pd *pd,
 		pr_debug("%s: Invalid QP type: %d\n", __func__,
 			init_attr->qp_type);
 		return ERR_PTR(-EINVAL);
-		break;
 	}
 
 	if (err) {

drivers/infiniband/hw/cxgb3/iwch_provider.c

@@ -1155,13 +1155,11 @@ static int iwch_query_port(struct ib_device *ibdev,
 			   u8 port, struct ib_port_attr *props)
 {
 	PDBG("%s ibdev %p\n", __func__, ibdev);
+
+	memset(props, 0, sizeof(struct ib_port_attr));
 	props->max_mtu = IB_MTU_4096;
-	props->lid = 0;
-	props->lmc = 0;
-	props->sm_lid = 0;
-	props->sm_sl = 0;
+	props->active_mtu = IB_MTU_2048;
 	props->state = IB_PORT_ACTIVE;
-	props->phys_state = 0;
 	props->port_cap_flags =
 	    IB_PORT_CM_SUP |
 	    IB_PORT_SNMP_TUNNEL_SUP |
@@ -1170,7 +1168,6 @@ static int iwch_query_port(struct ib_device *ibdev,
 	    IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
 	props->gid_tbl_len = 1;
 	props->pkey_tbl_len = 1;
-	props->qkey_viol_cntr = 0;
 	props->active_width = 2;
 	props->active_speed = 2;
 	props->max_msg_sz = -1;

drivers/infiniband/hw/ehca/ehca_classes.h

@@ -164,6 +164,13 @@ struct ehca_qmap_entry {
 	u16 reported;
 };
 
+struct ehca_queue_map {
+	struct ehca_qmap_entry *map;
+	unsigned int entries;
+	unsigned int tail;
+	unsigned int left_to_poll;
+};
+
 struct ehca_qp {
 	union {
 		struct ib_qp ib_qp;
@@ -173,8 +180,9 @@ struct ehca_qp {
 	enum ehca_ext_qp_type ext_type;
 	enum ib_qp_state state;
 	struct ipz_queue ipz_squeue;
-	struct ehca_qmap_entry *sq_map;
+	struct ehca_queue_map sq_map;
 	struct ipz_queue ipz_rqueue;
+	struct ehca_queue_map rq_map;
 	struct h_galpas galpas;
 	u32 qkey;
 	u32 real_qp_num;
@@ -204,6 +212,8 @@ struct ehca_qp {
 	atomic_t nr_events; /* events seen */
 	wait_queue_head_t wait_completion;
 	int mig_armed;
+	struct list_head sq_err_node;
+	struct list_head rq_err_node;
 };
 
 #define IS_SRQ(qp) (qp->ext_type == EQPT_SRQ)
@@ -233,6 +243,8 @@ struct ehca_cq {
 	/* mmap counter for resources mapped into user space */
 	u32 mm_count_queue;
 	u32 mm_count_galpa;
+	struct list_head sqp_err_list;
+	struct list_head rqp_err_list;
 };
 
 enum ehca_mr_flag {

drivers/infiniband/hw/ehca/ehca_cq.c

@@ -276,6 +276,9 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 	for (i = 0; i < QP_HASHTAB_LEN; i++)
 		INIT_HLIST_HEAD(&my_cq->qp_hashtab[i]);
 
+	INIT_LIST_HEAD(&my_cq->sqp_err_list);
+	INIT_LIST_HEAD(&my_cq->rqp_err_list);
+
 	if (context) {
 		struct ipz_queue *ipz_queue = &my_cq->ipz_queue;
 		struct ehca_create_cq_resp resp;

drivers/infiniband/hw/ehca/ehca_iverbs.h

@@ -197,6 +197,8 @@ void ehca_poll_eqs(unsigned long data);
 int ehca_calc_ipd(struct ehca_shca *shca, int port,
 		  enum ib_rate path_rate, u32 *ipd);
 
+void ehca_add_to_err_list(struct ehca_qp *qp, int on_sq);
+
 #ifdef CONFIG_PPC_64K_PAGES
 void *ehca_alloc_fw_ctrlblock(gfp_t flags);
 void ehca_free_fw_ctrlblock(void *ptr);

drivers/infiniband/hw/ehca/ehca_qp.c

@@ -396,6 +396,50 @@ static void ehca_determine_small_queue(struct ehca_alloc_queue_parms *queue,
 	queue->is_small = (queue->page_size != 0);
 }
 
+/* needs to be called with cq->spinlock held */
+void ehca_add_to_err_list(struct ehca_qp *qp, int on_sq)
+{
+	struct list_head *list, *node;
+
+	/* TODO: support low latency QPs */
+	if (qp->ext_type == EQPT_LLQP)
+		return;
+
+	if (on_sq) {
+		list = &qp->send_cq->sqp_err_list;
+		node = &qp->sq_err_node;
+	} else {
+		list = &qp->recv_cq->rqp_err_list;
+		node = &qp->rq_err_node;
+	}
+
+	if (list_empty(node))
+		list_add_tail(node, list);
+
+	return;
+}
+
+static void del_from_err_list(struct ehca_cq *cq, struct list_head *node)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&cq->spinlock, flags);
+
+	if (!list_empty(node))
+		list_del_init(node);
+
+	spin_unlock_irqrestore(&cq->spinlock, flags);
+}
+
+static void reset_queue_map(struct ehca_queue_map *qmap)
+{
+	int i;
+
+	qmap->tail = 0;
+	for (i = 0; i < qmap->entries; i++)
+		qmap->map[i].reported = 1;
+}
+
 /*
  * Create an ib_qp struct that is either a QP or an SRQ, depending on
  * the value of the is_srq parameter. If init_attr and srq_init_attr share
@@ -407,12 +451,11 @@ static struct ehca_qp *internal_create_qp(
 	struct ib_srq_init_attr *srq_init_attr,
 	struct ib_udata *udata, int is_srq)
 {
-	struct ehca_qp *my_qp;
+	struct ehca_qp *my_qp, *my_srq = NULL;
 	struct ehca_pd *my_pd = container_of(pd, struct ehca_pd, ib_pd);
 	struct ehca_shca *shca = container_of(pd->device, struct ehca_shca,
 					      ib_device);
 	struct ib_ucontext *context = NULL;
-	u32 nr_qes;
 	u64 h_ret;
 	int is_llqp = 0, has_srq = 0;
 	int qp_type, max_send_sge, max_recv_sge, ret;
@@ -457,8 +500,7 @@ static struct ehca_qp *internal_create_qp(
 
 	/* handle SRQ base QPs */
 	if (init_attr->srq) {
-		struct ehca_qp *my_srq =
-			container_of(init_attr->srq, struct ehca_qp, ib_srq);
+		my_srq = container_of(init_attr->srq, struct ehca_qp, ib_srq);
 
 		has_srq = 1;
 		parms.ext_type = EQPT_SRQBASE;
@@ -716,15 +758,19 @@ static struct ehca_qp *internal_create_qp(
 				 "and pages ret=%i", ret);
 			goto create_qp_exit2;
 		}
-		nr_qes = my_qp->ipz_squeue.queue_length /
+
+		my_qp->sq_map.entries = my_qp->ipz_squeue.queue_length /
 			 my_qp->ipz_squeue.qe_size;
-		my_qp->sq_map = vmalloc(nr_qes *
+		my_qp->sq_map.map = vmalloc(my_qp->sq_map.entries *
 					sizeof(struct ehca_qmap_entry));
-		if (!my_qp->sq_map) {
+		if (!my_qp->sq_map.map) {
 			ehca_err(pd->device, "Couldn't allocate squeue "
 				 "map ret=%i", ret);
 			goto create_qp_exit3;
 		}
+		INIT_LIST_HEAD(&my_qp->sq_err_node);
+		/* to avoid the generation of bogus flush CQEs */
+		reset_queue_map(&my_qp->sq_map);
 	}
 
 	if (HAS_RQ(my_qp)) {
@@ -736,6 +782,25 @@ static struct ehca_qp *internal_create_qp(
 				 "and pages ret=%i", ret);
 			goto create_qp_exit4;
 		}
+
+		my_qp->rq_map.entries = my_qp->ipz_rqueue.queue_length /
+			my_qp->ipz_rqueue.qe_size;
+		my_qp->rq_map.map = vmalloc(my_qp->rq_map.entries *
+				sizeof(struct ehca_qmap_entry));
+		if (!my_qp->rq_map.map) {
+			ehca_err(pd->device, "Couldn't allocate squeue "
+					"map ret=%i", ret);
+			goto create_qp_exit5;
+		}
+		INIT_LIST_HEAD(&my_qp->rq_err_node);
+		/* to avoid the generation of bogus flush CQEs */
+		reset_queue_map(&my_qp->rq_map);
+	} else if (init_attr->srq) {
+		/* this is a base QP, use the queue map of the SRQ */
+		my_qp->rq_map = my_srq->rq_map;
+		INIT_LIST_HEAD(&my_qp->rq_err_node);
+
+		my_qp->ipz_rqueue = my_srq->ipz_rqueue;
 	}
 
 	if (is_srq) {
@@ -799,7 +864,7 @@ static struct ehca_qp *internal_create_qp(
 		if (ret) {
 			ehca_err(pd->device,
 				 "Couldn't assign qp to send_cq ret=%i", ret);
-			goto create_qp_exit6;
+			goto create_qp_exit7;
 		}
 	}
 
@@ -825,25 +890,29 @@ static struct ehca_qp *internal_create_qp(
 		if (ib_copy_to_udata(udata, &resp, sizeof resp)) {
 			ehca_err(pd->device, "Copy to udata failed");
 			ret = -EINVAL;
-			goto create_qp_exit7;
+			goto create_qp_exit8;
 		}
 	}
 
 	return my_qp;
 
-create_qp_exit7:
+create_qp_exit8:
 	ehca_cq_unassign_qp(my_qp->send_cq, my_qp->real_qp_num);
 
-create_qp_exit6:
+create_qp_exit7:
 	kfree(my_qp->mod_qp_parm);
 
+create_qp_exit6:
+	if (HAS_RQ(my_qp))
+		vfree(my_qp->rq_map.map);
+
 create_qp_exit5:
 	if (HAS_RQ(my_qp))
 		ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue);
 
 create_qp_exit4:
 	if (HAS_SQ(my_qp))
-		vfree(my_qp->sq_map);
+		vfree(my_qp->sq_map.map);
 
 create_qp_exit3:
 	if (HAS_SQ(my_qp))
@@ -1035,6 +1104,101 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca,
 	return 0;
 }
 
+static int calc_left_cqes(u64 wqe_p, struct ipz_queue *ipz_queue,
+			  struct ehca_queue_map *qmap)
+{
+	void *wqe_v;
+	u64 q_ofs;
+	u32 wqe_idx;
+
+	/* convert real to abs address */
+	wqe_p = wqe_p & (~(1UL << 63));
+
+	wqe_v = abs_to_virt(wqe_p);
+
+	if (ipz_queue_abs_to_offset(ipz_queue, wqe_p, &q_ofs)) {
+		ehca_gen_err("Invalid offset for calculating left cqes "
+				"wqe_p=%#lx wqe_v=%p\n", wqe_p, wqe_v);
+		return -EFAULT;
+	}
+
+	wqe_idx = q_ofs / ipz_queue->qe_size;
+	if (wqe_idx < qmap->tail)
+		qmap->left_to_poll = (qmap->entries - qmap->tail) + wqe_idx;
+	else
+		qmap->left_to_poll = wqe_idx - qmap->tail;
+
+	return 0;
+}
+
+static int check_for_left_cqes(struct ehca_qp *my_qp, struct ehca_shca *shca)
+{
+	u64 h_ret;
+	void *send_wqe_p, *recv_wqe_p;
+	int ret;
+	unsigned long flags;
+	int qp_num = my_qp->ib_qp.qp_num;
+
+	/* this hcall is not supported on base QPs */
+	if (my_qp->ext_type != EQPT_SRQBASE) {
+		/* get send and receive wqe pointer */
+		h_ret = hipz_h_disable_and_get_wqe(shca->ipz_hca_handle,
+				my_qp->ipz_qp_handle, &my_qp->pf,
+				&send_wqe_p, &recv_wqe_p, 4);
+		if (h_ret != H_SUCCESS) {
+			ehca_err(&shca->ib_device, "disable_and_get_wqe() "
+				 "failed ehca_qp=%p qp_num=%x h_ret=%li",
+				 my_qp, qp_num, h_ret);
+			return ehca2ib_return_code(h_ret);
+		}
+
+		/*
+		 * acquire lock to ensure that nobody is polling the cq which
+		 * could mean that the qmap->tail pointer is in an
+		 * inconsistent state.
+		 */
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		ret = calc_left_cqes((u64)send_wqe_p, &my_qp->ipz_squeue,
+				&my_qp->sq_map);
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+		if (ret)
+			return ret;
+
+
+		spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+		ret = calc_left_cqes((u64)recv_wqe_p, &my_qp->ipz_rqueue,
+				&my_qp->rq_map);
+		spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags);
+		if (ret)
+			return ret;
+	} else {
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		my_qp->sq_map.left_to_poll = 0;
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+
+		spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+		my_qp->rq_map.left_to_poll = 0;
+		spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags);
+	}
+
+	/* this assures flush cqes being generated only for pending wqes */
+	if ((my_qp->sq_map.left_to_poll == 0) &&
+				(my_qp->rq_map.left_to_poll == 0)) {
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		ehca_add_to_err_list(my_qp, 1);
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+
+		if (HAS_RQ(my_qp)) {
+			spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+			ehca_add_to_err_list(my_qp, 0);
+			spin_unlock_irqrestore(&my_qp->recv_cq->spinlock,
+					flags);
+		}
+	}
+
+	return 0;
+}
+
 /*
  * internal_modify_qp with circumvention to handle aqp0 properly
  * smi_reset2init indicates if this is an internal reset-to-init-call for
@@ -1539,10 +1703,27 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 			goto modify_qp_exit2;
 		}
 	}
+	if ((qp_new_state == IB_QPS_ERR) && (qp_cur_state != IB_QPS_ERR)) {
+		ret = check_for_left_cqes(my_qp, shca);
+		if (ret)
+			goto modify_qp_exit2;
+	}
 
 	if (statetrans == IB_QPST_ANY2RESET) {
 		ipz_qeit_reset(&my_qp->ipz_rqueue);
 		ipz_qeit_reset(&my_qp->ipz_squeue);
+
+		if (qp_cur_state == IB_QPS_ERR) {
+			del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node);
+
+			if (HAS_RQ(my_qp))
+				del_from_err_list(my_qp->recv_cq,
+						  &my_qp->rq_err_node);
+		}
+		reset_queue_map(&my_qp->sq_map);
+
+		if (HAS_RQ(my_qp))
+			reset_queue_map(&my_qp->rq_map);
 	}
 
 	if (attr_mask & IB_QP_QKEY)
@@ -1958,6 +2139,16 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 	idr_remove(&ehca_qp_idr, my_qp->token);
 	write_unlock_irqrestore(&ehca_qp_idr_lock, flags);
 
+	/*
+	 * SRQs will never get into an error list and do not have a recv_cq,
+	 * so we need to skip them here.
+	 */
+	if (HAS_RQ(my_qp) && !IS_SRQ(my_qp))
+		del_from_err_list(my_qp->recv_cq, &my_qp->rq_err_node);
+
+	if (HAS_SQ(my_qp))
+		del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node);
+
 	/* now wait until all pending events have completed */
 	wait_event(my_qp->wait_completion, !atomic_read(&my_qp->nr_events));
 
@@ -1991,11 +2182,15 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 		ib_dispatch_event(&event);
 	}
 
-	if (HAS_RQ(my_qp))
+	if (HAS_RQ(my_qp)) {
 		ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue);
+
+		vfree(my_qp->rq_map.map);
+	}
 	if (HAS_SQ(my_qp)) {
 		ipz_queue_dtor(my_pd, &my_qp->ipz_squeue);
-		vfree(my_qp->sq_map);
+
+		vfree(my_qp->sq_map.map);
 	}
 	kmem_cache_free(qp_cache, my_qp);
 	atomic_dec(&shca->num_qps);

drivers/infiniband/hw/ehca/ehca_reqs.c

@@ -53,9 +53,25 @@
 /* in RC traffic, insert an empty RDMA READ every this many packets */
 #define ACK_CIRC_THRESHOLD 2000000
 
+static u64 replace_wr_id(u64 wr_id, u16 idx)
+{
+	u64 ret;
+
+	ret = wr_id & ~QMAP_IDX_MASK;
+	ret |= idx & QMAP_IDX_MASK;
+
+	return ret;
+}
+
+static u16 get_app_wr_id(u64 wr_id)
+{
+	return wr_id & QMAP_IDX_MASK;
+}
+
 static inline int ehca_write_rwqe(struct ipz_queue *ipz_rqueue,
 				  struct ehca_wqe *wqe_p,
-				  struct ib_recv_wr *recv_wr)
+				  struct ib_recv_wr *recv_wr,
+				  u32 rq_map_idx)
 {
 	u8 cnt_ds;
 	if (unlikely((recv_wr->num_sge < 0) ||
@@ -69,7 +85,7 @@ static inline int ehca_write_rwqe(struct ipz_queue *ipz_rqueue,
 	/* clear wqe header until sglist */
 	memset(wqe_p, 0, offsetof(struct ehca_wqe, u.ud_av.sg_list));
 
-	wqe_p->work_request_id = recv_wr->wr_id;
+	wqe_p->work_request_id = replace_wr_id(recv_wr->wr_id, rq_map_idx);
 	wqe_p->nr_of_data_seg = recv_wr->num_sge;
 
 	for (cnt_ds = 0; cnt_ds < recv_wr->num_sge; cnt_ds++) {
@@ -146,6 +162,7 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 	u64 dma_length;
 	struct ehca_av *my_av;
 	u32 remote_qkey = send_wr->wr.ud.remote_qkey;
+	struct ehca_qmap_entry *qmap_entry = &qp->sq_map.map[sq_map_idx];
 
 	if (unlikely((send_wr->num_sge < 0) ||
 		     (send_wr->num_sge > qp->ipz_squeue.act_nr_of_sg))) {
@@ -158,11 +175,10 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 	/* clear wqe header until sglist */
 	memset(wqe_p, 0, offsetof(struct ehca_wqe, u.ud_av.sg_list));
 
-	wqe_p->work_request_id = send_wr->wr_id & ~QMAP_IDX_MASK;
-	wqe_p->work_request_id |= sq_map_idx & QMAP_IDX_MASK;
+	wqe_p->work_request_id = replace_wr_id(send_wr->wr_id, sq_map_idx);
 
-	qp->sq_map[sq_map_idx].app_wr_id = send_wr->wr_id & QMAP_IDX_MASK;
-	qp->sq_map[sq_map_idx].reported = 0;
+	qmap_entry->app_wr_id = get_app_wr_id(send_wr->wr_id);
+	qmap_entry->reported = 0;
 
 	switch (send_wr->opcode) {
 	case IB_WR_SEND:
@@ -496,7 +512,9 @@ static int internal_post_recv(struct ehca_qp *my_qp,
 	struct ehca_wqe *wqe_p;
 	int wqe_cnt = 0;
 	int ret = 0;
+	u32 rq_map_idx;
 	unsigned long flags;
+	struct ehca_qmap_entry *qmap_entry;
 
 	if (unlikely(!HAS_RQ(my_qp))) {
 		ehca_err(dev, "QP has no RQ  ehca_qp=%p qp_num=%x ext_type=%d",
@@ -524,8 +542,15 @@ static int internal_post_recv(struct ehca_qp *my_qp,
 			}
 			goto post_recv_exit0;
 		}
+		/*
+		 * Get the index of the WQE in the recv queue. The same index
+		 * is used for writing into the rq_map.
+		 */
+		rq_map_idx = start_offset / my_qp->ipz_rqueue.qe_size;
+
 		/* write a RECV WQE into the QUEUE */
-		ret = ehca_write_rwqe(&my_qp->ipz_rqueue, wqe_p, cur_recv_wr);
+		ret = ehca_write_rwqe(&my_qp->ipz_rqueue, wqe_p, cur_recv_wr,
+				rq_map_idx);
 		/*
 		 * if something failed,
 		 * reset the free entry pointer to the start value
@@ -540,6 +565,11 @@ static int internal_post_recv(struct ehca_qp *my_qp,
 			}
 			goto post_recv_exit0;
 		}
+
+		qmap_entry = &my_qp->rq_map.map[rq_map_idx];
+		qmap_entry->app_wr_id = get_app_wr_id(cur_recv_wr->wr_id);
+		qmap_entry->reported = 0;
+
 		wqe_cnt++;
 	} /* eof for cur_recv_wr */
 
@@ -596,10 +626,12 @@ static const u8 ib_wc_opcode[255] = {
 /* internal function to poll one entry of cq */
 static inline int ehca_poll_cq_one(struct ib_cq *cq, struct ib_wc *wc)
 {
-	int ret = 0;
+	int ret = 0, qmap_tail_idx;
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
 	struct ehca_cqe *cqe;
 	struct ehca_qp *my_qp;
+	struct ehca_qmap_entry *qmap_entry;
+	struct ehca_queue_map *qmap;
 	int cqe_count = 0, is_error;
 
 repoll:
@@ -674,27 +706,52 @@ static inline int ehca_poll_cq_one(struct ib_cq *cq, struct ib_wc *wc)
 		goto repoll;
 	wc->qp = &my_qp->ib_qp;
 
-	if (!(cqe->w_completion_flags & WC_SEND_RECEIVE_BIT)) {
-		struct ehca_qmap_entry *qmap_entry;
+	if (is_error) {
 		/*
-		 * We got a send completion and need to restore the original
-		 * wr_id.
+		 * set left_to_poll to 0 because in error state, we will not
+		 * get any additional CQEs
 		 */
-		qmap_entry = &my_qp->sq_map[cqe->work_request_id &
-					    QMAP_IDX_MASK];
+		ehca_add_to_err_list(my_qp, 1);
+		my_qp->sq_map.left_to_poll = 0;
 
+		if (HAS_RQ(my_qp))
+			ehca_add_to_err_list(my_qp, 0);
+		my_qp->rq_map.left_to_poll = 0;
+	}
+
+	qmap_tail_idx = get_app_wr_id(cqe->work_request_id);
+	if (!(cqe->w_completion_flags & WC_SEND_RECEIVE_BIT))
+		/* We got a send completion. */
+		qmap = &my_qp->sq_map;
+	else
+		/* We got a receive completion. */
+		qmap = &my_qp->rq_map;
+
+	qmap_entry = &qmap->map[qmap_tail_idx];
 	if (qmap_entry->reported) {
 		ehca_warn(cq->device, "Double cqe on qp_num=%#x",
 				my_qp->real_qp_num);
 		/* found a double cqe, discard it and read next one */
 		goto repoll;
 	}
-		wc->wr_id = cqe->work_request_id & ~QMAP_IDX_MASK;
-		wc->wr_id |= qmap_entry->app_wr_id;
+
+	wc->wr_id = replace_wr_id(cqe->work_request_id, qmap_entry->app_wr_id);
 	qmap_entry->reported = 1;
-	} else
-		/* We got a receive completion. */
-		wc->wr_id = cqe->work_request_id;
+
+	/* this is a proper completion, we need to advance the tail pointer */
+	if (++qmap->tail == qmap->entries)
+		qmap->tail = 0;
+
+	/* if left_to_poll is decremented to 0, add the QP to the error list */
+	if (qmap->left_to_poll > 0) {
+		qmap->left_to_poll--;
+		if ((my_qp->sq_map.left_to_poll == 0) &&
+				(my_qp->rq_map.left_to_poll == 0)) {
+			ehca_add_to_err_list(my_qp, 1);
+			if (HAS_RQ(my_qp))
+				ehca_add_to_err_list(my_qp, 0);
+		}
+	}
 
 	/* eval ib_wc_opcode */
 	wc->opcode = ib_wc_opcode[cqe->optype]-1;
@@ -733,13 +790,88 @@ static inline int ehca_poll_cq_one(struct ib_cq *cq, struct ib_wc *wc)
 	return ret;
 }
 
+static int generate_flush_cqes(struct ehca_qp *my_qp, struct ib_cq *cq,
+			       struct ib_wc *wc, int num_entries,
+			       struct ipz_queue *ipz_queue, int on_sq)
+{
+	int nr = 0;
+	struct ehca_wqe *wqe;
+	u64 offset;
+	struct ehca_queue_map *qmap;
+	struct ehca_qmap_entry *qmap_entry;
+
+	if (on_sq)
+		qmap = &my_qp->sq_map;
+	else
+		qmap = &my_qp->rq_map;
+
+	qmap_entry = &qmap->map[qmap->tail];
+
+	while ((nr < num_entries) && (qmap_entry->reported == 0)) {
+		/* generate flush CQE */
+		memset(wc, 0, sizeof(*wc));
+
+		offset = qmap->tail * ipz_queue->qe_size;
+		wqe = (struct ehca_wqe *)ipz_qeit_calc(ipz_queue, offset);
+		if (!wqe) {
+			ehca_err(cq->device, "Invalid wqe offset=%#lx on "
+				 "qp_num=%#x", offset, my_qp->real_qp_num);
+			return nr;
+		}
+
+		wc->wr_id = replace_wr_id(wqe->work_request_id,
+					  qmap_entry->app_wr_id);
+
+		if (on_sq) {
+			switch (wqe->optype) {
+			case WQE_OPTYPE_SEND:
+				wc->opcode = IB_WC_SEND;
+				break;
+			case WQE_OPTYPE_RDMAWRITE:
+				wc->opcode = IB_WC_RDMA_WRITE;
+				break;
+			case WQE_OPTYPE_RDMAREAD:
+				wc->opcode = IB_WC_RDMA_READ;
+				break;
+			default:
+				ehca_err(cq->device, "Invalid optype=%x",
+						wqe->optype);
+				return nr;
+			}
+		} else
+			wc->opcode = IB_WC_RECV;
+
+		if (wqe->wr_flag & WQE_WRFLAG_IMM_DATA_PRESENT) {
+			wc->ex.imm_data = wqe->immediate_data;
+			wc->wc_flags |= IB_WC_WITH_IMM;
+		}
+
+		wc->status = IB_WC_WR_FLUSH_ERR;
+
+		wc->qp = &my_qp->ib_qp;
+
+		/* mark as reported and advance tail pointer */
+		qmap_entry->reported = 1;
+		if (++qmap->tail == qmap->entries)
+			qmap->tail = 0;
+		qmap_entry = &qmap->map[qmap->tail];
+
+		wc++; nr++;
+	}
+
+	return nr;
+
+}
+
 int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 {
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
 	int nr;
+	struct ehca_qp *err_qp;
 	struct ib_wc *current_wc = wc;
 	int ret = 0;
 	unsigned long flags;
+	int entries_left = num_entries;
 
 	if (num_entries < 1) {
 		ehca_err(cq->device, "Invalid num_entries=%d ehca_cq=%p "
@@ -749,15 +881,40 @@ int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 	}
 
 	spin_lock_irqsave(&my_cq->spinlock, flags);
-	for (nr = 0; nr < num_entries; nr++) {
+
+	/* generate flush cqes for send queues */
+	list_for_each_entry(err_qp, &my_cq->sqp_err_list, sq_err_node) {
+		nr = generate_flush_cqes(err_qp, cq, current_wc, entries_left,
+				&err_qp->ipz_squeue, 1);
+		entries_left -= nr;
+		current_wc += nr;
+
+		if (entries_left == 0)
+			break;
+	}
+
+	/* generate flush cqes for receive queues */
+	list_for_each_entry(err_qp, &my_cq->rqp_err_list, rq_err_node) {
+		nr = generate_flush_cqes(err_qp, cq, current_wc, entries_left,
+				&err_qp->ipz_rqueue, 0);
+		entries_left -= nr;
+		current_wc += nr;
+
+		if (entries_left == 0)
+			break;
+	}
+
+	for (nr = 0; nr < entries_left; nr++) {
 		ret = ehca_poll_cq_one(cq, current_wc);
 		if (ret)
 			break;
 		current_wc++;
 	} /* eof for nr */
+	entries_left -= nr;
+
 	spin_unlock_irqrestore(&my_cq->spinlock, flags);
 	if (ret == -EAGAIN  || !ret)
-		ret = nr;
+		ret = num_entries - entries_left;
 
 poll_cq_exit0:
 	return ret;

drivers/infiniband/hw/ipath/ipath_rc.c

@@ -675,7 +675,8 @@ static void send_rc_ack(struct ipath_qp *qp)
 	hdr.lrh[0] = cpu_to_be16(lrh0);
 	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
-	hdr.lrh[3] = cpu_to_be16(dd->ipath_lid);
+	hdr.lrh[3] = cpu_to_be16(dd->ipath_lid |
+				 qp->remote_ah_attr.src_path_bits);
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);

drivers/infiniband/hw/ipath/ipath_ruc.c

@@ -618,7 +618,8 @@ void ipath_make_ruc_header(struct ipath_ibdev *dev, struct ipath_qp *qp,
 	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
 	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
-	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
+	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid |
+				       qp->remote_ah_attr.src_path_bits);
 	bth0 |= ipath_get_pkey(dev->dd, qp->s_pkey_index);
 	bth0 |= extra_bytes << 20;
 	ohdr->bth[0] = cpu_to_be32(bth0 | (1 << 22));

drivers/infiniband/hw/ipath/ipath_verbs.c

@@ -340,9 +340,16 @@ static int ipath_post_one_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 	int acc;
 	int ret;
 	unsigned long flags;
+	struct ipath_devdata *dd = to_idev(qp->ibqp.device)->dd;
 
 	spin_lock_irqsave(&qp->s_lock, flags);
 
+	if (qp->ibqp.qp_type != IB_QPT_SMI &&
+	    !(dd->ipath_flags & IPATH_LINKACTIVE)) {
+		ret = -ENETDOWN;
+		goto bail;
+	}
+
 	/* Check that state is OK to post send. */
 	if (unlikely(!(ib_ipath_state_ops[qp->state] & IPATH_POST_SEND_OK)))
 		goto bail_inval;

drivers/infiniband/hw/mlx4/qp.c

@@ -1058,6 +1058,9 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	else
 		sqd_event = 0;
 
+	if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->rlkey |= (1 << 4);
+
 	/*
 	 * Before passing a kernel QP to the HW, make sure that the
 	 * ownership bits of the send queue are set and the SQ

drivers/infiniband/hw/mthca/mthca_catas.c

@@ -149,18 +149,10 @@ void mthca_start_catas_poll(struct mthca_dev *dev)
 		((pci_resource_len(dev->pdev, 0) - 1) &
 		 dev->catas_err.addr);
 
-	if (!request_mem_region(addr, dev->catas_err.size * 4,
-				DRV_NAME)) {
-		mthca_warn(dev, "couldn't request catastrophic error region "
-			   "at 0x%lx/0x%x\n", addr, dev->catas_err.size * 4);
-		return;
-	}
-
 	dev->catas_err.map = ioremap(addr, dev->catas_err.size * 4);
 	if (!dev->catas_err.map) {
 		mthca_warn(dev, "couldn't map catastrophic error region "
 			   "at 0x%lx/0x%x\n", addr, dev->catas_err.size * 4);
-		release_mem_region(addr, dev->catas_err.size * 4);
 		return;
 	}
 
@@ -175,13 +167,8 @@ void mthca_stop_catas_poll(struct mthca_dev *dev)
 {
 	del_timer_sync(&dev->catas_err.timer);
 
-	if (dev->catas_err.map) {
+	if (dev->catas_err.map)
 		iounmap(dev->catas_err.map);
-		release_mem_region(pci_resource_start(dev->pdev, 0) +
-				   ((pci_resource_len(dev->pdev, 0) - 1) &
-				    dev->catas_err.addr),
-				   dev->catas_err.size * 4);
-	}
 
 	spin_lock_irq(&catas_lock);
 	list_del(&dev->catas_err.list);

drivers/infiniband/hw/mthca/mthca_eq.c

@@ -652,27 +652,13 @@ static int mthca_map_reg(struct mthca_dev *dev,
 {
 	unsigned long base = pci_resource_start(dev->pdev, 0);
 
-	if (!request_mem_region(base + offset, size, DRV_NAME))
-		return -EBUSY;
-
 	*map = ioremap(base + offset, size);
-	if (!*map) {
-		release_mem_region(base + offset, size);
+	if (!*map)
 		return -ENOMEM;
-	}
 
 	return 0;
 }
 
-static void mthca_unmap_reg(struct mthca_dev *dev, unsigned long offset,
-			    unsigned long size, void __iomem *map)
-{
-	unsigned long base = pci_resource_start(dev->pdev, 0);
-
-	release_mem_region(base + offset, size);
-	iounmap(map);
-}
-
 static int mthca_map_eq_regs(struct mthca_dev *dev)
 {
 	if (mthca_is_memfree(dev)) {
@@ -699,9 +685,7 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 					dev->fw.arbel.eq_arm_base) + 4, 4,
 				  &dev->eq_regs.arbel.eq_arm)) {
 			mthca_err(dev, "Couldn't map EQ arm register, aborting.\n");
-			mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-					dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 
@@ -710,12 +694,8 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 				  MTHCA_EQ_SET_CI_SIZE,
 				  &dev->eq_regs.arbel.eq_set_ci_base)) {
 			mthca_err(dev, "Couldn't map EQ CI register, aborting.\n");
-			mthca_unmap_reg(dev, ((pci_resource_len(dev->pdev, 0) - 1) &
-					      dev->fw.arbel.eq_arm_base) + 4, 4,
-					dev->eq_regs.arbel.eq_arm);
-			mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-					dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->eq_regs.arbel.eq_arm);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 	} else {
@@ -731,8 +711,7 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 				  &dev->eq_regs.tavor.ecr_base)) {
 			mthca_err(dev, "Couldn't map ecr register, "
 				  "aborting.\n");
-			mthca_unmap_reg(dev, MTHCA_CLR_INT_BASE, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 	}
@@ -744,22 +723,12 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 static void mthca_unmap_eq_regs(struct mthca_dev *dev)
 {
 	if (mthca_is_memfree(dev)) {
-		mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-				dev->fw.arbel.eq_set_ci_base,
-				MTHCA_EQ_SET_CI_SIZE,
-				dev->eq_regs.arbel.eq_set_ci_base);
-		mthca_unmap_reg(dev, ((pci_resource_len(dev->pdev, 0) - 1) &
-				      dev->fw.arbel.eq_arm_base) + 4, 4,
-				dev->eq_regs.arbel.eq_arm);
-		mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-				dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-				dev->clr_base);
+		iounmap(dev->eq_regs.arbel.eq_set_ci_base);
+		iounmap(dev->eq_regs.arbel.eq_arm);
+		iounmap(dev->clr_base);
 	} else {
-		mthca_unmap_reg(dev, MTHCA_ECR_BASE,
-				MTHCA_ECR_SIZE + MTHCA_ECR_CLR_SIZE,
-				dev->eq_regs.tavor.ecr_base);
-		mthca_unmap_reg(dev, MTHCA_CLR_INT_BASE, MTHCA_CLR_INT_SIZE,
-				dev->clr_base);
+		iounmap(dev->eq_regs.tavor.ecr_base);
+		iounmap(dev->clr_base);
 	}
 }
 

drivers/infiniband/hw/mthca/mthca_main.c

@@ -921,58 +921,6 @@ static int mthca_setup_hca(struct mthca_dev *dev)
 	return err;
 }
 
-static int mthca_request_regions(struct pci_dev *pdev, int ddr_hidden)
-{
-	int err;
-
-	/*
-	 * We can't just use pci_request_regions() because the MSI-X
-	 * table is right in the middle of the first BAR.  If we did
-	 * pci_request_region and grab all of the first BAR, then
-	 * setting up MSI-X would fail, since the PCI core wants to do
-	 * request_mem_region on the MSI-X vector table.
-	 *
-	 * So just request what we need right now, and request any
-	 * other regions we need when setting up EQs.
-	 */
-	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-				MTHCA_HCR_SIZE, DRV_NAME))
-		return -EBUSY;
-
-	err = pci_request_region(pdev, 2, DRV_NAME);
-	if (err)
-		goto err_bar2_failed;
-
-	if (!ddr_hidden) {
-		err = pci_request_region(pdev, 4, DRV_NAME);
-		if (err)
-			goto err_bar4_failed;
-	}
-
-	return 0;
-
-err_bar4_failed:
-	pci_release_region(pdev, 2);
-
-err_bar2_failed:
-	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-			   MTHCA_HCR_SIZE);
-
-	return err;
-}
-
-static void mthca_release_regions(struct pci_dev *pdev,
-				  int ddr_hidden)
-{
-	if (!ddr_hidden)
-		pci_release_region(pdev, 4);
-
-	pci_release_region(pdev, 2);
-
-	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-			   MTHCA_HCR_SIZE);
-}
-
 static int mthca_enable_msi_x(struct mthca_dev *mdev)
 {
 	struct msix_entry entries[3];
@@ -1059,7 +1007,7 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
 		ddr_hidden = 1;
 
-	err = mthca_request_regions(pdev, ddr_hidden);
+	err = pci_request_regions(pdev, DRV_NAME);
 	if (err) {
 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
 			"aborting.\n");
@@ -1196,7 +1144,7 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 	ib_dealloc_device(&mdev->ib_dev);
 
 err_free_res:
-	mthca_release_regions(pdev, ddr_hidden);
+	pci_release_regions(pdev);
 
 err_disable_pdev:
 	pci_disable_device(pdev);
@@ -1240,8 +1188,7 @@ static void __mthca_remove_one(struct pci_dev *pdev)
 			pci_disable_msix(pdev);
 
 		ib_dealloc_device(&mdev->ib_dev);
-		mthca_release_regions(pdev, mdev->mthca_flags &
-				      MTHCA_FLAG_DDR_HIDDEN);
+		pci_release_regions(pdev);
 		pci_disable_device(pdev);
 		pci_set_drvdata(pdev, NULL);
 	}

drivers/infiniband/hw/nes/nes.c

@@ -70,27 +70,31 @@ int interrupt_mod_interval = 0;
 
 /* Interoperability */
 int mpa_version = 1;
-module_param(mpa_version, int, 0);
+module_param(mpa_version, int, 0644);
 MODULE_PARM_DESC(mpa_version, "MPA version to be used int MPA Req/Resp (0 or 1)");
 
 /* Interoperability */
 int disable_mpa_crc = 0;
-module_param(disable_mpa_crc, int, 0);
+module_param(disable_mpa_crc, int, 0644);
 MODULE_PARM_DESC(disable_mpa_crc, "Disable checking of MPA CRC");
 
 unsigned int send_first = 0;
-module_param(send_first, int, 0);
+module_param(send_first, int, 0644);
 MODULE_PARM_DESC(send_first, "Send RDMA Message First on Active Connection");
 
 
 unsigned int nes_drv_opt = 0;
-module_param(nes_drv_opt, int, 0);
+module_param(nes_drv_opt, int, 0644);
 MODULE_PARM_DESC(nes_drv_opt, "Driver option parameters");
 
 unsigned int nes_debug_level = 0;
 module_param_named(debug_level, nes_debug_level, uint, 0644);
 MODULE_PARM_DESC(debug_level, "Enable debug output level");
 
+unsigned int wqm_quanta = 0x10000;
+module_param(wqm_quanta, int, 0644);
+MODULE_PARM_DESC(wqm_quanta, "WQM quanta");
+
 LIST_HEAD(nes_adapter_list);
 static LIST_HEAD(nes_dev_list);
 
@@ -557,12 +561,32 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 		goto bail5;
 	}
 	nesdev->nesadapter->et_rx_coalesce_usecs_irq = interrupt_mod_interval;
+	nesdev->nesadapter->wqm_quanta = wqm_quanta;
 
 	/* nesdev->base_doorbell_index =
 			nesdev->nesadapter->pd_config_base[PCI_FUNC(nesdev->pcidev->devfn)]; */
 	nesdev->base_doorbell_index = 1;
 	nesdev->doorbell_start = nesdev->nesadapter->doorbell_start;
-	nesdev->mac_index = PCI_FUNC(nesdev->pcidev->devfn) % nesdev->nesadapter->port_count;
+	if (nesdev->nesadapter->phy_type[0] == NES_PHY_TYPE_PUMA_1G) {
+		switch (PCI_FUNC(nesdev->pcidev->devfn) %
+			nesdev->nesadapter->port_count) {
+		case 1:
+			nesdev->mac_index = 2;
+			break;
+		case 2:
+			nesdev->mac_index = 1;
+			break;
+		case 3:
+			nesdev->mac_index = 3;
+			break;
+		case 0:
+		default:
+			nesdev->mac_index = 0;
+		}
+	} else {
+		nesdev->mac_index = PCI_FUNC(nesdev->pcidev->devfn) %
+						nesdev->nesadapter->port_count;
+	}
 
 	tasklet_init(&nesdev->dpc_tasklet, nes_dpc, (unsigned long)nesdev);
 
@@ -581,7 +605,7 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 	nesdev->int_req = (0x101 << PCI_FUNC(nesdev->pcidev->devfn)) |
 			(1 << (PCI_FUNC(nesdev->pcidev->devfn)+16));
 	if (PCI_FUNC(nesdev->pcidev->devfn) < 4) {
-		nesdev->int_req |= (1 << (PCI_FUNC(nesdev->pcidev->devfn)+24));
+		nesdev->int_req |= (1 << (PCI_FUNC(nesdev->mac_index)+24));
 	}
 
 	/* TODO: This really should be the first driver to load, not function 0 */
@@ -772,14 +796,14 @@ static ssize_t nes_show_adapter(struct device_driver *ddp, char *buf)
 
 	list_for_each_entry(nesdev, &nes_dev_list, list) {
 		if (i == ee_flsh_adapter) {
-			devfn      = nesdev->nesadapter->devfn;
-			bus_number = nesdev->nesadapter->bus_number;
+			devfn = nesdev->pcidev->devfn;
+			bus_number = nesdev->pcidev->bus->number;
 			break;
 		}
 		i++;
 	}
 
-	return snprintf(buf, PAGE_SIZE, "%x:%x", bus_number, devfn);
+	return snprintf(buf, PAGE_SIZE, "%x:%x\n", bus_number, devfn);
 }
 
 static ssize_t nes_store_adapter(struct device_driver *ddp,
@@ -1050,6 +1074,55 @@ static ssize_t nes_store_idx_data(struct device_driver *ddp,
 	return strnlen(buf, count);
 }
 
+
+/**
+ * nes_show_wqm_quanta
+ */
+static ssize_t nes_show_wqm_quanta(struct device_driver *ddp, char *buf)
+{
+	u32 wqm_quanta_value = 0xdead;
+	u32 i = 0;
+	struct nes_device *nesdev;
+
+	list_for_each_entry(nesdev, &nes_dev_list, list) {
+		if (i == ee_flsh_adapter) {
+			wqm_quanta_value = nesdev->nesadapter->wqm_quanta;
+			break;
+		}
+		i++;
+	}
+
+	return  snprintf(buf, PAGE_SIZE, "0x%X\n", wqm_quanta);
+}
+
+
+/**
+ * nes_store_wqm_quanta
+ */
+static ssize_t nes_store_wqm_quanta(struct device_driver *ddp,
+					const char *buf, size_t count)
+{
+	unsigned long wqm_quanta_value;
+	u32 wqm_config1;
+	u32 i = 0;
+	struct nes_device *nesdev;
+
+	strict_strtoul(buf, 0, &wqm_quanta_value);
+	list_for_each_entry(nesdev, &nes_dev_list, list) {
+		if (i == ee_flsh_adapter) {
+			nesdev->nesadapter->wqm_quanta = wqm_quanta_value;
+			wqm_config1 = nes_read_indexed(nesdev,
+						NES_IDX_WQM_CONFIG1);
+			nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG1,
+					((wqm_quanta_value << 1) |
+					(wqm_config1 & 0x00000001)));
+			break;
+		}
+		i++;
+	}
+	return strnlen(buf, count);
+}
+
 static DRIVER_ATTR(adapter, S_IRUSR | S_IWUSR,
 		   nes_show_adapter, nes_store_adapter);
 static DRIVER_ATTR(eeprom_cmd, S_IRUSR | S_IWUSR,
@@ -1068,6 +1141,8 @@ static DRIVER_ATTR(idx_addr, S_IRUSR | S_IWUSR,
 		   nes_show_idx_addr, nes_store_idx_addr);
 static DRIVER_ATTR(idx_data, S_IRUSR | S_IWUSR,
 		   nes_show_idx_data, nes_store_idx_data);
+static DRIVER_ATTR(wqm_quanta, S_IRUSR | S_IWUSR,
+		   nes_show_wqm_quanta, nes_store_wqm_quanta);
 
 static int nes_create_driver_sysfs(struct pci_driver *drv)
 {
@@ -1081,6 +1156,7 @@ static int nes_create_driver_sysfs(struct pci_driver *drv)
 	error |= driver_create_file(&drv->driver, &driver_attr_nonidx_data);
 	error |= driver_create_file(&drv->driver, &driver_attr_idx_addr);
 	error |= driver_create_file(&drv->driver, &driver_attr_idx_data);
+	error |= driver_create_file(&drv->driver, &driver_attr_wqm_quanta);
 	return error;
 }
 
@@ -1095,6 +1171,7 @@ static void nes_remove_driver_sysfs(struct pci_driver *drv)
 	driver_remove_file(&drv->driver, &driver_attr_nonidx_data);
 	driver_remove_file(&drv->driver, &driver_attr_idx_addr);
 	driver_remove_file(&drv->driver, &driver_attr_idx_data);
+	driver_remove_file(&drv->driver, &driver_attr_wqm_quanta);
 }
 
 /**

drivers/infiniband/hw/nes/nes.h

@@ -169,7 +169,7 @@ extern int disable_mpa_crc;
 extern unsigned int send_first;
 extern unsigned int nes_drv_opt;
 extern unsigned int nes_debug_level;
-
+extern unsigned int wqm_quanta;
 extern struct list_head nes_adapter_list;
 
 extern atomic_t cm_connects;

drivers/infiniband/hw/nes/nes_cm.c

@@ -52,7 +52,7 @@
 #include <linux/random.h>
 #include <linux/list.h>
 #include <linux/threads.h>
-
+#include <net/arp.h>
 #include <net/neighbour.h>
 #include <net/route.h>
 #include <net/ip_fib.h>
@@ -1019,23 +1019,43 @@ static inline int mini_cm_accelerated(struct nes_cm_core *cm_core,
 
 
 /**
- * nes_addr_send_arp
+ * nes_addr_resolve_neigh
  */
-static void nes_addr_send_arp(u32 dst_ip)
+static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip)
 {
 	struct rtable *rt;
 	struct flowi fl;
+	struct neighbour *neigh;
+	int rc = -1;
+	DECLARE_MAC_BUF(mac);
 
 	memset(&fl, 0, sizeof fl);
 	fl.nl_u.ip4_u.daddr = htonl(dst_ip);
 	if (ip_route_output_key(&init_net, &rt, &fl)) {
 		printk("%s: ip_route_output_key failed for 0x%08X\n",
 				__func__, dst_ip);
-		return;
+		return rc;
 	}
 
+	neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, nesvnic->netdev);
+	if (neigh) {
+		if (neigh->nud_state & NUD_VALID) {
+			nes_debug(NES_DBG_CM, "Neighbor MAC address for 0x%08X"
+				  " is %s, Gateway is 0x%08X \n", dst_ip,
+				  print_mac(mac, neigh->ha), ntohl(rt->rt_gateway));
+			nes_manage_arp_cache(nesvnic->netdev, neigh->ha,
+					     dst_ip, NES_ARP_ADD);
+			rc = nes_arp_table(nesvnic->nesdev, dst_ip, NULL,
+					   NES_ARP_RESOLVE);
+		}
+		neigh_release(neigh);
+	}
+
+	if ((neigh == NULL) || (!(neigh->nud_state & NUD_VALID)))
 		neigh_event_send(rt->u.dst.neighbour, NULL);
+
 	ip_rt_put(rt);
+	return rc;
 }
 
 
@@ -1107,11 +1127,13 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
 	cm_node->loopbackpartner = NULL;
 	/* get the mac addr for the remote node */
 	arpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE);
+	if (arpindex < 0) {
+		arpindex = nes_addr_resolve_neigh(nesvnic, cm_info->rem_addr);
 		if (arpindex < 0) {
 			kfree(cm_node);
-		nes_addr_send_arp(cm_info->rem_addr);
 			return NULL;
 		}
+	}
 
 	/* copy the mac addr to node context */
 	memcpy(cm_node->rem_mac, nesadapter->arp_table[arpindex].mac_addr, ETH_ALEN);
@@ -1826,7 +1848,7 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
 /**
  * mini_cm_connect - make a connection node with params
  */
-struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
+static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
 	struct nes_vnic *nesvnic, u16 private_data_len,
 	void *private_data, struct nes_cm_info *cm_info)
 {
@@ -2007,7 +2029,6 @@ static int mini_cm_close(struct nes_cm_core *cm_core, struct nes_cm_node *cm_nod
 		ret = rem_ref_cm_node(cm_core, cm_node);
 		break;
 	}
-	cm_node->cm_id = NULL;
 	return ret;
 }
 

drivers/infiniband/hw/nes/nes_hw.h

@@ -156,6 +156,7 @@ enum indexed_regs {
 	NES_IDX_ENDNODE0_NSTAT_TX_OCTETS_HI = 0x7004,
 	NES_IDX_ENDNODE0_NSTAT_TX_FRAMES_LO = 0x7008,
 	NES_IDX_ENDNODE0_NSTAT_TX_FRAMES_HI = 0x700c,
+	NES_IDX_WQM_CONFIG1 = 0x5004,
 	NES_IDX_CM_CONFIG = 0x5100,
 	NES_IDX_NIC_LOGPORT_TO_PHYPORT = 0x6000,
 	NES_IDX_NIC_PHYPORT_TO_USW = 0x6008,
@@ -967,6 +968,7 @@ struct nes_arp_entry {
 #define DEFAULT_JUMBO_NES_QL_TARGET 40
 #define DEFAULT_JUMBO_NES_QL_HIGH   128
 #define NES_NIC_CQ_DOWNWARD_TREND   16
+#define NES_PFT_SIZE		    48
 
 struct nes_hw_tune_timer {
     /* u16 cq_count; */
@@ -1079,6 +1081,7 @@ struct nes_adapter {
 	u32 et_rx_max_coalesced_frames_high;
 	u32 et_rate_sample_interval;
 	u32 timer_int_limit;
+	u32 wqm_quanta;
 
 	/* Adapter base MAC address */
 	u32 mac_addr_low;
@@ -1094,12 +1097,14 @@ struct nes_adapter {
 	u16 pd_config_base[4];
 
 	u16 link_interrupt_count[4];
+	u8 crit_error_count[32];
 
 	/* the phy index for each port */
 	u8  phy_index[4];
 	u8  mac_sw_state[4];
 	u8  mac_link_down[4];
 	u8  phy_type[4];
+	u8  log_port;
 
 	/* PCI information */
 	unsigned int  devfn;
@@ -1113,6 +1118,7 @@ struct nes_adapter {
 	u8            virtwq;
 	u8            et_use_adaptive_rx_coalesce;
 	u8            adapter_fcn_count;
+	u8 pft_mcast_map[NES_PFT_SIZE];
 };
 
 struct nes_pbl {

drivers/infiniband/hw/nes/nes_nic.c

@@ -91,6 +91,7 @@ static struct nic_qp_map *nic_qp_mapping_per_function[] = {
 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK
 		| NETIF_MSG_IFUP | NETIF_MSG_IFDOWN;
 static int debug = -1;
+static int nics_per_function = 1;
 
 /**
  * nes_netdev_poll
@@ -201,7 +202,8 @@ static int nes_netdev_open(struct net_device *netdev)
 		nes_debug(NES_DBG_NETDEV, "i=%d, perfect filter table index= %d, PERF FILTER LOW"
 				" (Addr:%08X) = %08X, HIGH = %08X.\n",
 				i, nesvnic->qp_nic_index[i],
-				NES_IDX_PERFECT_FILTER_LOW+((nesvnic->perfect_filter_index + i) * 8),
+				NES_IDX_PERFECT_FILTER_LOW+
+					(nesvnic->qp_nic_index[i] * 8),
 				macaddr_low,
 				(u32)macaddr_high | NES_MAC_ADDR_VALID |
 				((((u32)nesvnic->nic_index) << 16)));
@@ -272,14 +274,18 @@ static int nes_netdev_stop(struct net_device *netdev)
 			break;
 	}
 
-	if (first_nesvnic->netdev_open == 0)
-		nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK+(0x200*nesdev->mac_index), 0xffffffff);
-	else if ((first_nesvnic != nesvnic) &&
-		 (PCI_FUNC(first_nesvnic->nesdev->pcidev->devfn) != PCI_FUNC(nesvnic->nesdev->pcidev->devfn))) {
-		nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK + (0x200 * nesdev->mac_index), 0xffffffff);
-		nes_write_indexed(first_nesvnic->nesdev, NES_IDX_MAC_INT_MASK + (0x200 * first_nesvnic->nesdev->mac_index),
+	if ((first_nesvnic->netdev_open == 1) && (first_nesvnic != nesvnic)  &&
+		(PCI_FUNC(first_nesvnic->nesdev->pcidev->devfn) !=
+		PCI_FUNC(nesvnic->nesdev->pcidev->devfn))) {
+			nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK+
+				(0x200*nesdev->mac_index), 0xffffffff);
+			nes_write_indexed(first_nesvnic->nesdev,
+				NES_IDX_MAC_INT_MASK+
+				(0x200*first_nesvnic->nesdev->mac_index),
 			~(NES_MAC_INT_LINK_STAT_CHG | NES_MAC_INT_XGMII_EXT |
 			NES_MAC_INT_TX_UNDERFLOW | NES_MAC_INT_TX_ERROR));
+	} else {
+		nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK+(0x200*nesdev->mac_index), 0xffffffff);
 	}
 
 	nic_active_mask = ~((u32)(1 << nesvnic->nic_index));
@@ -437,7 +443,7 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 	struct nes_hw_nic_sq_wqe *nic_sqe;
 	struct tcphdr *tcph;
 	/* struct udphdr *udph; */
-#define NES_MAX_TSO_FRAGS 18
+#define NES_MAX_TSO_FRAGS MAX_SKB_FRAGS
 	/* 64K segment plus overflow on each side */
 	dma_addr_t tso_bus_address[NES_MAX_TSO_FRAGS];
 	dma_addr_t bus_address;
@@ -605,6 +611,8 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 					wqe_fragment_length[wqe_fragment_index] = 0;
 					set_wqe_64bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_LOW_IDX,
 									bus_address);
+					tso_wqe_length += skb_headlen(skb) -
+							original_first_length;
 				}
 				while (wqe_fragment_index < 5) {
 					wqe_fragment_length[wqe_fragment_index] =
@@ -827,6 +835,7 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 {
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
 	struct nes_device *nesdev = nesvnic->nesdev;
+	struct nes_adapter *nesadapter = nesvnic->nesdev->nesadapter;
 	struct dev_mc_list *multicast_addr;
 	u32 nic_active_bit;
 	u32 nic_active;
@@ -836,7 +845,12 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 	u8 mc_all_on = 0;
 	u8 mc_index;
 	int mc_nic_index = -1;
+	u8 pft_entries_preallocated = max(nesadapter->adapter_fcn_count *
+					nics_per_function, 4);
+	u8 max_pft_entries_avaiable = NES_PFT_SIZE - pft_entries_preallocated;
+	unsigned long flags;
 
+	spin_lock_irqsave(&nesadapter->resource_lock, flags);
 	nic_active_bit = 1 << nesvnic->nic_index;
 
 	if (netdev->flags & IFF_PROMISC) {
@@ -847,7 +861,7 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 		nic_active |= nic_active_bit;
 		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
 		mc_all_on = 1;
-	} else if ((netdev->flags & IFF_ALLMULTI) || (netdev->mc_count > NES_MULTICAST_PF_MAX) ||
+	} else if ((netdev->flags & IFF_ALLMULTI) ||
 			   (nesvnic->nic_index > 3)) {
 		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL);
 		nic_active |= nic_active_bit;
@@ -866,17 +880,34 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 	}
 
 	nes_debug(NES_DBG_NIC_RX, "Number of MC entries = %d, Promiscous = %d, All Multicast = %d.\n",
-			  netdev->mc_count, (netdev->flags & IFF_PROMISC)?1:0,
-			  (netdev->flags & IFF_ALLMULTI)?1:0);
+		  netdev->mc_count, !!(netdev->flags & IFF_PROMISC),
+		  !!(netdev->flags & IFF_ALLMULTI));
 	if (!mc_all_on) {
 		multicast_addr = netdev->mc_list;
-		perfect_filter_register_address = NES_IDX_PERFECT_FILTER_LOW + 0x80;
-		perfect_filter_register_address += nesvnic->nic_index*0x40;
-		for (mc_index=0; mc_index < NES_MULTICAST_PF_MAX; mc_index++) {
-			while (multicast_addr && nesvnic->mcrq_mcast_filter && ((mc_nic_index = nesvnic->mcrq_mcast_filter(nesvnic, multicast_addr->dmi_addr)) == 0))
+		perfect_filter_register_address = NES_IDX_PERFECT_FILTER_LOW +
+						pft_entries_preallocated * 0x8;
+		for (mc_index = 0; mc_index < max_pft_entries_avaiable;
+		mc_index++) {
+			while (multicast_addr && nesvnic->mcrq_mcast_filter &&
+			((mc_nic_index = nesvnic->mcrq_mcast_filter(nesvnic,
+					multicast_addr->dmi_addr)) == 0)) {
 				multicast_addr = multicast_addr->next;
+			}
 			if (mc_nic_index < 0)
 				mc_nic_index = nesvnic->nic_index;
+			while (nesadapter->pft_mcast_map[mc_index] < 16 &&
+				nesadapter->pft_mcast_map[mc_index] !=
+					nesvnic->nic_index &&
+					mc_index < max_pft_entries_avaiable) {
+						nes_debug(NES_DBG_NIC_RX,
+					"mc_index=%d skipping nic_index=%d,\
+					used for=%d \n", mc_index,
+					nesvnic->nic_index,
+					nesadapter->pft_mcast_map[mc_index]);
+				mc_index++;
+			}
+			if (mc_index >= max_pft_entries_avaiable)
+				break;
 			if (multicast_addr) {
 				DECLARE_MAC_BUF(mac);
 				nes_debug(NES_DBG_NIC_RX, "Assigning MC Address %s to register 0x%04X nic_idx=%d\n",
@@ -897,15 +928,33 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 						(u32)macaddr_high | NES_MAC_ADDR_VALID |
 						((((u32)(1<<mc_nic_index)) << 16)));
 				multicast_addr = multicast_addr->next;
+				nesadapter->pft_mcast_map[mc_index] =
+							nesvnic->nic_index;
 			} else {
 				nes_debug(NES_DBG_NIC_RX, "Clearing MC Address at register 0x%04X\n",
 						  perfect_filter_register_address+(mc_index * 8));
 				nes_write_indexed(nesdev,
 						perfect_filter_register_address+4+(mc_index * 8),
 						0);
+				nesadapter->pft_mcast_map[mc_index] = 255;
 			}
 		}
+		/* PFT is not large enough */
+		if (multicast_addr && multicast_addr->next) {
+			nic_active = nes_read_indexed(nesdev,
+						NES_IDX_NIC_MULTICAST_ALL);
+			nic_active |= nic_active_bit;
+			nes_write_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL,
+								nic_active);
+			nic_active = nes_read_indexed(nesdev,
+						NES_IDX_NIC_UNICAST_ALL);
+			nic_active &= ~nic_active_bit;
+			nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL,
+								nic_active);
 		}
+	}
+
+	spin_unlock_irqrestore(&nesadapter->resource_lock, flags);
 }
 
 
@@ -918,6 +967,10 @@ static int nes_netdev_change_mtu(struct net_device *netdev, int new_mtu)
 	struct nes_device *nesdev = nesvnic->nesdev;
 	int ret = 0;
 	u8 jumbomode = 0;
+	u32 nic_active;
+	u32 nic_active_bit;
+	u32 uc_all_active;
+	u32 mc_all_active;
 
 	if ((new_mtu < ETH_ZLEN) || (new_mtu > max_mtu))
 		return -EINVAL;
@@ -931,8 +984,24 @@ static int nes_netdev_change_mtu(struct net_device *netdev, int new_mtu)
 	nes_nic_init_timer_defaults(nesdev, jumbomode);
 
 	if (netif_running(netdev)) {
+		nic_active_bit = 1 << nesvnic->nic_index;
+		mc_all_active = nes_read_indexed(nesdev,
+				NES_IDX_NIC_MULTICAST_ALL) & nic_active_bit;
+		uc_all_active = nes_read_indexed(nesdev,
+				NES_IDX_NIC_UNICAST_ALL)  & nic_active_bit;
+
 		nes_netdev_stop(netdev);
 		nes_netdev_open(netdev);
+
+		nic_active = nes_read_indexed(nesdev,
+					NES_IDX_NIC_MULTICAST_ALL);
+		nic_active |= mc_all_active;
+		nes_write_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL,
+							nic_active);
+
+		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL);
+		nic_active |= uc_all_active;
+		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
 	}
 
 	return ret;
@@ -1208,10 +1277,12 @@ static void nes_netdev_get_drvinfo(struct net_device *netdev,
 		struct ethtool_drvinfo *drvinfo)
 {
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
+	struct nes_adapter *nesadapter = nesvnic->nesdev->nesadapter;
 
 	strcpy(drvinfo->driver, DRV_NAME);
 	strcpy(drvinfo->bus_info, pci_name(nesvnic->nesdev->pcidev));
-	strcpy(drvinfo->fw_version, "TBD");
+	sprintf(drvinfo->fw_version, "%u.%u", nesadapter->firmware_version>>16,
+				nesadapter->firmware_version & 0x000000ff);
 	strcpy(drvinfo->version, DRV_VERSION);
 	drvinfo->n_stats = nes_netdev_get_stats_count(netdev);
 	drvinfo->testinfo_len = 0;
@@ -1587,7 +1658,9 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 			nesvnic, (unsigned long)netdev->features, nesvnic->nic.qp_id,
 			nesvnic->nic_index, nesvnic->logical_port,  nesdev->mac_index);
 
-	if (nesvnic->nesdev->nesadapter->port_count == 1) {
+	if (nesvnic->nesdev->nesadapter->port_count == 1 &&
+		nesvnic->nesdev->nesadapter->adapter_fcn_count == 1) {
+
 		nesvnic->qp_nic_index[0] = nesvnic->nic_index;
 		nesvnic->qp_nic_index[1] = nesvnic->nic_index + 1;
 		if (nes_drv_opt & NES_DRV_OPT_DUAL_LOGICAL_PORT) {
@@ -1598,9 +1671,12 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 			nesvnic->qp_nic_index[3] = nesvnic->nic_index + 3;
 		}
 	} else {
-		if (nesvnic->nesdev->nesadapter->port_count == 2) {
+		if (nesvnic->nesdev->nesadapter->port_count == 2 ||
+			(nesvnic->nesdev->nesadapter->port_count == 1 &&
+			nesvnic->nesdev->nesadapter->adapter_fcn_count == 2)) {
 				nesvnic->qp_nic_index[0] = nesvnic->nic_index;
-			nesvnic->qp_nic_index[1] = nesvnic->nic_index + 2;
+				nesvnic->qp_nic_index[1] = nesvnic->nic_index
+									+ 2;
 				nesvnic->qp_nic_index[2] = 0xf;
 				nesvnic->qp_nic_index[3] = 0xf;
 		} else {

drivers/infiniband/hw/nes/nes_verbs.c

@@ -1467,7 +1467,6 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 		default:
 			nes_debug(NES_DBG_QP, "Invalid QP type: %d\n", init_attr->qp_type);
 			return ERR_PTR(-EINVAL);
-			break;
 	}
 
 	/* update the QP table */
@@ -2498,7 +2497,6 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			nes_debug(NES_DBG_MR, "Leaving, ibmr=%p", ibmr);
 
 			return ibmr;
-			break;
 		case IWNES_MEMREG_TYPE_QP:
 		case IWNES_MEMREG_TYPE_CQ:
 			nespbl = kzalloc(sizeof(*nespbl), GFP_KERNEL);
@@ -2572,7 +2570,6 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			nesmr->ibmr.lkey = -1;
 			nesmr->mode = req.reg_type;
 			return &nesmr->ibmr;
-			break;
 	}
 
 	return ERR_PTR(-ENOSYS);

drivers/infiniband/ulp/ipoib/ipoib.h

@@ -268,10 +268,9 @@ struct ipoib_lro {
 };
 
 /*
- * Device private locking: tx_lock protects members used in TX fast
- * path (and we use LLTX so upper layers don't do extra locking).
- * lock protects everything else.  lock nests inside of tx_lock (ie
- * tx_lock must be acquired first if needed).
+ * Device private locking: network stack tx_lock protects members used
+ * in TX fast path, lock protects everything else.  lock nests inside
+ * of tx_lock (ie tx_lock must be acquired first if needed).
  */
 struct ipoib_dev_priv {
 	spinlock_t lock;
@@ -320,7 +319,6 @@ struct ipoib_dev_priv {
 
 	struct ipoib_rx_buf *rx_ring;
 
-	spinlock_t	     tx_lock;
 	struct ipoib_tx_buf *tx_ring;
 	unsigned	     tx_head;
 	unsigned	     tx_tail;

drivers/infiniband/ulp/ipoib/ipoib_cm.c

@@ -786,7 +786,8 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 
 	dev_kfree_skb_any(tx_req->skb);
 
-	spin_lock_irqsave(&priv->tx_lock, flags);
+	netif_tx_lock(dev);
+
 	++tx->tx_tail;
 	if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) &&
 	    netif_queue_stopped(dev) &&
@@ -801,7 +802,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 			   "(status=%d, wrid=%d vend_err %x)\n",
 			   wc->status, wr_id, wc->vendor_err);
 
-		spin_lock(&priv->lock);
+		spin_lock_irqsave(&priv->lock, flags);
 		neigh = tx->neigh;
 
 		if (neigh) {
@@ -821,10 +822,10 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 
 		clear_bit(IPOIB_FLAG_OPER_UP, &tx->flags);
 
-		spin_unlock(&priv->lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
 	}
 
-	spin_unlock_irqrestore(&priv->tx_lock, flags);
+	netif_tx_unlock(dev);
 }
 
 int ipoib_cm_dev_open(struct net_device *dev)
@@ -1149,7 +1150,6 @@ static void ipoib_cm_tx_destroy(struct ipoib_cm_tx *p)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
 	struct ipoib_cm_tx_buf *tx_req;
-	unsigned long flags;
 	unsigned long begin;
 
 	ipoib_dbg(priv, "Destroy active connection 0x%x head 0x%x tail 0x%x\n",
@@ -1180,12 +1180,12 @@ static void ipoib_cm_tx_destroy(struct ipoib_cm_tx *p)
 				    DMA_TO_DEVICE);
 		dev_kfree_skb_any(tx_req->skb);
 		++p->tx_tail;
-		spin_lock_irqsave(&priv->tx_lock, flags);
+		netif_tx_lock_bh(p->dev);
 		if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) &&
 		    netif_queue_stopped(p->dev) &&
 		    test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
 			netif_wake_queue(p->dev);
-		spin_unlock_irqrestore(&priv->tx_lock, flags);
+		netif_tx_unlock_bh(p->dev);
 	}
 
 	if (p->qp)
@@ -1202,6 +1202,7 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 	struct ipoib_dev_priv *priv = netdev_priv(tx->dev);
 	struct net_device *dev = priv->dev;
 	struct ipoib_neigh *neigh;
+	unsigned long flags;
 	int ret;
 
 	switch (event->event) {
@@ -1220,8 +1221,8 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 	case IB_CM_REJ_RECEIVED:
 	case IB_CM_TIMEWAIT_EXIT:
 		ipoib_dbg(priv, "CM error %d.\n", event->event);
-		spin_lock_irq(&priv->tx_lock);
-		spin_lock(&priv->lock);
+		netif_tx_lock_bh(dev);
+		spin_lock_irqsave(&priv->lock, flags);
 		neigh = tx->neigh;
 
 		if (neigh) {
@@ -1239,8 +1240,8 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 			queue_work(ipoib_workqueue, &priv->cm.reap_task);
 		}
 
-		spin_unlock(&priv->lock);
-		spin_unlock_irq(&priv->tx_lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		netif_tx_unlock_bh(dev);
 		break;
 	default:
 		break;
@@ -1294,19 +1295,24 @@ static void ipoib_cm_tx_start(struct work_struct *work)
 	struct ib_sa_path_rec pathrec;
 	u32 qpn;
 
-	spin_lock_irqsave(&priv->tx_lock, flags);
-	spin_lock(&priv->lock);
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
+
 	while (!list_empty(&priv->cm.start_list)) {
 		p = list_entry(priv->cm.start_list.next, typeof(*p), list);
 		list_del_init(&p->list);
 		neigh = p->neigh;
 		qpn = IPOIB_QPN(neigh->neighbour->ha);
 		memcpy(&pathrec, &p->path->pathrec, sizeof pathrec);
-		spin_unlock(&priv->lock);
-		spin_unlock_irqrestore(&priv->tx_lock, flags);
+
+		spin_unlock_irqrestore(&priv->lock, flags);
+		netif_tx_unlock_bh(dev);
+
 		ret = ipoib_cm_tx_init(p, qpn, &pathrec);
-		spin_lock_irqsave(&priv->tx_lock, flags);
-		spin_lock(&priv->lock);
+
+		netif_tx_lock_bh(dev);
+		spin_lock_irqsave(&priv->lock, flags);
+
 		if (ret) {
 			neigh = p->neigh;
 			if (neigh) {
@@ -1320,44 +1326,52 @@ static void ipoib_cm_tx_start(struct work_struct *work)
 			kfree(p);
 		}
 	}
-	spin_unlock(&priv->lock);
-	spin_unlock_irqrestore(&priv->tx_lock, flags);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
 static void ipoib_cm_tx_reap(struct work_struct *work)
 {
 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
 						   cm.reap_task);
+	struct net_device *dev = priv->dev;
 	struct ipoib_cm_tx *p;
+	unsigned long flags;
+
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
 
-	spin_lock_irq(&priv->tx_lock);
-	spin_lock(&priv->lock);
 	while (!list_empty(&priv->cm.reap_list)) {
 		p = list_entry(priv->cm.reap_list.next, typeof(*p), list);
 		list_del(&p->list);
-		spin_unlock(&priv->lock);
-		spin_unlock_irq(&priv->tx_lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		netif_tx_unlock_bh(dev);
 		ipoib_cm_tx_destroy(p);
-		spin_lock_irq(&priv->tx_lock);
-		spin_lock(&priv->lock);
+		netif_tx_lock_bh(dev);
+		spin_lock_irqsave(&priv->lock, flags);
 	}
-	spin_unlock(&priv->lock);
-	spin_unlock_irq(&priv->tx_lock);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
 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 long flags;
 	unsigned mtu = priv->mcast_mtu;
 
-	spin_lock_irq(&priv->tx_lock);
-	spin_lock(&priv->lock);
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
+
 	while ((skb = skb_dequeue(&priv->cm.skb_queue))) {
-		spin_unlock(&priv->lock);
-		spin_unlock_irq(&priv->tx_lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		netif_tx_unlock_bh(dev);
+
 		if (skb->protocol == htons(ETH_P_IP))
 			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
@@ -1365,11 +1379,13 @@ static void ipoib_cm_skb_reap(struct work_struct *work)
 			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, priv->dev);
 #endif
 		dev_kfree_skb_any(skb);
-		spin_lock_irq(&priv->tx_lock);
-		spin_lock(&priv->lock);
+
+		netif_tx_lock_bh(dev);
+		spin_lock_irqsave(&priv->lock, flags);
 	}
-	spin_unlock(&priv->lock);
-	spin_unlock_irq(&priv->tx_lock);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,

drivers/infiniband/ulp/ipoib/ipoib_ib.c

@@ -468,21 +468,22 @@ void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
 static void drain_tx_cq(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	unsigned long flags;
 
-	spin_lock_irqsave(&priv->tx_lock, flags);
+	netif_tx_lock(dev);
 	while (poll_tx(priv))
 		; /* nothing */
 
 	if (netif_queue_stopped(dev))
 		mod_timer(&priv->poll_timer, jiffies + 1);
 
-	spin_unlock_irqrestore(&priv->tx_lock, flags);
+	netif_tx_unlock(dev);
 }
 
 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr)
 {
-	drain_tx_cq((struct net_device *)dev_ptr);
+	struct ipoib_dev_priv *priv = netdev_priv(dev_ptr);
+
+	mod_timer(&priv->poll_timer, jiffies);
 }
 
 static inline int post_send(struct ipoib_dev_priv *priv,
@@ -614,17 +615,20 @@ static void __ipoib_reap_ah(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_ah *ah, *tah;
 	LIST_HEAD(remove_list);
+	unsigned long flags;
+
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
 
-	spin_lock_irq(&priv->tx_lock);
-	spin_lock(&priv->lock);
 	list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
 		if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
 			list_del(&ah->list);
 			ib_destroy_ah(ah->ah);
 			kfree(ah);
 		}
-	spin_unlock(&priv->lock);
-	spin_unlock_irq(&priv->tx_lock);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
 void ipoib_reap_ah(struct work_struct *work)
@@ -761,6 +765,14 @@ void ipoib_drain_cq(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int i, n;
+
+	/*
+	 * We call completion handling routines that expect to be
+	 * called from the BH-disabled NAPI poll context, so disable
+	 * BHs here too.
+	 */
+	local_bh_disable();
+
 	do {
 		n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc);
 		for (i = 0; i < n; ++i) {
@@ -784,6 +796,8 @@ void ipoib_drain_cq(struct net_device *dev)
 
 	while (poll_tx(priv))
 		; /* nothing */
+
+	local_bh_enable();
 }
 
 int ipoib_ib_dev_stop(struct net_device *dev, int flush)

drivers/infiniband/ulp/ipoib/ipoib_main.c

@@ -373,9 +373,10 @@ void ipoib_flush_paths(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_path *path, *tp;
 	LIST_HEAD(remove_list);
+	unsigned long flags;
 
-	spin_lock_irq(&priv->tx_lock);
-	spin_lock(&priv->lock);
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
 
 	list_splice_init(&priv->path_list, &remove_list);
 
@@ -385,15 +386,16 @@ void ipoib_flush_paths(struct net_device *dev)
 	list_for_each_entry_safe(path, tp, &remove_list, list) {
 		if (path->query)
 			ib_sa_cancel_query(path->query_id, path->query);
-		spin_unlock(&priv->lock);
-		spin_unlock_irq(&priv->tx_lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		netif_tx_unlock_bh(dev);
 		wait_for_completion(&path->done);
 		path_free(dev, path);
-		spin_lock_irq(&priv->tx_lock);
-		spin_lock(&priv->lock);
+		netif_tx_lock_bh(dev);
+		spin_lock_irqsave(&priv->lock, flags);
 	}
-	spin_unlock(&priv->lock);
-	spin_unlock_irq(&priv->tx_lock);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
 static void path_rec_completion(int status,
@@ -404,7 +406,7 @@ static void path_rec_completion(int status,
 	struct net_device *dev = path->dev;
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_ah *ah = NULL;
-	struct ipoib_ah *old_ah;
+	struct ipoib_ah *old_ah = NULL;
 	struct ipoib_neigh *neigh, *tn;
 	struct sk_buff_head skqueue;
 	struct sk_buff *skb;
@@ -428,12 +430,12 @@ static void path_rec_completion(int status,
 
 	spin_lock_irqsave(&priv->lock, flags);
 
-	old_ah   = path->ah;
-	path->ah = ah;
-
 	if (ah) {
 		path->pathrec = *pathrec;
 
+		old_ah   = path->ah;
+		path->ah = ah;
+
 		ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
 			  ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
 
@@ -555,6 +557,7 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_path *path;
 	struct ipoib_neigh *neigh;
+	unsigned long flags;
 
 	neigh = ipoib_neigh_alloc(skb->dst->neighbour, skb->dev);
 	if (!neigh) {
@@ -563,11 +566,7 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
 		return;
 	}
 
-	/*
-	 * We can only be called from ipoib_start_xmit, so we're
-	 * inside tx_lock -- no need to save/restore flags.
-	 */
-	spin_lock(&priv->lock);
+	spin_lock_irqsave(&priv->lock, flags);
 
 	path = __path_find(dev, skb->dst->neighbour->ha + 4);
 	if (!path) {
@@ -614,7 +613,7 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
 		__skb_queue_tail(&neigh->queue, skb);
 	}
 
-	spin_unlock(&priv->lock);
+	spin_unlock_irqrestore(&priv->lock, flags);
 	return;
 
 err_list:
@@ -626,7 +625,7 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
 	++dev->stats.tx_dropped;
 	dev_kfree_skb_any(skb);
 
-	spin_unlock(&priv->lock);
+	spin_unlock_irqrestore(&priv->lock, flags);
 }
 
 static void ipoib_path_lookup(struct sk_buff *skb, struct net_device *dev)
@@ -650,12 +649,9 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_path *path;
+	unsigned long flags;
 
-	/*
-	 * We can only be called from ipoib_start_xmit, so we're
-	 * inside tx_lock -- no need to save/restore flags.
-	 */
-	spin_lock(&priv->lock);
+	spin_lock_irqsave(&priv->lock, flags);
 
 	path = __path_find(dev, phdr->hwaddr + 4);
 	if (!path || !path->valid) {
@@ -667,7 +663,7 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 			__skb_queue_tail(&path->queue, skb);
 
 			if (path_rec_start(dev, path)) {
-				spin_unlock(&priv->lock);
+				spin_unlock_irqrestore(&priv->lock, flags);
 				path_free(dev, path);
 				return;
 			} else
@@ -677,7 +673,7 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 			dev_kfree_skb_any(skb);
 		}
 
-		spin_unlock(&priv->lock);
+		spin_unlock_irqrestore(&priv->lock, flags);
 		return;
 	}
 
@@ -696,7 +692,7 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 		dev_kfree_skb_any(skb);
 	}
 
-	spin_unlock(&priv->lock);
+	spin_unlock_irqrestore(&priv->lock, flags);
 }
 
 static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
@@ -705,13 +701,10 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct ipoib_neigh *neigh;
 	unsigned long flags;
 
-	if (unlikely(!spin_trylock_irqsave(&priv->tx_lock, flags)))
-		return NETDEV_TX_LOCKED;
-
 	if (likely(skb->dst && skb->dst->neighbour)) {
 		if (unlikely(!*to_ipoib_neigh(skb->dst->neighbour))) {
 			ipoib_path_lookup(skb, dev);
-			goto out;
+			return NETDEV_TX_OK;
 		}
 
 		neigh = *to_ipoib_neigh(skb->dst->neighbour);
@@ -721,7 +714,7 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 					    skb->dst->neighbour->ha + 4,
 					    sizeof(union ib_gid))) ||
 					 (neigh->dev != dev))) {
-				spin_lock(&priv->lock);
+				spin_lock_irqsave(&priv->lock, flags);
 				/*
 				 * It's safe to call ipoib_put_ah() inside
 				 * priv->lock here, because we know that
@@ -732,25 +725,25 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 				ipoib_put_ah(neigh->ah);
 				list_del(&neigh->list);
 				ipoib_neigh_free(dev, neigh);
-				spin_unlock(&priv->lock);
+				spin_unlock_irqrestore(&priv->lock, flags);
 				ipoib_path_lookup(skb, dev);
-				goto out;
+				return NETDEV_TX_OK;
 			}
 
 		if (ipoib_cm_get(neigh)) {
 			if (ipoib_cm_up(neigh)) {
 				ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
-				goto out;
+				return NETDEV_TX_OK;
 			}
 		} else if (neigh->ah) {
 			ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(skb->dst->neighbour->ha));
-			goto out;
+			return NETDEV_TX_OK;
 		}
 
 		if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
-			spin_lock(&priv->lock);
+			spin_lock_irqsave(&priv->lock, flags);
 			__skb_queue_tail(&neigh->queue, skb);
-			spin_unlock(&priv->lock);
+			spin_unlock_irqrestore(&priv->lock, flags);
 		} else {
 			++dev->stats.tx_dropped;
 			dev_kfree_skb_any(skb);
@@ -779,16 +772,13 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 					   IPOIB_GID_RAW_ARG(phdr->hwaddr + 4));
 				dev_kfree_skb_any(skb);
 				++dev->stats.tx_dropped;
-				goto out;
+				return NETDEV_TX_OK;
 			}
 
 			unicast_arp_send(skb, dev, phdr);
 		}
 	}
 
-out:
-	spin_unlock_irqrestore(&priv->tx_lock, flags);
-
 	return NETDEV_TX_OK;
 }
 
@@ -1052,7 +1042,6 @@ static void ipoib_setup(struct net_device *dev)
 	dev->type		 = ARPHRD_INFINIBAND;
 	dev->tx_queue_len	 = ipoib_sendq_size * 2;
 	dev->features		 = (NETIF_F_VLAN_CHALLENGED	|
-				    NETIF_F_LLTX		|
 				    NETIF_F_HIGHDMA);
 
 	memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
@@ -1064,7 +1053,6 @@ static void ipoib_setup(struct net_device *dev)
 	ipoib_lro_setup(priv);
 
 	spin_lock_init(&priv->lock);
-	spin_lock_init(&priv->tx_lock);
 
 	mutex_init(&priv->vlan_mutex);
 

drivers/infiniband/ulp/ipoib/ipoib_multicast.c

@@ -69,14 +69,13 @@ static void ipoib_mcast_free(struct ipoib_mcast *mcast)
 	struct net_device *dev = mcast->dev;
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_neigh *neigh, *tmp;
-	unsigned long flags;
 	int tx_dropped = 0;
 
 	ipoib_dbg_mcast(netdev_priv(dev),
 			"deleting multicast group " IPOIB_GID_FMT "\n",
 			IPOIB_GID_ARG(mcast->mcmember.mgid));
 
-	spin_lock_irqsave(&priv->lock, flags);
+	spin_lock_irq(&priv->lock);
 
 	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
 		/*
@@ -90,7 +89,7 @@ static void ipoib_mcast_free(struct ipoib_mcast *mcast)
 		ipoib_neigh_free(dev, neigh);
 	}
 
-	spin_unlock_irqrestore(&priv->lock, flags);
+	spin_unlock_irq(&priv->lock);
 
 	if (mcast->ah)
 		ipoib_put_ah(mcast->ah);
@@ -100,9 +99,9 @@ static void ipoib_mcast_free(struct ipoib_mcast *mcast)
 		dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
 	}
 
-	spin_lock_irqsave(&priv->tx_lock, flags);
+	netif_tx_lock_bh(dev);
 	dev->stats.tx_dropped += tx_dropped;
-	spin_unlock_irqrestore(&priv->tx_lock, flags);
+	netif_tx_unlock_bh(dev);
 
 	kfree(mcast);
 }
@@ -259,10 +258,10 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
 	}
 
 	/* actually send any queued packets */
-	spin_lock_irq(&priv->tx_lock);
+	netif_tx_lock_bh(dev);
 	while (!skb_queue_empty(&mcast->pkt_queue)) {
 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
-		spin_unlock_irq(&priv->tx_lock);
+		netif_tx_unlock_bh(dev);
 
 		skb->dev = dev;
 
@@ -273,9 +272,9 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
 
 		if (dev_queue_xmit(skb))
 			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
-		spin_lock_irq(&priv->tx_lock);
+		netif_tx_lock_bh(dev);
 	}
-	spin_unlock_irq(&priv->tx_lock);
+	netif_tx_unlock_bh(dev);
 
 	return 0;
 }
@@ -286,7 +285,6 @@ ipoib_mcast_sendonly_join_complete(int status,
 {
 	struct ipoib_mcast *mcast = multicast->context;
 	struct net_device *dev = mcast->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
 	/* We trap for port events ourselves. */
 	if (status == -ENETRESET)
@@ -302,12 +300,12 @@ ipoib_mcast_sendonly_join_complete(int status,
 					IPOIB_GID_ARG(mcast->mcmember.mgid), status);
 
 		/* Flush out any queued packets */
-		spin_lock_irq(&priv->tx_lock);
+		netif_tx_lock_bh(dev);
 		while (!skb_queue_empty(&mcast->pkt_queue)) {
 			++dev->stats.tx_dropped;
 			dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
 		}
-		spin_unlock_irq(&priv->tx_lock);
+		netif_tx_unlock_bh(dev);
 
 		/* Clear the busy flag so we try again */
 		status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
@@ -662,12 +660,9 @@ void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_mcast *mcast;
+	unsigned long flags;
 
-	/*
-	 * We can only be called from ipoib_start_xmit, so we're
-	 * inside tx_lock -- no need to save/restore flags.
-	 */
-	spin_lock(&priv->lock);
+	spin_lock_irqsave(&priv->lock, flags);
 
 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)		||
 	    !priv->broadcast					||
@@ -738,7 +733,7 @@ void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
 	}
 
 unlock:
-	spin_unlock(&priv->lock);
+	spin_unlock_irqrestore(&priv->lock, flags);
 }
 
 void ipoib_mcast_dev_flush(struct net_device *dev)