aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--drivers/infiniband/core/cm.c63
-rw-r--r--drivers/infiniband/core/cm_msgs.h32
-rw-r--r--drivers/infiniband/core/cma.c67
-rw-r--r--drivers/infiniband/core/mad.c3
-rw-r--r--drivers/infiniband/core/sysfs.c26
-rw-r--r--drivers/infiniband/core/ucm.c2
-rw-r--r--drivers/infiniband/core/ucma.c7
-rw-r--r--drivers/infiniband/core/user_mad.c5
-rw-r--r--drivers/infiniband/core/uverbs.h18
-rw-r--r--drivers/infiniband/core/uverbs_cmd.c703
-rw-r--r--drivers/infiniband/core/uverbs_main.c30
-rw-r--r--drivers/infiniband/core/verbs.c376
-rw-r--r--drivers/infiniband/hw/amso1100/c2_ae.c5
-rw-r--r--drivers/infiniband/hw/amso1100/c2_intr.c5
-rw-r--r--drivers/infiniband/hw/amso1100/c2_provider.c5
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_cm.c10
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_ev.c6
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_provider.c1
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_provider.h1
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_qp.c14
-rw-r--r--drivers/infiniband/hw/cxgb4/cm.c469
-rw-r--r--drivers/infiniband/hw/cxgb4/cq.c3
-rw-r--r--drivers/infiniband/hw/cxgb4/device.c41
-rw-r--r--drivers/infiniband/hw/cxgb4/ev.c10
-rw-r--r--drivers/infiniband/hw/cxgb4/iw_cxgb4.h23
-rw-r--r--drivers/infiniband/hw/cxgb4/qp.c39
-rw-r--r--drivers/infiniband/hw/ehca/ehca_eq.c4
-rw-r--r--drivers/infiniband/hw/ehca/ehca_qp.c3
-rw-r--r--drivers/infiniband/hw/ipath/ipath_init_chip.c1
-rw-r--r--drivers/infiniband/hw/ipath/ipath_srq.c5
-rw-r--r--drivers/infiniband/hw/mlx4/main.c106
-rw-r--r--drivers/infiniband/hw/mlx4/mlx4_ib.h13
-rw-r--r--drivers/infiniband/hw/mlx4/qp.c131
-rw-r--r--drivers/infiniband/hw/mlx4/srq.c10
-rw-r--r--drivers/infiniband/hw/mthca/mthca_provider.c3
-rw-r--r--drivers/infiniband/hw/nes/nes_cm.c1102
-rw-r--r--drivers/infiniband/hw/nes/nes_cm.h73
-rw-r--r--drivers/infiniband/hw/nes/nes_verbs.h3
-rw-r--r--drivers/infiniband/hw/qib/qib.h15
-rw-r--r--drivers/infiniband/hw/qib/qib_driver.c20
-rw-r--r--drivers/infiniband/hw/qib/qib_file_ops.c2
-rw-r--r--drivers/infiniband/hw/qib/qib_iba6120.c2
-rw-r--r--drivers/infiniband/hw/qib/qib_iba7220.c2
-rw-r--r--drivers/infiniband/hw/qib/qib_iba7322.c135
-rw-r--r--drivers/infiniband/hw/qib/qib_init.c8
-rw-r--r--drivers/infiniband/hw/qib/qib_qp.c90
-rw-r--r--drivers/infiniband/hw/qib/qib_qsfp.c25
-rw-r--r--drivers/infiniband/hw/qib/qib_qsfp.h3
-rw-r--r--drivers/infiniband/hw/qib/qib_rc.c36
-rw-r--r--drivers/infiniband/hw/qib/qib_ruc.c7
-rw-r--r--drivers/infiniband/hw/qib/qib_srq.c5
-rw-r--r--drivers/infiniband/hw/qib/qib_sysfs.c3
-rw-r--r--drivers/infiniband/hw/qib/qib_uc.c25
-rw-r--r--drivers/infiniband/hw/qib/qib_verbs.c36
-rw-r--r--drivers/infiniband/hw/qib/qib_verbs.h5
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_cm.c5
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_fs.c7
-rw-r--r--drivers/net/mlx4/eq.c2
-rw-r--r--drivers/net/mlx4/fw.c6
-rw-r--r--drivers/net/mlx4/fw.h2
-rw-r--r--drivers/net/mlx4/main.c36
-rw-r--r--drivers/net/mlx4/mlx4.h4
-rw-r--r--drivers/net/mlx4/mr.c2
-rw-r--r--drivers/net/mlx4/pd.c30
-rw-r--r--drivers/net/mlx4/port.c72
-rw-r--r--drivers/net/mlx4/qp.c3
-rw-r--r--drivers/net/mlx4/srq.c20
-rw-r--r--include/linux/mlx4/device.h16
-rw-r--r--include/linux/mlx4/qp.h3
-rw-r--r--include/rdma/ib_user_verbs.h48
-rw-r--r--include/rdma/ib_verbs.h106
-rw-r--r--include/rdma/iw_cm.h4
-rw-r--r--include/rdma/rdma_cm.h1
-rw-r--r--include/rdma/rdma_user_cm.h3
74 files changed, 3282 insertions, 925 deletions
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
index fc0f2bd9ca82..4104ea2427c2 100644
--- a/drivers/infiniband/core/cm.c
+++ b/drivers/infiniband/core/cm.c
@@ -889,6 +889,8 @@ retest:
889 break; 889 break;
890 case IB_CM_ESTABLISHED: 890 case IB_CM_ESTABLISHED:
891 spin_unlock_irq(&cm_id_priv->lock); 891 spin_unlock_irq(&cm_id_priv->lock);
892 if (cm_id_priv->qp_type == IB_QPT_XRC_TGT)
893 break;
892 ib_send_cm_dreq(cm_id, NULL, 0); 894 ib_send_cm_dreq(cm_id, NULL, 0);
893 goto retest; 895 goto retest;
894 case IB_CM_DREQ_SENT: 896 case IB_CM_DREQ_SENT:
@@ -1008,7 +1010,6 @@ static void cm_format_req(struct cm_req_msg *req_msg,
1008 req_msg->service_id = param->service_id; 1010 req_msg->service_id = param->service_id;
1009 req_msg->local_ca_guid = cm_id_priv->id.device->node_guid; 1011 req_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
1010 cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num)); 1012 cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num));
1011 cm_req_set_resp_res(req_msg, param->responder_resources);
1012 cm_req_set_init_depth(req_msg, param->initiator_depth); 1013 cm_req_set_init_depth(req_msg, param->initiator_depth);
1013 cm_req_set_remote_resp_timeout(req_msg, 1014 cm_req_set_remote_resp_timeout(req_msg,
1014 param->remote_cm_response_timeout); 1015 param->remote_cm_response_timeout);
@@ -1017,12 +1018,16 @@ static void cm_format_req(struct cm_req_msg *req_msg,
1017 cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn)); 1018 cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn));
1018 cm_req_set_local_resp_timeout(req_msg, 1019 cm_req_set_local_resp_timeout(req_msg,
1019 param->local_cm_response_timeout); 1020 param->local_cm_response_timeout);
1020 cm_req_set_retry_count(req_msg, param->retry_count);
1021 req_msg->pkey = param->primary_path->pkey; 1021 req_msg->pkey = param->primary_path->pkey;
1022 cm_req_set_path_mtu(req_msg, param->primary_path->mtu); 1022 cm_req_set_path_mtu(req_msg, param->primary_path->mtu);
1023 cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
1024 cm_req_set_max_cm_retries(req_msg, param->max_cm_retries); 1023 cm_req_set_max_cm_retries(req_msg, param->max_cm_retries);
1025 cm_req_set_srq(req_msg, param->srq); 1024
1025 if (param->qp_type != IB_QPT_XRC_INI) {
1026 cm_req_set_resp_res(req_msg, param->responder_resources);
1027 cm_req_set_retry_count(req_msg, param->retry_count);
1028 cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
1029 cm_req_set_srq(req_msg, param->srq);
1030 }
1026 1031
1027 if (pri_path->hop_limit <= 1) { 1032 if (pri_path->hop_limit <= 1) {
1028 req_msg->primary_local_lid = pri_path->slid; 1033 req_msg->primary_local_lid = pri_path->slid;
@@ -1080,7 +1085,8 @@ static int cm_validate_req_param(struct ib_cm_req_param *param)
1080 if (!param->primary_path) 1085 if (!param->primary_path)
1081 return -EINVAL; 1086 return -EINVAL;
1082 1087
1083 if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC) 1088 if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC &&
1089 param->qp_type != IB_QPT_XRC_INI)
1084 return -EINVAL; 1090 return -EINVAL;
1085 1091
1086 if (param->private_data && 1092 if (param->private_data &&
@@ -1601,18 +1607,24 @@ static void cm_format_rep(struct cm_rep_msg *rep_msg,
1601 cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid); 1607 cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid);
1602 rep_msg->local_comm_id = cm_id_priv->id.local_id; 1608 rep_msg->local_comm_id = cm_id_priv->id.local_id;
1603 rep_msg->remote_comm_id = cm_id_priv->id.remote_id; 1609 rep_msg->remote_comm_id = cm_id_priv->id.remote_id;
1604 cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
1605 cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn)); 1610 cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn));
1606 rep_msg->resp_resources = param->responder_resources; 1611 rep_msg->resp_resources = param->responder_resources;
1607 rep_msg->initiator_depth = param->initiator_depth;
1608 cm_rep_set_target_ack_delay(rep_msg, 1612 cm_rep_set_target_ack_delay(rep_msg,
1609 cm_id_priv->av.port->cm_dev->ack_delay); 1613 cm_id_priv->av.port->cm_dev->ack_delay);
1610 cm_rep_set_failover(rep_msg, param->failover_accepted); 1614 cm_rep_set_failover(rep_msg, param->failover_accepted);
1611 cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
1612 cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count); 1615 cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count);
1613 cm_rep_set_srq(rep_msg, param->srq);
1614 rep_msg->local_ca_guid = cm_id_priv->id.device->node_guid; 1616 rep_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
1615 1617
1618 if (cm_id_priv->qp_type != IB_QPT_XRC_TGT) {
1619 rep_msg->initiator_depth = param->initiator_depth;
1620 cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
1621 cm_rep_set_srq(rep_msg, param->srq);
1622 cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
1623 } else {
1624 cm_rep_set_srq(rep_msg, 1);
1625 cm_rep_set_local_eecn(rep_msg, cpu_to_be32(param->qp_num));
1626 }
1627
1616 if (param->private_data && param->private_data_len) 1628 if (param->private_data && param->private_data_len)
1617 memcpy(rep_msg->private_data, param->private_data, 1629 memcpy(rep_msg->private_data, param->private_data,
1618 param->private_data_len); 1630 param->private_data_len);
@@ -1660,7 +1672,7 @@ int ib_send_cm_rep(struct ib_cm_id *cm_id,
1660 cm_id_priv->initiator_depth = param->initiator_depth; 1672 cm_id_priv->initiator_depth = param->initiator_depth;
1661 cm_id_priv->responder_resources = param->responder_resources; 1673 cm_id_priv->responder_resources = param->responder_resources;
1662 cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg); 1674 cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg);
1663 cm_id_priv->local_qpn = cm_rep_get_local_qpn(rep_msg); 1675 cm_id_priv->local_qpn = cpu_to_be32(param->qp_num & 0xFFFFFF);
1664 1676
1665out: spin_unlock_irqrestore(&cm_id_priv->lock, flags); 1677out: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1666 return ret; 1678 return ret;
@@ -1731,7 +1743,7 @@ error: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1731} 1743}
1732EXPORT_SYMBOL(ib_send_cm_rtu); 1744EXPORT_SYMBOL(ib_send_cm_rtu);
1733 1745
1734static void cm_format_rep_event(struct cm_work *work) 1746static void cm_format_rep_event(struct cm_work *work, enum ib_qp_type qp_type)
1735{ 1747{
1736 struct cm_rep_msg *rep_msg; 1748 struct cm_rep_msg *rep_msg;
1737 struct ib_cm_rep_event_param *param; 1749 struct ib_cm_rep_event_param *param;
@@ -1740,7 +1752,7 @@ static void cm_format_rep_event(struct cm_work *work)
1740 param = &work->cm_event.param.rep_rcvd; 1752 param = &work->cm_event.param.rep_rcvd;
1741 param->remote_ca_guid = rep_msg->local_ca_guid; 1753 param->remote_ca_guid = rep_msg->local_ca_guid;
1742 param->remote_qkey = be32_to_cpu(rep_msg->local_qkey); 1754 param->remote_qkey = be32_to_cpu(rep_msg->local_qkey);
1743 param->remote_qpn = be32_to_cpu(cm_rep_get_local_qpn(rep_msg)); 1755 param->remote_qpn = be32_to_cpu(cm_rep_get_qpn(rep_msg, qp_type));
1744 param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg)); 1756 param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg));
1745 param->responder_resources = rep_msg->initiator_depth; 1757 param->responder_resources = rep_msg->initiator_depth;
1746 param->initiator_depth = rep_msg->resp_resources; 1758 param->initiator_depth = rep_msg->resp_resources;
@@ -1808,7 +1820,7 @@ static int cm_rep_handler(struct cm_work *work)
1808 return -EINVAL; 1820 return -EINVAL;
1809 } 1821 }
1810 1822
1811 cm_format_rep_event(work); 1823 cm_format_rep_event(work, cm_id_priv->qp_type);
1812 1824
1813 spin_lock_irq(&cm_id_priv->lock); 1825 spin_lock_irq(&cm_id_priv->lock);
1814 switch (cm_id_priv->id.state) { 1826 switch (cm_id_priv->id.state) {
@@ -1823,7 +1835,7 @@ static int cm_rep_handler(struct cm_work *work)
1823 1835
1824 cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id; 1836 cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id;
1825 cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid; 1837 cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid;
1826 cm_id_priv->timewait_info->remote_qpn = cm_rep_get_local_qpn(rep_msg); 1838 cm_id_priv->timewait_info->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type);
1827 1839
1828 spin_lock(&cm.lock); 1840 spin_lock(&cm.lock);
1829 /* Check for duplicate REP. */ 1841 /* Check for duplicate REP. */
@@ -1850,7 +1862,7 @@ static int cm_rep_handler(struct cm_work *work)
1850 1862
1851 cm_id_priv->id.state = IB_CM_REP_RCVD; 1863 cm_id_priv->id.state = IB_CM_REP_RCVD;
1852 cm_id_priv->id.remote_id = rep_msg->local_comm_id; 1864 cm_id_priv->id.remote_id = rep_msg->local_comm_id;
1853 cm_id_priv->remote_qpn = cm_rep_get_local_qpn(rep_msg); 1865 cm_id_priv->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type);
1854 cm_id_priv->initiator_depth = rep_msg->resp_resources; 1866 cm_id_priv->initiator_depth = rep_msg->resp_resources;
1855 cm_id_priv->responder_resources = rep_msg->initiator_depth; 1867 cm_id_priv->responder_resources = rep_msg->initiator_depth;
1856 cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg); 1868 cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg);
@@ -3492,7 +3504,8 @@ static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
3492 qp_attr->path_mtu = cm_id_priv->path_mtu; 3504 qp_attr->path_mtu = cm_id_priv->path_mtu;
3493 qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn); 3505 qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn);
3494 qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn); 3506 qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn);
3495 if (cm_id_priv->qp_type == IB_QPT_RC) { 3507 if (cm_id_priv->qp_type == IB_QPT_RC ||
3508 cm_id_priv->qp_type == IB_QPT_XRC_TGT) {
3496 *qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC | 3509 *qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC |
3497 IB_QP_MIN_RNR_TIMER; 3510 IB_QP_MIN_RNR_TIMER;
3498 qp_attr->max_dest_rd_atomic = 3511 qp_attr->max_dest_rd_atomic =
@@ -3537,15 +3550,21 @@ static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
3537 if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT) { 3550 if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT) {
3538 *qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN; 3551 *qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN;
3539 qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn); 3552 qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn);
3540 if (cm_id_priv->qp_type == IB_QPT_RC) { 3553 switch (cm_id_priv->qp_type) {
3541 *qp_attr_mask |= IB_QP_TIMEOUT | IB_QP_RETRY_CNT | 3554 case IB_QPT_RC:
3542 IB_QP_RNR_RETRY | 3555 case IB_QPT_XRC_INI:
3556 *qp_attr_mask |= IB_QP_RETRY_CNT | IB_QP_RNR_RETRY |
3543 IB_QP_MAX_QP_RD_ATOMIC; 3557 IB_QP_MAX_QP_RD_ATOMIC;
3544 qp_attr->timeout = cm_id_priv->av.timeout;
3545 qp_attr->retry_cnt = cm_id_priv->retry_count; 3558 qp_attr->retry_cnt = cm_id_priv->retry_count;
3546 qp_attr->rnr_retry = cm_id_priv->rnr_retry_count; 3559 qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
3547 qp_attr->max_rd_atomic = 3560 qp_attr->max_rd_atomic = cm_id_priv->initiator_depth;
3548 cm_id_priv->initiator_depth; 3561 /* fall through */
3562 case IB_QPT_XRC_TGT:
3563 *qp_attr_mask |= IB_QP_TIMEOUT;
3564 qp_attr->timeout = cm_id_priv->av.timeout;
3565 break;
3566 default:
3567 break;
3549 } 3568 }
3550 if (cm_id_priv->alt_av.ah_attr.dlid) { 3569 if (cm_id_priv->alt_av.ah_attr.dlid) {
3551 *qp_attr_mask |= IB_QP_PATH_MIG_STATE; 3570 *qp_attr_mask |= IB_QP_PATH_MIG_STATE;
diff --git a/drivers/infiniband/core/cm_msgs.h b/drivers/infiniband/core/cm_msgs.h
index 7e63c08f697c..505db2a59e7f 100644
--- a/drivers/infiniband/core/cm_msgs.h
+++ b/drivers/infiniband/core/cm_msgs.h
@@ -1,5 +1,5 @@
1/* 1/*
2 * Copyright (c) 2004 Intel Corporation. All rights reserved. 2 * Copyright (c) 2004, 2011 Intel Corporation. All rights reserved.
3 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 3 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
4 * Copyright (c) 2004 Voltaire Corporation. All rights reserved. 4 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
5 * 5 *
@@ -86,7 +86,7 @@ struct cm_req_msg {
86 __be16 pkey; 86 __be16 pkey;
87 /* path MTU:4, RDC exists:1, RNR retry count:3. */ 87 /* path MTU:4, RDC exists:1, RNR retry count:3. */
88 u8 offset50; 88 u8 offset50;
89 /* max CM Retries:4, SRQ:1, rsvd:3 */ 89 /* max CM Retries:4, SRQ:1, extended transport type:3 */
90 u8 offset51; 90 u8 offset51;
91 91
92 __be16 primary_local_lid; 92 __be16 primary_local_lid;
@@ -175,6 +175,11 @@ static inline enum ib_qp_type cm_req_get_qp_type(struct cm_req_msg *req_msg)
175 switch(transport_type) { 175 switch(transport_type) {
176 case 0: return IB_QPT_RC; 176 case 0: return IB_QPT_RC;
177 case 1: return IB_QPT_UC; 177 case 1: return IB_QPT_UC;
178 case 3:
179 switch (req_msg->offset51 & 0x7) {
180 case 1: return IB_QPT_XRC_TGT;
181 default: return 0;
182 }
178 default: return 0; 183 default: return 0;
179 } 184 }
180} 185}
@@ -188,6 +193,12 @@ static inline void cm_req_set_qp_type(struct cm_req_msg *req_msg,
188 req_msg->offset40) & 193 req_msg->offset40) &
189 0xFFFFFFF9) | 0x2); 194 0xFFFFFFF9) | 0x2);
190 break; 195 break;
196 case IB_QPT_XRC_INI:
197 req_msg->offset40 = cpu_to_be32((be32_to_cpu(
198 req_msg->offset40) &
199 0xFFFFFFF9) | 0x6);
200 req_msg->offset51 = (req_msg->offset51 & 0xF8) | 1;
201 break;
191 default: 202 default:
192 req_msg->offset40 = cpu_to_be32(be32_to_cpu( 203 req_msg->offset40 = cpu_to_be32(be32_to_cpu(
193 req_msg->offset40) & 204 req_msg->offset40) &
@@ -527,6 +538,23 @@ static inline void cm_rep_set_local_qpn(struct cm_rep_msg *rep_msg, __be32 qpn)
527 (be32_to_cpu(rep_msg->offset12) & 0x000000FF)); 538 (be32_to_cpu(rep_msg->offset12) & 0x000000FF));
528} 539}
529 540
541static inline __be32 cm_rep_get_local_eecn(struct cm_rep_msg *rep_msg)
542{
543 return cpu_to_be32(be32_to_cpu(rep_msg->offset16) >> 8);
544}
545
546static inline void cm_rep_set_local_eecn(struct cm_rep_msg *rep_msg, __be32 eecn)
547{
548 rep_msg->offset16 = cpu_to_be32((be32_to_cpu(eecn) << 8) |
549 (be32_to_cpu(rep_msg->offset16) & 0x000000FF));
550}
551
552static inline __be32 cm_rep_get_qpn(struct cm_rep_msg *rep_msg, enum ib_qp_type qp_type)
553{
554 return (qp_type == IB_QPT_XRC_INI) ?
555 cm_rep_get_local_eecn(rep_msg) : cm_rep_get_local_qpn(rep_msg);
556}
557
530static inline __be32 cm_rep_get_starting_psn(struct cm_rep_msg *rep_msg) 558static inline __be32 cm_rep_get_starting_psn(struct cm_rep_msg *rep_msg)
531{ 559{
532 return cpu_to_be32(be32_to_cpu(rep_msg->offset20) >> 8); 560 return cpu_to_be32(be32_to_cpu(rep_msg->offset20) >> 8);
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index ca4c5dcd7133..872b1842598a 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -81,6 +81,7 @@ static DEFINE_IDR(sdp_ps);
81static DEFINE_IDR(tcp_ps); 81static DEFINE_IDR(tcp_ps);
82static DEFINE_IDR(udp_ps); 82static DEFINE_IDR(udp_ps);
83static DEFINE_IDR(ipoib_ps); 83static DEFINE_IDR(ipoib_ps);
84static DEFINE_IDR(ib_ps);
84 85
85struct cma_device { 86struct cma_device {
86 struct list_head list; 87 struct list_head list;
@@ -1179,6 +1180,15 @@ static void cma_set_req_event_data(struct rdma_cm_event *event,
1179 event->param.conn.qp_num = req_data->remote_qpn; 1180 event->param.conn.qp_num = req_data->remote_qpn;
1180} 1181}
1181 1182
1183static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1184{
1185 return (((ib_event->event == IB_CM_REQ_RECEIVED) ||
1186 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1187 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1188 (id->qp_type == IB_QPT_UD)) ||
1189 (!id->qp_type));
1190}
1191
1182static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event) 1192static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1183{ 1193{
1184 struct rdma_id_private *listen_id, *conn_id; 1194 struct rdma_id_private *listen_id, *conn_id;
@@ -1186,13 +1196,16 @@ static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1186 int offset, ret; 1196 int offset, ret;
1187 1197
1188 listen_id = cm_id->context; 1198 listen_id = cm_id->context;
1199 if (!cma_check_req_qp_type(&listen_id->id, ib_event))
1200 return -EINVAL;
1201
1189 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN)) 1202 if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
1190 return -ECONNABORTED; 1203 return -ECONNABORTED;
1191 1204
1192 memset(&event, 0, sizeof event); 1205 memset(&event, 0, sizeof event);
1193 offset = cma_user_data_offset(listen_id->id.ps); 1206 offset = cma_user_data_offset(listen_id->id.ps);
1194 event.event = RDMA_CM_EVENT_CONNECT_REQUEST; 1207 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1195 if (listen_id->id.qp_type == IB_QPT_UD) { 1208 if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1196 conn_id = cma_new_udp_id(&listen_id->id, ib_event); 1209 conn_id = cma_new_udp_id(&listen_id->id, ib_event);
1197 event.param.ud.private_data = ib_event->private_data + offset; 1210 event.param.ud.private_data = ib_event->private_data + offset;
1198 event.param.ud.private_data_len = 1211 event.param.ud.private_data_len =
@@ -1328,6 +1341,8 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1328 switch (iw_event->status) { 1341 switch (iw_event->status) {
1329 case 0: 1342 case 0:
1330 event.event = RDMA_CM_EVENT_ESTABLISHED; 1343 event.event = RDMA_CM_EVENT_ESTABLISHED;
1344 event.param.conn.initiator_depth = iw_event->ird;
1345 event.param.conn.responder_resources = iw_event->ord;
1331 break; 1346 break;
1332 case -ECONNRESET: 1347 case -ECONNRESET:
1333 case -ECONNREFUSED: 1348 case -ECONNREFUSED:
@@ -1343,6 +1358,8 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
1343 break; 1358 break;
1344 case IW_CM_EVENT_ESTABLISHED: 1359 case IW_CM_EVENT_ESTABLISHED:
1345 event.event = RDMA_CM_EVENT_ESTABLISHED; 1360 event.event = RDMA_CM_EVENT_ESTABLISHED;
1361 event.param.conn.initiator_depth = iw_event->ird;
1362 event.param.conn.responder_resources = iw_event->ord;
1346 break; 1363 break;
1347 default: 1364 default:
1348 BUG_ON(1); 1365 BUG_ON(1);
@@ -1433,8 +1450,8 @@ static int iw_conn_req_handler(struct iw_cm_id *cm_id,
1433 event.event = RDMA_CM_EVENT_CONNECT_REQUEST; 1450 event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1434 event.param.conn.private_data = iw_event->private_data; 1451 event.param.conn.private_data = iw_event->private_data;
1435 event.param.conn.private_data_len = iw_event->private_data_len; 1452 event.param.conn.private_data_len = iw_event->private_data_len;
1436 event.param.conn.initiator_depth = attr.max_qp_init_rd_atom; 1453 event.param.conn.initiator_depth = iw_event->ird;
1437 event.param.conn.responder_resources = attr.max_qp_rd_atom; 1454 event.param.conn.responder_resources = iw_event->ord;
1438 1455
1439 /* 1456 /*
1440 * Protect against the user destroying conn_id from another thread 1457 * Protect against the user destroying conn_id from another thread
@@ -2234,6 +2251,9 @@ static int cma_get_port(struct rdma_id_private *id_priv)
2234 case RDMA_PS_IPOIB: 2251 case RDMA_PS_IPOIB:
2235 ps = &ipoib_ps; 2252 ps = &ipoib_ps;
2236 break; 2253 break;
2254 case RDMA_PS_IB:
2255 ps = &ib_ps;
2256 break;
2237 default: 2257 default:
2238 return -EPROTONOSUPPORT; 2258 return -EPROTONOSUPPORT;
2239 } 2259 }
@@ -2569,7 +2589,7 @@ static int cma_connect_ib(struct rdma_id_private *id_priv,
2569 req.service_id = cma_get_service_id(id_priv->id.ps, 2589 req.service_id = cma_get_service_id(id_priv->id.ps,
2570 (struct sockaddr *) &route->addr.dst_addr); 2590 (struct sockaddr *) &route->addr.dst_addr);
2571 req.qp_num = id_priv->qp_num; 2591 req.qp_num = id_priv->qp_num;
2572 req.qp_type = IB_QPT_RC; 2592 req.qp_type = id_priv->id.qp_type;
2573 req.starting_psn = id_priv->seq_num; 2593 req.starting_psn = id_priv->seq_num;
2574 req.responder_resources = conn_param->responder_resources; 2594 req.responder_resources = conn_param->responder_resources;
2575 req.initiator_depth = conn_param->initiator_depth; 2595 req.initiator_depth = conn_param->initiator_depth;
@@ -2616,14 +2636,16 @@ static int cma_connect_iw(struct rdma_id_private *id_priv,
2616 if (ret) 2636 if (ret)
2617 goto out; 2637 goto out;
2618 2638
2619 iw_param.ord = conn_param->initiator_depth; 2639 if (conn_param) {
2620 iw_param.ird = conn_param->responder_resources; 2640 iw_param.ord = conn_param->initiator_depth;
2621 iw_param.private_data = conn_param->private_data; 2641 iw_param.ird = conn_param->responder_resources;
2622 iw_param.private_data_len = conn_param->private_data_len; 2642 iw_param.private_data = conn_param->private_data;
2623 if (id_priv->id.qp) 2643 iw_param.private_data_len = conn_param->private_data_len;
2644 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
2645 } else {
2646 memset(&iw_param, 0, sizeof iw_param);
2624 iw_param.qpn = id_priv->qp_num; 2647 iw_param.qpn = id_priv->qp_num;
2625 else 2648 }
2626 iw_param.qpn = conn_param->qp_num;
2627 ret = iw_cm_connect(cm_id, &iw_param); 2649 ret = iw_cm_connect(cm_id, &iw_param);
2628out: 2650out:
2629 if (ret) { 2651 if (ret) {
@@ -2765,14 +2787,20 @@ int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
2765 2787
2766 switch (rdma_node_get_transport(id->device->node_type)) { 2788 switch (rdma_node_get_transport(id->device->node_type)) {
2767 case RDMA_TRANSPORT_IB: 2789 case RDMA_TRANSPORT_IB:
2768 if (id->qp_type == IB_QPT_UD) 2790 if (id->qp_type == IB_QPT_UD) {
2769 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS, 2791 if (conn_param)
2770 conn_param->private_data, 2792 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2771 conn_param->private_data_len); 2793 conn_param->private_data,
2772 else if (conn_param) 2794 conn_param->private_data_len);
2773 ret = cma_accept_ib(id_priv, conn_param); 2795 else
2774 else 2796 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
2775 ret = cma_rep_recv(id_priv); 2797 NULL, 0);
2798 } else {
2799 if (conn_param)
2800 ret = cma_accept_ib(id_priv, conn_param);
2801 else
2802 ret = cma_rep_recv(id_priv);
2803 }
2776 break; 2804 break;
2777 case RDMA_TRANSPORT_IWARP: 2805 case RDMA_TRANSPORT_IWARP:
2778 ret = cma_accept_iw(id_priv, conn_param); 2806 ret = cma_accept_iw(id_priv, conn_param);
@@ -3460,6 +3488,7 @@ static void __exit cma_cleanup(void)
3460 idr_destroy(&tcp_ps); 3488 idr_destroy(&tcp_ps);
3461 idr_destroy(&udp_ps); 3489 idr_destroy(&udp_ps);
3462 idr_destroy(&ipoib_ps); 3490 idr_destroy(&ipoib_ps);
3491 idr_destroy(&ib_ps);
3463} 3492}
3464 3493
3465module_init(cma_init); 3494module_init(cma_init);
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index b4d8672a3e4e..056389229ea7 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -1596,6 +1596,9 @@ find_mad_agent(struct ib_mad_port_private *port_priv,
1596 mad->mad_hdr.class_version].class; 1596 mad->mad_hdr.class_version].class;
1597 if (!class) 1597 if (!class)
1598 goto out; 1598 goto out;
1599 if (convert_mgmt_class(mad->mad_hdr.mgmt_class) >=
1600 IB_MGMT_MAX_METHODS)
1601 goto out;
1599 method = class->method_table[convert_mgmt_class( 1602 method = class->method_table[convert_mgmt_class(
1600 mad->mad_hdr.mgmt_class)]; 1603 mad->mad_hdr.mgmt_class)];
1601 if (method) 1604 if (method)
diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
index 9ab5df72df7b..2b59b72b57f9 100644
--- a/drivers/infiniband/core/sysfs.c
+++ b/drivers/infiniband/core/sysfs.c
@@ -185,17 +185,35 @@ static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
185 if (ret) 185 if (ret)
186 return ret; 186 return ret;
187 187
188 rate = (25 * attr.active_speed) / 10;
189
188 switch (attr.active_speed) { 190 switch (attr.active_speed) {
189 case 2: speed = " DDR"; break; 191 case 2:
190 case 4: speed = " QDR"; break; 192 speed = " DDR";
193 break;
194 case 4:
195 speed = " QDR";
196 break;
197 case 8:
198 speed = " FDR10";
199 rate = 10;
200 break;
201 case 16:
202 speed = " FDR";
203 rate = 14;
204 break;
205 case 32:
206 speed = " EDR";
207 rate = 25;
208 break;
191 } 209 }
192 210
193 rate = 25 * ib_width_enum_to_int(attr.active_width) * attr.active_speed; 211 rate *= ib_width_enum_to_int(attr.active_width);
194 if (rate < 0) 212 if (rate < 0)
195 return -EINVAL; 213 return -EINVAL;
196 214
197 return sprintf(buf, "%d%s Gb/sec (%dX%s)\n", 215 return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
198 rate / 10, rate % 10 ? ".5" : "", 216 rate, (attr.active_speed == 1) ? ".5" : "",
199 ib_width_enum_to_int(attr.active_width), speed); 217 ib_width_enum_to_int(attr.active_width), speed);
200} 218}
201 219
diff --git a/drivers/infiniband/core/ucm.c b/drivers/infiniband/core/ucm.c
index 08f948df8fa9..b8a0b4a7811b 100644
--- a/drivers/infiniband/core/ucm.c
+++ b/drivers/infiniband/core/ucm.c
@@ -1122,7 +1122,7 @@ static ssize_t ib_ucm_write(struct file *filp, const char __user *buf,
1122 if (copy_from_user(&hdr, buf, sizeof(hdr))) 1122 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1123 return -EFAULT; 1123 return -EFAULT;
1124 1124
1125 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucm_cmd_table)) 1125 if (hdr.cmd >= ARRAY_SIZE(ucm_cmd_table))
1126 return -EINVAL; 1126 return -EINVAL;
1127 1127
1128 if (hdr.in + sizeof(hdr) > len) 1128 if (hdr.in + sizeof(hdr) > len)
diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c
index 71be5eebd683..b69307f4f6d0 100644
--- a/drivers/infiniband/core/ucma.c
+++ b/drivers/infiniband/core/ucma.c
@@ -276,7 +276,7 @@ static int ucma_event_handler(struct rdma_cm_id *cm_id,
276 ucma_set_event_context(ctx, event, uevent); 276 ucma_set_event_context(ctx, event, uevent);
277 uevent->resp.event = event->event; 277 uevent->resp.event = event->event;
278 uevent->resp.status = event->status; 278 uevent->resp.status = event->status;
279 if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB) 279 if (cm_id->qp_type == IB_QPT_UD)
280 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud); 280 ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
281 else 281 else
282 ucma_copy_conn_event(&uevent->resp.param.conn, 282 ucma_copy_conn_event(&uevent->resp.param.conn,
@@ -377,6 +377,9 @@ static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_
377 case RDMA_PS_IPOIB: 377 case RDMA_PS_IPOIB:
378 *qp_type = IB_QPT_UD; 378 *qp_type = IB_QPT_UD;
379 return 0; 379 return 0;
380 case RDMA_PS_IB:
381 *qp_type = cmd->qp_type;
382 return 0;
380 default: 383 default:
381 return -EINVAL; 384 return -EINVAL;
382 } 385 }
@@ -1270,7 +1273,7 @@ static ssize_t ucma_write(struct file *filp, const char __user *buf,
1270 if (copy_from_user(&hdr, buf, sizeof(hdr))) 1273 if (copy_from_user(&hdr, buf, sizeof(hdr)))
1271 return -EFAULT; 1274 return -EFAULT;
1272 1275
1273 if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table)) 1276 if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
1274 return -EINVAL; 1277 return -EINVAL;
1275 1278
1276 if (hdr.in + sizeof(hdr) > len) 1279 if (hdr.in + sizeof(hdr) > len)
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index 8d261b6ea5fe..07db22997e97 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -458,8 +458,7 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
458 goto err; 458 goto err;
459 } 459 }
460 460
461 if (packet->mad.hdr.id < 0 || 461 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
462 packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
463 ret = -EINVAL; 462 ret = -EINVAL;
464 goto err; 463 goto err;
465 } 464 }
@@ -703,7 +702,7 @@ static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
703 mutex_lock(&file->port->file_mutex); 702 mutex_lock(&file->port->file_mutex);
704 mutex_lock(&file->mutex); 703 mutex_lock(&file->mutex);
705 704
706 if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) { 705 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
707 ret = -EINVAL; 706 ret = -EINVAL;
708 goto out; 707 goto out;
709 } 708 }
diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h
index a078e5624d22..5bcb2afd3dcb 100644
--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h
@@ -76,6 +76,8 @@ struct ib_uverbs_device {
76 struct ib_device *ib_dev; 76 struct ib_device *ib_dev;
77 int devnum; 77 int devnum;
78 struct cdev cdev; 78 struct cdev cdev;
79 struct rb_root xrcd_tree;
80 struct mutex xrcd_tree_mutex;
79}; 81};
80 82
81struct ib_uverbs_event_file { 83struct ib_uverbs_event_file {
@@ -120,6 +122,16 @@ struct ib_uevent_object {
120 u32 events_reported; 122 u32 events_reported;
121}; 123};
122 124
125struct ib_uxrcd_object {
126 struct ib_uobject uobject;
127 atomic_t refcnt;
128};
129
130struct ib_usrq_object {
131 struct ib_uevent_object uevent;
132 struct ib_uxrcd_object *uxrcd;
133};
134
123struct ib_uqp_object { 135struct ib_uqp_object {
124 struct ib_uevent_object uevent; 136 struct ib_uevent_object uevent;
125 struct list_head mcast_list; 137 struct list_head mcast_list;
@@ -142,6 +154,7 @@ extern struct idr ib_uverbs_ah_idr;
142extern struct idr ib_uverbs_cq_idr; 154extern struct idr ib_uverbs_cq_idr;
143extern struct idr ib_uverbs_qp_idr; 155extern struct idr ib_uverbs_qp_idr;
144extern struct idr ib_uverbs_srq_idr; 156extern struct idr ib_uverbs_srq_idr;
157extern struct idr ib_uverbs_xrcd_idr;
145 158
146void idr_remove_uobj(struct idr *idp, struct ib_uobject *uobj); 159void idr_remove_uobj(struct idr *idp, struct ib_uobject *uobj);
147 160
@@ -161,6 +174,7 @@ void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr);
161void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr); 174void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr);
162void ib_uverbs_event_handler(struct ib_event_handler *handler, 175void ib_uverbs_event_handler(struct ib_event_handler *handler,
163 struct ib_event *event); 176 struct ib_event *event);
177void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev, struct ib_xrcd *xrcd);
164 178
165#define IB_UVERBS_DECLARE_CMD(name) \ 179#define IB_UVERBS_DECLARE_CMD(name) \
166 ssize_t ib_uverbs_##name(struct ib_uverbs_file *file, \ 180 ssize_t ib_uverbs_##name(struct ib_uverbs_file *file, \
@@ -181,6 +195,7 @@ IB_UVERBS_DECLARE_CMD(poll_cq);
181IB_UVERBS_DECLARE_CMD(req_notify_cq); 195IB_UVERBS_DECLARE_CMD(req_notify_cq);
182IB_UVERBS_DECLARE_CMD(destroy_cq); 196IB_UVERBS_DECLARE_CMD(destroy_cq);
183IB_UVERBS_DECLARE_CMD(create_qp); 197IB_UVERBS_DECLARE_CMD(create_qp);
198IB_UVERBS_DECLARE_CMD(open_qp);
184IB_UVERBS_DECLARE_CMD(query_qp); 199IB_UVERBS_DECLARE_CMD(query_qp);
185IB_UVERBS_DECLARE_CMD(modify_qp); 200IB_UVERBS_DECLARE_CMD(modify_qp);
186IB_UVERBS_DECLARE_CMD(destroy_qp); 201IB_UVERBS_DECLARE_CMD(destroy_qp);
@@ -195,5 +210,8 @@ IB_UVERBS_DECLARE_CMD(create_srq);
195IB_UVERBS_DECLARE_CMD(modify_srq); 210IB_UVERBS_DECLARE_CMD(modify_srq);
196IB_UVERBS_DECLARE_CMD(query_srq); 211IB_UVERBS_DECLARE_CMD(query_srq);
197IB_UVERBS_DECLARE_CMD(destroy_srq); 212IB_UVERBS_DECLARE_CMD(destroy_srq);
213IB_UVERBS_DECLARE_CMD(create_xsrq);
214IB_UVERBS_DECLARE_CMD(open_xrcd);
215IB_UVERBS_DECLARE_CMD(close_xrcd);
198 216
199#endif /* UVERBS_H */ 217#endif /* UVERBS_H */
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index c42699285f8e..254f1649c734 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -47,6 +47,7 @@ static struct lock_class_key cq_lock_key;
47static struct lock_class_key qp_lock_key; 47static struct lock_class_key qp_lock_key;
48static struct lock_class_key ah_lock_key; 48static struct lock_class_key ah_lock_key;
49static struct lock_class_key srq_lock_key; 49static struct lock_class_key srq_lock_key;
50static struct lock_class_key xrcd_lock_key;
50 51
51#define INIT_UDATA(udata, ibuf, obuf, ilen, olen) \ 52#define INIT_UDATA(udata, ibuf, obuf, ilen, olen) \
52 do { \ 53 do { \
@@ -255,6 +256,18 @@ static void put_srq_read(struct ib_srq *srq)
255 put_uobj_read(srq->uobject); 256 put_uobj_read(srq->uobject);
256} 257}
257 258
259static struct ib_xrcd *idr_read_xrcd(int xrcd_handle, struct ib_ucontext *context,
260 struct ib_uobject **uobj)
261{
262 *uobj = idr_read_uobj(&ib_uverbs_xrcd_idr, xrcd_handle, context, 0);
263 return *uobj ? (*uobj)->object : NULL;
264}
265
266static void put_xrcd_read(struct ib_uobject *uobj)
267{
268 put_uobj_read(uobj);
269}
270
258ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file, 271ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
259 const char __user *buf, 272 const char __user *buf,
260 int in_len, int out_len) 273 int in_len, int out_len)
@@ -298,6 +311,7 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
298 INIT_LIST_HEAD(&ucontext->qp_list); 311 INIT_LIST_HEAD(&ucontext->qp_list);
299 INIT_LIST_HEAD(&ucontext->srq_list); 312 INIT_LIST_HEAD(&ucontext->srq_list);
300 INIT_LIST_HEAD(&ucontext->ah_list); 313 INIT_LIST_HEAD(&ucontext->ah_list);
314 INIT_LIST_HEAD(&ucontext->xrcd_list);
301 ucontext->closing = 0; 315 ucontext->closing = 0;
302 316
303 resp.num_comp_vectors = file->device->num_comp_vectors; 317 resp.num_comp_vectors = file->device->num_comp_vectors;
@@ -579,6 +593,310 @@ ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
579 return in_len; 593 return in_len;
580} 594}
581 595
596struct xrcd_table_entry {
597 struct rb_node node;
598 struct ib_xrcd *xrcd;
599 struct inode *inode;
600};
601
602static int xrcd_table_insert(struct ib_uverbs_device *dev,
603 struct inode *inode,
604 struct ib_xrcd *xrcd)
605{
606 struct xrcd_table_entry *entry, *scan;
607 struct rb_node **p = &dev->xrcd_tree.rb_node;
608 struct rb_node *parent = NULL;
609
610 entry = kmalloc(sizeof *entry, GFP_KERNEL);
611 if (!entry)
612 return -ENOMEM;
613
614 entry->xrcd = xrcd;
615 entry->inode = inode;
616
617 while (*p) {
618 parent = *p;
619 scan = rb_entry(parent, struct xrcd_table_entry, node);
620
621 if (inode < scan->inode) {
622 p = &(*p)->rb_left;
623 } else if (inode > scan->inode) {
624 p = &(*p)->rb_right;
625 } else {
626 kfree(entry);
627 return -EEXIST;
628 }
629 }
630
631 rb_link_node(&entry->node, parent, p);
632 rb_insert_color(&entry->node, &dev->xrcd_tree);
633 igrab(inode);
634 return 0;
635}
636
637static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
638 struct inode *inode)
639{
640 struct xrcd_table_entry *entry;
641 struct rb_node *p = dev->xrcd_tree.rb_node;
642
643 while (p) {
644 entry = rb_entry(p, struct xrcd_table_entry, node);
645
646 if (inode < entry->inode)
647 p = p->rb_left;
648 else if (inode > entry->inode)
649 p = p->rb_right;
650 else
651 return entry;
652 }
653
654 return NULL;
655}
656
657static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
658{
659 struct xrcd_table_entry *entry;
660
661 entry = xrcd_table_search(dev, inode);
662 if (!entry)
663 return NULL;
664
665 return entry->xrcd;
666}
667
668static void xrcd_table_delete(struct ib_uverbs_device *dev,
669 struct inode *inode)
670{
671 struct xrcd_table_entry *entry;
672
673 entry = xrcd_table_search(dev, inode);
674 if (entry) {
675 iput(inode);
676 rb_erase(&entry->node, &dev->xrcd_tree);
677 kfree(entry);
678 }
679}
680
681ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
682 const char __user *buf, int in_len,
683 int out_len)
684{
685 struct ib_uverbs_open_xrcd cmd;
686 struct ib_uverbs_open_xrcd_resp resp;
687 struct ib_udata udata;
688 struct ib_uxrcd_object *obj;
689 struct ib_xrcd *xrcd = NULL;
690 struct file *f = NULL;
691 struct inode *inode = NULL;
692 int ret = 0;
693 int new_xrcd = 0;
694
695 if (out_len < sizeof resp)
696 return -ENOSPC;
697
698 if (copy_from_user(&cmd, buf, sizeof cmd))
699 return -EFAULT;
700
701 INIT_UDATA(&udata, buf + sizeof cmd,
702 (unsigned long) cmd.response + sizeof resp,
703 in_len - sizeof cmd, out_len - sizeof resp);
704
705 mutex_lock(&file->device->xrcd_tree_mutex);
706
707 if (cmd.fd != -1) {
708 /* search for file descriptor */
709 f = fget(cmd.fd);
710 if (!f) {
711 ret = -EBADF;
712 goto err_tree_mutex_unlock;
713 }
714
715 inode = f->f_dentry->d_inode;
716 if (!inode) {
717 ret = -EBADF;
718 goto err_tree_mutex_unlock;
719 }
720
721 xrcd = find_xrcd(file->device, inode);
722 if (!xrcd && !(cmd.oflags & O_CREAT)) {
723 /* no file descriptor. Need CREATE flag */
724 ret = -EAGAIN;
725 goto err_tree_mutex_unlock;
726 }
727
728 if (xrcd && cmd.oflags & O_EXCL) {
729 ret = -EINVAL;
730 goto err_tree_mutex_unlock;
731 }
732 }
733
734 obj = kmalloc(sizeof *obj, GFP_KERNEL);
735 if (!obj) {
736 ret = -ENOMEM;
737 goto err_tree_mutex_unlock;
738 }
739
740 init_uobj(&obj->uobject, 0, file->ucontext, &xrcd_lock_key);
741
742 down_write(&obj->uobject.mutex);
743
744 if (!xrcd) {
745 xrcd = file->device->ib_dev->alloc_xrcd(file->device->ib_dev,
746 file->ucontext, &udata);
747 if (IS_ERR(xrcd)) {
748 ret = PTR_ERR(xrcd);
749 goto err;
750 }
751
752 xrcd->inode = inode;
753 xrcd->device = file->device->ib_dev;
754 atomic_set(&xrcd->usecnt, 0);
755 mutex_init(&xrcd->tgt_qp_mutex);
756 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
757 new_xrcd = 1;
758 }
759
760 atomic_set(&obj->refcnt, 0);
761 obj->uobject.object = xrcd;
762 ret = idr_add_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
763 if (ret)
764 goto err_idr;
765
766 memset(&resp, 0, sizeof resp);
767 resp.xrcd_handle = obj->uobject.id;
768
769 if (inode) {
770 if (new_xrcd) {
771 /* create new inode/xrcd table entry */
772 ret = xrcd_table_insert(file->device, inode, xrcd);
773 if (ret)
774 goto err_insert_xrcd;
775 }
776 atomic_inc(&xrcd->usecnt);
777 }
778
779 if (copy_to_user((void __user *) (unsigned long) cmd.response,
780 &resp, sizeof resp)) {
781 ret = -EFAULT;
782 goto err_copy;
783 }
784
785 if (f)
786 fput(f);
787
788 mutex_lock(&file->mutex);
789 list_add_tail(&obj->uobject.list, &file->ucontext->xrcd_list);
790 mutex_unlock(&file->mutex);
791
792 obj->uobject.live = 1;
793 up_write(&obj->uobject.mutex);
794
795 mutex_unlock(&file->device->xrcd_tree_mutex);
796 return in_len;
797
798err_copy:
799 if (inode) {
800 if (new_xrcd)
801 xrcd_table_delete(file->device, inode);
802 atomic_dec(&xrcd->usecnt);
803 }
804
805err_insert_xrcd:
806 idr_remove_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
807
808err_idr:
809 ib_dealloc_xrcd(xrcd);
810
811err:
812 put_uobj_write(&obj->uobject);
813
814err_tree_mutex_unlock:
815 if (f)
816 fput(f);
817
818 mutex_unlock(&file->device->xrcd_tree_mutex);
819
820 return ret;
821}
822
823ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
824 const char __user *buf, int in_len,
825 int out_len)
826{
827 struct ib_uverbs_close_xrcd cmd;
828 struct ib_uobject *uobj;
829 struct ib_xrcd *xrcd = NULL;
830 struct inode *inode = NULL;
831 struct ib_uxrcd_object *obj;
832 int live;
833 int ret = 0;
834
835 if (copy_from_user(&cmd, buf, sizeof cmd))
836 return -EFAULT;
837
838 mutex_lock(&file->device->xrcd_tree_mutex);
839 uobj = idr_write_uobj(&ib_uverbs_xrcd_idr, cmd.xrcd_handle, file->ucontext);
840 if (!uobj) {
841 ret = -EINVAL;
842 goto out;
843 }
844
845 xrcd = uobj->object;
846 inode = xrcd->inode;
847 obj = container_of(uobj, struct ib_uxrcd_object, uobject);
848 if (atomic_read(&obj->refcnt)) {
849 put_uobj_write(uobj);
850 ret = -EBUSY;
851 goto out;
852 }
853
854 if (!inode || atomic_dec_and_test(&xrcd->usecnt)) {
855 ret = ib_dealloc_xrcd(uobj->object);
856 if (!ret)
857 uobj->live = 0;
858 }
859
860 live = uobj->live;
861 if (inode && ret)
862 atomic_inc(&xrcd->usecnt);
863
864 put_uobj_write(uobj);
865
866 if (ret)
867 goto out;
868
869 if (inode && !live)
870 xrcd_table_delete(file->device, inode);
871
872 idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
873 mutex_lock(&file->mutex);
874 list_del(&uobj->list);
875 mutex_unlock(&file->mutex);
876
877 put_uobj(uobj);
878 ret = in_len;
879
880out:
881 mutex_unlock(&file->device->xrcd_tree_mutex);
882 return ret;
883}
884
885void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
886 struct ib_xrcd *xrcd)
887{
888 struct inode *inode;
889
890 inode = xrcd->inode;
891 if (inode && !atomic_dec_and_test(&xrcd->usecnt))
892 return;
893
894 ib_dealloc_xrcd(xrcd);
895
896 if (inode)
897 xrcd_table_delete(dev, inode);
898}
899
582ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file, 900ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
583 const char __user *buf, int in_len, 901 const char __user *buf, int in_len,
584 int out_len) 902 int out_len)
@@ -1052,9 +1370,12 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1052 struct ib_uverbs_create_qp_resp resp; 1370 struct ib_uverbs_create_qp_resp resp;
1053 struct ib_udata udata; 1371 struct ib_udata udata;
1054 struct ib_uqp_object *obj; 1372 struct ib_uqp_object *obj;
1055 struct ib_pd *pd; 1373 struct ib_device *device;
1056 struct ib_cq *scq, *rcq; 1374 struct ib_pd *pd = NULL;
1057 struct ib_srq *srq; 1375 struct ib_xrcd *xrcd = NULL;
1376 struct ib_uobject *uninitialized_var(xrcd_uobj);
1377 struct ib_cq *scq = NULL, *rcq = NULL;
1378 struct ib_srq *srq = NULL;
1058 struct ib_qp *qp; 1379 struct ib_qp *qp;
1059 struct ib_qp_init_attr attr; 1380 struct ib_qp_init_attr attr;
1060 int ret; 1381 int ret;
@@ -1076,15 +1397,39 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1076 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key); 1397 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
1077 down_write(&obj->uevent.uobject.mutex); 1398 down_write(&obj->uevent.uobject.mutex);
1078 1399
1079 srq = cmd.is_srq ? idr_read_srq(cmd.srq_handle, file->ucontext) : NULL; 1400 if (cmd.qp_type == IB_QPT_XRC_TGT) {
1080 pd = idr_read_pd(cmd.pd_handle, file->ucontext); 1401 xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1081 scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, 0); 1402 if (!xrcd) {
1082 rcq = cmd.recv_cq_handle == cmd.send_cq_handle ? 1403 ret = -EINVAL;
1083 scq : idr_read_cq(cmd.recv_cq_handle, file->ucontext, 1); 1404 goto err_put;
1405 }
1406 device = xrcd->device;
1407 } else {
1408 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1409 scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, 0);
1410 if (!pd || !scq) {
1411 ret = -EINVAL;
1412 goto err_put;
1413 }
1084 1414
1085 if (!pd || !scq || !rcq || (cmd.is_srq && !srq)) { 1415 if (cmd.qp_type == IB_QPT_XRC_INI) {
1086 ret = -EINVAL; 1416 cmd.max_recv_wr = cmd.max_recv_sge = 0;
1087 goto err_put; 1417 } else {
1418 if (cmd.is_srq) {
1419 srq = idr_read_srq(cmd.srq_handle, file->ucontext);
1420 if (!srq || srq->srq_type != IB_SRQT_BASIC) {
1421 ret = -EINVAL;
1422 goto err_put;
1423 }
1424 }
1425 rcq = (cmd.recv_cq_handle == cmd.send_cq_handle) ?
1426 scq : idr_read_cq(cmd.recv_cq_handle, file->ucontext, 1);
1427 if (!rcq) {
1428 ret = -EINVAL;
1429 goto err_put;
1430 }
1431 }
1432 device = pd->device;
1088 } 1433 }
1089 1434
1090 attr.event_handler = ib_uverbs_qp_event_handler; 1435 attr.event_handler = ib_uverbs_qp_event_handler;
@@ -1092,6 +1437,7 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1092 attr.send_cq = scq; 1437 attr.send_cq = scq;
1093 attr.recv_cq = rcq; 1438 attr.recv_cq = rcq;
1094 attr.srq = srq; 1439 attr.srq = srq;
1440 attr.xrcd = xrcd;
1095 attr.sq_sig_type = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR; 1441 attr.sq_sig_type = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
1096 attr.qp_type = cmd.qp_type; 1442 attr.qp_type = cmd.qp_type;
1097 attr.create_flags = 0; 1443 attr.create_flags = 0;
@@ -1106,26 +1452,34 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1106 INIT_LIST_HEAD(&obj->uevent.event_list); 1452 INIT_LIST_HEAD(&obj->uevent.event_list);
1107 INIT_LIST_HEAD(&obj->mcast_list); 1453 INIT_LIST_HEAD(&obj->mcast_list);
1108 1454
1109 qp = pd->device->create_qp(pd, &attr, &udata); 1455 if (cmd.qp_type == IB_QPT_XRC_TGT)
1456 qp = ib_create_qp(pd, &attr);
1457 else
1458 qp = device->create_qp(pd, &attr, &udata);
1459
1110 if (IS_ERR(qp)) { 1460 if (IS_ERR(qp)) {
1111 ret = PTR_ERR(qp); 1461 ret = PTR_ERR(qp);
1112 goto err_put; 1462 goto err_put;
1113 } 1463 }
1114 1464
1115 qp->device = pd->device; 1465 if (cmd.qp_type != IB_QPT_XRC_TGT) {
1116 qp->pd = pd; 1466 qp->real_qp = qp;
1117 qp->send_cq = attr.send_cq; 1467 qp->device = device;
1118 qp->recv_cq = attr.recv_cq; 1468 qp->pd = pd;
1119 qp->srq = attr.srq; 1469 qp->send_cq = attr.send_cq;
1120 qp->uobject = &obj->uevent.uobject; 1470 qp->recv_cq = attr.recv_cq;
1121 qp->event_handler = attr.event_handler; 1471 qp->srq = attr.srq;
1122 qp->qp_context = attr.qp_context; 1472 qp->event_handler = attr.event_handler;
1123 qp->qp_type = attr.qp_type; 1473 qp->qp_context = attr.qp_context;
1124 atomic_inc(&pd->usecnt); 1474 qp->qp_type = attr.qp_type;
1125 atomic_inc(&attr.send_cq->usecnt); 1475 atomic_inc(&pd->usecnt);
1126 atomic_inc(&attr.recv_cq->usecnt); 1476 atomic_inc(&attr.send_cq->usecnt);
1127 if (attr.srq) 1477 if (attr.recv_cq)
1128 atomic_inc(&attr.srq->usecnt); 1478 atomic_inc(&attr.recv_cq->usecnt);
1479 if (attr.srq)
1480 atomic_inc(&attr.srq->usecnt);
1481 }
1482 qp->uobject = &obj->uevent.uobject;
1129 1483
1130 obj->uevent.uobject.object = qp; 1484 obj->uevent.uobject.object = qp;
1131 ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject); 1485 ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
@@ -1147,9 +1501,13 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1147 goto err_copy; 1501 goto err_copy;
1148 } 1502 }
1149 1503
1150 put_pd_read(pd); 1504 if (xrcd)
1151 put_cq_read(scq); 1505 put_xrcd_read(xrcd_uobj);
1152 if (rcq != scq) 1506 if (pd)
1507 put_pd_read(pd);
1508 if (scq)
1509 put_cq_read(scq);
1510 if (rcq && rcq != scq)
1153 put_cq_read(rcq); 1511 put_cq_read(rcq);
1154 if (srq) 1512 if (srq)
1155 put_srq_read(srq); 1513 put_srq_read(srq);
@@ -1171,6 +1529,8 @@ err_destroy:
1171 ib_destroy_qp(qp); 1529 ib_destroy_qp(qp);
1172 1530
1173err_put: 1531err_put:
1532 if (xrcd)
1533 put_xrcd_read(xrcd_uobj);
1174 if (pd) 1534 if (pd)
1175 put_pd_read(pd); 1535 put_pd_read(pd);
1176 if (scq) 1536 if (scq)
@@ -1184,6 +1544,98 @@ err_put:
1184 return ret; 1544 return ret;
1185} 1545}
1186 1546
1547ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
1548 const char __user *buf, int in_len, int out_len)
1549{
1550 struct ib_uverbs_open_qp cmd;
1551 struct ib_uverbs_create_qp_resp resp;
1552 struct ib_udata udata;
1553 struct ib_uqp_object *obj;
1554 struct ib_xrcd *xrcd;
1555 struct ib_uobject *uninitialized_var(xrcd_uobj);
1556 struct ib_qp *qp;
1557 struct ib_qp_open_attr attr;
1558 int ret;
1559
1560 if (out_len < sizeof resp)
1561 return -ENOSPC;
1562
1563 if (copy_from_user(&cmd, buf, sizeof cmd))
1564 return -EFAULT;
1565
1566 INIT_UDATA(&udata, buf + sizeof cmd,
1567 (unsigned long) cmd.response + sizeof resp,
1568 in_len - sizeof cmd, out_len - sizeof resp);
1569
1570 obj = kmalloc(sizeof *obj, GFP_KERNEL);
1571 if (!obj)
1572 return -ENOMEM;
1573
1574 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
1575 down_write(&obj->uevent.uobject.mutex);
1576
1577 xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
1578 if (!xrcd) {
1579 ret = -EINVAL;
1580 goto err_put;
1581 }
1582
1583 attr.event_handler = ib_uverbs_qp_event_handler;
1584 attr.qp_context = file;
1585 attr.qp_num = cmd.qpn;
1586 attr.qp_type = cmd.qp_type;
1587
1588 obj->uevent.events_reported = 0;
1589 INIT_LIST_HEAD(&obj->uevent.event_list);
1590 INIT_LIST_HEAD(&obj->mcast_list);
1591
1592 qp = ib_open_qp(xrcd, &attr);
1593 if (IS_ERR(qp)) {
1594 ret = PTR_ERR(qp);
1595 goto err_put;
1596 }
1597
1598 qp->uobject = &obj->uevent.uobject;
1599
1600 obj->uevent.uobject.object = qp;
1601 ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1602 if (ret)
1603 goto err_destroy;
1604
1605 memset(&resp, 0, sizeof resp);
1606 resp.qpn = qp->qp_num;
1607 resp.qp_handle = obj->uevent.uobject.id;
1608
1609 if (copy_to_user((void __user *) (unsigned long) cmd.response,
1610 &resp, sizeof resp)) {
1611 ret = -EFAULT;
1612 goto err_remove;
1613 }
1614
1615 put_xrcd_read(xrcd_uobj);
1616
1617 mutex_lock(&file->mutex);
1618 list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
1619 mutex_unlock(&file->mutex);
1620
1621 obj->uevent.uobject.live = 1;
1622
1623 up_write(&obj->uevent.uobject.mutex);
1624
1625 return in_len;
1626
1627err_remove:
1628 idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
1629
1630err_destroy:
1631 ib_destroy_qp(qp);
1632
1633err_put:
1634 put_xrcd_read(xrcd_uobj);
1635 put_uobj_write(&obj->uevent.uobject);
1636 return ret;
1637}
1638
1187ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file, 1639ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
1188 const char __user *buf, int in_len, 1640 const char __user *buf, int in_len,
1189 int out_len) 1641 int out_len)
@@ -1284,6 +1736,20 @@ out:
1284 return ret ? ret : in_len; 1736 return ret ? ret : in_len;
1285} 1737}
1286 1738
1739/* Remove ignored fields set in the attribute mask */
1740static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
1741{
1742 switch (qp_type) {
1743 case IB_QPT_XRC_INI:
1744 return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
1745 case IB_QPT_XRC_TGT:
1746 return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
1747 IB_QP_RNR_RETRY);
1748 default:
1749 return mask;
1750 }
1751}
1752
1287ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file, 1753ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
1288 const char __user *buf, int in_len, 1754 const char __user *buf, int in_len,
1289 int out_len) 1755 int out_len)
@@ -1356,7 +1822,12 @@ ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
1356 attr->alt_ah_attr.ah_flags = cmd.alt_dest.is_global ? IB_AH_GRH : 0; 1822 attr->alt_ah_attr.ah_flags = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
1357 attr->alt_ah_attr.port_num = cmd.alt_dest.port_num; 1823 attr->alt_ah_attr.port_num = cmd.alt_dest.port_num;
1358 1824
1359 ret = qp->device->modify_qp(qp, attr, cmd.attr_mask, &udata); 1825 if (qp->real_qp == qp) {
1826 ret = qp->device->modify_qp(qp, attr,
1827 modify_qp_mask(qp->qp_type, cmd.attr_mask), &udata);
1828 } else {
1829 ret = ib_modify_qp(qp, attr, modify_qp_mask(qp->qp_type, cmd.attr_mask));
1830 }
1360 1831
1361 put_qp_read(qp); 1832 put_qp_read(qp);
1362 1833
@@ -1553,7 +2024,7 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
1553 } 2024 }
1554 2025
1555 resp.bad_wr = 0; 2026 resp.bad_wr = 0;
1556 ret = qp->device->post_send(qp, wr, &bad_wr); 2027 ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
1557 if (ret) 2028 if (ret)
1558 for (next = wr; next; next = next->next) { 2029 for (next = wr; next; next = next->next) {
1559 ++resp.bad_wr; 2030 ++resp.bad_wr;
@@ -1691,7 +2162,7 @@ ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
1691 goto out; 2162 goto out;
1692 2163
1693 resp.bad_wr = 0; 2164 resp.bad_wr = 0;
1694 ret = qp->device->post_recv(qp, wr, &bad_wr); 2165 ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
1695 2166
1696 put_qp_read(qp); 2167 put_qp_read(qp);
1697 2168
@@ -1975,107 +2446,199 @@ out_put:
1975 return ret ? ret : in_len; 2446 return ret ? ret : in_len;
1976} 2447}
1977 2448
1978ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file, 2449int __uverbs_create_xsrq(struct ib_uverbs_file *file,
1979 const char __user *buf, int in_len, 2450 struct ib_uverbs_create_xsrq *cmd,
1980 int out_len) 2451 struct ib_udata *udata)
1981{ 2452{
1982 struct ib_uverbs_create_srq cmd;
1983 struct ib_uverbs_create_srq_resp resp; 2453 struct ib_uverbs_create_srq_resp resp;
1984 struct ib_udata udata; 2454 struct ib_usrq_object *obj;
1985 struct ib_uevent_object *obj;
1986 struct ib_pd *pd; 2455 struct ib_pd *pd;
1987 struct ib_srq *srq; 2456 struct ib_srq *srq;
2457 struct ib_uobject *uninitialized_var(xrcd_uobj);
1988 struct ib_srq_init_attr attr; 2458 struct ib_srq_init_attr attr;
1989 int ret; 2459 int ret;
1990 2460
1991 if (out_len < sizeof resp)
1992 return -ENOSPC;
1993
1994 if (copy_from_user(&cmd, buf, sizeof cmd))
1995 return -EFAULT;
1996
1997 INIT_UDATA(&udata, buf + sizeof cmd,
1998 (unsigned long) cmd.response + sizeof resp,
1999 in_len - sizeof cmd, out_len - sizeof resp);
2000
2001 obj = kmalloc(sizeof *obj, GFP_KERNEL); 2461 obj = kmalloc(sizeof *obj, GFP_KERNEL);
2002 if (!obj) 2462 if (!obj)
2003 return -ENOMEM; 2463 return -ENOMEM;
2004 2464
2005 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &srq_lock_key); 2465 init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext, &srq_lock_key);
2006 down_write(&obj->uobject.mutex); 2466 down_write(&obj->uevent.uobject.mutex);
2007 2467
2008 pd = idr_read_pd(cmd.pd_handle, file->ucontext); 2468 pd = idr_read_pd(cmd->pd_handle, file->ucontext);
2009 if (!pd) { 2469 if (!pd) {
2010 ret = -EINVAL; 2470 ret = -EINVAL;
2011 goto err; 2471 goto err;
2012 } 2472 }
2013 2473
2474 if (cmd->srq_type == IB_SRQT_XRC) {
2475 attr.ext.xrc.cq = idr_read_cq(cmd->cq_handle, file->ucontext, 0);
2476 if (!attr.ext.xrc.cq) {
2477 ret = -EINVAL;
2478 goto err_put_pd;
2479 }
2480
2481 attr.ext.xrc.xrcd = idr_read_xrcd(cmd->xrcd_handle, file->ucontext, &xrcd_uobj);
2482 if (!attr.ext.xrc.xrcd) {
2483 ret = -EINVAL;
2484 goto err_put_cq;
2485 }
2486
2487 obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
2488 atomic_inc(&obj->uxrcd->refcnt);
2489 }
2490
2014 attr.event_handler = ib_uverbs_srq_event_handler; 2491 attr.event_handler = ib_uverbs_srq_event_handler;
2015 attr.srq_context = file; 2492 attr.srq_context = file;
2016 attr.attr.max_wr = cmd.max_wr; 2493 attr.srq_type = cmd->srq_type;
2017 attr.attr.max_sge = cmd.max_sge; 2494 attr.attr.max_wr = cmd->max_wr;
2018 attr.attr.srq_limit = cmd.srq_limit; 2495 attr.attr.max_sge = cmd->max_sge;
2496 attr.attr.srq_limit = cmd->srq_limit;
2019 2497
2020 obj->events_reported = 0; 2498 obj->uevent.events_reported = 0;
2021 INIT_LIST_HEAD(&obj->event_list); 2499 INIT_LIST_HEAD(&obj->uevent.event_list);
2022 2500
2023 srq = pd->device->create_srq(pd, &attr, &udata); 2501 srq = pd->device->create_srq(pd, &attr, udata);
2024 if (IS_ERR(srq)) { 2502 if (IS_ERR(srq)) {
2025 ret = PTR_ERR(srq); 2503 ret = PTR_ERR(srq);
2026 goto err_put; 2504 goto err_put;
2027 } 2505 }
2028 2506
2029 srq->device = pd->device; 2507 srq->device = pd->device;
2030 srq->pd = pd; 2508 srq->pd = pd;
2031 srq->uobject = &obj->uobject; 2509 srq->srq_type = cmd->srq_type;
2510 srq->uobject = &obj->uevent.uobject;
2032 srq->event_handler = attr.event_handler; 2511 srq->event_handler = attr.event_handler;
2033 srq->srq_context = attr.srq_context; 2512 srq->srq_context = attr.srq_context;
2513
2514 if (cmd->srq_type == IB_SRQT_XRC) {
2515 srq->ext.xrc.cq = attr.ext.xrc.cq;
2516 srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
2517 atomic_inc(&attr.ext.xrc.cq->usecnt);
2518 atomic_inc(&attr.ext.xrc.xrcd->usecnt);
2519 }
2520
2034 atomic_inc(&pd->usecnt); 2521 atomic_inc(&pd->usecnt);
2035 atomic_set(&srq->usecnt, 0); 2522 atomic_set(&srq->usecnt, 0);
2036 2523
2037 obj->uobject.object = srq; 2524 obj->uevent.uobject.object = srq;
2038 ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uobject); 2525 ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
2039 if (ret) 2526 if (ret)
2040 goto err_destroy; 2527 goto err_destroy;
2041 2528
2042 memset(&resp, 0, sizeof resp); 2529 memset(&resp, 0, sizeof resp);
2043 resp.srq_handle = obj->uobject.id; 2530 resp.srq_handle = obj->uevent.uobject.id;
2044 resp.max_wr = attr.attr.max_wr; 2531 resp.max_wr = attr.attr.max_wr;
2045 resp.max_sge = attr.attr.max_sge; 2532 resp.max_sge = attr.attr.max_sge;
2533 if (cmd->srq_type == IB_SRQT_XRC)
2534 resp.srqn = srq->ext.xrc.srq_num;
2046 2535
2047 if (copy_to_user((void __user *) (unsigned long) cmd.response, 2536 if (copy_to_user((void __user *) (unsigned long) cmd->response,
2048 &resp, sizeof resp)) { 2537 &resp, sizeof resp)) {
2049 ret = -EFAULT; 2538 ret = -EFAULT;
2050 goto err_copy; 2539 goto err_copy;
2051 } 2540 }
2052 2541
2542 if (cmd->srq_type == IB_SRQT_XRC) {
2543 put_uobj_read(xrcd_uobj);
2544 put_cq_read(attr.ext.xrc.cq);
2545 }
2053 put_pd_read(pd); 2546 put_pd_read(pd);
2054 2547
2055 mutex_lock(&file->mutex); 2548 mutex_lock(&file->mutex);
2056 list_add_tail(&obj->uobject.list, &file->ucontext->srq_list); 2549 list_add_tail(&obj->uevent.uobject.list, &file->ucontext->srq_list);
2057 mutex_unlock(&file->mutex); 2550 mutex_unlock(&file->mutex);
2058 2551
2059 obj->uobject.live = 1; 2552 obj->uevent.uobject.live = 1;
2060 2553
2061 up_write(&obj->uobject.mutex); 2554 up_write(&obj->uevent.uobject.mutex);
2062 2555
2063 return in_len; 2556 return 0;
2064 2557
2065err_copy: 2558err_copy:
2066 idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uobject); 2559 idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
2067 2560
2068err_destroy: 2561err_destroy:
2069 ib_destroy_srq(srq); 2562 ib_destroy_srq(srq);
2070 2563
2071err_put: 2564err_put:
2565 if (cmd->srq_type == IB_SRQT_XRC) {
2566 atomic_dec(&obj->uxrcd->refcnt);
2567 put_uobj_read(xrcd_uobj);
2568 }
2569
2570err_put_cq:
2571 if (cmd->srq_type == IB_SRQT_XRC)
2572 put_cq_read(attr.ext.xrc.cq);
2573
2574err_put_pd:
2072 put_pd_read(pd); 2575 put_pd_read(pd);
2073 2576
2074err: 2577err:
2075 put_uobj_write(&obj->uobject); 2578 put_uobj_write(&obj->uevent.uobject);
2076 return ret; 2579 return ret;
2077} 2580}
2078 2581
2582ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
2583 const char __user *buf, int in_len,
2584 int out_len)
2585{
2586 struct ib_uverbs_create_srq cmd;
2587 struct ib_uverbs_create_xsrq xcmd;
2588 struct ib_uverbs_create_srq_resp resp;
2589 struct ib_udata udata;
2590 int ret;
2591
2592 if (out_len < sizeof resp)
2593 return -ENOSPC;
2594
2595 if (copy_from_user(&cmd, buf, sizeof cmd))
2596 return -EFAULT;
2597
2598 xcmd.response = cmd.response;
2599 xcmd.user_handle = cmd.user_handle;
2600 xcmd.srq_type = IB_SRQT_BASIC;
2601 xcmd.pd_handle = cmd.pd_handle;
2602 xcmd.max_wr = cmd.max_wr;
2603 xcmd.max_sge = cmd.max_sge;
2604 xcmd.srq_limit = cmd.srq_limit;
2605
2606 INIT_UDATA(&udata, buf + sizeof cmd,
2607 (unsigned long) cmd.response + sizeof resp,
2608 in_len - sizeof cmd, out_len - sizeof resp);
2609
2610 ret = __uverbs_create_xsrq(file, &xcmd, &udata);
2611 if (ret)
2612 return ret;
2613
2614 return in_len;
2615}
2616
2617ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
2618 const char __user *buf, int in_len, int out_len)
2619{
2620 struct ib_uverbs_create_xsrq cmd;
2621 struct ib_uverbs_create_srq_resp resp;
2622 struct ib_udata udata;
2623 int ret;
2624
2625 if (out_len < sizeof resp)
2626 return -ENOSPC;
2627
2628 if (copy_from_user(&cmd, buf, sizeof cmd))
2629 return -EFAULT;
2630
2631 INIT_UDATA(&udata, buf + sizeof cmd,
2632 (unsigned long) cmd.response + sizeof resp,
2633 in_len - sizeof cmd, out_len - sizeof resp);
2634
2635 ret = __uverbs_create_xsrq(file, &cmd, &udata);
2636 if (ret)
2637 return ret;
2638
2639 return in_len;
2640}
2641
2079ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file, 2642ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
2080 const char __user *buf, int in_len, 2643 const char __user *buf, int in_len,
2081 int out_len) 2644 int out_len)
diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index 56898b6578a4..879636746373 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -72,6 +72,7 @@ DEFINE_IDR(ib_uverbs_ah_idr);
72DEFINE_IDR(ib_uverbs_cq_idr); 72DEFINE_IDR(ib_uverbs_cq_idr);
73DEFINE_IDR(ib_uverbs_qp_idr); 73DEFINE_IDR(ib_uverbs_qp_idr);
74DEFINE_IDR(ib_uverbs_srq_idr); 74DEFINE_IDR(ib_uverbs_srq_idr);
75DEFINE_IDR(ib_uverbs_xrcd_idr);
75 76
76static DEFINE_SPINLOCK(map_lock); 77static DEFINE_SPINLOCK(map_lock);
77static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES); 78static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
@@ -107,6 +108,10 @@ static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
107 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq, 108 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
108 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq, 109 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq,
109 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq, 110 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
111 [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd,
112 [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd,
113 [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq,
114 [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp
110}; 115};
111 116
112static void ib_uverbs_add_one(struct ib_device *device); 117static void ib_uverbs_add_one(struct ib_device *device);
@@ -202,8 +207,12 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
202 container_of(uobj, struct ib_uqp_object, uevent.uobject); 207 container_of(uobj, struct ib_uqp_object, uevent.uobject);
203 208
204 idr_remove_uobj(&ib_uverbs_qp_idr, uobj); 209 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
205 ib_uverbs_detach_umcast(qp, uqp); 210 if (qp != qp->real_qp) {
206 ib_destroy_qp(qp); 211 ib_close_qp(qp);
212 } else {
213 ib_uverbs_detach_umcast(qp, uqp);
214 ib_destroy_qp(qp);
215 }
207 ib_uverbs_release_uevent(file, &uqp->uevent); 216 ib_uverbs_release_uevent(file, &uqp->uevent);
208 kfree(uqp); 217 kfree(uqp);
209 } 218 }
@@ -241,6 +250,18 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
241 kfree(uobj); 250 kfree(uobj);
242 } 251 }
243 252
253 mutex_lock(&file->device->xrcd_tree_mutex);
254 list_for_each_entry_safe(uobj, tmp, &context->xrcd_list, list) {
255 struct ib_xrcd *xrcd = uobj->object;
256 struct ib_uxrcd_object *uxrcd =
257 container_of(uobj, struct ib_uxrcd_object, uobject);
258
259 idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
260 ib_uverbs_dealloc_xrcd(file->device, xrcd);
261 kfree(uxrcd);
262 }
263 mutex_unlock(&file->device->xrcd_tree_mutex);
264
244 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) { 265 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
245 struct ib_pd *pd = uobj->object; 266 struct ib_pd *pd = uobj->object;
246 267
@@ -557,8 +578,7 @@ static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
557 if (hdr.in_words * 4 != count) 578 if (hdr.in_words * 4 != count)
558 return -EINVAL; 579 return -EINVAL;
559 580
560 if (hdr.command < 0 || 581 if (hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
561 hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
562 !uverbs_cmd_table[hdr.command]) 582 !uverbs_cmd_table[hdr.command])
563 return -EINVAL; 583 return -EINVAL;
564 584
@@ -741,6 +761,8 @@ static void ib_uverbs_add_one(struct ib_device *device)
741 761
742 kref_init(&uverbs_dev->ref); 762 kref_init(&uverbs_dev->ref);
743 init_completion(&uverbs_dev->comp); 763 init_completion(&uverbs_dev->comp);
764 uverbs_dev->xrcd_tree = RB_ROOT;
765 mutex_init(&uverbs_dev->xrcd_tree_mutex);
744 766
745 spin_lock(&map_lock); 767 spin_lock(&map_lock);
746 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES); 768 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index af7a8b08b2e9..42517500b223 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -39,6 +39,7 @@
39#include <linux/errno.h> 39#include <linux/errno.h>
40#include <linux/err.h> 40#include <linux/err.h>
41#include <linux/string.h> 41#include <linux/string.h>
42#include <linux/slab.h>
42 43
43#include <rdma/ib_verbs.h> 44#include <rdma/ib_verbs.h>
44#include <rdma/ib_cache.h> 45#include <rdma/ib_cache.h>
@@ -77,6 +78,31 @@ enum ib_rate mult_to_ib_rate(int mult)
77} 78}
78EXPORT_SYMBOL(mult_to_ib_rate); 79EXPORT_SYMBOL(mult_to_ib_rate);
79 80
81int ib_rate_to_mbps(enum ib_rate rate)
82{
83 switch (rate) {
84 case IB_RATE_2_5_GBPS: return 2500;
85 case IB_RATE_5_GBPS: return 5000;
86 case IB_RATE_10_GBPS: return 10000;
87 case IB_RATE_20_GBPS: return 20000;
88 case IB_RATE_30_GBPS: return 30000;
89 case IB_RATE_40_GBPS: return 40000;
90 case IB_RATE_60_GBPS: return 60000;
91 case IB_RATE_80_GBPS: return 80000;
92 case IB_RATE_120_GBPS: return 120000;
93 case IB_RATE_14_GBPS: return 14062;
94 case IB_RATE_56_GBPS: return 56250;
95 case IB_RATE_112_GBPS: return 112500;
96 case IB_RATE_168_GBPS: return 168750;
97 case IB_RATE_25_GBPS: return 25781;
98 case IB_RATE_100_GBPS: return 103125;
99 case IB_RATE_200_GBPS: return 206250;
100 case IB_RATE_300_GBPS: return 309375;
101 default: return -1;
102 }
103}
104EXPORT_SYMBOL(ib_rate_to_mbps);
105
80enum rdma_transport_type 106enum rdma_transport_type
81rdma_node_get_transport(enum rdma_node_type node_type) 107rdma_node_get_transport(enum rdma_node_type node_type)
82{ 108{
@@ -250,6 +276,13 @@ struct ib_srq *ib_create_srq(struct ib_pd *pd,
250 srq->uobject = NULL; 276 srq->uobject = NULL;
251 srq->event_handler = srq_init_attr->event_handler; 277 srq->event_handler = srq_init_attr->event_handler;
252 srq->srq_context = srq_init_attr->srq_context; 278 srq->srq_context = srq_init_attr->srq_context;
279 srq->srq_type = srq_init_attr->srq_type;
280 if (srq->srq_type == IB_SRQT_XRC) {
281 srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
282 srq->ext.xrc.cq = srq_init_attr->ext.xrc.cq;
283 atomic_inc(&srq->ext.xrc.xrcd->usecnt);
284 atomic_inc(&srq->ext.xrc.cq->usecnt);
285 }
253 atomic_inc(&pd->usecnt); 286 atomic_inc(&pd->usecnt);
254 atomic_set(&srq->usecnt, 0); 287 atomic_set(&srq->usecnt, 0);
255 } 288 }
@@ -279,16 +312,29 @@ EXPORT_SYMBOL(ib_query_srq);
279int ib_destroy_srq(struct ib_srq *srq) 312int ib_destroy_srq(struct ib_srq *srq)
280{ 313{
281 struct ib_pd *pd; 314 struct ib_pd *pd;
315 enum ib_srq_type srq_type;
316 struct ib_xrcd *uninitialized_var(xrcd);
317 struct ib_cq *uninitialized_var(cq);
282 int ret; 318 int ret;
283 319
284 if (atomic_read(&srq->usecnt)) 320 if (atomic_read(&srq->usecnt))
285 return -EBUSY; 321 return -EBUSY;
286 322
287 pd = srq->pd; 323 pd = srq->pd;
324 srq_type = srq->srq_type;
325 if (srq_type == IB_SRQT_XRC) {
326 xrcd = srq->ext.xrc.xrcd;
327 cq = srq->ext.xrc.cq;
328 }
288 329
289 ret = srq->device->destroy_srq(srq); 330 ret = srq->device->destroy_srq(srq);
290 if (!ret) 331 if (!ret) {
291 atomic_dec(&pd->usecnt); 332 atomic_dec(&pd->usecnt);
333 if (srq_type == IB_SRQT_XRC) {
334 atomic_dec(&xrcd->usecnt);
335 atomic_dec(&cq->usecnt);
336 }
337 }
292 338
293 return ret; 339 return ret;
294} 340}
@@ -296,28 +342,123 @@ EXPORT_SYMBOL(ib_destroy_srq);
296 342
297/* Queue pairs */ 343/* Queue pairs */
298 344
345static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
346{
347 struct ib_qp *qp = context;
348
349 list_for_each_entry(event->element.qp, &qp->open_list, open_list)
350 event->element.qp->event_handler(event, event->element.qp->qp_context);
351}
352
353static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
354{
355 mutex_lock(&xrcd->tgt_qp_mutex);
356 list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
357 mutex_unlock(&xrcd->tgt_qp_mutex);
358}
359
360static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
361 void (*event_handler)(struct ib_event *, void *),
362 void *qp_context)
363{
364 struct ib_qp *qp;
365 unsigned long flags;
366
367 qp = kzalloc(sizeof *qp, GFP_KERNEL);
368 if (!qp)
369 return ERR_PTR(-ENOMEM);
370
371 qp->real_qp = real_qp;
372 atomic_inc(&real_qp->usecnt);
373 qp->device = real_qp->device;
374 qp->event_handler = event_handler;
375 qp->qp_context = qp_context;
376 qp->qp_num = real_qp->qp_num;
377 qp->qp_type = real_qp->qp_type;
378
379 spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
380 list_add(&qp->open_list, &real_qp->open_list);
381 spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
382
383 return qp;
384}
385
386struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
387 struct ib_qp_open_attr *qp_open_attr)
388{
389 struct ib_qp *qp, *real_qp;
390
391 if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
392 return ERR_PTR(-EINVAL);
393
394 qp = ERR_PTR(-EINVAL);
395 mutex_lock(&xrcd->tgt_qp_mutex);
396 list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
397 if (real_qp->qp_num == qp_open_attr->qp_num) {
398 qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
399 qp_open_attr->qp_context);
400 break;
401 }
402 }
403 mutex_unlock(&xrcd->tgt_qp_mutex);
404 return qp;
405}
406EXPORT_SYMBOL(ib_open_qp);
407
299struct ib_qp *ib_create_qp(struct ib_pd *pd, 408struct ib_qp *ib_create_qp(struct ib_pd *pd,
300 struct ib_qp_init_attr *qp_init_attr) 409 struct ib_qp_init_attr *qp_init_attr)
301{ 410{
302 struct ib_qp *qp; 411 struct ib_qp *qp, *real_qp;
412 struct ib_device *device;
303 413
304 qp = pd->device->create_qp(pd, qp_init_attr, NULL); 414 device = pd ? pd->device : qp_init_attr->xrcd->device;
415 qp = device->create_qp(pd, qp_init_attr, NULL);
305 416
306 if (!IS_ERR(qp)) { 417 if (!IS_ERR(qp)) {
307 qp->device = pd->device; 418 qp->device = device;
308 qp->pd = pd; 419 qp->real_qp = qp;
309 qp->send_cq = qp_init_attr->send_cq; 420 qp->uobject = NULL;
310 qp->recv_cq = qp_init_attr->recv_cq; 421 qp->qp_type = qp_init_attr->qp_type;
311 qp->srq = qp_init_attr->srq; 422
312 qp->uobject = NULL; 423 if (qp_init_attr->qp_type == IB_QPT_XRC_TGT) {
313 qp->event_handler = qp_init_attr->event_handler; 424 qp->event_handler = __ib_shared_qp_event_handler;
314 qp->qp_context = qp_init_attr->qp_context; 425 qp->qp_context = qp;
315 qp->qp_type = qp_init_attr->qp_type; 426 qp->pd = NULL;
316 atomic_inc(&pd->usecnt); 427 qp->send_cq = qp->recv_cq = NULL;
317 atomic_inc(&qp_init_attr->send_cq->usecnt); 428 qp->srq = NULL;
318 atomic_inc(&qp_init_attr->recv_cq->usecnt); 429 qp->xrcd = qp_init_attr->xrcd;
319 if (qp_init_attr->srq) 430 atomic_inc(&qp_init_attr->xrcd->usecnt);
320 atomic_inc(&qp_init_attr->srq->usecnt); 431 INIT_LIST_HEAD(&qp->open_list);
432 atomic_set(&qp->usecnt, 0);
433
434 real_qp = qp;
435 qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
436 qp_init_attr->qp_context);
437 if (!IS_ERR(qp))
438 __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
439 else
440 real_qp->device->destroy_qp(real_qp);
441 } else {
442 qp->event_handler = qp_init_attr->event_handler;
443 qp->qp_context = qp_init_attr->qp_context;
444 if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
445 qp->recv_cq = NULL;
446 qp->srq = NULL;
447 } else {
448 qp->recv_cq = qp_init_attr->recv_cq;
449 atomic_inc(&qp_init_attr->recv_cq->usecnt);
450 qp->srq = qp_init_attr->srq;
451 if (qp->srq)
452 atomic_inc(&qp_init_attr->srq->usecnt);
453 }
454
455 qp->pd = pd;
456 qp->send_cq = qp_init_attr->send_cq;
457 qp->xrcd = NULL;
458
459 atomic_inc(&pd->usecnt);
460 atomic_inc(&qp_init_attr->send_cq->usecnt);
461 }
321 } 462 }
322 463
323 return qp; 464 return qp;
@@ -326,8 +467,8 @@ EXPORT_SYMBOL(ib_create_qp);
326 467
327static const struct { 468static const struct {
328 int valid; 469 int valid;
329 enum ib_qp_attr_mask req_param[IB_QPT_RAW_ETHERTYPE + 1]; 470 enum ib_qp_attr_mask req_param[IB_QPT_MAX];
330 enum ib_qp_attr_mask opt_param[IB_QPT_RAW_ETHERTYPE + 1]; 471 enum ib_qp_attr_mask opt_param[IB_QPT_MAX];
331} qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = { 472} qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
332 [IB_QPS_RESET] = { 473 [IB_QPS_RESET] = {
333 [IB_QPS_RESET] = { .valid = 1 }, 474 [IB_QPS_RESET] = { .valid = 1 },
@@ -343,6 +484,12 @@ static const struct {
343 [IB_QPT_RC] = (IB_QP_PKEY_INDEX | 484 [IB_QPT_RC] = (IB_QP_PKEY_INDEX |
344 IB_QP_PORT | 485 IB_QP_PORT |
345 IB_QP_ACCESS_FLAGS), 486 IB_QP_ACCESS_FLAGS),
487 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX |
488 IB_QP_PORT |
489 IB_QP_ACCESS_FLAGS),
490 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX |
491 IB_QP_PORT |
492 IB_QP_ACCESS_FLAGS),
346 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX | 493 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
347 IB_QP_QKEY), 494 IB_QP_QKEY),
348 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX | 495 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
@@ -365,6 +512,12 @@ static const struct {
365 [IB_QPT_RC] = (IB_QP_PKEY_INDEX | 512 [IB_QPT_RC] = (IB_QP_PKEY_INDEX |
366 IB_QP_PORT | 513 IB_QP_PORT |
367 IB_QP_ACCESS_FLAGS), 514 IB_QP_ACCESS_FLAGS),
515 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX |
516 IB_QP_PORT |
517 IB_QP_ACCESS_FLAGS),
518 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX |
519 IB_QP_PORT |
520 IB_QP_ACCESS_FLAGS),
368 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX | 521 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
369 IB_QP_QKEY), 522 IB_QP_QKEY),
370 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX | 523 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
@@ -384,6 +537,16 @@ static const struct {
384 IB_QP_RQ_PSN | 537 IB_QP_RQ_PSN |
385 IB_QP_MAX_DEST_RD_ATOMIC | 538 IB_QP_MAX_DEST_RD_ATOMIC |
386 IB_QP_MIN_RNR_TIMER), 539 IB_QP_MIN_RNR_TIMER),
540 [IB_QPT_XRC_INI] = (IB_QP_AV |
541 IB_QP_PATH_MTU |
542 IB_QP_DEST_QPN |
543 IB_QP_RQ_PSN),
544 [IB_QPT_XRC_TGT] = (IB_QP_AV |
545 IB_QP_PATH_MTU |
546 IB_QP_DEST_QPN |
547 IB_QP_RQ_PSN |
548 IB_QP_MAX_DEST_RD_ATOMIC |
549 IB_QP_MIN_RNR_TIMER),
387 }, 550 },
388 .opt_param = { 551 .opt_param = {
389 [IB_QPT_UD] = (IB_QP_PKEY_INDEX | 552 [IB_QPT_UD] = (IB_QP_PKEY_INDEX |
@@ -394,6 +557,12 @@ static const struct {
394 [IB_QPT_RC] = (IB_QP_ALT_PATH | 557 [IB_QPT_RC] = (IB_QP_ALT_PATH |
395 IB_QP_ACCESS_FLAGS | 558 IB_QP_ACCESS_FLAGS |
396 IB_QP_PKEY_INDEX), 559 IB_QP_PKEY_INDEX),
560 [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH |
561 IB_QP_ACCESS_FLAGS |
562 IB_QP_PKEY_INDEX),
563 [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH |
564 IB_QP_ACCESS_FLAGS |
565 IB_QP_PKEY_INDEX),
397 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX | 566 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
398 IB_QP_QKEY), 567 IB_QP_QKEY),
399 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX | 568 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
@@ -414,6 +583,13 @@ static const struct {
414 IB_QP_RNR_RETRY | 583 IB_QP_RNR_RETRY |
415 IB_QP_SQ_PSN | 584 IB_QP_SQ_PSN |
416 IB_QP_MAX_QP_RD_ATOMIC), 585 IB_QP_MAX_QP_RD_ATOMIC),
586 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT |
587 IB_QP_RETRY_CNT |
588 IB_QP_RNR_RETRY |
589 IB_QP_SQ_PSN |
590 IB_QP_MAX_QP_RD_ATOMIC),
591 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT |
592 IB_QP_SQ_PSN),
417 [IB_QPT_SMI] = IB_QP_SQ_PSN, 593 [IB_QPT_SMI] = IB_QP_SQ_PSN,
418 [IB_QPT_GSI] = IB_QP_SQ_PSN, 594 [IB_QPT_GSI] = IB_QP_SQ_PSN,
419 }, 595 },
@@ -429,6 +605,15 @@ static const struct {
429 IB_QP_ACCESS_FLAGS | 605 IB_QP_ACCESS_FLAGS |
430 IB_QP_MIN_RNR_TIMER | 606 IB_QP_MIN_RNR_TIMER |
431 IB_QP_PATH_MIG_STATE), 607 IB_QP_PATH_MIG_STATE),
608 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
609 IB_QP_ALT_PATH |
610 IB_QP_ACCESS_FLAGS |
611 IB_QP_PATH_MIG_STATE),
612 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
613 IB_QP_ALT_PATH |
614 IB_QP_ACCESS_FLAGS |
615 IB_QP_MIN_RNR_TIMER |
616 IB_QP_PATH_MIG_STATE),
432 [IB_QPT_SMI] = (IB_QP_CUR_STATE | 617 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
433 IB_QP_QKEY), 618 IB_QP_QKEY),
434 [IB_QPT_GSI] = (IB_QP_CUR_STATE | 619 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
@@ -453,6 +638,15 @@ static const struct {
453 IB_QP_ALT_PATH | 638 IB_QP_ALT_PATH |
454 IB_QP_PATH_MIG_STATE | 639 IB_QP_PATH_MIG_STATE |
455 IB_QP_MIN_RNR_TIMER), 640 IB_QP_MIN_RNR_TIMER),
641 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
642 IB_QP_ACCESS_FLAGS |
643 IB_QP_ALT_PATH |
644 IB_QP_PATH_MIG_STATE),
645 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
646 IB_QP_ACCESS_FLAGS |
647 IB_QP_ALT_PATH |
648 IB_QP_PATH_MIG_STATE |
649 IB_QP_MIN_RNR_TIMER),
456 [IB_QPT_SMI] = (IB_QP_CUR_STATE | 650 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
457 IB_QP_QKEY), 651 IB_QP_QKEY),
458 [IB_QPT_GSI] = (IB_QP_CUR_STATE | 652 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
@@ -465,6 +659,8 @@ static const struct {
465 [IB_QPT_UD] = IB_QP_EN_SQD_ASYNC_NOTIFY, 659 [IB_QPT_UD] = IB_QP_EN_SQD_ASYNC_NOTIFY,
466 [IB_QPT_UC] = IB_QP_EN_SQD_ASYNC_NOTIFY, 660 [IB_QPT_UC] = IB_QP_EN_SQD_ASYNC_NOTIFY,
467 [IB_QPT_RC] = IB_QP_EN_SQD_ASYNC_NOTIFY, 661 [IB_QPT_RC] = IB_QP_EN_SQD_ASYNC_NOTIFY,
662 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
663 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
468 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY, 664 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
469 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY 665 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
470 } 666 }
@@ -487,6 +683,15 @@ static const struct {
487 IB_QP_ACCESS_FLAGS | 683 IB_QP_ACCESS_FLAGS |
488 IB_QP_MIN_RNR_TIMER | 684 IB_QP_MIN_RNR_TIMER |
489 IB_QP_PATH_MIG_STATE), 685 IB_QP_PATH_MIG_STATE),
686 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
687 IB_QP_ALT_PATH |
688 IB_QP_ACCESS_FLAGS |
689 IB_QP_PATH_MIG_STATE),
690 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
691 IB_QP_ALT_PATH |
692 IB_QP_ACCESS_FLAGS |
693 IB_QP_MIN_RNR_TIMER |
694 IB_QP_PATH_MIG_STATE),
490 [IB_QPT_SMI] = (IB_QP_CUR_STATE | 695 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
491 IB_QP_QKEY), 696 IB_QP_QKEY),
492 [IB_QPT_GSI] = (IB_QP_CUR_STATE | 697 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
@@ -515,6 +720,25 @@ static const struct {
515 IB_QP_PKEY_INDEX | 720 IB_QP_PKEY_INDEX |
516 IB_QP_MIN_RNR_TIMER | 721 IB_QP_MIN_RNR_TIMER |
517 IB_QP_PATH_MIG_STATE), 722 IB_QP_PATH_MIG_STATE),
723 [IB_QPT_XRC_INI] = (IB_QP_PORT |
724 IB_QP_AV |
725 IB_QP_TIMEOUT |
726 IB_QP_RETRY_CNT |
727 IB_QP_RNR_RETRY |
728 IB_QP_MAX_QP_RD_ATOMIC |
729 IB_QP_ALT_PATH |
730 IB_QP_ACCESS_FLAGS |
731 IB_QP_PKEY_INDEX |
732 IB_QP_PATH_MIG_STATE),
733 [IB_QPT_XRC_TGT] = (IB_QP_PORT |
734 IB_QP_AV |
735 IB_QP_TIMEOUT |
736 IB_QP_MAX_DEST_RD_ATOMIC |
737 IB_QP_ALT_PATH |
738 IB_QP_ACCESS_FLAGS |
739 IB_QP_PKEY_INDEX |
740 IB_QP_MIN_RNR_TIMER |
741 IB_QP_PATH_MIG_STATE),
518 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX | 742 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
519 IB_QP_QKEY), 743 IB_QP_QKEY),
520 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX | 744 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
@@ -579,7 +803,7 @@ int ib_modify_qp(struct ib_qp *qp,
579 struct ib_qp_attr *qp_attr, 803 struct ib_qp_attr *qp_attr,
580 int qp_attr_mask) 804 int qp_attr_mask)
581{ 805{
582 return qp->device->modify_qp(qp, qp_attr, qp_attr_mask, NULL); 806 return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
583} 807}
584EXPORT_SYMBOL(ib_modify_qp); 808EXPORT_SYMBOL(ib_modify_qp);
585 809
@@ -589,11 +813,59 @@ int ib_query_qp(struct ib_qp *qp,
589 struct ib_qp_init_attr *qp_init_attr) 813 struct ib_qp_init_attr *qp_init_attr)
590{ 814{
591 return qp->device->query_qp ? 815 return qp->device->query_qp ?
592 qp->device->query_qp(qp, qp_attr, qp_attr_mask, qp_init_attr) : 816 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
593 -ENOSYS; 817 -ENOSYS;
594} 818}
595EXPORT_SYMBOL(ib_query_qp); 819EXPORT_SYMBOL(ib_query_qp);
596 820
821int ib_close_qp(struct ib_qp *qp)
822{
823 struct ib_qp *real_qp;
824 unsigned long flags;
825
826 real_qp = qp->real_qp;
827 if (real_qp == qp)
828 return -EINVAL;
829
830 spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
831 list_del(&qp->open_list);
832 spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
833
834 atomic_dec(&real_qp->usecnt);
835 kfree(qp);
836
837 return 0;
838}
839EXPORT_SYMBOL(ib_close_qp);
840
841static int __ib_destroy_shared_qp(struct ib_qp *qp)
842{
843 struct ib_xrcd *xrcd;
844 struct ib_qp *real_qp;
845 int ret;
846
847 real_qp = qp->real_qp;
848 xrcd = real_qp->xrcd;
849
850 mutex_lock(&xrcd->tgt_qp_mutex);
851 ib_close_qp(qp);
852 if (atomic_read(&real_qp->usecnt) == 0)
853 list_del(&real_qp->xrcd_list);
854 else
855 real_qp = NULL;
856 mutex_unlock(&xrcd->tgt_qp_mutex);
857
858 if (real_qp) {
859 ret = ib_destroy_qp(real_qp);
860 if (!ret)
861 atomic_dec(&xrcd->usecnt);
862 else
863 __ib_insert_xrcd_qp(xrcd, real_qp);
864 }
865
866 return 0;
867}
868
597int ib_destroy_qp(struct ib_qp *qp) 869int ib_destroy_qp(struct ib_qp *qp)
598{ 870{
599 struct ib_pd *pd; 871 struct ib_pd *pd;
@@ -601,16 +873,25 @@ int ib_destroy_qp(struct ib_qp *qp)
601 struct ib_srq *srq; 873 struct ib_srq *srq;
602 int ret; 874 int ret;
603 875
604 pd = qp->pd; 876 if (atomic_read(&qp->usecnt))
605 scq = qp->send_cq; 877 return -EBUSY;
606 rcq = qp->recv_cq; 878
607 srq = qp->srq; 879 if (qp->real_qp != qp)
880 return __ib_destroy_shared_qp(qp);
881
882 pd = qp->pd;
883 scq = qp->send_cq;
884 rcq = qp->recv_cq;
885 srq = qp->srq;
608 886
609 ret = qp->device->destroy_qp(qp); 887 ret = qp->device->destroy_qp(qp);
610 if (!ret) { 888 if (!ret) {
611 atomic_dec(&pd->usecnt); 889 if (pd)
612 atomic_dec(&scq->usecnt); 890 atomic_dec(&pd->usecnt);
613 atomic_dec(&rcq->usecnt); 891 if (scq)
892 atomic_dec(&scq->usecnt);
893 if (rcq)
894 atomic_dec(&rcq->usecnt);
614 if (srq) 895 if (srq)
615 atomic_dec(&srq->usecnt); 896 atomic_dec(&srq->usecnt);
616 } 897 }
@@ -920,3 +1201,42 @@ int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
920 return qp->device->detach_mcast(qp, gid, lid); 1201 return qp->device->detach_mcast(qp, gid, lid);
921} 1202}
922EXPORT_SYMBOL(ib_detach_mcast); 1203EXPORT_SYMBOL(ib_detach_mcast);
1204
1205struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1206{
1207 struct ib_xrcd *xrcd;
1208
1209 if (!device->alloc_xrcd)
1210 return ERR_PTR(-ENOSYS);
1211
1212 xrcd = device->alloc_xrcd(device, NULL, NULL);
1213 if (!IS_ERR(xrcd)) {
1214 xrcd->device = device;
1215 xrcd->inode = NULL;
1216 atomic_set(&xrcd->usecnt, 0);
1217 mutex_init(&xrcd->tgt_qp_mutex);
1218 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1219 }
1220
1221 return xrcd;
1222}
1223EXPORT_SYMBOL(ib_alloc_xrcd);
1224
1225int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1226{
1227 struct ib_qp *qp;
1228 int ret;
1229
1230 if (atomic_read(&xrcd->usecnt))
1231 return -EBUSY;
1232
1233 while (!list_empty(&xrcd->tgt_qp_list)) {
1234 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1235 ret = ib_destroy_qp(qp);
1236 if (ret)
1237 return ret;
1238 }
1239
1240 return xrcd->device->dealloc_xrcd(xrcd);
1241}
1242EXPORT_SYMBOL(ib_dealloc_xrcd);
diff --git a/drivers/infiniband/hw/amso1100/c2_ae.c b/drivers/infiniband/hw/amso1100/c2_ae.c
index 24f9e3a90e8e..32d34e88d5cf 100644
--- a/drivers/infiniband/hw/amso1100/c2_ae.c
+++ b/drivers/infiniband/hw/amso1100/c2_ae.c
@@ -288,6 +288,11 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
288 cm_event.private_data_len = 288 cm_event.private_data_len =
289 be32_to_cpu(req->private_data_length); 289 be32_to_cpu(req->private_data_length);
290 cm_event.private_data = req->private_data; 290 cm_event.private_data = req->private_data;
291 /*
292 * Until ird/ord negotiation via MPAv2 support is added, send
293 * max supported values
294 */
295 cm_event.ird = cm_event.ord = 128;
291 296
292 if (cm_id->event_handler) 297 if (cm_id->event_handler)
293 cm_id->event_handler(cm_id, &cm_event); 298 cm_id->event_handler(cm_id, &cm_event);
diff --git a/drivers/infiniband/hw/amso1100/c2_intr.c b/drivers/infiniband/hw/amso1100/c2_intr.c
index 0ebe4e806b86..8951db4ae29d 100644
--- a/drivers/infiniband/hw/amso1100/c2_intr.c
+++ b/drivers/infiniband/hw/amso1100/c2_intr.c
@@ -183,6 +183,11 @@ static void handle_vq(struct c2_dev *c2dev, u32 mq_index)
183 case IW_CM_EVENT_ESTABLISHED: 183 case IW_CM_EVENT_ESTABLISHED:
184 c2_set_qp_state(req->qp, 184 c2_set_qp_state(req->qp,
185 C2_QP_STATE_RTS); 185 C2_QP_STATE_RTS);
186 /*
187 * Until ird/ord negotiation via MPAv2 support is added, send
188 * max supported values
189 */
190 cm_event.ird = cm_event.ord = 128;
186 case IW_CM_EVENT_CLOSE: 191 case IW_CM_EVENT_CLOSE:
187 192
188 /* 193 /*
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c
index f101bb73be63..12f923d64e42 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -753,10 +753,7 @@ static struct net_device *c2_pseudo_netdev_init(struct c2_dev *c2dev)
753 memcpy_fromio(netdev->dev_addr, c2dev->kva + C2_REGS_RDMA_ENADDR, 6); 753 memcpy_fromio(netdev->dev_addr, c2dev->kva + C2_REGS_RDMA_ENADDR, 6);
754 754
755 /* Print out the MAC address */ 755 /* Print out the MAC address */
756 pr_debug("%s: MAC %02X:%02X:%02X:%02X:%02X:%02X\n", 756 pr_debug("%s: MAC %pM\n", netdev->name, netdev->dev_addr);
757 netdev->name,
758 netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
759 netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
760 757
761#if 0 758#if 0
762 /* Disable network packets */ 759 /* Disable network packets */
diff --git a/drivers/infiniband/hw/cxgb3/iwch_cm.c b/drivers/infiniband/hw/cxgb3/iwch_cm.c
index 6cd642aaa4de..de6d0774e609 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_cm.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_cm.c
@@ -753,6 +753,11 @@ static void connect_request_upcall(struct iwch_ep *ep)
753 event.private_data_len = ep->plen; 753 event.private_data_len = ep->plen;
754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
755 event.provider_data = ep; 755 event.provider_data = ep;
756 /*
757 * Until ird/ord negotiation via MPAv2 support is added, send max
758 * supported values
759 */
760 event.ird = event.ord = 8;
756 if (state_read(&ep->parent_ep->com) != DEAD) { 761 if (state_read(&ep->parent_ep->com) != DEAD) {
757 get_ep(&ep->com); 762 get_ep(&ep->com);
758 ep->parent_ep->com.cm_id->event_handler( 763 ep->parent_ep->com.cm_id->event_handler(
@@ -770,6 +775,11 @@ static void established_upcall(struct iwch_ep *ep)
770 PDBG("%s ep %p\n", __func__, ep); 775 PDBG("%s ep %p\n", __func__, ep);
771 memset(&event, 0, sizeof(event)); 776 memset(&event, 0, sizeof(event));
772 event.event = IW_CM_EVENT_ESTABLISHED; 777 event.event = IW_CM_EVENT_ESTABLISHED;
778 /*
779 * Until ird/ord negotiation via MPAv2 support is added, send max
780 * supported values
781 */
782 event.ird = event.ord = 8;
773 if (ep->com.cm_id) { 783 if (ep->com.cm_id) {
774 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid); 784 PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
775 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 785 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
diff --git a/drivers/infiniband/hw/cxgb3/iwch_ev.c b/drivers/infiniband/hw/cxgb3/iwch_ev.c
index 71e0d845da3d..abcc9e76962b 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_ev.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_ev.c
@@ -46,6 +46,7 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
46 struct ib_event event; 46 struct ib_event event;
47 struct iwch_qp_attributes attrs; 47 struct iwch_qp_attributes attrs;
48 struct iwch_qp *qhp; 48 struct iwch_qp *qhp;
49 unsigned long flag;
49 50
50 spin_lock(&rnicp->lock); 51 spin_lock(&rnicp->lock);
51 qhp = get_qhp(rnicp, CQE_QPID(rsp_msg->cqe)); 52 qhp = get_qhp(rnicp, CQE_QPID(rsp_msg->cqe));
@@ -94,7 +95,9 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
94 if (qhp->ibqp.event_handler) 95 if (qhp->ibqp.event_handler)
95 (*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context); 96 (*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
96 97
98 spin_lock_irqsave(&chp->comp_handler_lock, flag);
97 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context); 99 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
100 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
98 101
99 if (atomic_dec_and_test(&qhp->refcnt)) 102 if (atomic_dec_and_test(&qhp->refcnt))
100 wake_up(&qhp->wait); 103 wake_up(&qhp->wait);
@@ -107,6 +110,7 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
107 struct iwch_cq *chp; 110 struct iwch_cq *chp;
108 struct iwch_qp *qhp; 111 struct iwch_qp *qhp;
109 u32 cqid = RSPQ_CQID(rsp_msg); 112 u32 cqid = RSPQ_CQID(rsp_msg);
113 unsigned long flag;
110 114
111 rnicp = (struct iwch_dev *) rdev_p->ulp; 115 rnicp = (struct iwch_dev *) rdev_p->ulp;
112 spin_lock(&rnicp->lock); 116 spin_lock(&rnicp->lock);
@@ -170,7 +174,9 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
170 */ 174 */
171 if (qhp->ep && SQ_TYPE(rsp_msg->cqe)) 175 if (qhp->ep && SQ_TYPE(rsp_msg->cqe))
172 dst_confirm(qhp->ep->dst); 176 dst_confirm(qhp->ep->dst);
177 spin_lock_irqsave(&chp->comp_handler_lock, flag);
173 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context); 178 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
179 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
174 break; 180 break;
175 181
176 case TPT_ERR_STAG: 182 case TPT_ERR_STAG:
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c
index c7d9411f2954..37c224fc3ad9 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c
@@ -190,6 +190,7 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve
190 chp->rhp = rhp; 190 chp->rhp = rhp;
191 chp->ibcq.cqe = 1 << chp->cq.size_log2; 191 chp->ibcq.cqe = 1 << chp->cq.size_log2;
192 spin_lock_init(&chp->lock); 192 spin_lock_init(&chp->lock);
193 spin_lock_init(&chp->comp_handler_lock);
193 atomic_set(&chp->refcnt, 1); 194 atomic_set(&chp->refcnt, 1);
194 init_waitqueue_head(&chp->wait); 195 init_waitqueue_head(&chp->wait);
195 if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) { 196 if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.h b/drivers/infiniband/hw/cxgb3/iwch_provider.h
index 9a342c9b220d..87c14b0c5ac0 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.h
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.h
@@ -103,6 +103,7 @@ struct iwch_cq {
103 struct iwch_dev *rhp; 103 struct iwch_dev *rhp;
104 struct t3_cq cq; 104 struct t3_cq cq;
105 spinlock_t lock; 105 spinlock_t lock;
106 spinlock_t comp_handler_lock;
106 atomic_t refcnt; 107 atomic_t refcnt;
107 wait_queue_head_t wait; 108 wait_queue_head_t wait;
108 u32 __user *user_rptr_addr; 109 u32 __user *user_rptr_addr;
diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c
index ecd313f359a4..bea5839d89ee 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_qp.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c
@@ -822,8 +822,11 @@ static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
822 flushed = cxio_flush_rq(&qhp->wq, &rchp->cq, count); 822 flushed = cxio_flush_rq(&qhp->wq, &rchp->cq, count);
823 spin_unlock(&qhp->lock); 823 spin_unlock(&qhp->lock);
824 spin_unlock_irqrestore(&rchp->lock, *flag); 824 spin_unlock_irqrestore(&rchp->lock, *flag);
825 if (flushed) 825 if (flushed) {
826 spin_lock_irqsave(&rchp->comp_handler_lock, *flag);
826 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); 827 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
828 spin_unlock_irqrestore(&rchp->comp_handler_lock, *flag);
829 }
827 830
828 /* locking hierarchy: cq lock first, then qp lock. */ 831 /* locking hierarchy: cq lock first, then qp lock. */
829 spin_lock_irqsave(&schp->lock, *flag); 832 spin_lock_irqsave(&schp->lock, *flag);
@@ -833,8 +836,11 @@ static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
833 flushed = cxio_flush_sq(&qhp->wq, &schp->cq, count); 836 flushed = cxio_flush_sq(&qhp->wq, &schp->cq, count);
834 spin_unlock(&qhp->lock); 837 spin_unlock(&qhp->lock);
835 spin_unlock_irqrestore(&schp->lock, *flag); 838 spin_unlock_irqrestore(&schp->lock, *flag);
836 if (flushed) 839 if (flushed) {
840 spin_lock_irqsave(&schp->comp_handler_lock, *flag);
837 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); 841 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
842 spin_unlock_irqrestore(&schp->comp_handler_lock, *flag);
843 }
838 844
839 /* deref */ 845 /* deref */
840 if (atomic_dec_and_test(&qhp->refcnt)) 846 if (atomic_dec_and_test(&qhp->refcnt))
@@ -853,11 +859,15 @@ static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
853 if (qhp->ibqp.uobject) { 859 if (qhp->ibqp.uobject) {
854 cxio_set_wq_in_error(&qhp->wq); 860 cxio_set_wq_in_error(&qhp->wq);
855 cxio_set_cq_in_error(&rchp->cq); 861 cxio_set_cq_in_error(&rchp->cq);
862 spin_lock_irqsave(&rchp->comp_handler_lock, *flag);
856 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); 863 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
864 spin_unlock_irqrestore(&rchp->comp_handler_lock, *flag);
857 if (schp != rchp) { 865 if (schp != rchp) {
858 cxio_set_cq_in_error(&schp->cq); 866 cxio_set_cq_in_error(&schp->cq);
867 spin_lock_irqsave(&schp->comp_handler_lock, *flag);
859 (*schp->ibcq.comp_handler)(&schp->ibcq, 868 (*schp->ibcq.comp_handler)(&schp->ibcq,
860 schp->ibcq.cq_context); 869 schp->ibcq.cq_context);
870 spin_unlock_irqrestore(&schp->comp_handler_lock, *flag);
861 } 871 }
862 return; 872 return;
863 } 873 }
diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index 77f769d9227d..b36cdac9c558 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -103,7 +103,8 @@ MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
103static int mpa_rev = 1; 103static int mpa_rev = 1;
104module_param(mpa_rev, int, 0644); 104module_param(mpa_rev, int, 0644);
105MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, " 105MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
106 "1 is spec compliant. (default=1)"); 106 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
107 " compliant (default=1)");
107 108
108static int markers_enabled; 109static int markers_enabled;
109module_param(markers_enabled, int, 0644); 110module_param(markers_enabled, int, 0644);
@@ -497,17 +498,21 @@ static int send_connect(struct c4iw_ep *ep)
497 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 498 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
498} 499}
499 500
500static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb) 501static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
502 u8 mpa_rev_to_use)
501{ 503{
502 int mpalen, wrlen; 504 int mpalen, wrlen;
503 struct fw_ofld_tx_data_wr *req; 505 struct fw_ofld_tx_data_wr *req;
504 struct mpa_message *mpa; 506 struct mpa_message *mpa;
507 struct mpa_v2_conn_params mpa_v2_params;
505 508
506 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 509 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
507 510
508 BUG_ON(skb_cloned(skb)); 511 BUG_ON(skb_cloned(skb));
509 512
510 mpalen = sizeof(*mpa) + ep->plen; 513 mpalen = sizeof(*mpa) + ep->plen;
514 if (mpa_rev_to_use == 2)
515 mpalen += sizeof(struct mpa_v2_conn_params);
511 wrlen = roundup(mpalen + sizeof *req, 16); 516 wrlen = roundup(mpalen + sizeof *req, 16);
512 skb = get_skb(skb, wrlen, GFP_KERNEL); 517 skb = get_skb(skb, wrlen, GFP_KERNEL);
513 if (!skb) { 518 if (!skb) {
@@ -533,12 +538,39 @@ static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
533 mpa = (struct mpa_message *)(req + 1); 538 mpa = (struct mpa_message *)(req + 1);
534 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 539 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
535 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 540 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
536 (markers_enabled ? MPA_MARKERS : 0); 541 (markers_enabled ? MPA_MARKERS : 0) |
542 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
537 mpa->private_data_size = htons(ep->plen); 543 mpa->private_data_size = htons(ep->plen);
538 mpa->revision = mpa_rev; 544 mpa->revision = mpa_rev_to_use;
545 if (mpa_rev_to_use == 1)
546 ep->tried_with_mpa_v1 = 1;
547
548 if (mpa_rev_to_use == 2) {
549 mpa->private_data_size +=
550 htons(sizeof(struct mpa_v2_conn_params));
551 mpa_v2_params.ird = htons((u16)ep->ird);
552 mpa_v2_params.ord = htons((u16)ep->ord);
553
554 if (peer2peer) {
555 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
556 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
557 mpa_v2_params.ord |=
558 htons(MPA_V2_RDMA_WRITE_RTR);
559 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
560 mpa_v2_params.ord |=
561 htons(MPA_V2_RDMA_READ_RTR);
562 }
563 memcpy(mpa->private_data, &mpa_v2_params,
564 sizeof(struct mpa_v2_conn_params));
539 565
540 if (ep->plen) 566 if (ep->plen)
541 memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen); 567 memcpy(mpa->private_data +
568 sizeof(struct mpa_v2_conn_params),
569 ep->mpa_pkt + sizeof(*mpa), ep->plen);
570 } else
571 if (ep->plen)
572 memcpy(mpa->private_data,
573 ep->mpa_pkt + sizeof(*mpa), ep->plen);
542 574
543 /* 575 /*
544 * Reference the mpa skb. This ensures the data area 576 * Reference the mpa skb. This ensures the data area
@@ -562,10 +594,13 @@ static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
562 struct fw_ofld_tx_data_wr *req; 594 struct fw_ofld_tx_data_wr *req;
563 struct mpa_message *mpa; 595 struct mpa_message *mpa;
564 struct sk_buff *skb; 596 struct sk_buff *skb;
597 struct mpa_v2_conn_params mpa_v2_params;
565 598
566 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 599 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
567 600
568 mpalen = sizeof(*mpa) + plen; 601 mpalen = sizeof(*mpa) + plen;
602 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
603 mpalen += sizeof(struct mpa_v2_conn_params);
569 wrlen = roundup(mpalen + sizeof *req, 16); 604 wrlen = roundup(mpalen + sizeof *req, 16);
570 605
571 skb = get_skb(NULL, wrlen, GFP_KERNEL); 606 skb = get_skb(NULL, wrlen, GFP_KERNEL);
@@ -595,8 +630,29 @@ static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
595 mpa->flags = MPA_REJECT; 630 mpa->flags = MPA_REJECT;
596 mpa->revision = mpa_rev; 631 mpa->revision = mpa_rev;
597 mpa->private_data_size = htons(plen); 632 mpa->private_data_size = htons(plen);
598 if (plen) 633
599 memcpy(mpa->private_data, pdata, plen); 634 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
635 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
636 mpa->private_data_size +=
637 htons(sizeof(struct mpa_v2_conn_params));
638 mpa_v2_params.ird = htons(((u16)ep->ird) |
639 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
640 0));
641 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
642 (p2p_type ==
643 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
644 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
645 FW_RI_INIT_P2PTYPE_READ_REQ ?
646 MPA_V2_RDMA_READ_RTR : 0) : 0));
647 memcpy(mpa->private_data, &mpa_v2_params,
648 sizeof(struct mpa_v2_conn_params));
649
650 if (ep->plen)
651 memcpy(mpa->private_data +
652 sizeof(struct mpa_v2_conn_params), pdata, plen);
653 } else
654 if (plen)
655 memcpy(mpa->private_data, pdata, plen);
600 656
601 /* 657 /*
602 * Reference the mpa skb again. This ensures the data area 658 * Reference the mpa skb again. This ensures the data area
@@ -617,10 +673,13 @@ static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
617 struct fw_ofld_tx_data_wr *req; 673 struct fw_ofld_tx_data_wr *req;
618 struct mpa_message *mpa; 674 struct mpa_message *mpa;
619 struct sk_buff *skb; 675 struct sk_buff *skb;
676 struct mpa_v2_conn_params mpa_v2_params;
620 677
621 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 678 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
622 679
623 mpalen = sizeof(*mpa) + plen; 680 mpalen = sizeof(*mpa) + plen;
681 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
682 mpalen += sizeof(struct mpa_v2_conn_params);
624 wrlen = roundup(mpalen + sizeof *req, 16); 683 wrlen = roundup(mpalen + sizeof *req, 16);
625 684
626 skb = get_skb(NULL, wrlen, GFP_KERNEL); 685 skb = get_skb(NULL, wrlen, GFP_KERNEL);
@@ -649,10 +708,36 @@ static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
649 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 708 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
650 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 709 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
651 (markers_enabled ? MPA_MARKERS : 0); 710 (markers_enabled ? MPA_MARKERS : 0);
652 mpa->revision = mpa_rev; 711 mpa->revision = ep->mpa_attr.version;
653 mpa->private_data_size = htons(plen); 712 mpa->private_data_size = htons(plen);
654 if (plen) 713
655 memcpy(mpa->private_data, pdata, plen); 714 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
715 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
716 mpa->private_data_size +=
717 htons(sizeof(struct mpa_v2_conn_params));
718 mpa_v2_params.ird = htons((u16)ep->ird);
719 mpa_v2_params.ord = htons((u16)ep->ord);
720 if (peer2peer && (ep->mpa_attr.p2p_type !=
721 FW_RI_INIT_P2PTYPE_DISABLED)) {
722 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
723
724 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
725 mpa_v2_params.ord |=
726 htons(MPA_V2_RDMA_WRITE_RTR);
727 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
728 mpa_v2_params.ord |=
729 htons(MPA_V2_RDMA_READ_RTR);
730 }
731
732 memcpy(mpa->private_data, &mpa_v2_params,
733 sizeof(struct mpa_v2_conn_params));
734
735 if (ep->plen)
736 memcpy(mpa->private_data +
737 sizeof(struct mpa_v2_conn_params), pdata, plen);
738 } else
739 if (plen)
740 memcpy(mpa->private_data, pdata, plen);
656 741
657 /* 742 /*
658 * Reference the mpa skb. This ensures the data area 743 * Reference the mpa skb. This ensures the data area
@@ -695,7 +780,10 @@ static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
695 780
696 /* start MPA negotiation */ 781 /* start MPA negotiation */
697 send_flowc(ep, NULL); 782 send_flowc(ep, NULL);
698 send_mpa_req(ep, skb); 783 if (ep->retry_with_mpa_v1)
784 send_mpa_req(ep, skb, 1);
785 else
786 send_mpa_req(ep, skb, mpa_rev);
699 787
700 return 0; 788 return 0;
701} 789}
@@ -769,8 +857,19 @@ static void connect_reply_upcall(struct c4iw_ep *ep, int status)
769 event.remote_addr = ep->com.remote_addr; 857 event.remote_addr = ep->com.remote_addr;
770 858
771 if ((status == 0) || (status == -ECONNREFUSED)) { 859 if ((status == 0) || (status == -ECONNREFUSED)) {
772 event.private_data_len = ep->plen; 860 if (!ep->tried_with_mpa_v1) {
773 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 861 /* this means MPA_v2 is used */
862 event.private_data_len = ep->plen -
863 sizeof(struct mpa_v2_conn_params);
864 event.private_data = ep->mpa_pkt +
865 sizeof(struct mpa_message) +
866 sizeof(struct mpa_v2_conn_params);
867 } else {
868 /* this means MPA_v1 is used */
869 event.private_data_len = ep->plen;
870 event.private_data = ep->mpa_pkt +
871 sizeof(struct mpa_message);
872 }
774 } 873 }
775 874
776 PDBG("%s ep %p tid %u status %d\n", __func__, ep, 875 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
@@ -793,9 +892,22 @@ static void connect_request_upcall(struct c4iw_ep *ep)
793 event.event = IW_CM_EVENT_CONNECT_REQUEST; 892 event.event = IW_CM_EVENT_CONNECT_REQUEST;
794 event.local_addr = ep->com.local_addr; 893 event.local_addr = ep->com.local_addr;
795 event.remote_addr = ep->com.remote_addr; 894 event.remote_addr = ep->com.remote_addr;
796 event.private_data_len = ep->plen;
797 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
798 event.provider_data = ep; 895 event.provider_data = ep;
896 if (!ep->tried_with_mpa_v1) {
897 /* this means MPA_v2 is used */
898 event.ord = ep->ord;
899 event.ird = ep->ird;
900 event.private_data_len = ep->plen -
901 sizeof(struct mpa_v2_conn_params);
902 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
903 sizeof(struct mpa_v2_conn_params);
904 } else {
905 /* this means MPA_v1 is used. Send max supported */
906 event.ord = c4iw_max_read_depth;
907 event.ird = c4iw_max_read_depth;
908 event.private_data_len = ep->plen;
909 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
910 }
799 if (state_read(&ep->parent_ep->com) != DEAD) { 911 if (state_read(&ep->parent_ep->com) != DEAD) {
800 c4iw_get_ep(&ep->com); 912 c4iw_get_ep(&ep->com);
801 ep->parent_ep->com.cm_id->event_handler( 913 ep->parent_ep->com.cm_id->event_handler(
@@ -813,6 +925,8 @@ static void established_upcall(struct c4iw_ep *ep)
813 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 925 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
814 memset(&event, 0, sizeof(event)); 926 memset(&event, 0, sizeof(event));
815 event.event = IW_CM_EVENT_ESTABLISHED; 927 event.event = IW_CM_EVENT_ESTABLISHED;
928 event.ird = ep->ird;
929 event.ord = ep->ord;
816 if (ep->com.cm_id) { 930 if (ep->com.cm_id) {
817 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 931 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
818 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 932 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
@@ -848,7 +962,10 @@ static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
848static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb) 962static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
849{ 963{
850 struct mpa_message *mpa; 964 struct mpa_message *mpa;
965 struct mpa_v2_conn_params *mpa_v2_params;
851 u16 plen; 966 u16 plen;
967 u16 resp_ird, resp_ord;
968 u8 rtr_mismatch = 0, insuff_ird = 0;
852 struct c4iw_qp_attributes attrs; 969 struct c4iw_qp_attributes attrs;
853 enum c4iw_qp_attr_mask mask; 970 enum c4iw_qp_attr_mask mask;
854 int err; 971 int err;
@@ -888,7 +1005,9 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
888 mpa = (struct mpa_message *) ep->mpa_pkt; 1005 mpa = (struct mpa_message *) ep->mpa_pkt;
889 1006
890 /* Validate MPA header. */ 1007 /* Validate MPA header. */
891 if (mpa->revision != mpa_rev) { 1008 if (mpa->revision > mpa_rev) {
1009 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1010 " Received = %d\n", __func__, mpa_rev, mpa->revision);
892 err = -EPROTO; 1011 err = -EPROTO;
893 goto err; 1012 goto err;
894 } 1013 }
@@ -938,13 +1057,66 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
938 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1057 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
939 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1058 ep->mpa_attr.recv_marker_enabled = markers_enabled;
940 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1059 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
941 ep->mpa_attr.version = mpa_rev; 1060 ep->mpa_attr.version = mpa->revision;
942 ep->mpa_attr.p2p_type = peer2peer ? p2p_type : 1061 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
943 FW_RI_INIT_P2PTYPE_DISABLED; 1062
1063 if (mpa->revision == 2) {
1064 ep->mpa_attr.enhanced_rdma_conn =
1065 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1066 if (ep->mpa_attr.enhanced_rdma_conn) {
1067 mpa_v2_params = (struct mpa_v2_conn_params *)
1068 (ep->mpa_pkt + sizeof(*mpa));
1069 resp_ird = ntohs(mpa_v2_params->ird) &
1070 MPA_V2_IRD_ORD_MASK;
1071 resp_ord = ntohs(mpa_v2_params->ord) &
1072 MPA_V2_IRD_ORD_MASK;
1073
1074 /*
1075 * This is a double-check. Ideally, below checks are
1076 * not required since ird/ord stuff has been taken
1077 * care of in c4iw_accept_cr
1078 */
1079 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1080 err = -ENOMEM;
1081 ep->ird = resp_ord;
1082 ep->ord = resp_ird;
1083 insuff_ird = 1;
1084 }
1085
1086 if (ntohs(mpa_v2_params->ird) &
1087 MPA_V2_PEER2PEER_MODEL) {
1088 if (ntohs(mpa_v2_params->ord) &
1089 MPA_V2_RDMA_WRITE_RTR)
1090 ep->mpa_attr.p2p_type =
1091 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1092 else if (ntohs(mpa_v2_params->ord) &
1093 MPA_V2_RDMA_READ_RTR)
1094 ep->mpa_attr.p2p_type =
1095 FW_RI_INIT_P2PTYPE_READ_REQ;
1096 }
1097 }
1098 } else if (mpa->revision == 1)
1099 if (peer2peer)
1100 ep->mpa_attr.p2p_type = p2p_type;
1101
944 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1102 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
945 "xmit_marker_enabled=%d, version=%d\n", __func__, 1103 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
946 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1104 "%d\n", __func__, ep->mpa_attr.crc_enabled,
947 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); 1105 ep->mpa_attr.recv_marker_enabled,
1106 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1107 ep->mpa_attr.p2p_type, p2p_type);
1108
1109 /*
1110 * If responder's RTR does not match with that of initiator, assign
1111 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1112 * generated when moving QP to RTS state.
1113 * A TERM message will be sent after QP has moved to RTS state
1114 */
1115 if ((ep->mpa_attr.version == 2) &&
1116 (ep->mpa_attr.p2p_type != p2p_type)) {
1117 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1118 rtr_mismatch = 1;
1119 }
948 1120
949 attrs.mpa_attr = ep->mpa_attr; 1121 attrs.mpa_attr = ep->mpa_attr;
950 attrs.max_ird = ep->ird; 1122 attrs.max_ird = ep->ird;
@@ -961,6 +1133,39 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
961 ep->com.qp, mask, &attrs, 1); 1133 ep->com.qp, mask, &attrs, 1);
962 if (err) 1134 if (err)
963 goto err; 1135 goto err;
1136
1137 /*
1138 * If responder's RTR requirement did not match with what initiator
1139 * supports, generate TERM message
1140 */
1141 if (rtr_mismatch) {
1142 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1143 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1144 attrs.ecode = MPA_NOMATCH_RTR;
1145 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1146 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1147 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1148 err = -ENOMEM;
1149 goto out;
1150 }
1151
1152 /*
1153 * Generate TERM if initiator IRD is not sufficient for responder
1154 * provided ORD. Currently, we do the same behaviour even when
1155 * responder provided IRD is also not sufficient as regards to
1156 * initiator ORD.
1157 */
1158 if (insuff_ird) {
1159 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1160 __func__);
1161 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1162 attrs.ecode = MPA_INSUFF_IRD;
1163 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1164 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1165 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1166 err = -ENOMEM;
1167 goto out;
1168 }
964 goto out; 1169 goto out;
965err: 1170err:
966 state_set(&ep->com, ABORTING); 1171 state_set(&ep->com, ABORTING);
@@ -973,6 +1178,7 @@ out:
973static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) 1178static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
974{ 1179{
975 struct mpa_message *mpa; 1180 struct mpa_message *mpa;
1181 struct mpa_v2_conn_params *mpa_v2_params;
976 u16 plen; 1182 u16 plen;
977 1183
978 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1184 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
@@ -1013,7 +1219,9 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1013 /* 1219 /*
1014 * Validate MPA Header. 1220 * Validate MPA Header.
1015 */ 1221 */
1016 if (mpa->revision != mpa_rev) { 1222 if (mpa->revision > mpa_rev) {
1223 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1224 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1017 abort_connection(ep, skb, GFP_KERNEL); 1225 abort_connection(ep, skb, GFP_KERNEL);
1018 return; 1226 return;
1019 } 1227 }
@@ -1056,9 +1264,37 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1056 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1264 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1057 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1265 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1058 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1266 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1059 ep->mpa_attr.version = mpa_rev; 1267 ep->mpa_attr.version = mpa->revision;
1060 ep->mpa_attr.p2p_type = peer2peer ? p2p_type : 1268 if (mpa->revision == 1)
1061 FW_RI_INIT_P2PTYPE_DISABLED; 1269 ep->tried_with_mpa_v1 = 1;
1270 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1271
1272 if (mpa->revision == 2) {
1273 ep->mpa_attr.enhanced_rdma_conn =
1274 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1275 if (ep->mpa_attr.enhanced_rdma_conn) {
1276 mpa_v2_params = (struct mpa_v2_conn_params *)
1277 (ep->mpa_pkt + sizeof(*mpa));
1278 ep->ird = ntohs(mpa_v2_params->ird) &
1279 MPA_V2_IRD_ORD_MASK;
1280 ep->ord = ntohs(mpa_v2_params->ord) &
1281 MPA_V2_IRD_ORD_MASK;
1282 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1283 if (peer2peer) {
1284 if (ntohs(mpa_v2_params->ord) &
1285 MPA_V2_RDMA_WRITE_RTR)
1286 ep->mpa_attr.p2p_type =
1287 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1288 else if (ntohs(mpa_v2_params->ord) &
1289 MPA_V2_RDMA_READ_RTR)
1290 ep->mpa_attr.p2p_type =
1291 FW_RI_INIT_P2PTYPE_READ_REQ;
1292 }
1293 }
1294 } else if (mpa->revision == 1)
1295 if (peer2peer)
1296 ep->mpa_attr.p2p_type = p2p_type;
1297
1062 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1298 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1063 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__, 1299 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1064 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1300 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
@@ -1550,6 +1786,112 @@ static int is_neg_adv_abort(unsigned int status)
1550 status == CPL_ERR_PERSIST_NEG_ADVICE; 1786 status == CPL_ERR_PERSIST_NEG_ADVICE;
1551} 1787}
1552 1788
1789static int c4iw_reconnect(struct c4iw_ep *ep)
1790{
1791 int err = 0;
1792 struct rtable *rt;
1793 struct net_device *pdev;
1794 struct neighbour *neigh;
1795 int step;
1796
1797 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1798 init_timer(&ep->timer);
1799
1800 /*
1801 * Allocate an active TID to initiate a TCP connection.
1802 */
1803 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1804 if (ep->atid == -1) {
1805 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1806 err = -ENOMEM;
1807 goto fail2;
1808 }
1809
1810 /* find a route */
1811 rt = find_route(ep->com.dev,
1812 ep->com.cm_id->local_addr.sin_addr.s_addr,
1813 ep->com.cm_id->remote_addr.sin_addr.s_addr,
1814 ep->com.cm_id->local_addr.sin_port,
1815 ep->com.cm_id->remote_addr.sin_port, 0);
1816 if (!rt) {
1817 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1818 err = -EHOSTUNREACH;
1819 goto fail3;
1820 }
1821 ep->dst = &rt->dst;
1822
1823 neigh = dst_get_neighbour(ep->dst);
1824
1825 /* get a l2t entry */
1826 if (neigh->dev->flags & IFF_LOOPBACK) {
1827 PDBG("%s LOOPBACK\n", __func__);
1828 pdev = ip_dev_find(&init_net,
1829 ep->com.cm_id->remote_addr.sin_addr.s_addr);
1830 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1831 neigh, pdev, 0);
1832 ep->mtu = pdev->mtu;
1833 ep->tx_chan = cxgb4_port_chan(pdev);
1834 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1835 step = ep->com.dev->rdev.lldi.ntxq /
1836 ep->com.dev->rdev.lldi.nchan;
1837 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1838 step = ep->com.dev->rdev.lldi.nrxq /
1839 ep->com.dev->rdev.lldi.nchan;
1840 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1841 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1842 cxgb4_port_idx(pdev) * step];
1843 dev_put(pdev);
1844 } else {
1845 ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
1846 neigh, neigh->dev, 0);
1847 ep->mtu = dst_mtu(ep->dst);
1848 ep->tx_chan = cxgb4_port_chan(neigh->dev);
1849 ep->smac_idx = (cxgb4_port_viid(neigh->dev) & 0x7F) << 1;
1850 step = ep->com.dev->rdev.lldi.ntxq /
1851 ep->com.dev->rdev.lldi.nchan;
1852 ep->txq_idx = cxgb4_port_idx(neigh->dev) * step;
1853 ep->ctrlq_idx = cxgb4_port_idx(neigh->dev);
1854 step = ep->com.dev->rdev.lldi.nrxq /
1855 ep->com.dev->rdev.lldi.nchan;
1856 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1857 cxgb4_port_idx(neigh->dev) * step];
1858 }
1859 if (!ep->l2t) {
1860 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1861 err = -ENOMEM;
1862 goto fail4;
1863 }
1864
1865 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
1866 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
1867 ep->l2t->idx);
1868
1869 state_set(&ep->com, CONNECTING);
1870 ep->tos = 0;
1871
1872 /* send connect request to rnic */
1873 err = send_connect(ep);
1874 if (!err)
1875 goto out;
1876
1877 cxgb4_l2t_release(ep->l2t);
1878fail4:
1879 dst_release(ep->dst);
1880fail3:
1881 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
1882fail2:
1883 /*
1884 * remember to send notification to upper layer.
1885 * We are in here so the upper layer is not aware that this is
1886 * re-connect attempt and so, upper layer is still waiting for
1887 * response of 1st connect request.
1888 */
1889 connect_reply_upcall(ep, -ECONNRESET);
1890 c4iw_put_ep(&ep->com);
1891out:
1892 return err;
1893}
1894
1553static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) 1895static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1554{ 1896{
1555 struct cpl_abort_req_rss *req = cplhdr(skb); 1897 struct cpl_abort_req_rss *req = cplhdr(skb);
@@ -1573,8 +1915,11 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1573 1915
1574 /* 1916 /*
1575 * Wake up any threads in rdma_init() or rdma_fini(). 1917 * Wake up any threads in rdma_init() or rdma_fini().
1918 * However, this is not needed if com state is just
1919 * MPA_REQ_SENT
1576 */ 1920 */
1577 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1921 if (ep->com.state != MPA_REQ_SENT)
1922 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1578 1923
1579 mutex_lock(&ep->com.mutex); 1924 mutex_lock(&ep->com.mutex);
1580 switch (ep->com.state) { 1925 switch (ep->com.state) {
@@ -1585,7 +1930,21 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1585 break; 1930 break;
1586 case MPA_REQ_SENT: 1931 case MPA_REQ_SENT:
1587 stop_ep_timer(ep); 1932 stop_ep_timer(ep);
1588 connect_reply_upcall(ep, -ECONNRESET); 1933 if (mpa_rev == 2 && ep->tried_with_mpa_v1)
1934 connect_reply_upcall(ep, -ECONNRESET);
1935 else {
1936 /*
1937 * we just don't send notification upwards because we
1938 * want to retry with mpa_v1 without upper layers even
1939 * knowing it.
1940 *
1941 * do some housekeeping so as to re-initiate the
1942 * connection
1943 */
1944 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
1945 mpa_rev);
1946 ep->retry_with_mpa_v1 = 1;
1947 }
1589 break; 1948 break;
1590 case MPA_REP_SENT: 1949 case MPA_REP_SENT:
1591 break; 1950 break;
@@ -1621,7 +1980,9 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1621 dst_confirm(ep->dst); 1980 dst_confirm(ep->dst);
1622 if (ep->com.state != ABORTING) { 1981 if (ep->com.state != ABORTING) {
1623 __state_set(&ep->com, DEAD); 1982 __state_set(&ep->com, DEAD);
1624 release = 1; 1983 /* we don't release if we want to retry with mpa_v1 */
1984 if (!ep->retry_with_mpa_v1)
1985 release = 1;
1625 } 1986 }
1626 mutex_unlock(&ep->com.mutex); 1987 mutex_unlock(&ep->com.mutex);
1627 1988
@@ -1641,6 +2002,15 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
1641out: 2002out:
1642 if (release) 2003 if (release)
1643 release_ep_resources(ep); 2004 release_ep_resources(ep);
2005
2006 /* retry with mpa-v1 */
2007 if (ep && ep->retry_with_mpa_v1) {
2008 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2009 dst_release(ep->dst);
2010 cxgb4_l2t_release(ep->l2t);
2011 c4iw_reconnect(ep);
2012 }
2013
1644 return 0; 2014 return 0;
1645} 2015}
1646 2016
@@ -1792,18 +2162,40 @@ int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1792 goto err; 2162 goto err;
1793 } 2163 }
1794 2164
1795 cm_id->add_ref(cm_id); 2165 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1796 ep->com.cm_id = cm_id; 2166 if (conn_param->ord > ep->ird) {
1797 ep->com.qp = qp; 2167 ep->ird = conn_param->ird;
2168 ep->ord = conn_param->ord;
2169 send_mpa_reject(ep, conn_param->private_data,
2170 conn_param->private_data_len);
2171 abort_connection(ep, NULL, GFP_KERNEL);
2172 err = -ENOMEM;
2173 goto err;
2174 }
2175 if (conn_param->ird > ep->ord) {
2176 if (!ep->ord)
2177 conn_param->ird = 1;
2178 else {
2179 abort_connection(ep, NULL, GFP_KERNEL);
2180 err = -ENOMEM;
2181 goto err;
2182 }
2183 }
1798 2184
2185 }
1799 ep->ird = conn_param->ird; 2186 ep->ird = conn_param->ird;
1800 ep->ord = conn_param->ord; 2187 ep->ord = conn_param->ord;
1801 2188
1802 if (peer2peer && ep->ird == 0) 2189 if (ep->mpa_attr.version != 2)
1803 ep->ird = 1; 2190 if (peer2peer && ep->ird == 0)
2191 ep->ird = 1;
1804 2192
1805 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord); 2193 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1806 2194
2195 cm_id->add_ref(cm_id);
2196 ep->com.cm_id = cm_id;
2197 ep->com.qp = qp;
2198
1807 /* bind QP to EP and move to RTS */ 2199 /* bind QP to EP and move to RTS */
1808 attrs.mpa_attr = ep->mpa_attr; 2200 attrs.mpa_attr = ep->mpa_attr;
1809 attrs.max_ird = ep->ird; 2201 attrs.max_ird = ep->ird;
@@ -1944,6 +2336,8 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1944 ep->com.dev->rdev.lldi.nchan; 2336 ep->com.dev->rdev.lldi.nchan;
1945 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[ 2337 ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
1946 cxgb4_port_idx(neigh->dev) * step]; 2338 cxgb4_port_idx(neigh->dev) * step];
2339 ep->retry_with_mpa_v1 = 0;
2340 ep->tried_with_mpa_v1 = 0;
1947 } 2341 }
1948 if (!ep->l2t) { 2342 if (!ep->l2t) {
1949 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__); 2343 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
@@ -2323,8 +2717,11 @@ static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
2323 2717
2324 /* 2718 /*
2325 * Wake up any threads in rdma_init() or rdma_fini(). 2719 * Wake up any threads in rdma_init() or rdma_fini().
2720 * However, this is not needed if com state is just
2721 * MPA_REQ_SENT
2326 */ 2722 */
2327 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2723 if (ep->com.state != MPA_REQ_SENT)
2724 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2328 sched(dev, skb); 2725 sched(dev, skb);
2329 return 0; 2726 return 0;
2330} 2727}
diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index 1720dc790d13..f35a935267e7 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -185,7 +185,7 @@ static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
185 V_CQE_OPCODE(FW_RI_SEND) | 185 V_CQE_OPCODE(FW_RI_SEND) |
186 V_CQE_TYPE(0) | 186 V_CQE_TYPE(0) |
187 V_CQE_SWCQE(1) | 187 V_CQE_SWCQE(1) |
188 V_CQE_QPID(wq->rq.qid)); 188 V_CQE_QPID(wq->sq.qid));
189 cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen)); 189 cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
190 cq->sw_queue[cq->sw_pidx] = cqe; 190 cq->sw_queue[cq->sw_pidx] = cqe;
191 t4_swcq_produce(cq); 191 t4_swcq_produce(cq);
@@ -818,6 +818,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
818 chp->cq.size--; /* status page */ 818 chp->cq.size--; /* status page */
819 chp->ibcq.cqe = entries - 2; 819 chp->ibcq.cqe = entries - 2;
820 spin_lock_init(&chp->lock); 820 spin_lock_init(&chp->lock);
821 spin_lock_init(&chp->comp_handler_lock);
821 atomic_set(&chp->refcnt, 1); 822 atomic_set(&chp->refcnt, 1);
822 init_waitqueue_head(&chp->wait); 823 init_waitqueue_head(&chp->wait);
823 ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid); 824 ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c
index 40a13cc633a3..6d0df6ec161b 100644
--- a/drivers/infiniband/hw/cxgb4/device.c
+++ b/drivers/infiniband/hw/cxgb4/device.c
@@ -376,10 +376,8 @@ struct uld_ctx {
376 struct c4iw_dev *dev; 376 struct c4iw_dev *dev;
377}; 377};
378 378
379static void c4iw_remove(struct uld_ctx *ctx) 379static void c4iw_dealloc(struct uld_ctx *ctx)
380{ 380{
381 PDBG("%s c4iw_dev %p\n", __func__, ctx->dev);
382 c4iw_unregister_device(ctx->dev);
383 c4iw_rdev_close(&ctx->dev->rdev); 381 c4iw_rdev_close(&ctx->dev->rdev);
384 idr_destroy(&ctx->dev->cqidr); 382 idr_destroy(&ctx->dev->cqidr);
385 idr_destroy(&ctx->dev->qpidr); 383 idr_destroy(&ctx->dev->qpidr);
@@ -389,11 +387,30 @@ static void c4iw_remove(struct uld_ctx *ctx)
389 ctx->dev = NULL; 387 ctx->dev = NULL;
390} 388}
391 389
390static void c4iw_remove(struct uld_ctx *ctx)
391{
392 PDBG("%s c4iw_dev %p\n", __func__, ctx->dev);
393 c4iw_unregister_device(ctx->dev);
394 c4iw_dealloc(ctx);
395}
396
397static int rdma_supported(const struct cxgb4_lld_info *infop)
398{
399 return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
400 infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
401 infop->vr->cq.size > 0 && infop->vr->ocq.size > 0;
402}
403
392static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop) 404static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
393{ 405{
394 struct c4iw_dev *devp; 406 struct c4iw_dev *devp;
395 int ret; 407 int ret;
396 408
409 if (!rdma_supported(infop)) {
410 printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
411 pci_name(infop->pdev));
412 return ERR_PTR(-ENOSYS);
413 }
397 devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp)); 414 devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
398 if (!devp) { 415 if (!devp) {
399 printk(KERN_ERR MOD "Cannot allocate ib device\n"); 416 printk(KERN_ERR MOD "Cannot allocate ib device\n");
@@ -414,7 +431,6 @@ static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
414 431
415 ret = c4iw_rdev_open(&devp->rdev); 432 ret = c4iw_rdev_open(&devp->rdev);
416 if (ret) { 433 if (ret) {
417 mutex_unlock(&dev_mutex);
418 printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret); 434 printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
419 ib_dealloc_device(&devp->ibdev); 435 ib_dealloc_device(&devp->ibdev);
420 return ERR_PTR(ret); 436 return ERR_PTR(ret);
@@ -519,15 +535,24 @@ static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
519 case CXGB4_STATE_UP: 535 case CXGB4_STATE_UP:
520 printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev)); 536 printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
521 if (!ctx->dev) { 537 if (!ctx->dev) {
522 int ret = 0; 538 int ret;
523 539
524 ctx->dev = c4iw_alloc(&ctx->lldi); 540 ctx->dev = c4iw_alloc(&ctx->lldi);
525 if (!IS_ERR(ctx->dev)) 541 if (IS_ERR(ctx->dev)) {
526 ret = c4iw_register_device(ctx->dev); 542 printk(KERN_ERR MOD
527 if (IS_ERR(ctx->dev) || ret) 543 "%s: initialization failed: %ld\n",
544 pci_name(ctx->lldi.pdev),
545 PTR_ERR(ctx->dev));
546 ctx->dev = NULL;
547 break;
548 }
549 ret = c4iw_register_device(ctx->dev);
550 if (ret) {
528 printk(KERN_ERR MOD 551 printk(KERN_ERR MOD
529 "%s: RDMA registration failed: %d\n", 552 "%s: RDMA registration failed: %d\n",
530 pci_name(ctx->lldi.pdev), ret); 553 pci_name(ctx->lldi.pdev), ret);
554 c4iw_dealloc(ctx);
555 }
531 } 556 }
532 break; 557 break;
533 case CXGB4_STATE_DOWN: 558 case CXGB4_STATE_DOWN:
diff --git a/drivers/infiniband/hw/cxgb4/ev.c b/drivers/infiniband/hw/cxgb4/ev.c
index c13041a0aeba..397cb36cf103 100644
--- a/drivers/infiniband/hw/cxgb4/ev.c
+++ b/drivers/infiniband/hw/cxgb4/ev.c
@@ -42,6 +42,7 @@ static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
42{ 42{
43 struct ib_event event; 43 struct ib_event event;
44 struct c4iw_qp_attributes attrs; 44 struct c4iw_qp_attributes attrs;
45 unsigned long flag;
45 46
46 if ((qhp->attr.state == C4IW_QP_STATE_ERROR) || 47 if ((qhp->attr.state == C4IW_QP_STATE_ERROR) ||
47 (qhp->attr.state == C4IW_QP_STATE_TERMINATE)) { 48 (qhp->attr.state == C4IW_QP_STATE_TERMINATE)) {
@@ -72,7 +73,9 @@ static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
72 if (qhp->ibqp.event_handler) 73 if (qhp->ibqp.event_handler)
73 (*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context); 74 (*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
74 75
76 spin_lock_irqsave(&chp->comp_handler_lock, flag);
75 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context); 77 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
78 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
76} 79}
77 80
78void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe) 81void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
@@ -183,11 +186,14 @@ out:
183int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid) 186int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid)
184{ 187{
185 struct c4iw_cq *chp; 188 struct c4iw_cq *chp;
189 unsigned long flag;
186 190
187 chp = get_chp(dev, qid); 191 chp = get_chp(dev, qid);
188 if (chp) 192 if (chp) {
193 spin_lock_irqsave(&chp->comp_handler_lock, flag);
189 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context); 194 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
190 else 195 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
196 } else
191 PDBG("%s unknown cqid 0x%x\n", __func__, qid); 197 PDBG("%s unknown cqid 0x%x\n", __func__, qid);
192 return 0; 198 return 0;
193} 199}
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index 4f045375c8e2..1357c5bf209b 100644
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -309,6 +309,7 @@ struct c4iw_cq {
309 struct c4iw_dev *rhp; 309 struct c4iw_dev *rhp;
310 struct t4_cq cq; 310 struct t4_cq cq;
311 spinlock_t lock; 311 spinlock_t lock;
312 spinlock_t comp_handler_lock;
312 atomic_t refcnt; 313 atomic_t refcnt;
313 wait_queue_head_t wait; 314 wait_queue_head_t wait;
314}; 315};
@@ -323,6 +324,7 @@ struct c4iw_mpa_attributes {
323 u8 recv_marker_enabled; 324 u8 recv_marker_enabled;
324 u8 xmit_marker_enabled; 325 u8 xmit_marker_enabled;
325 u8 crc_enabled; 326 u8 crc_enabled;
327 u8 enhanced_rdma_conn;
326 u8 version; 328 u8 version;
327 u8 p2p_type; 329 u8 p2p_type;
328}; 330};
@@ -349,6 +351,8 @@ struct c4iw_qp_attributes {
349 u8 is_terminate_local; 351 u8 is_terminate_local;
350 struct c4iw_mpa_attributes mpa_attr; 352 struct c4iw_mpa_attributes mpa_attr;
351 struct c4iw_ep *llp_stream_handle; 353 struct c4iw_ep *llp_stream_handle;
354 u8 layer_etype;
355 u8 ecode;
352}; 356};
353 357
354struct c4iw_qp { 358struct c4iw_qp {
@@ -501,11 +505,18 @@ enum c4iw_mmid_state {
501#define MPA_KEY_REP "MPA ID Rep Frame" 505#define MPA_KEY_REP "MPA ID Rep Frame"
502 506
503#define MPA_MAX_PRIVATE_DATA 256 507#define MPA_MAX_PRIVATE_DATA 256
508#define MPA_ENHANCED_RDMA_CONN 0x10
504#define MPA_REJECT 0x20 509#define MPA_REJECT 0x20
505#define MPA_CRC 0x40 510#define MPA_CRC 0x40
506#define MPA_MARKERS 0x80 511#define MPA_MARKERS 0x80
507#define MPA_FLAGS_MASK 0xE0 512#define MPA_FLAGS_MASK 0xE0
508 513
514#define MPA_V2_PEER2PEER_MODEL 0x8000
515#define MPA_V2_ZERO_LEN_FPDU_RTR 0x4000
516#define MPA_V2_RDMA_WRITE_RTR 0x8000
517#define MPA_V2_RDMA_READ_RTR 0x4000
518#define MPA_V2_IRD_ORD_MASK 0x3FFF
519
509#define c4iw_put_ep(ep) { \ 520#define c4iw_put_ep(ep) { \
510 PDBG("put_ep (via %s:%u) ep %p refcnt %d\n", __func__, __LINE__, \ 521 PDBG("put_ep (via %s:%u) ep %p refcnt %d\n", __func__, __LINE__, \
511 ep, atomic_read(&((ep)->kref.refcount))); \ 522 ep, atomic_read(&((ep)->kref.refcount))); \
@@ -528,6 +539,11 @@ struct mpa_message {
528 u8 private_data[0]; 539 u8 private_data[0];
529}; 540};
530 541
542struct mpa_v2_conn_params {
543 __be16 ird;
544 __be16 ord;
545};
546
531struct terminate_message { 547struct terminate_message {
532 u8 layer_etype; 548 u8 layer_etype;
533 u8 ecode; 549 u8 ecode;
@@ -580,7 +596,10 @@ enum c4iw_ddp_ecodes {
580 596
581enum c4iw_mpa_ecodes { 597enum c4iw_mpa_ecodes {
582 MPA_CRC_ERR = 0x02, 598 MPA_CRC_ERR = 0x02,
583 MPA_MARKER_ERR = 0x03 599 MPA_MARKER_ERR = 0x03,
600 MPA_LOCAL_CATA = 0x05,
601 MPA_INSUFF_IRD = 0x06,
602 MPA_NOMATCH_RTR = 0x07,
584}; 603};
585 604
586enum c4iw_ep_state { 605enum c4iw_ep_state {
@@ -651,6 +670,8 @@ struct c4iw_ep {
651 u16 txq_idx; 670 u16 txq_idx;
652 u16 ctrlq_idx; 671 u16 ctrlq_idx;
653 u8 tos; 672 u8 tos;
673 u8 retry_with_mpa_v1;
674 u8 tried_with_mpa_v1;
654}; 675};
655 676
656static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id) 677static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id)
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index a41578e48c7b..d6ccc7e84802 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -917,7 +917,11 @@ static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
917 wqe->u.terminate.type = FW_RI_TYPE_TERMINATE; 917 wqe->u.terminate.type = FW_RI_TYPE_TERMINATE;
918 wqe->u.terminate.immdlen = cpu_to_be32(sizeof *term); 918 wqe->u.terminate.immdlen = cpu_to_be32(sizeof *term);
919 term = (struct terminate_message *)wqe->u.terminate.termmsg; 919 term = (struct terminate_message *)wqe->u.terminate.termmsg;
920 build_term_codes(err_cqe, &term->layer_etype, &term->ecode); 920 if (qhp->attr.layer_etype == (LAYER_MPA|DDP_LLP)) {
921 term->layer_etype = qhp->attr.layer_etype;
922 term->ecode = qhp->attr.ecode;
923 } else
924 build_term_codes(err_cqe, &term->layer_etype, &term->ecode);
921 c4iw_ofld_send(&qhp->rhp->rdev, skb); 925 c4iw_ofld_send(&qhp->rhp->rdev, skb);
922} 926}
923 927
@@ -941,8 +945,11 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
941 flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count); 945 flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
942 spin_unlock(&qhp->lock); 946 spin_unlock(&qhp->lock);
943 spin_unlock_irqrestore(&rchp->lock, flag); 947 spin_unlock_irqrestore(&rchp->lock, flag);
944 if (flushed) 948 if (flushed) {
949 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
945 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context); 950 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
951 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
952 }
946 953
947 /* locking hierarchy: cq lock first, then qp lock. */ 954 /* locking hierarchy: cq lock first, then qp lock. */
948 spin_lock_irqsave(&schp->lock, flag); 955 spin_lock_irqsave(&schp->lock, flag);
@@ -952,13 +959,17 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
952 flushed = c4iw_flush_sq(&qhp->wq, &schp->cq, count); 959 flushed = c4iw_flush_sq(&qhp->wq, &schp->cq, count);
953 spin_unlock(&qhp->lock); 960 spin_unlock(&qhp->lock);
954 spin_unlock_irqrestore(&schp->lock, flag); 961 spin_unlock_irqrestore(&schp->lock, flag);
955 if (flushed) 962 if (flushed) {
963 spin_lock_irqsave(&schp->comp_handler_lock, flag);
956 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); 964 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
965 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
966 }
957} 967}
958 968
959static void flush_qp(struct c4iw_qp *qhp) 969static void flush_qp(struct c4iw_qp *qhp)
960{ 970{
961 struct c4iw_cq *rchp, *schp; 971 struct c4iw_cq *rchp, *schp;
972 unsigned long flag;
962 973
963 rchp = get_chp(qhp->rhp, qhp->attr.rcq); 974 rchp = get_chp(qhp->rhp, qhp->attr.rcq);
964 schp = get_chp(qhp->rhp, qhp->attr.scq); 975 schp = get_chp(qhp->rhp, qhp->attr.scq);
@@ -966,8 +977,16 @@ static void flush_qp(struct c4iw_qp *qhp)
966 if (qhp->ibqp.uobject) { 977 if (qhp->ibqp.uobject) {
967 t4_set_wq_in_error(&qhp->wq); 978 t4_set_wq_in_error(&qhp->wq);
968 t4_set_cq_in_error(&rchp->cq); 979 t4_set_cq_in_error(&rchp->cq);
969 if (schp != rchp) 980 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
981 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
982 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
983 if (schp != rchp) {
970 t4_set_cq_in_error(&schp->cq); 984 t4_set_cq_in_error(&schp->cq);
985 spin_lock_irqsave(&schp->comp_handler_lock, flag);
986 (*schp->ibcq.comp_handler)(&schp->ibcq,
987 schp->ibcq.cq_context);
988 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
989 }
971 return; 990 return;
972 } 991 }
973 __flush_qp(qhp, rchp, schp); 992 __flush_qp(qhp, rchp, schp);
@@ -1012,6 +1031,7 @@ out:
1012 1031
1013static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init) 1032static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init)
1014{ 1033{
1034 PDBG("%s p2p_type = %d\n", __func__, p2p_type);
1015 memset(&init->u, 0, sizeof init->u); 1035 memset(&init->u, 0, sizeof init->u);
1016 switch (p2p_type) { 1036 switch (p2p_type) {
1017 case FW_RI_INIT_P2PTYPE_RDMA_WRITE: 1037 case FW_RI_INIT_P2PTYPE_RDMA_WRITE:
@@ -1206,12 +1226,16 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
1206 disconnect = 1; 1226 disconnect = 1;
1207 c4iw_get_ep(&qhp->ep->com); 1227 c4iw_get_ep(&qhp->ep->com);
1208 } 1228 }
1229 if (qhp->ibqp.uobject)
1230 t4_set_wq_in_error(&qhp->wq);
1209 ret = rdma_fini(rhp, qhp, ep); 1231 ret = rdma_fini(rhp, qhp, ep);
1210 if (ret) 1232 if (ret)
1211 goto err; 1233 goto err;
1212 break; 1234 break;
1213 case C4IW_QP_STATE_TERMINATE: 1235 case C4IW_QP_STATE_TERMINATE:
1214 set_state(qhp, C4IW_QP_STATE_TERMINATE); 1236 set_state(qhp, C4IW_QP_STATE_TERMINATE);
1237 qhp->attr.layer_etype = attrs->layer_etype;
1238 qhp->attr.ecode = attrs->ecode;
1215 if (qhp->ibqp.uobject) 1239 if (qhp->ibqp.uobject)
1216 t4_set_wq_in_error(&qhp->wq); 1240 t4_set_wq_in_error(&qhp->wq);
1217 ep = qhp->ep; 1241 ep = qhp->ep;
@@ -1222,6 +1246,8 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
1222 break; 1246 break;
1223 case C4IW_QP_STATE_ERROR: 1247 case C4IW_QP_STATE_ERROR:
1224 set_state(qhp, C4IW_QP_STATE_ERROR); 1248 set_state(qhp, C4IW_QP_STATE_ERROR);
1249 if (qhp->ibqp.uobject)
1250 t4_set_wq_in_error(&qhp->wq);
1225 if (!internal) { 1251 if (!internal) {
1226 abort = 1; 1252 abort = 1;
1227 disconnect = 1; 1253 disconnect = 1;
@@ -1334,7 +1360,10 @@ int c4iw_destroy_qp(struct ib_qp *ib_qp)
1334 rhp = qhp->rhp; 1360 rhp = qhp->rhp;
1335 1361
1336 attrs.next_state = C4IW_QP_STATE_ERROR; 1362 attrs.next_state = C4IW_QP_STATE_ERROR;
1337 c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0); 1363 if (qhp->attr.state == C4IW_QP_STATE_TERMINATE)
1364 c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1365 else
1366 c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1338 wait_event(qhp->wait, !qhp->ep); 1367 wait_event(qhp->wait, !qhp->ep);
1339 1368
1340 remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid); 1369 remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid);
diff --git a/drivers/infiniband/hw/ehca/ehca_eq.c b/drivers/infiniband/hw/ehca/ehca_eq.c
index d9b1bb40f480..818d721fc448 100644
--- a/drivers/infiniband/hw/ehca/ehca_eq.c
+++ b/drivers/infiniband/hw/ehca/ehca_eq.c
@@ -125,7 +125,7 @@ int ehca_create_eq(struct ehca_shca *shca,
125 tasklet_init(&eq->interrupt_task, ehca_tasklet_eq, (long)shca); 125 tasklet_init(&eq->interrupt_task, ehca_tasklet_eq, (long)shca);
126 126
127 ret = ibmebus_request_irq(eq->ist, ehca_interrupt_eq, 127 ret = ibmebus_request_irq(eq->ist, ehca_interrupt_eq,
128 IRQF_DISABLED, "ehca_eq", 128 0, "ehca_eq",
129 (void *)shca); 129 (void *)shca);
130 if (ret < 0) 130 if (ret < 0)
131 ehca_err(ib_dev, "Can't map interrupt handler."); 131 ehca_err(ib_dev, "Can't map interrupt handler.");
@@ -133,7 +133,7 @@ int ehca_create_eq(struct ehca_shca *shca,
133 tasklet_init(&eq->interrupt_task, ehca_tasklet_neq, (long)shca); 133 tasklet_init(&eq->interrupt_task, ehca_tasklet_neq, (long)shca);
134 134
135 ret = ibmebus_request_irq(eq->ist, ehca_interrupt_neq, 135 ret = ibmebus_request_irq(eq->ist, ehca_interrupt_neq,
136 IRQF_DISABLED, "ehca_neq", 136 0, "ehca_neq",
137 (void *)shca); 137 (void *)shca);
138 if (ret < 0) 138 if (ret < 0)
139 ehca_err(ib_dev, "Can't map interrupt handler."); 139 ehca_err(ib_dev, "Can't map interrupt handler.");
diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c
index 32fb34201aba..964f85520798 100644
--- a/drivers/infiniband/hw/ehca/ehca_qp.c
+++ b/drivers/infiniband/hw/ehca/ehca_qp.c
@@ -977,6 +977,9 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
977 struct hcp_modify_qp_control_block *mqpcb; 977 struct hcp_modify_qp_control_block *mqpcb;
978 u64 hret, update_mask; 978 u64 hret, update_mask;
979 979
980 if (srq_init_attr->srq_type != IB_SRQT_BASIC)
981 return ERR_PTR(-ENOSYS);
982
980 /* For common attributes, internal_create_qp() takes its info 983 /* For common attributes, internal_create_qp() takes its info
981 * out of qp_init_attr, so copy all common attrs there. 984 * out of qp_init_attr, so copy all common attrs there.
982 */ 985 */
diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c
index 7c1eebe8c7c9..824a4d508836 100644
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c
@@ -34,6 +34,7 @@
34#include <linux/pci.h> 34#include <linux/pci.h>
35#include <linux/netdevice.h> 35#include <linux/netdevice.h>
36#include <linux/slab.h> 36#include <linux/slab.h>
37#include <linux/stat.h>
37#include <linux/vmalloc.h> 38#include <linux/vmalloc.h>
38 39
39#include "ipath_kernel.h" 40#include "ipath_kernel.h"
diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c
index 386e2c717c53..26271984b717 100644
--- a/drivers/infiniband/hw/ipath/ipath_srq.c
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c
@@ -107,6 +107,11 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
107 u32 sz; 107 u32 sz;
108 struct ib_srq *ret; 108 struct ib_srq *ret;
109 109
110 if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
111 ret = ERR_PTR(-ENOSYS);
112 goto done;
113 }
114
110 if (srq_init_attr->attr.max_wr == 0) { 115 if (srq_init_attr->attr.max_wr == 0) {
111 ret = ERR_PTR(-EINVAL); 116 ret = ERR_PTR(-EINVAL);
112 goto done; 117 goto done;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index fa643f4f4e28..77f3dbc0aaa1 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -128,6 +128,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
128 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) && 128 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR)) 129 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; 130 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
131 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
132 props->device_cap_flags |= IB_DEVICE_XRC;
131 133
132 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) & 134 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
133 0xffffff; 135 0xffffff;
@@ -181,8 +183,12 @@ mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
181 183
182static int ib_link_query_port(struct ib_device *ibdev, u8 port, 184static int ib_link_query_port(struct ib_device *ibdev, u8 port,
183 struct ib_port_attr *props, 185 struct ib_port_attr *props,
186 struct ib_smp *in_mad,
184 struct ib_smp *out_mad) 187 struct ib_smp *out_mad)
185{ 188{
189 int ext_active_speed;
190 int err;
191
186 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16)); 192 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
187 props->lmc = out_mad->data[34] & 0x7; 193 props->lmc = out_mad->data[34] & 0x7;
188 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18)); 194 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
@@ -203,6 +209,39 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
203 props->max_vl_num = out_mad->data[37] >> 4; 209 props->max_vl_num = out_mad->data[37] >> 4;
204 props->init_type_reply = out_mad->data[41] >> 4; 210 props->init_type_reply = out_mad->data[41] >> 4;
205 211
212 /* Check if extended speeds (EDR/FDR/...) are supported */
213 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
214 ext_active_speed = out_mad->data[62] >> 4;
215
216 switch (ext_active_speed) {
217 case 1:
218 props->active_speed = 16; /* FDR */
219 break;
220 case 2:
221 props->active_speed = 32; /* EDR */
222 break;
223 }
224 }
225
226 /* If reported active speed is QDR, check if is FDR-10 */
227 if (props->active_speed == 4) {
228 if (to_mdev(ibdev)->dev->caps.ext_port_cap[port] &
229 MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
230 init_query_mad(in_mad);
231 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
232 in_mad->attr_mod = cpu_to_be32(port);
233
234 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
235 NULL, NULL, in_mad, out_mad);
236 if (err)
237 return err;
238
239 /* Checking LinkSpeedActive for FDR-10 */
240 if (out_mad->data[15] & 0x1)
241 props->active_speed = 8;
242 }
243 }
244
206 return 0; 245 return 0;
207} 246}
208 247
@@ -227,7 +266,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
227 props->pkey_tbl_len = 1; 266 props->pkey_tbl_len = 1;
228 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46)); 267 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
229 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48)); 268 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
230 props->max_mtu = IB_MTU_2048; 269 props->max_mtu = IB_MTU_4096;
231 props->subnet_timeout = 0; 270 props->subnet_timeout = 0;
232 props->max_vl_num = out_mad->data[37] >> 4; 271 props->max_vl_num = out_mad->data[37] >> 4;
233 props->init_type_reply = 0; 272 props->init_type_reply = 0;
@@ -274,7 +313,7 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
274 goto out; 313 goto out;
275 314
276 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ? 315 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
277 ib_link_query_port(ibdev, port, props, out_mad) : 316 ib_link_query_port(ibdev, port, props, in_mad, out_mad) :
278 eth_link_query_port(ibdev, port, props, out_mad); 317 eth_link_query_port(ibdev, port, props, out_mad);
279 318
280out: 319out:
@@ -566,6 +605,57 @@ static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
566 return 0; 605 return 0;
567} 606}
568 607
608static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
609 struct ib_ucontext *context,
610 struct ib_udata *udata)
611{
612 struct mlx4_ib_xrcd *xrcd;
613 int err;
614
615 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
616 return ERR_PTR(-ENOSYS);
617
618 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
619 if (!xrcd)
620 return ERR_PTR(-ENOMEM);
621
622 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
623 if (err)
624 goto err1;
625
626 xrcd->pd = ib_alloc_pd(ibdev);
627 if (IS_ERR(xrcd->pd)) {
628 err = PTR_ERR(xrcd->pd);
629 goto err2;
630 }
631
632 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
633 if (IS_ERR(xrcd->cq)) {
634 err = PTR_ERR(xrcd->cq);
635 goto err3;
636 }
637
638 return &xrcd->ibxrcd;
639
640err3:
641 ib_dealloc_pd(xrcd->pd);
642err2:
643 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
644err1:
645 kfree(xrcd);
646 return ERR_PTR(err);
647}
648
649static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
650{
651 ib_destroy_cq(to_mxrcd(xrcd)->cq);
652 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
653 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
654 kfree(xrcd);
655
656 return 0;
657}
658
569static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid) 659static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
570{ 660{
571 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 661 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
@@ -1044,7 +1134,9 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
1044 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | 1134 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1045 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | 1135 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1046 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | 1136 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1047 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ); 1137 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1138 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1139 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1048 1140
1049 ibdev->ib_dev.query_device = mlx4_ib_query_device; 1141 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1050 ibdev->ib_dev.query_port = mlx4_ib_query_port; 1142 ibdev->ib_dev.query_port = mlx4_ib_query_port;
@@ -1093,6 +1185,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
1093 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr; 1185 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1094 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; 1186 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1095 1187
1188 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1189 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1190 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1191 ibdev->ib_dev.uverbs_cmd_mask |=
1192 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1193 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1194 }
1195
1096 spin_lock_init(&iboe->lock); 1196 spin_lock_init(&iboe->lock);
1097 1197
1098 if (init_node_data(ibdev)) 1198 if (init_node_data(ibdev))
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index e4bf2cff8662..ed80345c99ae 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -56,6 +56,13 @@ struct mlx4_ib_pd {
56 u32 pdn; 56 u32 pdn;
57}; 57};
58 58
59struct mlx4_ib_xrcd {
60 struct ib_xrcd ibxrcd;
61 u32 xrcdn;
62 struct ib_pd *pd;
63 struct ib_cq *cq;
64};
65
59struct mlx4_ib_cq_buf { 66struct mlx4_ib_cq_buf {
60 struct mlx4_buf buf; 67 struct mlx4_buf buf;
61 struct mlx4_mtt mtt; 68 struct mlx4_mtt mtt;
@@ -138,6 +145,7 @@ struct mlx4_ib_qp {
138 struct mlx4_mtt mtt; 145 struct mlx4_mtt mtt;
139 int buf_size; 146 int buf_size;
140 struct mutex mutex; 147 struct mutex mutex;
148 u16 xrcdn;
141 u32 flags; 149 u32 flags;
142 u8 port; 150 u8 port;
143 u8 alt_port; 151 u8 alt_port;
@@ -211,6 +219,11 @@ static inline struct mlx4_ib_pd *to_mpd(struct ib_pd *ibpd)
211 return container_of(ibpd, struct mlx4_ib_pd, ibpd); 219 return container_of(ibpd, struct mlx4_ib_pd, ibpd);
212} 220}
213 221
222static inline struct mlx4_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
223{
224 return container_of(ibxrcd, struct mlx4_ib_xrcd, ibxrcd);
225}
226
214static inline struct mlx4_ib_cq *to_mcq(struct ib_cq *ibcq) 227static inline struct mlx4_ib_cq *to_mcq(struct ib_cq *ibcq)
215{ 228{
216 return container_of(ibcq, struct mlx4_ib_cq, ibcq); 229 return container_of(ibcq, struct mlx4_ib_cq, ibcq);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 3a91d9d8dc51..a16f0c8e6f3f 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -302,15 +302,14 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
302} 302}
303 303
304static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap, 304static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
305 int is_user, int has_srq, struct mlx4_ib_qp *qp) 305 int is_user, int has_rq, struct mlx4_ib_qp *qp)
306{ 306{
307 /* Sanity check RQ size before proceeding */ 307 /* Sanity check RQ size before proceeding */
308 if (cap->max_recv_wr > dev->dev->caps.max_wqes || 308 if (cap->max_recv_wr > dev->dev->caps.max_wqes ||
309 cap->max_recv_sge > dev->dev->caps.max_rq_sg) 309 cap->max_recv_sge > dev->dev->caps.max_rq_sg)
310 return -EINVAL; 310 return -EINVAL;
311 311
312 if (has_srq) { 312 if (!has_rq) {
313 /* QPs attached to an SRQ should have no RQ */
314 if (cap->max_recv_wr) 313 if (cap->max_recv_wr)
315 return -EINVAL; 314 return -EINVAL;
316 315
@@ -463,6 +462,14 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev,
463 return 0; 462 return 0;
464} 463}
465 464
465static int qp_has_rq(struct ib_qp_init_attr *attr)
466{
467 if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
468 return 0;
469
470 return !attr->srq;
471}
472
466static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, 473static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
467 struct ib_qp_init_attr *init_attr, 474 struct ib_qp_init_attr *init_attr,
468 struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp) 475 struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp)
@@ -479,7 +486,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
479 if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) 486 if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
480 qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE); 487 qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
481 488
482 err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, !!init_attr->srq, qp); 489 err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, qp_has_rq(init_attr), qp);
483 if (err) 490 if (err)
484 goto err; 491 goto err;
485 492
@@ -513,7 +520,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
513 if (err) 520 if (err)
514 goto err_mtt; 521 goto err_mtt;
515 522
516 if (!init_attr->srq) { 523 if (qp_has_rq(init_attr)) {
517 err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), 524 err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
518 ucmd.db_addr, &qp->db); 525 ucmd.db_addr, &qp->db);
519 if (err) 526 if (err)
@@ -532,7 +539,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
532 if (err) 539 if (err)
533 goto err; 540 goto err;
534 541
535 if (!init_attr->srq) { 542 if (qp_has_rq(init_attr)) {
536 err = mlx4_db_alloc(dev->dev, &qp->db, 0); 543 err = mlx4_db_alloc(dev->dev, &qp->db, 0);
537 if (err) 544 if (err)
538 goto err; 545 goto err;
@@ -575,6 +582,9 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
575 if (err) 582 if (err)
576 goto err_qpn; 583 goto err_qpn;
577 584
585 if (init_attr->qp_type == IB_QPT_XRC_TGT)
586 qp->mqp.qpn |= (1 << 23);
587
578 /* 588 /*
579 * Hardware wants QPN written in big-endian order (after 589 * Hardware wants QPN written in big-endian order (after
580 * shifting) for send doorbell. Precompute this value to save 590 * shifting) for send doorbell. Precompute this value to save
@@ -592,9 +602,8 @@ err_qpn:
592 602
593err_wrid: 603err_wrid:
594 if (pd->uobject) { 604 if (pd->uobject) {
595 if (!init_attr->srq) 605 if (qp_has_rq(init_attr))
596 mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), 606 mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db);
597 &qp->db);
598 } else { 607 } else {
599 kfree(qp->sq.wrid); 608 kfree(qp->sq.wrid);
600 kfree(qp->rq.wrid); 609 kfree(qp->rq.wrid);
@@ -610,7 +619,7 @@ err_buf:
610 mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf); 619 mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
611 620
612err_db: 621err_db:
613 if (!pd->uobject && !init_attr->srq) 622 if (!pd->uobject && qp_has_rq(init_attr))
614 mlx4_db_free(dev->dev, &qp->db); 623 mlx4_db_free(dev->dev, &qp->db);
615 624
616err: 625err:
@@ -671,6 +680,33 @@ static void del_gid_entries(struct mlx4_ib_qp *qp)
671 } 680 }
672} 681}
673 682
683static struct mlx4_ib_pd *get_pd(struct mlx4_ib_qp *qp)
684{
685 if (qp->ibqp.qp_type == IB_QPT_XRC_TGT)
686 return to_mpd(to_mxrcd(qp->ibqp.xrcd)->pd);
687 else
688 return to_mpd(qp->ibqp.pd);
689}
690
691static void get_cqs(struct mlx4_ib_qp *qp,
692 struct mlx4_ib_cq **send_cq, struct mlx4_ib_cq **recv_cq)
693{
694 switch (qp->ibqp.qp_type) {
695 case IB_QPT_XRC_TGT:
696 *send_cq = to_mcq(to_mxrcd(qp->ibqp.xrcd)->cq);
697 *recv_cq = *send_cq;
698 break;
699 case IB_QPT_XRC_INI:
700 *send_cq = to_mcq(qp->ibqp.send_cq);
701 *recv_cq = *send_cq;
702 break;
703 default:
704 *send_cq = to_mcq(qp->ibqp.send_cq);
705 *recv_cq = to_mcq(qp->ibqp.recv_cq);
706 break;
707 }
708}
709
674static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, 710static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
675 int is_user) 711 int is_user)
676{ 712{
@@ -682,8 +718,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
682 printk(KERN_WARNING "mlx4_ib: modify QP %06x to RESET failed.\n", 718 printk(KERN_WARNING "mlx4_ib: modify QP %06x to RESET failed.\n",
683 qp->mqp.qpn); 719 qp->mqp.qpn);
684 720
685 send_cq = to_mcq(qp->ibqp.send_cq); 721 get_cqs(qp, &send_cq, &recv_cq);
686 recv_cq = to_mcq(qp->ibqp.recv_cq);
687 722
688 mlx4_ib_lock_cqs(send_cq, recv_cq); 723 mlx4_ib_lock_cqs(send_cq, recv_cq);
689 724
@@ -706,7 +741,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
706 mlx4_mtt_cleanup(dev->dev, &qp->mtt); 741 mlx4_mtt_cleanup(dev->dev, &qp->mtt);
707 742
708 if (is_user) { 743 if (is_user) {
709 if (!qp->ibqp.srq) 744 if (qp->rq.wqe_cnt)
710 mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context), 745 mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context),
711 &qp->db); 746 &qp->db);
712 ib_umem_release(qp->umem); 747 ib_umem_release(qp->umem);
@@ -714,7 +749,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
714 kfree(qp->sq.wrid); 749 kfree(qp->sq.wrid);
715 kfree(qp->rq.wrid); 750 kfree(qp->rq.wrid);
716 mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf); 751 mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
717 if (!qp->ibqp.srq) 752 if (qp->rq.wqe_cnt)
718 mlx4_db_free(dev->dev, &qp->db); 753 mlx4_db_free(dev->dev, &qp->db);
719 } 754 }
720 755
@@ -725,10 +760,10 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
725 struct ib_qp_init_attr *init_attr, 760 struct ib_qp_init_attr *init_attr,
726 struct ib_udata *udata) 761 struct ib_udata *udata)
727{ 762{
728 struct mlx4_ib_dev *dev = to_mdev(pd->device);
729 struct mlx4_ib_sqp *sqp; 763 struct mlx4_ib_sqp *sqp;
730 struct mlx4_ib_qp *qp; 764 struct mlx4_ib_qp *qp;
731 int err; 765 int err;
766 u16 xrcdn = 0;
732 767
733 /* 768 /*
734 * We only support LSO and multicast loopback blocking, and 769 * We only support LSO and multicast loopback blocking, and
@@ -739,10 +774,20 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
739 return ERR_PTR(-EINVAL); 774 return ERR_PTR(-EINVAL);
740 775
741 if (init_attr->create_flags && 776 if (init_attr->create_flags &&
742 (pd->uobject || init_attr->qp_type != IB_QPT_UD)) 777 (udata || init_attr->qp_type != IB_QPT_UD))
743 return ERR_PTR(-EINVAL); 778 return ERR_PTR(-EINVAL);
744 779
745 switch (init_attr->qp_type) { 780 switch (init_attr->qp_type) {
781 case IB_QPT_XRC_TGT:
782 pd = to_mxrcd(init_attr->xrcd)->pd;
783 xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
784 init_attr->send_cq = to_mxrcd(init_attr->xrcd)->cq;
785 /* fall through */
786 case IB_QPT_XRC_INI:
787 if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
788 return ERR_PTR(-ENOSYS);
789 init_attr->recv_cq = init_attr->send_cq;
790 /* fall through */
746 case IB_QPT_RC: 791 case IB_QPT_RC:
747 case IB_QPT_UC: 792 case IB_QPT_UC:
748 case IB_QPT_UD: 793 case IB_QPT_UD:
@@ -751,13 +796,14 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
751 if (!qp) 796 if (!qp)
752 return ERR_PTR(-ENOMEM); 797 return ERR_PTR(-ENOMEM);
753 798
754 err = create_qp_common(dev, pd, init_attr, udata, 0, qp); 799 err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, 0, qp);
755 if (err) { 800 if (err) {
756 kfree(qp); 801 kfree(qp);
757 return ERR_PTR(err); 802 return ERR_PTR(err);
758 } 803 }
759 804
760 qp->ibqp.qp_num = qp->mqp.qpn; 805 qp->ibqp.qp_num = qp->mqp.qpn;
806 qp->xrcdn = xrcdn;
761 807
762 break; 808 break;
763 } 809 }
@@ -765,7 +811,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
765 case IB_QPT_GSI: 811 case IB_QPT_GSI:
766 { 812 {
767 /* Userspace is not allowed to create special QPs: */ 813 /* Userspace is not allowed to create special QPs: */
768 if (pd->uobject) 814 if (udata)
769 return ERR_PTR(-EINVAL); 815 return ERR_PTR(-EINVAL);
770 816
771 sqp = kzalloc(sizeof *sqp, GFP_KERNEL); 817 sqp = kzalloc(sizeof *sqp, GFP_KERNEL);
@@ -774,8 +820,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
774 820
775 qp = &sqp->qp; 821 qp = &sqp->qp;
776 822
777 err = create_qp_common(dev, pd, init_attr, udata, 823 err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
778 dev->dev->caps.sqp_start + 824 to_mdev(pd->device)->dev->caps.sqp_start +
779 (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) + 825 (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
780 init_attr->port_num - 1, 826 init_attr->port_num - 1,
781 qp); 827 qp);
@@ -801,11 +847,13 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
801{ 847{
802 struct mlx4_ib_dev *dev = to_mdev(qp->device); 848 struct mlx4_ib_dev *dev = to_mdev(qp->device);
803 struct mlx4_ib_qp *mqp = to_mqp(qp); 849 struct mlx4_ib_qp *mqp = to_mqp(qp);
850 struct mlx4_ib_pd *pd;
804 851
805 if (is_qp0(dev, mqp)) 852 if (is_qp0(dev, mqp))
806 mlx4_CLOSE_PORT(dev->dev, mqp->port); 853 mlx4_CLOSE_PORT(dev->dev, mqp->port);
807 854
808 destroy_qp_common(dev, mqp, !!qp->pd->uobject); 855 pd = get_pd(mqp);
856 destroy_qp_common(dev, mqp, !!pd->ibpd.uobject);
809 857
810 if (is_sqp(dev, mqp)) 858 if (is_sqp(dev, mqp))
811 kfree(to_msqp(mqp)); 859 kfree(to_msqp(mqp));
@@ -821,6 +869,8 @@ static int to_mlx4_st(enum ib_qp_type type)
821 case IB_QPT_RC: return MLX4_QP_ST_RC; 869 case IB_QPT_RC: return MLX4_QP_ST_RC;
822 case IB_QPT_UC: return MLX4_QP_ST_UC; 870 case IB_QPT_UC: return MLX4_QP_ST_UC;
823 case IB_QPT_UD: return MLX4_QP_ST_UD; 871 case IB_QPT_UD: return MLX4_QP_ST_UD;
872 case IB_QPT_XRC_INI:
873 case IB_QPT_XRC_TGT: return MLX4_QP_ST_XRC;
824 case IB_QPT_SMI: 874 case IB_QPT_SMI:
825 case IB_QPT_GSI: return MLX4_QP_ST_MLX; 875 case IB_QPT_GSI: return MLX4_QP_ST_MLX;
826 default: return -1; 876 default: return -1;
@@ -959,6 +1009,8 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
959{ 1009{
960 struct mlx4_ib_dev *dev = to_mdev(ibqp->device); 1010 struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
961 struct mlx4_ib_qp *qp = to_mqp(ibqp); 1011 struct mlx4_ib_qp *qp = to_mqp(ibqp);
1012 struct mlx4_ib_pd *pd;
1013 struct mlx4_ib_cq *send_cq, *recv_cq;
962 struct mlx4_qp_context *context; 1014 struct mlx4_qp_context *context;
963 enum mlx4_qp_optpar optpar = 0; 1015 enum mlx4_qp_optpar optpar = 0;
964 int sqd_event; 1016 int sqd_event;
@@ -1014,8 +1066,10 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1014 context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3; 1066 context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3;
1015 context->sq_size_stride |= qp->sq.wqe_shift - 4; 1067 context->sq_size_stride |= qp->sq.wqe_shift - 4;
1016 1068
1017 if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) 1069 if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
1018 context->sq_size_stride |= !!qp->sq_no_prefetch << 7; 1070 context->sq_size_stride |= !!qp->sq_no_prefetch << 7;
1071 context->xrcd = cpu_to_be32((u32) qp->xrcdn);
1072 }
1019 1073
1020 if (qp->ibqp.uobject) 1074 if (qp->ibqp.uobject)
1021 context->usr_page = cpu_to_be32(to_mucontext(ibqp->uobject->context)->uar.index); 1075 context->usr_page = cpu_to_be32(to_mucontext(ibqp->uobject->context)->uar.index);
@@ -1079,8 +1133,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1079 optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH; 1133 optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH;
1080 } 1134 }
1081 1135
1082 context->pd = cpu_to_be32(to_mpd(ibqp->pd)->pdn); 1136 pd = get_pd(qp);
1083 context->params1 = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28); 1137 get_cqs(qp, &send_cq, &recv_cq);
1138 context->pd = cpu_to_be32(pd->pdn);
1139 context->cqn_send = cpu_to_be32(send_cq->mcq.cqn);
1140 context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn);
1141 context->params1 = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
1084 1142
1085 /* Set "fast registration enabled" for all kernel QPs */ 1143 /* Set "fast registration enabled" for all kernel QPs */
1086 if (!qp->ibqp.uobject) 1144 if (!qp->ibqp.uobject)
@@ -1106,8 +1164,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1106 if (attr_mask & IB_QP_SQ_PSN) 1164 if (attr_mask & IB_QP_SQ_PSN)
1107 context->next_send_psn = cpu_to_be32(attr->sq_psn); 1165 context->next_send_psn = cpu_to_be32(attr->sq_psn);
1108 1166
1109 context->cqn_send = cpu_to_be32(to_mcq(ibqp->send_cq)->mcq.cqn);
1110
1111 if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) { 1167 if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1112 if (attr->max_dest_rd_atomic) 1168 if (attr->max_dest_rd_atomic)
1113 context->params2 |= 1169 context->params2 |=
@@ -1130,8 +1186,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1130 if (attr_mask & IB_QP_RQ_PSN) 1186 if (attr_mask & IB_QP_RQ_PSN)
1131 context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn); 1187 context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
1132 1188
1133 context->cqn_recv = cpu_to_be32(to_mcq(ibqp->recv_cq)->mcq.cqn);
1134
1135 if (attr_mask & IB_QP_QKEY) { 1189 if (attr_mask & IB_QP_QKEY) {
1136 context->qkey = cpu_to_be32(attr->qkey); 1190 context->qkey = cpu_to_be32(attr->qkey);
1137 optpar |= MLX4_QP_OPTPAR_Q_KEY; 1191 optpar |= MLX4_QP_OPTPAR_Q_KEY;
@@ -1140,7 +1194,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1140 if (ibqp->srq) 1194 if (ibqp->srq)
1141 context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn); 1195 context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn);
1142 1196
1143 if (!ibqp->srq && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) 1197 if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
1144 context->db_rec_addr = cpu_to_be64(qp->db.dma); 1198 context->db_rec_addr = cpu_to_be64(qp->db.dma);
1145 1199
1146 if (cur_state == IB_QPS_INIT && 1200 if (cur_state == IB_QPS_INIT &&
@@ -1225,17 +1279,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
1225 * entries and reinitialize the QP. 1279 * entries and reinitialize the QP.
1226 */ 1280 */
1227 if (new_state == IB_QPS_RESET && !ibqp->uobject) { 1281 if (new_state == IB_QPS_RESET && !ibqp->uobject) {
1228 mlx4_ib_cq_clean(to_mcq(ibqp->recv_cq), qp->mqp.qpn, 1282 mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
1229 ibqp->srq ? to_msrq(ibqp->srq): NULL); 1283 ibqp->srq ? to_msrq(ibqp->srq): NULL);
1230 if (ibqp->send_cq != ibqp->recv_cq) 1284 if (send_cq != recv_cq)
1231 mlx4_ib_cq_clean(to_mcq(ibqp->send_cq), qp->mqp.qpn, NULL); 1285 mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
1232 1286
1233 qp->rq.head = 0; 1287 qp->rq.head = 0;
1234 qp->rq.tail = 0; 1288 qp->rq.tail = 0;
1235 qp->sq.head = 0; 1289 qp->sq.head = 0;
1236 qp->sq.tail = 0; 1290 qp->sq.tail = 0;
1237 qp->sq_next_wqe = 0; 1291 qp->sq_next_wqe = 0;
1238 if (!ibqp->srq) 1292 if (qp->rq.wqe_cnt)
1239 *qp->db.db = 0; 1293 *qp->db.db = 0;
1240 } 1294 }
1241 1295
@@ -1547,14 +1601,13 @@ static void set_masked_atomic_seg(struct mlx4_wqe_masked_atomic_seg *aseg,
1547} 1601}
1548 1602
1549static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg, 1603static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg,
1550 struct ib_send_wr *wr, __be16 *vlan) 1604 struct ib_send_wr *wr)
1551{ 1605{
1552 memcpy(dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof (struct mlx4_av)); 1606 memcpy(dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof (struct mlx4_av));
1553 dseg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn); 1607 dseg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
1554 dseg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey); 1608 dseg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
1555 dseg->vlan = to_mah(wr->wr.ud.ah)->av.eth.vlan; 1609 dseg->vlan = to_mah(wr->wr.ud.ah)->av.eth.vlan;
1556 memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6); 1610 memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6);
1557 *vlan = dseg->vlan;
1558} 1611}
1559 1612
1560static void set_mlx_icrc_seg(void *dseg) 1613static void set_mlx_icrc_seg(void *dseg)
@@ -1657,7 +1710,6 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1657 __be32 uninitialized_var(lso_hdr_sz); 1710 __be32 uninitialized_var(lso_hdr_sz);
1658 __be32 blh; 1711 __be32 blh;
1659 int i; 1712 int i;
1660 __be16 vlan = cpu_to_be16(0xffff);
1661 1713
1662 spin_lock_irqsave(&qp->sq.lock, flags); 1714 spin_lock_irqsave(&qp->sq.lock, flags);
1663 1715
@@ -1761,7 +1813,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1761 break; 1813 break;
1762 1814
1763 case IB_QPT_UD: 1815 case IB_QPT_UD:
1764 set_datagram_seg(wqe, wr, &vlan); 1816 set_datagram_seg(wqe, wr);
1765 wqe += sizeof (struct mlx4_wqe_datagram_seg); 1817 wqe += sizeof (struct mlx4_wqe_datagram_seg);
1766 size += sizeof (struct mlx4_wqe_datagram_seg) / 16; 1818 size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
1767 1819
@@ -1824,11 +1876,6 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1824 ctrl->fence_size = (wr->send_flags & IB_SEND_FENCE ? 1876 ctrl->fence_size = (wr->send_flags & IB_SEND_FENCE ?
1825 MLX4_WQE_CTRL_FENCE : 0) | size; 1877 MLX4_WQE_CTRL_FENCE : 0) | size;
1826 1878
1827 if (be16_to_cpu(vlan) < 0x1000) {
1828 ctrl->ins_vlan = 1 << 6;
1829 ctrl->vlan_tag = vlan;
1830 }
1831
1832 /* 1879 /*
1833 * Make sure descriptor is fully written before 1880 * Make sure descriptor is fully written before
1834 * setting ownership bit (because HW can start 1881 * setting ownership bit (because HW can start
diff --git a/drivers/infiniband/hw/mlx4/srq.c b/drivers/infiniband/hw/mlx4/srq.c
index 818b7ecace5e..39542f3703b8 100644
--- a/drivers/infiniband/hw/mlx4/srq.c
+++ b/drivers/infiniband/hw/mlx4/srq.c
@@ -76,6 +76,8 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
76 struct mlx4_ib_srq *srq; 76 struct mlx4_ib_srq *srq;
77 struct mlx4_wqe_srq_next_seg *next; 77 struct mlx4_wqe_srq_next_seg *next;
78 struct mlx4_wqe_data_seg *scatter; 78 struct mlx4_wqe_data_seg *scatter;
79 u32 cqn;
80 u16 xrcdn;
79 int desc_size; 81 int desc_size;
80 int buf_size; 82 int buf_size;
81 int err; 83 int err;
@@ -174,12 +176,18 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
174 } 176 }
175 } 177 }
176 178
177 err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt, 179 cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
180 to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
181 xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
182 to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
183 (u16) dev->dev->caps.reserved_xrcds;
184 err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
178 srq->db.dma, &srq->msrq); 185 srq->db.dma, &srq->msrq);
179 if (err) 186 if (err)
180 goto err_wrid; 187 goto err_wrid;
181 188
182 srq->msrq.event = mlx4_ib_srq_event; 189 srq->msrq.event = mlx4_ib_srq_event;
190 srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
183 191
184 if (pd->uobject) 192 if (pd->uobject)
185 if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) { 193 if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 365fe0e14192..cb9a0b976804 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -438,6 +438,9 @@ static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
438 struct mthca_srq *srq; 438 struct mthca_srq *srq;
439 int err; 439 int err;
440 440
441 if (init_attr->srq_type != IB_SRQT_BASIC)
442 return ERR_PTR(-ENOSYS);
443
441 srq = kmalloc(sizeof *srq, GFP_KERNEL); 444 srq = kmalloc(sizeof *srq, GFP_KERNEL);
442 if (!srq) 445 if (!srq)
443 return ERR_PTR(-ENOMEM); 446 return ERR_PTR(-ENOMEM);
diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c
index 401b7bb828d0..dfce9ea98a39 100644
--- a/drivers/infiniband/hw/nes/nes_cm.c
+++ b/drivers/infiniband/hw/nes/nes_cm.c
@@ -77,26 +77,19 @@ atomic_t cm_nodes_destroyed;
77atomic_t cm_accel_dropped_pkts; 77atomic_t cm_accel_dropped_pkts;
78atomic_t cm_resets_recvd; 78atomic_t cm_resets_recvd;
79 79
80static inline int mini_cm_accelerated(struct nes_cm_core *, 80static inline int mini_cm_accelerated(struct nes_cm_core *, struct nes_cm_node *);
81 struct nes_cm_node *); 81static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *, struct nes_vnic *, struct nes_cm_info *);
82static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *,
83 struct nes_vnic *, struct nes_cm_info *);
84static int mini_cm_del_listen(struct nes_cm_core *, struct nes_cm_listener *); 82static int mini_cm_del_listen(struct nes_cm_core *, struct nes_cm_listener *);
85static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *, 83static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *, struct nes_vnic *, u16, void *, struct nes_cm_info *);
86 struct nes_vnic *, u16, void *, struct nes_cm_info *);
87static int mini_cm_close(struct nes_cm_core *, struct nes_cm_node *); 84static int mini_cm_close(struct nes_cm_core *, struct nes_cm_node *);
88static int mini_cm_accept(struct nes_cm_core *, struct ietf_mpa_frame *, 85static int mini_cm_accept(struct nes_cm_core *, struct nes_cm_node *);
89 struct nes_cm_node *); 86static int mini_cm_reject(struct nes_cm_core *, struct nes_cm_node *);
90static int mini_cm_reject(struct nes_cm_core *, struct ietf_mpa_frame *, 87static int mini_cm_recv_pkt(struct nes_cm_core *, struct nes_vnic *, struct sk_buff *);
91 struct nes_cm_node *);
92static int mini_cm_recv_pkt(struct nes_cm_core *, struct nes_vnic *,
93 struct sk_buff *);
94static int mini_cm_dealloc_core(struct nes_cm_core *); 88static int mini_cm_dealloc_core(struct nes_cm_core *);
95static int mini_cm_get(struct nes_cm_core *); 89static int mini_cm_get(struct nes_cm_core *);
96static int mini_cm_set(struct nes_cm_core *, u32, u32); 90static int mini_cm_set(struct nes_cm_core *, u32, u32);
97 91
98static void form_cm_frame(struct sk_buff *, struct nes_cm_node *, 92static void form_cm_frame(struct sk_buff *, struct nes_cm_node *, void *, u32, void *, u32, u8);
99 void *, u32, void *, u32, u8);
100static int add_ref_cm_node(struct nes_cm_node *); 93static int add_ref_cm_node(struct nes_cm_node *);
101static int rem_ref_cm_node(struct nes_cm_core *, struct nes_cm_node *); 94static int rem_ref_cm_node(struct nes_cm_core *, struct nes_cm_node *);
102 95
@@ -111,16 +104,14 @@ static int send_syn(struct nes_cm_node *, u32, struct sk_buff *);
111static int send_reset(struct nes_cm_node *, struct sk_buff *); 104static int send_reset(struct nes_cm_node *, struct sk_buff *);
112static int send_ack(struct nes_cm_node *cm_node, struct sk_buff *skb); 105static int send_ack(struct nes_cm_node *cm_node, struct sk_buff *skb);
113static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb); 106static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb);
114static void process_packet(struct nes_cm_node *, struct sk_buff *, 107static void process_packet(struct nes_cm_node *, struct sk_buff *, struct nes_cm_core *);
115 struct nes_cm_core *);
116 108
117static void active_open_err(struct nes_cm_node *, struct sk_buff *, int); 109static void active_open_err(struct nes_cm_node *, struct sk_buff *, int);
118static void passive_open_err(struct nes_cm_node *, struct sk_buff *, int); 110static void passive_open_err(struct nes_cm_node *, struct sk_buff *, int);
119static void cleanup_retrans_entry(struct nes_cm_node *); 111static void cleanup_retrans_entry(struct nes_cm_node *);
120static void handle_rcv_mpa(struct nes_cm_node *, struct sk_buff *); 112static void handle_rcv_mpa(struct nes_cm_node *, struct sk_buff *);
121static void free_retrans_entry(struct nes_cm_node *cm_node); 113static void free_retrans_entry(struct nes_cm_node *cm_node);
122static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph, 114static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph, struct sk_buff *skb, int optionsize, int passive);
123 struct sk_buff *skb, int optionsize, int passive);
124 115
125/* CM event handler functions */ 116/* CM event handler functions */
126static void cm_event_connected(struct nes_cm_event *); 117static void cm_event_connected(struct nes_cm_event *);
@@ -130,6 +121,12 @@ static void cm_event_mpa_req(struct nes_cm_event *);
130static void cm_event_mpa_reject(struct nes_cm_event *); 121static void cm_event_mpa_reject(struct nes_cm_event *);
131static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node); 122static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node);
132 123
124/* MPA build functions */
125static int cm_build_mpa_frame(struct nes_cm_node *, u8 **, u16 *, u8 *, u8);
126static void build_mpa_v2(struct nes_cm_node *, void *, u8);
127static void build_mpa_v1(struct nes_cm_node *, void *, u8);
128static void build_rdma0_msg(struct nes_cm_node *, struct nes_qp **);
129
133static void print_core(struct nes_cm_core *core); 130static void print_core(struct nes_cm_core *core);
134 131
135/* External CM API Interface */ 132/* External CM API Interface */
@@ -172,8 +169,8 @@ int nes_rem_ref_cm_node(struct nes_cm_node *cm_node)
172/** 169/**
173 * create_event 170 * create_event
174 */ 171 */
175static struct nes_cm_event *create_event(struct nes_cm_node *cm_node, 172static struct nes_cm_event *create_event(struct nes_cm_node * cm_node,
176 enum nes_cm_event_type type) 173 enum nes_cm_event_type type)
177{ 174{
178 struct nes_cm_event *event; 175 struct nes_cm_event *event;
179 176
@@ -195,10 +192,10 @@ static struct nes_cm_event *create_event(struct nes_cm_node *cm_node,
195 event->cm_info.cm_id = cm_node->cm_id; 192 event->cm_info.cm_id = cm_node->cm_id;
196 193
197 nes_debug(NES_DBG_CM, "cm_node=%p Created event=%p, type=%u, " 194 nes_debug(NES_DBG_CM, "cm_node=%p Created event=%p, type=%u, "
198 "dst_addr=%08x[%x], src_addr=%08x[%x]\n", 195 "dst_addr=%08x[%x], src_addr=%08x[%x]\n",
199 cm_node, event, type, event->cm_info.loc_addr, 196 cm_node, event, type, event->cm_info.loc_addr,
200 event->cm_info.loc_port, event->cm_info.rem_addr, 197 event->cm_info.loc_port, event->cm_info.rem_addr,
201 event->cm_info.rem_port); 198 event->cm_info.rem_port);
202 199
203 nes_cm_post_event(event); 200 nes_cm_post_event(event);
204 return event; 201 return event;
@@ -210,14 +207,19 @@ static struct nes_cm_event *create_event(struct nes_cm_node *cm_node,
210 */ 207 */
211static int send_mpa_request(struct nes_cm_node *cm_node, struct sk_buff *skb) 208static int send_mpa_request(struct nes_cm_node *cm_node, struct sk_buff *skb)
212{ 209{
210 u8 start_addr = 0;
211 u8 *start_ptr = &start_addr;
212 u8 **start_buff = &start_ptr;
213 u16 buff_len = 0;
214
213 if (!skb) { 215 if (!skb) {
214 nes_debug(NES_DBG_CM, "skb set to NULL\n"); 216 nes_debug(NES_DBG_CM, "skb set to NULL\n");
215 return -1; 217 return -1;
216 } 218 }
217 219
218 /* send an MPA Request frame */ 220 /* send an MPA Request frame */
219 form_cm_frame(skb, cm_node, NULL, 0, &cm_node->mpa_frame, 221 cm_build_mpa_frame(cm_node, start_buff, &buff_len, NULL, MPA_KEY_REQUEST);
220 cm_node->mpa_frame_size, SET_ACK); 222 form_cm_frame(skb, cm_node, NULL, 0, *start_buff, buff_len, SET_ACK);
221 223
222 return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0); 224 return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
223} 225}
@@ -226,7 +228,11 @@ static int send_mpa_request(struct nes_cm_node *cm_node, struct sk_buff *skb)
226 228
227static int send_mpa_reject(struct nes_cm_node *cm_node) 229static int send_mpa_reject(struct nes_cm_node *cm_node)
228{ 230{
229 struct sk_buff *skb = NULL; 231 struct sk_buff *skb = NULL;
232 u8 start_addr = 0;
233 u8 *start_ptr = &start_addr;
234 u8 **start_buff = &start_ptr;
235 u16 buff_len = 0;
230 236
231 skb = dev_alloc_skb(MAX_CM_BUFFER); 237 skb = dev_alloc_skb(MAX_CM_BUFFER);
232 if (!skb) { 238 if (!skb) {
@@ -235,8 +241,8 @@ static int send_mpa_reject(struct nes_cm_node *cm_node)
235 } 241 }
236 242
237 /* send an MPA reject frame */ 243 /* send an MPA reject frame */
238 form_cm_frame(skb, cm_node, NULL, 0, &cm_node->mpa_frame, 244 cm_build_mpa_frame(cm_node, start_buff, &buff_len, NULL, MPA_KEY_REPLY);
239 cm_node->mpa_frame_size, SET_ACK | SET_FIN); 245 form_cm_frame(skb, cm_node, NULL, 0, *start_buff, buff_len, SET_ACK | SET_FIN);
240 246
241 cm_node->state = NES_CM_STATE_FIN_WAIT1; 247 cm_node->state = NES_CM_STATE_FIN_WAIT1;
242 return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0); 248 return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
@@ -248,24 +254,31 @@ static int send_mpa_reject(struct nes_cm_node *cm_node)
248 * IETF MPA frame 254 * IETF MPA frame
249 */ 255 */
250static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type, 256static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type,
251 u32 len) 257 u32 len)
252{ 258{
253 struct ietf_mpa_frame *mpa_frame; 259 struct ietf_mpa_v1 *mpa_frame;
260 struct ietf_mpa_v2 *mpa_v2_frame;
261 struct ietf_rtr_msg *rtr_msg;
262 int mpa_hdr_len;
263 int priv_data_len;
254 264
255 *type = NES_MPA_REQUEST_ACCEPT; 265 *type = NES_MPA_REQUEST_ACCEPT;
256 266
257 /* assume req frame is in tcp data payload */ 267 /* assume req frame is in tcp data payload */
258 if (len < sizeof(struct ietf_mpa_frame)) { 268 if (len < sizeof(struct ietf_mpa_v1)) {
259 nes_debug(NES_DBG_CM, "The received ietf buffer was too small (%x)\n", len); 269 nes_debug(NES_DBG_CM, "The received ietf buffer was too small (%x)\n", len);
260 return -EINVAL; 270 return -EINVAL;
261 } 271 }
262 272
263 mpa_frame = (struct ietf_mpa_frame *)buffer; 273 /* points to the beginning of the frame, which could be MPA V1 or V2 */
264 cm_node->mpa_frame_size = ntohs(mpa_frame->priv_data_len); 274 mpa_frame = (struct ietf_mpa_v1 *)buffer;
275 mpa_hdr_len = sizeof(struct ietf_mpa_v1);
276 priv_data_len = ntohs(mpa_frame->priv_data_len);
277
265 /* make sure mpa private data len is less than 512 bytes */ 278 /* make sure mpa private data len is less than 512 bytes */
266 if (cm_node->mpa_frame_size > IETF_MAX_PRIV_DATA_LEN) { 279 if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) {
267 nes_debug(NES_DBG_CM, "The received Length of Private" 280 nes_debug(NES_DBG_CM, "The received Length of Private"
268 " Data field exceeds 512 octets\n"); 281 " Data field exceeds 512 octets\n");
269 return -EINVAL; 282 return -EINVAL;
270 } 283 }
271 /* 284 /*
@@ -273,11 +286,22 @@ static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type,
273 * received MPA version and MPA key information 286 * received MPA version and MPA key information
274 * 287 *
275 */ 288 */
276 if (mpa_frame->rev != mpa_version) { 289 if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) {
290 nes_debug(NES_DBG_CM, "The received mpa version"
291 " is not supported\n");
292 return -EINVAL;
293 }
294 /*
295 * backwards compatibility only
296 */
297 if (mpa_frame->rev > cm_node->mpa_frame_rev) {
277 nes_debug(NES_DBG_CM, "The received mpa version" 298 nes_debug(NES_DBG_CM, "The received mpa version"
278 " can not be interoperated\n"); 299 " can not be interoperated\n");
279 return -EINVAL; 300 return -EINVAL;
301 } else {
302 cm_node->mpa_frame_rev = mpa_frame->rev;
280 } 303 }
304
281 if (cm_node->state != NES_CM_STATE_MPAREQ_SENT) { 305 if (cm_node->state != NES_CM_STATE_MPAREQ_SENT) {
282 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE)) { 306 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE)) {
283 nes_debug(NES_DBG_CM, "Unexpected MPA Key received \n"); 307 nes_debug(NES_DBG_CM, "Unexpected MPA Key received \n");
@@ -290,25 +314,75 @@ static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type,
290 } 314 }
291 } 315 }
292 316
293 if (cm_node->mpa_frame_size + sizeof(struct ietf_mpa_frame) != len) { 317
318 if (priv_data_len + mpa_hdr_len != len) {
294 nes_debug(NES_DBG_CM, "The received ietf buffer was not right" 319 nes_debug(NES_DBG_CM, "The received ietf buffer was not right"
295 " complete (%x + %x != %x)\n", 320 " complete (%x + %x != %x)\n",
296 cm_node->mpa_frame_size, 321 priv_data_len, mpa_hdr_len, len);
297 (u32)sizeof(struct ietf_mpa_frame), len);
298 return -EINVAL; 322 return -EINVAL;
299 } 323 }
300 /* make sure it does not exceed the max size */ 324 /* make sure it does not exceed the max size */
301 if (len > MAX_CM_BUFFER) { 325 if (len > MAX_CM_BUFFER) {
302 nes_debug(NES_DBG_CM, "The received ietf buffer was too large" 326 nes_debug(NES_DBG_CM, "The received ietf buffer was too large"
303 " (%x + %x != %x)\n", 327 " (%x + %x != %x)\n",
304 cm_node->mpa_frame_size, 328 priv_data_len, mpa_hdr_len, len);
305 (u32)sizeof(struct ietf_mpa_frame), len);
306 return -EINVAL; 329 return -EINVAL;
307 } 330 }
308 331
332 cm_node->mpa_frame_size = priv_data_len;
333
334 switch (mpa_frame->rev) {
335 case IETF_MPA_V2: {
336 u16 ird_size;
337 u16 ord_size;
338 mpa_v2_frame = (struct ietf_mpa_v2 *)buffer;
339 mpa_hdr_len += IETF_RTR_MSG_SIZE;
340 cm_node->mpa_frame_size -= IETF_RTR_MSG_SIZE;
341 rtr_msg = &mpa_v2_frame->rtr_msg;
342
343 /* parse rtr message */
344 rtr_msg->ctrl_ird = ntohs(rtr_msg->ctrl_ird);
345 rtr_msg->ctrl_ord = ntohs(rtr_msg->ctrl_ord);
346 ird_size = rtr_msg->ctrl_ird & IETF_NO_IRD_ORD;
347 ord_size = rtr_msg->ctrl_ord & IETF_NO_IRD_ORD;
348
349 if (!(rtr_msg->ctrl_ird & IETF_PEER_TO_PEER)) {
350 /* send reset */
351 return -EINVAL;
352 }
353
354 if (cm_node->state != NES_CM_STATE_MPAREQ_SENT) {
355 /* responder */
356 if (cm_node->ord_size > ird_size)
357 cm_node->ord_size = ird_size;
358 } else {
359 /* initiator */
360 if (cm_node->ord_size > ird_size)
361 cm_node->ord_size = ird_size;
362
363 if (cm_node->ird_size < ord_size) {
364 /* no resources available */
365 /* send terminate message */
366 return -EINVAL;
367 }
368 }
369
370 if (rtr_msg->ctrl_ord & IETF_RDMA0_READ) {
371 cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
372 } else if (rtr_msg->ctrl_ord & IETF_RDMA0_WRITE) {
373 cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO;
374 } else { /* Not supported RDMA0 operation */
375 return -EINVAL;
376 }
377 break;
378 }
379 case IETF_MPA_V1:
380 default:
381 break;
382 }
383
309 /* copy entire MPA frame to our cm_node's frame */ 384 /* copy entire MPA frame to our cm_node's frame */
310 memcpy(cm_node->mpa_frame_buf, buffer + sizeof(struct ietf_mpa_frame), 385 memcpy(cm_node->mpa_frame_buf, buffer + mpa_hdr_len, cm_node->mpa_frame_size);
311 cm_node->mpa_frame_size);
312 386
313 if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT) 387 if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT)
314 *type = NES_MPA_REQUEST_REJECT; 388 *type = NES_MPA_REQUEST_REJECT;
@@ -321,8 +395,8 @@ static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type,
321 * node info to build. 395 * node info to build.
322 */ 396 */
323static void form_cm_frame(struct sk_buff *skb, 397static void form_cm_frame(struct sk_buff *skb,
324 struct nes_cm_node *cm_node, void *options, u32 optionsize, 398 struct nes_cm_node *cm_node, void *options, u32 optionsize,
325 void *data, u32 datasize, u8 flags) 399 void *data, u32 datasize, u8 flags)
326{ 400{
327 struct tcphdr *tcph; 401 struct tcphdr *tcph;
328 struct iphdr *iph; 402 struct iphdr *iph;
@@ -331,14 +405,14 @@ static void form_cm_frame(struct sk_buff *skb,
331 u16 packetsize = sizeof(*iph); 405 u16 packetsize = sizeof(*iph);
332 406
333 packetsize += sizeof(*tcph); 407 packetsize += sizeof(*tcph);
334 packetsize += optionsize + datasize; 408 packetsize += optionsize + datasize;
335 409
410 skb_trim(skb, 0);
336 memset(skb->data, 0x00, ETH_HLEN + sizeof(*iph) + sizeof(*tcph)); 411 memset(skb->data, 0x00, ETH_HLEN + sizeof(*iph) + sizeof(*tcph));
337 412
338 skb->len = 0;
339 buf = skb_put(skb, packetsize + ETH_HLEN); 413 buf = skb_put(skb, packetsize + ETH_HLEN);
340 414
341 ethh = (struct ethhdr *) buf; 415 ethh = (struct ethhdr *)buf;
342 buf += ETH_HLEN; 416 buf += ETH_HLEN;
343 417
344 iph = (struct iphdr *)buf; 418 iph = (struct iphdr *)buf;
@@ -346,7 +420,7 @@ static void form_cm_frame(struct sk_buff *skb,
346 tcph = (struct tcphdr *)buf; 420 tcph = (struct tcphdr *)buf;
347 skb_reset_mac_header(skb); 421 skb_reset_mac_header(skb);
348 skb_set_network_header(skb, ETH_HLEN); 422 skb_set_network_header(skb, ETH_HLEN);
349 skb_set_transport_header(skb, ETH_HLEN+sizeof(*iph)); 423 skb_set_transport_header(skb, ETH_HLEN + sizeof(*iph));
350 buf += sizeof(*tcph); 424 buf += sizeof(*tcph);
351 425
352 skb->ip_summed = CHECKSUM_PARTIAL; 426 skb->ip_summed = CHECKSUM_PARTIAL;
@@ -359,14 +433,14 @@ static void form_cm_frame(struct sk_buff *skb,
359 ethh->h_proto = htons(0x0800); 433 ethh->h_proto = htons(0x0800);
360 434
361 iph->version = IPVERSION; 435 iph->version = IPVERSION;
362 iph->ihl = 5; /* 5 * 4Byte words, IP headr len */ 436 iph->ihl = 5; /* 5 * 4Byte words, IP headr len */
363 iph->tos = 0; 437 iph->tos = 0;
364 iph->tot_len = htons(packetsize); 438 iph->tot_len = htons(packetsize);
365 iph->id = htons(++cm_node->tcp_cntxt.loc_id); 439 iph->id = htons(++cm_node->tcp_cntxt.loc_id);
366 440
367 iph->frag_off = htons(0x4000); 441 iph->frag_off = htons(0x4000);
368 iph->ttl = 0x40; 442 iph->ttl = 0x40;
369 iph->protocol = 0x06; /* IPPROTO_TCP */ 443 iph->protocol = 0x06; /* IPPROTO_TCP */
370 444
371 iph->saddr = htonl(cm_node->loc_addr); 445 iph->saddr = htonl(cm_node->loc_addr);
372 iph->daddr = htonl(cm_node->rem_addr); 446 iph->daddr = htonl(cm_node->rem_addr);
@@ -379,14 +453,16 @@ static void form_cm_frame(struct sk_buff *skb,
379 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt; 453 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
380 tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num); 454 tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
381 tcph->ack = 1; 455 tcph->ack = 1;
382 } else 456 } else {
383 tcph->ack_seq = 0; 457 tcph->ack_seq = 0;
458 }
384 459
385 if (flags & SET_SYN) { 460 if (flags & SET_SYN) {
386 cm_node->tcp_cntxt.loc_seq_num++; 461 cm_node->tcp_cntxt.loc_seq_num++;
387 tcph->syn = 1; 462 tcph->syn = 1;
388 } else 463 } else {
389 cm_node->tcp_cntxt.loc_seq_num += datasize; 464 cm_node->tcp_cntxt.loc_seq_num += datasize;
465 }
390 466
391 if (flags & SET_FIN) { 467 if (flags & SET_FIN) {
392 cm_node->tcp_cntxt.loc_seq_num++; 468 cm_node->tcp_cntxt.loc_seq_num++;
@@ -407,10 +483,8 @@ static void form_cm_frame(struct sk_buff *skb,
407 483
408 skb_shinfo(skb)->nr_frags = 0; 484 skb_shinfo(skb)->nr_frags = 0;
409 cm_packets_created++; 485 cm_packets_created++;
410
411} 486}
412 487
413
414/** 488/**
415 * print_core - dump a cm core 489 * print_core - dump a cm core
416 */ 490 */
@@ -422,7 +496,7 @@ static void print_core(struct nes_cm_core *core)
422 return; 496 return;
423 nes_debug(NES_DBG_CM, "---------------------------------------------\n"); 497 nes_debug(NES_DBG_CM, "---------------------------------------------\n");
424 498
425 nes_debug(NES_DBG_CM, "State : %u \n", core->state); 499 nes_debug(NES_DBG_CM, "State : %u \n", core->state);
426 500
427 nes_debug(NES_DBG_CM, "Listen Nodes : %u \n", atomic_read(&core->listen_node_cnt)); 501 nes_debug(NES_DBG_CM, "Listen Nodes : %u \n", atomic_read(&core->listen_node_cnt));
428 nes_debug(NES_DBG_CM, "Active Nodes : %u \n", atomic_read(&core->node_cnt)); 502 nes_debug(NES_DBG_CM, "Active Nodes : %u \n", atomic_read(&core->node_cnt));
@@ -432,6 +506,147 @@ static void print_core(struct nes_cm_core *core)
432 nes_debug(NES_DBG_CM, "-------------- end core ---------------\n"); 506 nes_debug(NES_DBG_CM, "-------------- end core ---------------\n");
433} 507}
434 508
509/**
510 * cm_build_mpa_frame - build a MPA V1 frame or MPA V2 frame
511 */
512static int cm_build_mpa_frame(struct nes_cm_node *cm_node, u8 **start_buff,
513 u16 *buff_len, u8 *pci_mem, u8 mpa_key)
514{
515 int ret = 0;
516
517 *start_buff = (pci_mem) ? pci_mem : &cm_node->mpa_frame_buf[0];
518
519 switch (cm_node->mpa_frame_rev) {
520 case IETF_MPA_V1:
521 *start_buff = (u8 *)*start_buff + sizeof(struct ietf_rtr_msg);
522 *buff_len = sizeof(struct ietf_mpa_v1) + cm_node->mpa_frame_size;
523 build_mpa_v1(cm_node, *start_buff, mpa_key);
524 break;
525 case IETF_MPA_V2:
526 *buff_len = sizeof(struct ietf_mpa_v2) + cm_node->mpa_frame_size;
527 build_mpa_v2(cm_node, *start_buff, mpa_key);
528 break;
529 default:
530 ret = -EINVAL;
531 }
532 return ret;
533}
534
535/**
536 * build_mpa_v2 - build a MPA V2 frame
537 */
538static void build_mpa_v2(struct nes_cm_node *cm_node,
539 void *start_addr, u8 mpa_key)
540{
541 struct ietf_mpa_v2 *mpa_frame = (struct ietf_mpa_v2 *)start_addr;
542 struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg;
543
544 /* initialize the upper 5 bytes of the frame */
545 build_mpa_v1(cm_node, start_addr, mpa_key);
546 mpa_frame->flags |= IETF_MPA_V2_FLAG; /* set a bit to indicate MPA V2 */
547 mpa_frame->priv_data_len += htons(IETF_RTR_MSG_SIZE);
548
549 /* initialize RTR msg */
550 rtr_msg->ctrl_ird = (cm_node->ird_size > IETF_NO_IRD_ORD) ?
551 IETF_NO_IRD_ORD : cm_node->ird_size;
552 rtr_msg->ctrl_ord = (cm_node->ord_size > IETF_NO_IRD_ORD) ?
553 IETF_NO_IRD_ORD : cm_node->ord_size;
554
555 rtr_msg->ctrl_ird |= IETF_PEER_TO_PEER;
556 rtr_msg->ctrl_ird |= IETF_FLPDU_ZERO_LEN;
557
558 switch (mpa_key) {
559 case MPA_KEY_REQUEST:
560 rtr_msg->ctrl_ord |= IETF_RDMA0_WRITE;
561 rtr_msg->ctrl_ord |= IETF_RDMA0_READ;
562 break;
563 case MPA_KEY_REPLY:
564 switch (cm_node->send_rdma0_op) {
565 case SEND_RDMA_WRITE_ZERO:
566 rtr_msg->ctrl_ord |= IETF_RDMA0_WRITE;
567 break;
568 case SEND_RDMA_READ_ZERO:
569 rtr_msg->ctrl_ord |= IETF_RDMA0_READ;
570 break;
571 }
572 }
573 rtr_msg->ctrl_ird = htons(rtr_msg->ctrl_ird);
574 rtr_msg->ctrl_ord = htons(rtr_msg->ctrl_ord);
575}
576
577/**
578 * build_mpa_v1 - build a MPA V1 frame
579 */
580static void build_mpa_v1(struct nes_cm_node *cm_node, void *start_addr, u8 mpa_key)
581{
582 struct ietf_mpa_v1 *mpa_frame = (struct ietf_mpa_v1 *)start_addr;
583
584 switch (mpa_key) {
585 case MPA_KEY_REQUEST:
586 memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
587 break;
588 case MPA_KEY_REPLY:
589 memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
590 break;
591 }
592 mpa_frame->flags = IETF_MPA_FLAGS_CRC;
593 mpa_frame->rev = cm_node->mpa_frame_rev;
594 mpa_frame->priv_data_len = htons(cm_node->mpa_frame_size);
595}
596
597static void build_rdma0_msg(struct nes_cm_node *cm_node, struct nes_qp **nesqp_addr)
598{
599 u64 u64temp;
600 struct nes_qp *nesqp = *nesqp_addr;
601 struct nes_hw_qp_wqe *wqe = &nesqp->hwqp.sq_vbase[0];
602
603 u64temp = (unsigned long)nesqp;
604 u64temp |= NES_SW_CONTEXT_ALIGN >> 1;
605 set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX, u64temp);
606
607 wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_LOW_IDX] = 0;
608 wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_HIGH_IDX] = 0;
609
610 switch (cm_node->send_rdma0_op) {
611 case SEND_RDMA_WRITE_ZERO:
612 nes_debug(NES_DBG_CM, "Sending first write.\n");
613 wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
614 cpu_to_le32(NES_IWARP_SQ_OP_RDMAW);
615 wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] = 0;
616 wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] = 0;
617 wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 0;
618 break;
619
620 case SEND_RDMA_READ_ZERO:
621 default:
622 if (cm_node->send_rdma0_op != SEND_RDMA_READ_ZERO) {
623 printk(KERN_ERR "%s[%u]: Unsupported RDMA0 len operation=%u\n",
624 __func__, __LINE__, cm_node->send_rdma0_op);
625 WARN_ON(1);
626 }
627 nes_debug(NES_DBG_CM, "Sending first rdma operation.\n");
628 wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
629 cpu_to_le32(NES_IWARP_SQ_OP_RDMAR);
630 wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX] = 1;
631 wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_TO_HIGH_IDX] = 0;
632 wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX] = 0;
633 wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_STAG_IDX] = 1;
634 wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 1;
635 break;
636 }
637
638 if (nesqp->sq_kmapped) {
639 nesqp->sq_kmapped = 0;
640 kunmap(nesqp->page);
641 }
642
643 /*use the reserved spot on the WQ for the extra first WQE*/
644 nesqp->nesqp_context->ird_ord_sizes &= cpu_to_le32(~(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
645 NES_QPCONTEXT_ORDIRD_WRPDU |
646 NES_QPCONTEXT_ORDIRD_ALSMM));
647 nesqp->skip_lsmm = 1;
648 nesqp->hwqp.sq_tail = 0;
649}
435 650
436/** 651/**
437 * schedule_nes_timer 652 * schedule_nes_timer
@@ -439,10 +654,10 @@ static void print_core(struct nes_cm_core *core)
439 * rem_ref_cm_node(cm_core, cm_node);add_ref_cm_node(cm_node); 654 * rem_ref_cm_node(cm_core, cm_node);add_ref_cm_node(cm_node);
440 */ 655 */
441int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb, 656int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb,
442 enum nes_timer_type type, int send_retrans, 657 enum nes_timer_type type, int send_retrans,
443 int close_when_complete) 658 int close_when_complete)
444{ 659{
445 unsigned long flags; 660 unsigned long flags;
446 struct nes_cm_core *cm_core = cm_node->cm_core; 661 struct nes_cm_core *cm_core = cm_node->cm_core;
447 struct nes_timer_entry *new_send; 662 struct nes_timer_entry *new_send;
448 int ret = 0; 663 int ret = 0;
@@ -463,7 +678,7 @@ int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb,
463 new_send->close_when_complete = close_when_complete; 678 new_send->close_when_complete = close_when_complete;
464 679
465 if (type == NES_TIMER_TYPE_CLOSE) { 680 if (type == NES_TIMER_TYPE_CLOSE) {
466 new_send->timetosend += (HZ/10); 681 new_send->timetosend += (HZ / 10);
467 if (cm_node->recv_entry) { 682 if (cm_node->recv_entry) {
468 kfree(new_send); 683 kfree(new_send);
469 WARN_ON(1); 684 WARN_ON(1);
@@ -484,7 +699,7 @@ int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb,
484 ret = nes_nic_cm_xmit(new_send->skb, cm_node->netdev); 699 ret = nes_nic_cm_xmit(new_send->skb, cm_node->netdev);
485 if (ret != NETDEV_TX_OK) { 700 if (ret != NETDEV_TX_OK) {
486 nes_debug(NES_DBG_CM, "Error sending packet %p " 701 nes_debug(NES_DBG_CM, "Error sending packet %p "
487 "(jiffies = %lu)\n", new_send, jiffies); 702 "(jiffies = %lu)\n", new_send, jiffies);
488 new_send->timetosend = jiffies; 703 new_send->timetosend = jiffies;
489 ret = NETDEV_TX_OK; 704 ret = NETDEV_TX_OK;
490 } else { 705 } else {
@@ -513,6 +728,7 @@ static void nes_retrans_expired(struct nes_cm_node *cm_node)
513 struct iw_cm_id *cm_id = cm_node->cm_id; 728 struct iw_cm_id *cm_id = cm_node->cm_id;
514 enum nes_cm_node_state state = cm_node->state; 729 enum nes_cm_node_state state = cm_node->state;
515 cm_node->state = NES_CM_STATE_CLOSED; 730 cm_node->state = NES_CM_STATE_CLOSED;
731
516 switch (state) { 732 switch (state) {
517 case NES_CM_STATE_SYN_RCVD: 733 case NES_CM_STATE_SYN_RCVD:
518 case NES_CM_STATE_CLOSING: 734 case NES_CM_STATE_CLOSING:
@@ -545,10 +761,10 @@ static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node)
545 spin_lock_irqsave(&nesqp->lock, qplockflags); 761 spin_lock_irqsave(&nesqp->lock, qplockflags);
546 if (nesqp->cm_id) { 762 if (nesqp->cm_id) {
547 nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, " 763 nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, "
548 "refcount = %d: HIT A " 764 "refcount = %d: HIT A "
549 "NES_TIMER_TYPE_CLOSE with something " 765 "NES_TIMER_TYPE_CLOSE with something "
550 "to do!!!\n", nesqp->hwqp.qp_id, cm_id, 766 "to do!!!\n", nesqp->hwqp.qp_id, cm_id,
551 atomic_read(&nesqp->refcount)); 767 atomic_read(&nesqp->refcount));
552 nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED; 768 nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
553 nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT; 769 nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
554 nesqp->ibqp_state = IB_QPS_ERR; 770 nesqp->ibqp_state = IB_QPS_ERR;
@@ -557,10 +773,10 @@ static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node)
557 } else { 773 } else {
558 spin_unlock_irqrestore(&nesqp->lock, qplockflags); 774 spin_unlock_irqrestore(&nesqp->lock, qplockflags);
559 nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, " 775 nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, "
560 "refcount = %d: HIT A " 776 "refcount = %d: HIT A "
561 "NES_TIMER_TYPE_CLOSE with nothing " 777 "NES_TIMER_TYPE_CLOSE with nothing "
562 "to do!!!\n", nesqp->hwqp.qp_id, cm_id, 778 "to do!!!\n", nesqp->hwqp.qp_id, cm_id,
563 atomic_read(&nesqp->refcount)); 779 atomic_read(&nesqp->refcount));
564 } 780 }
565 } else if (rem_node) { 781 } else if (rem_node) {
566 /* TIME_WAIT state */ 782 /* TIME_WAIT state */
@@ -589,11 +805,12 @@ static void nes_cm_timer_tick(unsigned long pass)
589 int ret = NETDEV_TX_OK; 805 int ret = NETDEV_TX_OK;
590 806
591 struct list_head timer_list; 807 struct list_head timer_list;
808
592 INIT_LIST_HEAD(&timer_list); 809 INIT_LIST_HEAD(&timer_list);
593 spin_lock_irqsave(&cm_core->ht_lock, flags); 810 spin_lock_irqsave(&cm_core->ht_lock, flags);
594 811
595 list_for_each_safe(list_node, list_core_temp, 812 list_for_each_safe(list_node, list_core_temp,
596 &cm_core->connected_nodes) { 813 &cm_core->connected_nodes) {
597 cm_node = container_of(list_node, struct nes_cm_node, list); 814 cm_node = container_of(list_node, struct nes_cm_node, list);
598 if ((cm_node->recv_entry) || (cm_node->send_entry)) { 815 if ((cm_node->recv_entry) || (cm_node->send_entry)) {
599 add_ref_cm_node(cm_node); 816 add_ref_cm_node(cm_node);
@@ -604,18 +821,19 @@ static void nes_cm_timer_tick(unsigned long pass)
604 821
605 list_for_each_safe(list_node, list_core_temp, &timer_list) { 822 list_for_each_safe(list_node, list_core_temp, &timer_list) {
606 cm_node = container_of(list_node, struct nes_cm_node, 823 cm_node = container_of(list_node, struct nes_cm_node,
607 timer_entry); 824 timer_entry);
608 recv_entry = cm_node->recv_entry; 825 recv_entry = cm_node->recv_entry;
609 826
610 if (recv_entry) { 827 if (recv_entry) {
611 if (time_after(recv_entry->timetosend, jiffies)) { 828 if (time_after(recv_entry->timetosend, jiffies)) {
612 if (nexttimeout > recv_entry->timetosend || 829 if (nexttimeout > recv_entry->timetosend ||
613 !settimer) { 830 !settimer) {
614 nexttimeout = recv_entry->timetosend; 831 nexttimeout = recv_entry->timetosend;
615 settimer = 1; 832 settimer = 1;
616 } 833 }
617 } else 834 } else {
618 handle_recv_entry(cm_node, 1); 835 handle_recv_entry(cm_node, 1);
836 }
619 } 837 }
620 838
621 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 839 spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
@@ -626,8 +844,8 @@ static void nes_cm_timer_tick(unsigned long pass)
626 if (time_after(send_entry->timetosend, jiffies)) { 844 if (time_after(send_entry->timetosend, jiffies)) {
627 if (cm_node->state != NES_CM_STATE_TSA) { 845 if (cm_node->state != NES_CM_STATE_TSA) {
628 if ((nexttimeout > 846 if ((nexttimeout >
629 send_entry->timetosend) || 847 send_entry->timetosend) ||
630 !settimer) { 848 !settimer) {
631 nexttimeout = 849 nexttimeout =
632 send_entry->timetosend; 850 send_entry->timetosend;
633 settimer = 1; 851 settimer = 1;
@@ -639,13 +857,13 @@ static void nes_cm_timer_tick(unsigned long pass)
639 } 857 }
640 858
641 if ((cm_node->state == NES_CM_STATE_TSA) || 859 if ((cm_node->state == NES_CM_STATE_TSA) ||
642 (cm_node->state == NES_CM_STATE_CLOSED)) { 860 (cm_node->state == NES_CM_STATE_CLOSED)) {
643 free_retrans_entry(cm_node); 861 free_retrans_entry(cm_node);
644 break; 862 break;
645 } 863 }
646 864
647 if (!send_entry->retranscount || 865 if (!send_entry->retranscount ||
648 !send_entry->retrycount) { 866 !send_entry->retrycount) {
649 cm_packets_dropped++; 867 cm_packets_dropped++;
650 free_retrans_entry(cm_node); 868 free_retrans_entry(cm_node);
651 869
@@ -654,28 +872,28 @@ static void nes_cm_timer_tick(unsigned long pass)
654 nes_retrans_expired(cm_node); 872 nes_retrans_expired(cm_node);
655 cm_node->state = NES_CM_STATE_CLOSED; 873 cm_node->state = NES_CM_STATE_CLOSED;
656 spin_lock_irqsave(&cm_node->retrans_list_lock, 874 spin_lock_irqsave(&cm_node->retrans_list_lock,
657 flags); 875 flags);
658 break; 876 break;
659 } 877 }
660 atomic_inc(&send_entry->skb->users); 878 atomic_inc(&send_entry->skb->users);
661 cm_packets_retrans++; 879 cm_packets_retrans++;
662 nes_debug(NES_DBG_CM, "Retransmitting send_entry %p " 880 nes_debug(NES_DBG_CM, "Retransmitting send_entry %p "
663 "for node %p, jiffies = %lu, time to send = " 881 "for node %p, jiffies = %lu, time to send = "
664 "%lu, retranscount = %u, send_entry->seq_num = " 882 "%lu, retranscount = %u, send_entry->seq_num = "
665 "0x%08X, cm_node->tcp_cntxt.rem_ack_num = " 883 "0x%08X, cm_node->tcp_cntxt.rem_ack_num = "
666 "0x%08X\n", send_entry, cm_node, jiffies, 884 "0x%08X\n", send_entry, cm_node, jiffies,
667 send_entry->timetosend, 885 send_entry->timetosend,
668 send_entry->retranscount, 886 send_entry->retranscount,
669 send_entry->seq_num, 887 send_entry->seq_num,
670 cm_node->tcp_cntxt.rem_ack_num); 888 cm_node->tcp_cntxt.rem_ack_num);
671 889
672 spin_unlock_irqrestore(&cm_node->retrans_list_lock, 890 spin_unlock_irqrestore(&cm_node->retrans_list_lock,
673 flags); 891 flags);
674 ret = nes_nic_cm_xmit(send_entry->skb, cm_node->netdev); 892 ret = nes_nic_cm_xmit(send_entry->skb, cm_node->netdev);
675 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 893 spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
676 if (ret != NETDEV_TX_OK) { 894 if (ret != NETDEV_TX_OK) {
677 nes_debug(NES_DBG_CM, "rexmit failed for " 895 nes_debug(NES_DBG_CM, "rexmit failed for "
678 "node=%p\n", cm_node); 896 "node=%p\n", cm_node);
679 cm_packets_bounced++; 897 cm_packets_bounced++;
680 send_entry->retrycount--; 898 send_entry->retrycount--;
681 nexttimeout = jiffies + NES_SHORT_TIME; 899 nexttimeout = jiffies + NES_SHORT_TIME;
@@ -685,18 +903,18 @@ static void nes_cm_timer_tick(unsigned long pass)
685 cm_packets_sent++; 903 cm_packets_sent++;
686 } 904 }
687 nes_debug(NES_DBG_CM, "Packet Sent: retrans count = " 905 nes_debug(NES_DBG_CM, "Packet Sent: retrans count = "
688 "%u, retry count = %u.\n", 906 "%u, retry count = %u.\n",
689 send_entry->retranscount, 907 send_entry->retranscount,
690 send_entry->retrycount); 908 send_entry->retrycount);
691 if (send_entry->send_retrans) { 909 if (send_entry->send_retrans) {
692 send_entry->retranscount--; 910 send_entry->retranscount--;
693 timetosend = (NES_RETRY_TIMEOUT << 911 timetosend = (NES_RETRY_TIMEOUT <<
694 (NES_DEFAULT_RETRANS - send_entry->retranscount)); 912 (NES_DEFAULT_RETRANS - send_entry->retranscount));
695 913
696 send_entry->timetosend = jiffies + 914 send_entry->timetosend = jiffies +
697 min(timetosend, NES_MAX_TIMEOUT); 915 min(timetosend, NES_MAX_TIMEOUT);
698 if (nexttimeout > send_entry->timetosend || 916 if (nexttimeout > send_entry->timetosend ||
699 !settimer) { 917 !settimer) {
700 nexttimeout = send_entry->timetosend; 918 nexttimeout = send_entry->timetosend;
701 settimer = 1; 919 settimer = 1;
702 } 920 }
@@ -705,11 +923,11 @@ static void nes_cm_timer_tick(unsigned long pass)
705 close_when_complete = 923 close_when_complete =
706 send_entry->close_when_complete; 924 send_entry->close_when_complete;
707 nes_debug(NES_DBG_CM, "cm_node=%p state=%d\n", 925 nes_debug(NES_DBG_CM, "cm_node=%p state=%d\n",
708 cm_node, cm_node->state); 926 cm_node, cm_node->state);
709 free_retrans_entry(cm_node); 927 free_retrans_entry(cm_node);
710 if (close_when_complete) 928 if (close_when_complete)
711 rem_ref_cm_node(cm_node->cm_core, 929 rem_ref_cm_node(cm_node->cm_core,
712 cm_node); 930 cm_node);
713 } 931 }
714 } while (0); 932 } while (0);
715 933
@@ -719,7 +937,7 @@ static void nes_cm_timer_tick(unsigned long pass)
719 937
720 if (settimer) { 938 if (settimer) {
721 if (!timer_pending(&cm_core->tcp_timer)) { 939 if (!timer_pending(&cm_core->tcp_timer)) {
722 cm_core->tcp_timer.expires = nexttimeout; 940 cm_core->tcp_timer.expires = nexttimeout;
723 add_timer(&cm_core->tcp_timer); 941 add_timer(&cm_core->tcp_timer);
724 } 942 }
725 } 943 }
@@ -730,13 +948,13 @@ static void nes_cm_timer_tick(unsigned long pass)
730 * send_syn 948 * send_syn
731 */ 949 */
732static int send_syn(struct nes_cm_node *cm_node, u32 sendack, 950static int send_syn(struct nes_cm_node *cm_node, u32 sendack,
733 struct sk_buff *skb) 951 struct sk_buff *skb)
734{ 952{
735 int ret; 953 int ret;
736 int flags = SET_SYN; 954 int flags = SET_SYN;
737 char optionsbuffer[sizeof(struct option_mss) + 955 char optionsbuffer[sizeof(struct option_mss) +
738 sizeof(struct option_windowscale) + sizeof(struct option_base) + 956 sizeof(struct option_windowscale) + sizeof(struct option_base) +
739 TCP_OPTIONS_PADDING]; 957 TCP_OPTIONS_PADDING];
740 958
741 int optionssize = 0; 959 int optionssize = 0;
742 /* Sending MSS option */ 960 /* Sending MSS option */
@@ -863,7 +1081,7 @@ static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb)
863 * find_node - find a cm node that matches the reference cm node 1081 * find_node - find a cm node that matches the reference cm node
864 */ 1082 */
865static struct nes_cm_node *find_node(struct nes_cm_core *cm_core, 1083static struct nes_cm_node *find_node(struct nes_cm_core *cm_core,
866 u16 rem_port, nes_addr_t rem_addr, u16 loc_port, nes_addr_t loc_addr) 1084 u16 rem_port, nes_addr_t rem_addr, u16 loc_port, nes_addr_t loc_addr)
867{ 1085{
868 unsigned long flags; 1086 unsigned long flags;
869 struct list_head *hte; 1087 struct list_head *hte;
@@ -877,12 +1095,12 @@ static struct nes_cm_node *find_node(struct nes_cm_core *cm_core,
877 list_for_each_entry(cm_node, hte, list) { 1095 list_for_each_entry(cm_node, hte, list) {
878 /* compare quad, return node handle if a match */ 1096 /* compare quad, return node handle if a match */
879 nes_debug(NES_DBG_CM, "finding node %x:%x =? %x:%x ^ %x:%x =? %x:%x\n", 1097 nes_debug(NES_DBG_CM, "finding node %x:%x =? %x:%x ^ %x:%x =? %x:%x\n",
880 cm_node->loc_addr, cm_node->loc_port, 1098 cm_node->loc_addr, cm_node->loc_port,
881 loc_addr, loc_port, 1099 loc_addr, loc_port,
882 cm_node->rem_addr, cm_node->rem_port, 1100 cm_node->rem_addr, cm_node->rem_port,
883 rem_addr, rem_port); 1101 rem_addr, rem_port);
884 if ((cm_node->loc_addr == loc_addr) && (cm_node->loc_port == loc_port) && 1102 if ((cm_node->loc_addr == loc_addr) && (cm_node->loc_port == loc_port) &&
885 (cm_node->rem_addr == rem_addr) && (cm_node->rem_port == rem_port)) { 1103 (cm_node->rem_addr == rem_addr) && (cm_node->rem_port == rem_port)) {
886 add_ref_cm_node(cm_node); 1104 add_ref_cm_node(cm_node);
887 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 1105 spin_unlock_irqrestore(&cm_core->ht_lock, flags);
888 return cm_node; 1106 return cm_node;
@@ -899,7 +1117,7 @@ static struct nes_cm_node *find_node(struct nes_cm_core *cm_core,
899 * find_listener - find a cm node listening on this addr-port pair 1117 * find_listener - find a cm node listening on this addr-port pair
900 */ 1118 */
901static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core, 1119static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core,
902 nes_addr_t dst_addr, u16 dst_port, enum nes_cm_listener_state listener_state) 1120 nes_addr_t dst_addr, u16 dst_port, enum nes_cm_listener_state listener_state)
903{ 1121{
904 unsigned long flags; 1122 unsigned long flags;
905 struct nes_cm_listener *listen_node; 1123 struct nes_cm_listener *listen_node;
@@ -909,9 +1127,9 @@ static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core,
909 list_for_each_entry(listen_node, &cm_core->listen_list.list, list) { 1127 list_for_each_entry(listen_node, &cm_core->listen_list.list, list) {
910 /* compare node pair, return node handle if a match */ 1128 /* compare node pair, return node handle if a match */
911 if (((listen_node->loc_addr == dst_addr) || 1129 if (((listen_node->loc_addr == dst_addr) ||
912 listen_node->loc_addr == 0x00000000) && 1130 listen_node->loc_addr == 0x00000000) &&
913 (listen_node->loc_port == dst_port) && 1131 (listen_node->loc_port == dst_port) &&
914 (listener_state & listen_node->listener_state)) { 1132 (listener_state & listen_node->listener_state)) {
915 atomic_inc(&listen_node->ref_count); 1133 atomic_inc(&listen_node->ref_count);
916 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 1134 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
917 return listen_node; 1135 return listen_node;
@@ -936,7 +1154,7 @@ static int add_hte_node(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node
936 return -EINVAL; 1154 return -EINVAL;
937 1155
938 nes_debug(NES_DBG_CM, "Adding Node %p to Active Connection HT\n", 1156 nes_debug(NES_DBG_CM, "Adding Node %p to Active Connection HT\n",
939 cm_node); 1157 cm_node);
940 1158
941 spin_lock_irqsave(&cm_core->ht_lock, flags); 1159 spin_lock_irqsave(&cm_core->ht_lock, flags);
942 1160
@@ -955,7 +1173,7 @@ static int add_hte_node(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node
955 * mini_cm_dec_refcnt_listen 1173 * mini_cm_dec_refcnt_listen
956 */ 1174 */
957static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core, 1175static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
958 struct nes_cm_listener *listener, int free_hanging_nodes) 1176 struct nes_cm_listener *listener, int free_hanging_nodes)
959{ 1177{
960 int ret = -EINVAL; 1178 int ret = -EINVAL;
961 int err = 0; 1179 int err = 0;
@@ -966,8 +1184,8 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
966 struct list_head reset_list; 1184 struct list_head reset_list;
967 1185
968 nes_debug(NES_DBG_CM, "attempting listener= %p free_nodes= %d, " 1186 nes_debug(NES_DBG_CM, "attempting listener= %p free_nodes= %d, "
969 "refcnt=%d\n", listener, free_hanging_nodes, 1187 "refcnt=%d\n", listener, free_hanging_nodes,
970 atomic_read(&listener->ref_count)); 1188 atomic_read(&listener->ref_count));
971 /* free non-accelerated child nodes for this listener */ 1189 /* free non-accelerated child nodes for this listener */
972 INIT_LIST_HEAD(&reset_list); 1190 INIT_LIST_HEAD(&reset_list);
973 if (free_hanging_nodes) { 1191 if (free_hanging_nodes) {
@@ -975,7 +1193,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
975 list_for_each_safe(list_pos, list_temp, 1193 list_for_each_safe(list_pos, list_temp,
976 &g_cm_core->connected_nodes) { 1194 &g_cm_core->connected_nodes) {
977 cm_node = container_of(list_pos, struct nes_cm_node, 1195 cm_node = container_of(list_pos, struct nes_cm_node,
978 list); 1196 list);
979 if ((cm_node->listener == listener) && 1197 if ((cm_node->listener == listener) &&
980 (!cm_node->accelerated)) { 1198 (!cm_node->accelerated)) {
981 add_ref_cm_node(cm_node); 1199 add_ref_cm_node(cm_node);
@@ -987,7 +1205,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
987 1205
988 list_for_each_safe(list_pos, list_temp, &reset_list) { 1206 list_for_each_safe(list_pos, list_temp, &reset_list) {
989 cm_node = container_of(list_pos, struct nes_cm_node, 1207 cm_node = container_of(list_pos, struct nes_cm_node,
990 reset_entry); 1208 reset_entry);
991 { 1209 {
992 struct nes_cm_node *loopback = cm_node->loopbackpartner; 1210 struct nes_cm_node *loopback = cm_node->loopbackpartner;
993 enum nes_cm_node_state old_state; 1211 enum nes_cm_node_state old_state;
@@ -999,7 +1217,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
999 err = send_reset(cm_node, NULL); 1217 err = send_reset(cm_node, NULL);
1000 if (err) { 1218 if (err) {
1001 cm_node->state = 1219 cm_node->state =
1002 NES_CM_STATE_CLOSED; 1220 NES_CM_STATE_CLOSED;
1003 WARN_ON(1); 1221 WARN_ON(1);
1004 } else { 1222 } else {
1005 old_state = cm_node->state; 1223 old_state = cm_node->state;
@@ -1044,10 +1262,9 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
1044 1262
1045 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 1263 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
1046 1264
1047 if (listener->nesvnic) { 1265 if (listener->nesvnic)
1048 nes_manage_apbvt(listener->nesvnic, listener->loc_port, 1266 nes_manage_apbvt(listener->nesvnic, listener->loc_port,
1049 PCI_FUNC(listener->nesvnic->nesdev->pcidev->devfn), NES_MANAGE_APBVT_DEL); 1267 PCI_FUNC(listener->nesvnic->nesdev->pcidev->devfn), NES_MANAGE_APBVT_DEL);
1050 }
1051 1268
1052 nes_debug(NES_DBG_CM, "destroying listener (%p)\n", listener); 1269 nes_debug(NES_DBG_CM, "destroying listener (%p)\n", listener);
1053 1270
@@ -1061,8 +1278,8 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
1061 if (listener) { 1278 if (listener) {
1062 if (atomic_read(&listener->pend_accepts_cnt) > 0) 1279 if (atomic_read(&listener->pend_accepts_cnt) > 0)
1063 nes_debug(NES_DBG_CM, "destroying listener (%p)" 1280 nes_debug(NES_DBG_CM, "destroying listener (%p)"
1064 " with non-zero pending accepts=%u\n", 1281 " with non-zero pending accepts=%u\n",
1065 listener, atomic_read(&listener->pend_accepts_cnt)); 1282 listener, atomic_read(&listener->pend_accepts_cnt));
1066 } 1283 }
1067 1284
1068 return ret; 1285 return ret;
@@ -1073,7 +1290,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
1073 * mini_cm_del_listen 1290 * mini_cm_del_listen
1074 */ 1291 */
1075static int mini_cm_del_listen(struct nes_cm_core *cm_core, 1292static int mini_cm_del_listen(struct nes_cm_core *cm_core,
1076 struct nes_cm_listener *listener) 1293 struct nes_cm_listener *listener)
1077{ 1294{
1078 listener->listener_state = NES_CM_LISTENER_PASSIVE_STATE; 1295 listener->listener_state = NES_CM_LISTENER_PASSIVE_STATE;
1079 listener->cm_id = NULL; /* going to be destroyed pretty soon */ 1296 listener->cm_id = NULL; /* going to be destroyed pretty soon */
@@ -1085,9 +1302,10 @@ static int mini_cm_del_listen(struct nes_cm_core *cm_core,
1085 * mini_cm_accelerated 1302 * mini_cm_accelerated
1086 */ 1303 */
1087static inline int mini_cm_accelerated(struct nes_cm_core *cm_core, 1304static inline int mini_cm_accelerated(struct nes_cm_core *cm_core,
1088 struct nes_cm_node *cm_node) 1305 struct nes_cm_node *cm_node)
1089{ 1306{
1090 u32 was_timer_set; 1307 u32 was_timer_set;
1308
1091 cm_node->accelerated = 1; 1309 cm_node->accelerated = 1;
1092 1310
1093 if (cm_node->accept_pend) { 1311 if (cm_node->accept_pend) {
@@ -1121,7 +1339,7 @@ static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpi
1121 rt = ip_route_output(&init_net, htonl(dst_ip), 0, 0, 0); 1339 rt = ip_route_output(&init_net, htonl(dst_ip), 0, 0, 0);
1122 if (IS_ERR(rt)) { 1340 if (IS_ERR(rt)) {
1123 printk(KERN_ERR "%s: ip_route_output_key failed for 0x%08X\n", 1341 printk(KERN_ERR "%s: ip_route_output_key failed for 0x%08X\n",
1124 __func__, dst_ip); 1342 __func__, dst_ip);
1125 return rc; 1343 return rc;
1126 } 1344 }
1127 1345
@@ -1139,7 +1357,7 @@ static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpi
1139 1357
1140 if (arpindex >= 0) { 1358 if (arpindex >= 0) {
1141 if (!memcmp(nesadapter->arp_table[arpindex].mac_addr, 1359 if (!memcmp(nesadapter->arp_table[arpindex].mac_addr,
1142 neigh->ha, ETH_ALEN)){ 1360 neigh->ha, ETH_ALEN)) {
1143 /* Mac address same as in nes_arp_table */ 1361 /* Mac address same as in nes_arp_table */
1144 neigh_release(neigh); 1362 neigh_release(neigh);
1145 ip_rt_put(rt); 1363 ip_rt_put(rt);
@@ -1147,8 +1365,8 @@ static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpi
1147 } 1365 }
1148 1366
1149 nes_manage_arp_cache(nesvnic->netdev, 1367 nes_manage_arp_cache(nesvnic->netdev,
1150 nesadapter->arp_table[arpindex].mac_addr, 1368 nesadapter->arp_table[arpindex].mac_addr,
1151 dst_ip, NES_ARP_DELETE); 1369 dst_ip, NES_ARP_DELETE);
1152 } 1370 }
1153 1371
1154 nes_manage_arp_cache(nesvnic->netdev, neigh->ha, 1372 nes_manage_arp_cache(nesvnic->netdev, neigh->ha,
@@ -1170,8 +1388,8 @@ static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpi
1170 * make_cm_node - create a new instance of a cm node 1388 * make_cm_node - create a new instance of a cm node
1171 */ 1389 */
1172static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core, 1390static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
1173 struct nes_vnic *nesvnic, struct nes_cm_info *cm_info, 1391 struct nes_vnic *nesvnic, struct nes_cm_info *cm_info,
1174 struct nes_cm_listener *listener) 1392 struct nes_cm_listener *listener)
1175{ 1393{
1176 struct nes_cm_node *cm_node; 1394 struct nes_cm_node *cm_node;
1177 struct timespec ts; 1395 struct timespec ts;
@@ -1190,7 +1408,12 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
1190 cm_node->rem_addr = cm_info->rem_addr; 1408 cm_node->rem_addr = cm_info->rem_addr;
1191 cm_node->loc_port = cm_info->loc_port; 1409 cm_node->loc_port = cm_info->loc_port;
1192 cm_node->rem_port = cm_info->rem_port; 1410 cm_node->rem_port = cm_info->rem_port;
1193 cm_node->send_write0 = send_first; 1411
1412 cm_node->mpa_frame_rev = mpa_version;
1413 cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
1414 cm_node->ird_size = IETF_NO_IRD_ORD;
1415 cm_node->ord_size = IETF_NO_IRD_ORD;
1416
1194 nes_debug(NES_DBG_CM, "Make node addresses : loc = %pI4:%x, rem = %pI4:%x\n", 1417 nes_debug(NES_DBG_CM, "Make node addresses : loc = %pI4:%x, rem = %pI4:%x\n",
1195 &cm_node->loc_addr, cm_node->loc_port, 1418 &cm_node->loc_addr, cm_node->loc_port,
1196 &cm_node->rem_addr, cm_node->rem_port); 1419 &cm_node->rem_addr, cm_node->rem_port);
@@ -1200,7 +1423,7 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
1200 memcpy(cm_node->loc_mac, nesvnic->netdev->dev_addr, ETH_ALEN); 1423 memcpy(cm_node->loc_mac, nesvnic->netdev->dev_addr, ETH_ALEN);
1201 1424
1202 nes_debug(NES_DBG_CM, "listener=%p, cm_id=%p\n", cm_node->listener, 1425 nes_debug(NES_DBG_CM, "listener=%p, cm_id=%p\n", cm_node->listener,
1203 cm_node->cm_id); 1426 cm_node->cm_id);
1204 1427
1205 spin_lock_init(&cm_node->retrans_list_lock); 1428 spin_lock_init(&cm_node->retrans_list_lock);
1206 1429
@@ -1211,11 +1434,11 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
1211 cm_node->tcp_cntxt.loc_id = NES_CM_DEF_LOCAL_ID; 1434 cm_node->tcp_cntxt.loc_id = NES_CM_DEF_LOCAL_ID;
1212 cm_node->tcp_cntxt.rcv_wscale = NES_CM_DEFAULT_RCV_WND_SCALE; 1435 cm_node->tcp_cntxt.rcv_wscale = NES_CM_DEFAULT_RCV_WND_SCALE;
1213 cm_node->tcp_cntxt.rcv_wnd = NES_CM_DEFAULT_RCV_WND_SCALED >> 1436 cm_node->tcp_cntxt.rcv_wnd = NES_CM_DEFAULT_RCV_WND_SCALED >>
1214 NES_CM_DEFAULT_RCV_WND_SCALE; 1437 NES_CM_DEFAULT_RCV_WND_SCALE;
1215 ts = current_kernel_time(); 1438 ts = current_kernel_time();
1216 cm_node->tcp_cntxt.loc_seq_num = htonl(ts.tv_nsec); 1439 cm_node->tcp_cntxt.loc_seq_num = htonl(ts.tv_nsec);
1217 cm_node->tcp_cntxt.mss = nesvnic->max_frame_size - sizeof(struct iphdr) - 1440 cm_node->tcp_cntxt.mss = nesvnic->max_frame_size - sizeof(struct iphdr) -
1218 sizeof(struct tcphdr) - ETH_HLEN - VLAN_HLEN; 1441 sizeof(struct tcphdr) - ETH_HLEN - VLAN_HLEN;
1219 cm_node->tcp_cntxt.rcv_nxt = 0; 1442 cm_node->tcp_cntxt.rcv_nxt = 0;
1220 /* get a unique session ID , add thread_id to an upcounter to handle race */ 1443 /* get a unique session ID , add thread_id to an upcounter to handle race */
1221 atomic_inc(&cm_core->node_cnt); 1444 atomic_inc(&cm_core->node_cnt);
@@ -1231,12 +1454,11 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
1231 cm_node->loopbackpartner = NULL; 1454 cm_node->loopbackpartner = NULL;
1232 1455
1233 /* get the mac addr for the remote node */ 1456 /* get the mac addr for the remote node */
1234 if (ipv4_is_loopback(htonl(cm_node->rem_addr))) 1457 if (ipv4_is_loopback(htonl(cm_node->rem_addr))) {
1235 arpindex = nes_arp_table(nesdev, ntohl(nesvnic->local_ipaddr), NULL, NES_ARP_RESOLVE); 1458 arpindex = nes_arp_table(nesdev, ntohl(nesvnic->local_ipaddr), NULL, NES_ARP_RESOLVE);
1236 else { 1459 } else {
1237 oldarpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE); 1460 oldarpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE);
1238 arpindex = nes_addr_resolve_neigh(nesvnic, cm_info->rem_addr, oldarpindex); 1461 arpindex = nes_addr_resolve_neigh(nesvnic, cm_info->rem_addr, oldarpindex);
1239
1240 } 1462 }
1241 if (arpindex < 0) { 1463 if (arpindex < 0) {
1242 kfree(cm_node); 1464 kfree(cm_node);
@@ -1269,7 +1491,7 @@ static int add_ref_cm_node(struct nes_cm_node *cm_node)
1269 * rem_ref_cm_node - destroy an instance of a cm node 1491 * rem_ref_cm_node - destroy an instance of a cm node
1270 */ 1492 */
1271static int rem_ref_cm_node(struct nes_cm_core *cm_core, 1493static int rem_ref_cm_node(struct nes_cm_core *cm_core,
1272 struct nes_cm_node *cm_node) 1494 struct nes_cm_node *cm_node)
1273{ 1495{
1274 unsigned long flags; 1496 unsigned long flags;
1275 struct nes_qp *nesqp; 1497 struct nes_qp *nesqp;
@@ -1300,9 +1522,9 @@ static int rem_ref_cm_node(struct nes_cm_core *cm_core,
1300 } else { 1522 } else {
1301 if (cm_node->apbvt_set && cm_node->nesvnic) { 1523 if (cm_node->apbvt_set && cm_node->nesvnic) {
1302 nes_manage_apbvt(cm_node->nesvnic, cm_node->loc_port, 1524 nes_manage_apbvt(cm_node->nesvnic, cm_node->loc_port,
1303 PCI_FUNC( 1525 PCI_FUNC(
1304 cm_node->nesvnic->nesdev->pcidev->devfn), 1526 cm_node->nesvnic->nesdev->pcidev->devfn),
1305 NES_MANAGE_APBVT_DEL); 1527 NES_MANAGE_APBVT_DEL);
1306 } 1528 }
1307 } 1529 }
1308 1530
@@ -1323,7 +1545,7 @@ static int rem_ref_cm_node(struct nes_cm_core *cm_core,
1323 * process_options 1545 * process_options
1324 */ 1546 */
1325static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc, 1547static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc,
1326 u32 optionsize, u32 syn_packet) 1548 u32 optionsize, u32 syn_packet)
1327{ 1549{
1328 u32 tmp; 1550 u32 tmp;
1329 u32 offset = 0; 1551 u32 offset = 0;
@@ -1341,15 +1563,15 @@ static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc,
1341 continue; 1563 continue;
1342 case OPTION_NUMBER_MSS: 1564 case OPTION_NUMBER_MSS:
1343 nes_debug(NES_DBG_CM, "%s: MSS Length: %d Offset: %d " 1565 nes_debug(NES_DBG_CM, "%s: MSS Length: %d Offset: %d "
1344 "Size: %d\n", __func__, 1566 "Size: %d\n", __func__,
1345 all_options->as_mss.length, offset, optionsize); 1567 all_options->as_mss.length, offset, optionsize);
1346 got_mss_option = 1; 1568 got_mss_option = 1;
1347 if (all_options->as_mss.length != 4) { 1569 if (all_options->as_mss.length != 4) {
1348 return 1; 1570 return 1;
1349 } else { 1571 } else {
1350 tmp = ntohs(all_options->as_mss.mss); 1572 tmp = ntohs(all_options->as_mss.mss);
1351 if (tmp > 0 && tmp < 1573 if (tmp > 0 && tmp <
1352 cm_node->tcp_cntxt.mss) 1574 cm_node->tcp_cntxt.mss)
1353 cm_node->tcp_cntxt.mss = tmp; 1575 cm_node->tcp_cntxt.mss = tmp;
1354 } 1576 }
1355 break; 1577 break;
@@ -1357,12 +1579,9 @@ static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc,
1357 cm_node->tcp_cntxt.snd_wscale = 1579 cm_node->tcp_cntxt.snd_wscale =
1358 all_options->as_windowscale.shiftcount; 1580 all_options->as_windowscale.shiftcount;
1359 break; 1581 break;
1360 case OPTION_NUMBER_WRITE0:
1361 cm_node->send_write0 = 1;
1362 break;
1363 default: 1582 default:
1364 nes_debug(NES_DBG_CM, "TCP Option not understood: %x\n", 1583 nes_debug(NES_DBG_CM, "TCP Option not understood: %x\n",
1365 all_options->as_base.optionnum); 1584 all_options->as_base.optionnum);
1366 break; 1585 break;
1367 } 1586 }
1368 offset += all_options->as_base.length; 1587 offset += all_options->as_base.length;
@@ -1381,8 +1600,8 @@ static void drop_packet(struct sk_buff *skb)
1381static void handle_fin_pkt(struct nes_cm_node *cm_node) 1600static void handle_fin_pkt(struct nes_cm_node *cm_node)
1382{ 1601{
1383 nes_debug(NES_DBG_CM, "Received FIN, cm_node = %p, state = %u. " 1602 nes_debug(NES_DBG_CM, "Received FIN, cm_node = %p, state = %u. "
1384 "refcnt=%d\n", cm_node, cm_node->state, 1603 "refcnt=%d\n", cm_node, cm_node->state,
1385 atomic_read(&cm_node->ref_count)); 1604 atomic_read(&cm_node->ref_count));
1386 switch (cm_node->state) { 1605 switch (cm_node->state) {
1387 case NES_CM_STATE_SYN_RCVD: 1606 case NES_CM_STATE_SYN_RCVD:
1388 case NES_CM_STATE_SYN_SENT: 1607 case NES_CM_STATE_SYN_SENT:
@@ -1448,7 +1667,20 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1448 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p " 1667 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
1449 "listener=%p state=%d\n", __func__, __LINE__, cm_node, 1668 "listener=%p state=%d\n", __func__, __LINE__, cm_node,
1450 cm_node->listener, cm_node->state); 1669 cm_node->listener, cm_node->state);
1451 active_open_err(cm_node, skb, reset); 1670 switch (cm_node->mpa_frame_rev) {
1671 case IETF_MPA_V2:
1672 cm_node->mpa_frame_rev = IETF_MPA_V1;
1673 /* send a syn and goto syn sent state */
1674 cm_node->state = NES_CM_STATE_SYN_SENT;
1675 if (send_syn(cm_node, 0, NULL)) {
1676 active_open_err(cm_node, skb, reset);
1677 }
1678 break;
1679 case IETF_MPA_V1:
1680 default:
1681 active_open_err(cm_node, skb, reset);
1682 break;
1683 }
1452 break; 1684 break;
1453 case NES_CM_STATE_MPAREQ_RCVD: 1685 case NES_CM_STATE_MPAREQ_RCVD:
1454 atomic_inc(&cm_node->passive_state); 1686 atomic_inc(&cm_node->passive_state);
@@ -1484,21 +1716,21 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1484 1716
1485static void handle_rcv_mpa(struct nes_cm_node *cm_node, struct sk_buff *skb) 1717static void handle_rcv_mpa(struct nes_cm_node *cm_node, struct sk_buff *skb)
1486{ 1718{
1487 1719 int ret = 0;
1488 int ret = 0;
1489 int datasize = skb->len; 1720 int datasize = skb->len;
1490 u8 *dataloc = skb->data; 1721 u8 *dataloc = skb->data;
1491 1722
1492 enum nes_cm_event_type type = NES_CM_EVENT_UNKNOWN; 1723 enum nes_cm_event_type type = NES_CM_EVENT_UNKNOWN;
1493 u32 res_type; 1724 u32 res_type;
1725
1494 ret = parse_mpa(cm_node, dataloc, &res_type, datasize); 1726 ret = parse_mpa(cm_node, dataloc, &res_type, datasize);
1495 if (ret) { 1727 if (ret) {
1496 nes_debug(NES_DBG_CM, "didn't like MPA Request\n"); 1728 nes_debug(NES_DBG_CM, "didn't like MPA Request\n");
1497 if (cm_node->state == NES_CM_STATE_MPAREQ_SENT) { 1729 if (cm_node->state == NES_CM_STATE_MPAREQ_SENT) {
1498 nes_debug(NES_DBG_CM, "%s[%u] create abort for " 1730 nes_debug(NES_DBG_CM, "%s[%u] create abort for "
1499 "cm_node=%p listener=%p state=%d\n", __func__, 1731 "cm_node=%p listener=%p state=%d\n", __func__,
1500 __LINE__, cm_node, cm_node->listener, 1732 __LINE__, cm_node, cm_node->listener,
1501 cm_node->state); 1733 cm_node->state);
1502 active_open_err(cm_node, skb, 1); 1734 active_open_err(cm_node, skb, 1);
1503 } else { 1735 } else {
1504 passive_open_err(cm_node, skb, 1); 1736 passive_open_err(cm_node, skb, 1);
@@ -1508,16 +1740,15 @@ static void handle_rcv_mpa(struct nes_cm_node *cm_node, struct sk_buff *skb)
1508 1740
1509 switch (cm_node->state) { 1741 switch (cm_node->state) {
1510 case NES_CM_STATE_ESTABLISHED: 1742 case NES_CM_STATE_ESTABLISHED:
1511 if (res_type == NES_MPA_REQUEST_REJECT) { 1743 if (res_type == NES_MPA_REQUEST_REJECT)
1512 /*BIG problem as we are receiving the MPA.. So should 1744 /*BIG problem as we are receiving the MPA.. So should
1513 * not be REJECT.. This is Passive Open.. We can 1745 * not be REJECT.. This is Passive Open.. We can
1514 * only receive it Reject for Active Open...*/ 1746 * only receive it Reject for Active Open...*/
1515 WARN_ON(1); 1747 WARN_ON(1);
1516 }
1517 cm_node->state = NES_CM_STATE_MPAREQ_RCVD; 1748 cm_node->state = NES_CM_STATE_MPAREQ_RCVD;
1518 type = NES_CM_EVENT_MPA_REQ; 1749 type = NES_CM_EVENT_MPA_REQ;
1519 atomic_set(&cm_node->passive_state, 1750 atomic_set(&cm_node->passive_state,
1520 NES_PASSIVE_STATE_INDICATED); 1751 NES_PASSIVE_STATE_INDICATED);
1521 break; 1752 break;
1522 case NES_CM_STATE_MPAREQ_SENT: 1753 case NES_CM_STATE_MPAREQ_SENT:
1523 cleanup_retrans_entry(cm_node); 1754 cleanup_retrans_entry(cm_node);
@@ -1544,8 +1775,8 @@ static void indicate_pkt_err(struct nes_cm_node *cm_node, struct sk_buff *skb)
1544 case NES_CM_STATE_SYN_SENT: 1775 case NES_CM_STATE_SYN_SENT:
1545 case NES_CM_STATE_MPAREQ_SENT: 1776 case NES_CM_STATE_MPAREQ_SENT:
1546 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p " 1777 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
1547 "listener=%p state=%d\n", __func__, __LINE__, cm_node, 1778 "listener=%p state=%d\n", __func__, __LINE__, cm_node,
1548 cm_node->listener, cm_node->state); 1779 cm_node->listener, cm_node->state);
1549 active_open_err(cm_node, skb, 1); 1780 active_open_err(cm_node, skb, 1);
1550 break; 1781 break;
1551 case NES_CM_STATE_ESTABLISHED: 1782 case NES_CM_STATE_ESTABLISHED:
@@ -1559,11 +1790,11 @@ static void indicate_pkt_err(struct nes_cm_node *cm_node, struct sk_buff *skb)
1559} 1790}
1560 1791
1561static int check_syn(struct nes_cm_node *cm_node, struct tcphdr *tcph, 1792static int check_syn(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1562 struct sk_buff *skb) 1793 struct sk_buff *skb)
1563{ 1794{
1564 int err; 1795 int err;
1565 1796
1566 err = ((ntohl(tcph->ack_seq) == cm_node->tcp_cntxt.loc_seq_num))? 0 : 1; 1797 err = ((ntohl(tcph->ack_seq) == cm_node->tcp_cntxt.loc_seq_num)) ? 0 : 1;
1567 if (err) 1798 if (err)
1568 active_open_err(cm_node, skb, 1); 1799 active_open_err(cm_node, skb, 1);
1569 1800
@@ -1571,7 +1802,7 @@ static int check_syn(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1571} 1802}
1572 1803
1573static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph, 1804static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1574 struct sk_buff *skb) 1805 struct sk_buff *skb)
1575{ 1806{
1576 int err = 0; 1807 int err = 0;
1577 u32 seq; 1808 u32 seq;
@@ -1579,21 +1810,22 @@ static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1579 u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num; 1810 u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num;
1580 u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt; 1811 u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
1581 u32 rcv_wnd; 1812 u32 rcv_wnd;
1813
1582 seq = ntohl(tcph->seq); 1814 seq = ntohl(tcph->seq);
1583 ack_seq = ntohl(tcph->ack_seq); 1815 ack_seq = ntohl(tcph->ack_seq);
1584 rcv_wnd = cm_node->tcp_cntxt.rcv_wnd; 1816 rcv_wnd = cm_node->tcp_cntxt.rcv_wnd;
1585 if (ack_seq != loc_seq_num) 1817 if (ack_seq != loc_seq_num)
1586 err = 1; 1818 err = 1;
1587 else if (!between(seq, rcv_nxt, (rcv_nxt+rcv_wnd))) 1819 else if (!between(seq, rcv_nxt, (rcv_nxt + rcv_wnd)))
1588 err = 1; 1820 err = 1;
1589 if (err) { 1821 if (err) {
1590 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p " 1822 nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
1591 "listener=%p state=%d\n", __func__, __LINE__, cm_node, 1823 "listener=%p state=%d\n", __func__, __LINE__, cm_node,
1592 cm_node->listener, cm_node->state); 1824 cm_node->listener, cm_node->state);
1593 indicate_pkt_err(cm_node, skb); 1825 indicate_pkt_err(cm_node, skb);
1594 nes_debug(NES_DBG_CM, "seq ERROR cm_node =%p seq=0x%08X " 1826 nes_debug(NES_DBG_CM, "seq ERROR cm_node =%p seq=0x%08X "
1595 "rcv_nxt=0x%08X rcv_wnd=0x%x\n", cm_node, seq, rcv_nxt, 1827 "rcv_nxt=0x%08X rcv_wnd=0x%x\n", cm_node, seq, rcv_nxt,
1596 rcv_wnd); 1828 rcv_wnd);
1597 } 1829 }
1598 return err; 1830 return err;
1599} 1831}
@@ -1603,9 +1835,8 @@ static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1603 * is created with a listener or it may comein as rexmitted packet which in 1835 * is created with a listener or it may comein as rexmitted packet which in
1604 * that case will be just dropped. 1836 * that case will be just dropped.
1605 */ 1837 */
1606
1607static void handle_syn_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, 1838static void handle_syn_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1608 struct tcphdr *tcph) 1839 struct tcphdr *tcph)
1609{ 1840{
1610 int ret; 1841 int ret;
1611 u32 inc_sequence; 1842 u32 inc_sequence;
@@ -1624,15 +1855,15 @@ static void handle_syn_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1624 case NES_CM_STATE_LISTENING: 1855 case NES_CM_STATE_LISTENING:
1625 /* Passive OPEN */ 1856 /* Passive OPEN */
1626 if (atomic_read(&cm_node->listener->pend_accepts_cnt) > 1857 if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
1627 cm_node->listener->backlog) { 1858 cm_node->listener->backlog) {
1628 nes_debug(NES_DBG_CM, "drop syn due to backlog " 1859 nes_debug(NES_DBG_CM, "drop syn due to backlog "
1629 "pressure \n"); 1860 "pressure \n");
1630 cm_backlog_drops++; 1861 cm_backlog_drops++;
1631 passive_open_err(cm_node, skb, 0); 1862 passive_open_err(cm_node, skb, 0);
1632 break; 1863 break;
1633 } 1864 }
1634 ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 1865 ret = handle_tcp_options(cm_node, tcph, skb, optionsize,
1635 1); 1866 1);
1636 if (ret) { 1867 if (ret) {
1637 passive_open_err(cm_node, skb, 0); 1868 passive_open_err(cm_node, skb, 0);
1638 /* drop pkt */ 1869 /* drop pkt */
@@ -1666,9 +1897,8 @@ static void handle_syn_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1666} 1897}
1667 1898
1668static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, 1899static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1669 struct tcphdr *tcph) 1900 struct tcphdr *tcph)
1670{ 1901{
1671
1672 int ret; 1902 int ret;
1673 u32 inc_sequence; 1903 u32 inc_sequence;
1674 int optionsize; 1904 int optionsize;
@@ -1687,7 +1917,7 @@ static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1687 ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 0); 1917 ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 0);
1688 if (ret) { 1918 if (ret) {
1689 nes_debug(NES_DBG_CM, "cm_node=%p tcp_options failed\n", 1919 nes_debug(NES_DBG_CM, "cm_node=%p tcp_options failed\n",
1690 cm_node); 1920 cm_node);
1691 break; 1921 break;
1692 } 1922 }
1693 cleanup_retrans_entry(cm_node); 1923 cleanup_retrans_entry(cm_node);
@@ -1726,12 +1956,13 @@ static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1726} 1956}
1727 1957
1728static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, 1958static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1729 struct tcphdr *tcph) 1959 struct tcphdr *tcph)
1730{ 1960{
1731 int datasize = 0; 1961 int datasize = 0;
1732 u32 inc_sequence; 1962 u32 inc_sequence;
1733 int ret = 0; 1963 int ret = 0;
1734 int optionsize; 1964 int optionsize;
1965
1735 optionsize = (tcph->doff << 2) - sizeof(struct tcphdr); 1966 optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
1736 1967
1737 if (check_seq(cm_node, tcph, skb)) 1968 if (check_seq(cm_node, tcph, skb))
@@ -1752,8 +1983,9 @@ static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1752 if (datasize) { 1983 if (datasize) {
1753 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 1984 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
1754 handle_rcv_mpa(cm_node, skb); 1985 handle_rcv_mpa(cm_node, skb);
1755 } else /* rcvd ACK only */ 1986 } else { /* rcvd ACK only */
1756 dev_kfree_skb_any(skb); 1987 dev_kfree_skb_any(skb);
1988 }
1757 break; 1989 break;
1758 case NES_CM_STATE_ESTABLISHED: 1990 case NES_CM_STATE_ESTABLISHED:
1759 /* Passive OPEN */ 1991 /* Passive OPEN */
@@ -1761,16 +1993,18 @@ static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1761 if (datasize) { 1993 if (datasize) {
1762 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 1994 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
1763 handle_rcv_mpa(cm_node, skb); 1995 handle_rcv_mpa(cm_node, skb);
1764 } else 1996 } else {
1765 drop_packet(skb); 1997 drop_packet(skb);
1998 }
1766 break; 1999 break;
1767 case NES_CM_STATE_MPAREQ_SENT: 2000 case NES_CM_STATE_MPAREQ_SENT:
1768 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); 2001 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
1769 if (datasize) { 2002 if (datasize) {
1770 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 2003 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
1771 handle_rcv_mpa(cm_node, skb); 2004 handle_rcv_mpa(cm_node, skb);
1772 } else /* Could be just an ack pkt.. */ 2005 } else { /* Could be just an ack pkt.. */
1773 dev_kfree_skb_any(skb); 2006 dev_kfree_skb_any(skb);
2007 }
1774 break; 2008 break;
1775 case NES_CM_STATE_LISTENING: 2009 case NES_CM_STATE_LISTENING:
1776 cleanup_retrans_entry(cm_node); 2010 cleanup_retrans_entry(cm_node);
@@ -1811,14 +2045,15 @@ static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
1811 2045
1812 2046
1813static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph, 2047static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1814 struct sk_buff *skb, int optionsize, int passive) 2048 struct sk_buff *skb, int optionsize, int passive)
1815{ 2049{
1816 u8 *optionsloc = (u8 *)&tcph[1]; 2050 u8 *optionsloc = (u8 *)&tcph[1];
2051
1817 if (optionsize) { 2052 if (optionsize) {
1818 if (process_options(cm_node, optionsloc, optionsize, 2053 if (process_options(cm_node, optionsloc, optionsize,
1819 (u32)tcph->syn)) { 2054 (u32)tcph->syn)) {
1820 nes_debug(NES_DBG_CM, "%s: Node %p, Sending RESET\n", 2055 nes_debug(NES_DBG_CM, "%s: Node %p, Sending RESET\n",
1821 __func__, cm_node); 2056 __func__, cm_node);
1822 if (passive) 2057 if (passive)
1823 passive_open_err(cm_node, skb, 1); 2058 passive_open_err(cm_node, skb, 1);
1824 else 2059 else
@@ -1828,7 +2063,7 @@ static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1828 } 2063 }
1829 2064
1830 cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window) << 2065 cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window) <<
1831 cm_node->tcp_cntxt.snd_wscale; 2066 cm_node->tcp_cntxt.snd_wscale;
1832 2067
1833 if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd) 2068 if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd)
1834 cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd; 2069 cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
@@ -1839,18 +2074,18 @@ static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph,
1839 * active_open_err() will send reset() if flag set.. 2074 * active_open_err() will send reset() if flag set..
1840 * It will also send ABORT event. 2075 * It will also send ABORT event.
1841 */ 2076 */
1842
1843static void active_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb, 2077static void active_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
1844 int reset) 2078 int reset)
1845{ 2079{
1846 cleanup_retrans_entry(cm_node); 2080 cleanup_retrans_entry(cm_node);
1847 if (reset) { 2081 if (reset) {
1848 nes_debug(NES_DBG_CM, "ERROR active err called for cm_node=%p, " 2082 nes_debug(NES_DBG_CM, "ERROR active err called for cm_node=%p, "
1849 "state=%d\n", cm_node, cm_node->state); 2083 "state=%d\n", cm_node, cm_node->state);
1850 add_ref_cm_node(cm_node); 2084 add_ref_cm_node(cm_node);
1851 send_reset(cm_node, skb); 2085 send_reset(cm_node, skb);
1852 } else 2086 } else {
1853 dev_kfree_skb_any(skb); 2087 dev_kfree_skb_any(skb);
2088 }
1854 2089
1855 cm_node->state = NES_CM_STATE_CLOSED; 2090 cm_node->state = NES_CM_STATE_CLOSED;
1856 create_event(cm_node, NES_CM_EVENT_ABORTED); 2091 create_event(cm_node, NES_CM_EVENT_ABORTED);
@@ -1860,15 +2095,14 @@ static void active_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
1860 * passive_open_err() will either do a reset() or will free up the skb and 2095 * passive_open_err() will either do a reset() or will free up the skb and
1861 * remove the cm_node. 2096 * remove the cm_node.
1862 */ 2097 */
1863
1864static void passive_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb, 2098static void passive_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
1865 int reset) 2099 int reset)
1866{ 2100{
1867 cleanup_retrans_entry(cm_node); 2101 cleanup_retrans_entry(cm_node);
1868 cm_node->state = NES_CM_STATE_CLOSED; 2102 cm_node->state = NES_CM_STATE_CLOSED;
1869 if (reset) { 2103 if (reset) {
1870 nes_debug(NES_DBG_CM, "passive_open_err sending RST for " 2104 nes_debug(NES_DBG_CM, "passive_open_err sending RST for "
1871 "cm_node=%p state =%d\n", cm_node, cm_node->state); 2105 "cm_node=%p state =%d\n", cm_node, cm_node->state);
1872 send_reset(cm_node, skb); 2106 send_reset(cm_node, skb);
1873 } else { 2107 } else {
1874 dev_kfree_skb_any(skb); 2108 dev_kfree_skb_any(skb);
@@ -1883,6 +2117,7 @@ static void passive_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
1883static void free_retrans_entry(struct nes_cm_node *cm_node) 2117static void free_retrans_entry(struct nes_cm_node *cm_node)
1884{ 2118{
1885 struct nes_timer_entry *send_entry; 2119 struct nes_timer_entry *send_entry;
2120
1886 send_entry = cm_node->send_entry; 2121 send_entry = cm_node->send_entry;
1887 if (send_entry) { 2122 if (send_entry) {
1888 cm_node->send_entry = NULL; 2123 cm_node->send_entry = NULL;
@@ -1906,26 +2141,28 @@ static void cleanup_retrans_entry(struct nes_cm_node *cm_node)
1906 * Returns skb if to be freed, else it will return NULL if already used.. 2141 * Returns skb if to be freed, else it will return NULL if already used..
1907 */ 2142 */
1908static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb, 2143static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb,
1909 struct nes_cm_core *cm_core) 2144 struct nes_cm_core *cm_core)
1910{ 2145{
1911 enum nes_tcpip_pkt_type pkt_type = NES_PKT_TYPE_UNKNOWN; 2146 enum nes_tcpip_pkt_type pkt_type = NES_PKT_TYPE_UNKNOWN;
1912 struct tcphdr *tcph = tcp_hdr(skb); 2147 struct tcphdr *tcph = tcp_hdr(skb);
1913 u32 fin_set = 0; 2148 u32 fin_set = 0;
1914 int ret = 0; 2149 int ret = 0;
2150
1915 skb_pull(skb, ip_hdr(skb)->ihl << 2); 2151 skb_pull(skb, ip_hdr(skb)->ihl << 2);
1916 2152
1917 nes_debug(NES_DBG_CM, "process_packet: cm_node=%p state =%d syn=%d " 2153 nes_debug(NES_DBG_CM, "process_packet: cm_node=%p state =%d syn=%d "
1918 "ack=%d rst=%d fin=%d\n", cm_node, cm_node->state, tcph->syn, 2154 "ack=%d rst=%d fin=%d\n", cm_node, cm_node->state, tcph->syn,
1919 tcph->ack, tcph->rst, tcph->fin); 2155 tcph->ack, tcph->rst, tcph->fin);
1920 2156
1921 if (tcph->rst) 2157 if (tcph->rst) {
1922 pkt_type = NES_PKT_TYPE_RST; 2158 pkt_type = NES_PKT_TYPE_RST;
1923 else if (tcph->syn) { 2159 } else if (tcph->syn) {
1924 pkt_type = NES_PKT_TYPE_SYN; 2160 pkt_type = NES_PKT_TYPE_SYN;
1925 if (tcph->ack) 2161 if (tcph->ack)
1926 pkt_type = NES_PKT_TYPE_SYNACK; 2162 pkt_type = NES_PKT_TYPE_SYNACK;
1927 } else if (tcph->ack) 2163 } else if (tcph->ack) {
1928 pkt_type = NES_PKT_TYPE_ACK; 2164 pkt_type = NES_PKT_TYPE_ACK;
2165 }
1929 if (tcph->fin) 2166 if (tcph->fin)
1930 fin_set = 1; 2167 fin_set = 1;
1931 2168
@@ -1956,17 +2193,17 @@ static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb,
1956 * mini_cm_listen - create a listen node with params 2193 * mini_cm_listen - create a listen node with params
1957 */ 2194 */
1958static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core, 2195static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
1959 struct nes_vnic *nesvnic, struct nes_cm_info *cm_info) 2196 struct nes_vnic *nesvnic, struct nes_cm_info *cm_info)
1960{ 2197{
1961 struct nes_cm_listener *listener; 2198 struct nes_cm_listener *listener;
1962 unsigned long flags; 2199 unsigned long flags;
1963 2200
1964 nes_debug(NES_DBG_CM, "Search for 0x%08x : 0x%04x\n", 2201 nes_debug(NES_DBG_CM, "Search for 0x%08x : 0x%04x\n",
1965 cm_info->loc_addr, cm_info->loc_port); 2202 cm_info->loc_addr, cm_info->loc_port);
1966 2203
1967 /* cannot have multiple matching listeners */ 2204 /* cannot have multiple matching listeners */
1968 listener = find_listener(cm_core, htonl(cm_info->loc_addr), 2205 listener = find_listener(cm_core, htonl(cm_info->loc_addr),
1969 htons(cm_info->loc_port), NES_CM_LISTENER_EITHER_STATE); 2206 htons(cm_info->loc_port), NES_CM_LISTENER_EITHER_STATE);
1970 if (listener && listener->listener_state == NES_CM_LISTENER_ACTIVE_STATE) { 2207 if (listener && listener->listener_state == NES_CM_LISTENER_ACTIVE_STATE) {
1971 /* find automatically incs ref count ??? */ 2208 /* find automatically incs ref count ??? */
1972 atomic_dec(&listener->ref_count); 2209 atomic_dec(&listener->ref_count);
@@ -2012,9 +2249,9 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
2012 } 2249 }
2013 2250
2014 nes_debug(NES_DBG_CM, "Api - listen(): addr=0x%08X, port=0x%04x," 2251 nes_debug(NES_DBG_CM, "Api - listen(): addr=0x%08X, port=0x%04x,"
2015 " listener = %p, backlog = %d, cm_id = %p.\n", 2252 " listener = %p, backlog = %d, cm_id = %p.\n",
2016 cm_info->loc_addr, cm_info->loc_port, 2253 cm_info->loc_addr, cm_info->loc_port,
2017 listener, listener->backlog, listener->cm_id); 2254 listener, listener->backlog, listener->cm_id);
2018 2255
2019 return listener; 2256 return listener;
2020} 2257}
@@ -2024,26 +2261,20 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
2024 * mini_cm_connect - make a connection node with params 2261 * mini_cm_connect - make a connection node with params
2025 */ 2262 */
2026static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core, 2263static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2027 struct nes_vnic *nesvnic, u16 private_data_len, 2264 struct nes_vnic *nesvnic, u16 private_data_len,
2028 void *private_data, struct nes_cm_info *cm_info) 2265 void *private_data, struct nes_cm_info *cm_info)
2029{ 2266{
2030 int ret = 0; 2267 int ret = 0;
2031 struct nes_cm_node *cm_node; 2268 struct nes_cm_node *cm_node;
2032 struct nes_cm_listener *loopbackremotelistener; 2269 struct nes_cm_listener *loopbackremotelistener;
2033 struct nes_cm_node *loopbackremotenode; 2270 struct nes_cm_node *loopbackremotenode;
2034 struct nes_cm_info loopback_cm_info; 2271 struct nes_cm_info loopback_cm_info;
2035 u16 mpa_frame_size = sizeof(struct ietf_mpa_frame) + private_data_len; 2272 u8 *start_buff;
2036 struct ietf_mpa_frame *mpa_frame = NULL;
2037 2273
2038 /* create a CM connection node */ 2274 /* create a CM connection node */
2039 cm_node = make_cm_node(cm_core, nesvnic, cm_info, NULL); 2275 cm_node = make_cm_node(cm_core, nesvnic, cm_info, NULL);
2040 if (!cm_node) 2276 if (!cm_node)
2041 return NULL; 2277 return NULL;
2042 mpa_frame = &cm_node->mpa_frame;
2043 memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
2044 mpa_frame->flags = IETF_MPA_FLAGS_CRC;
2045 mpa_frame->rev = IETF_MPA_VERSION;
2046 mpa_frame->priv_data_len = htons(private_data_len);
2047 2278
2048 /* set our node side to client (active) side */ 2279 /* set our node side to client (active) side */
2049 cm_node->tcp_cntxt.client = 1; 2280 cm_node->tcp_cntxt.client = 1;
@@ -2051,8 +2282,8 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2051 2282
2052 if (cm_info->loc_addr == cm_info->rem_addr) { 2283 if (cm_info->loc_addr == cm_info->rem_addr) {
2053 loopbackremotelistener = find_listener(cm_core, 2284 loopbackremotelistener = find_listener(cm_core,
2054 ntohl(nesvnic->local_ipaddr), cm_node->rem_port, 2285 ntohl(nesvnic->local_ipaddr), cm_node->rem_port,
2055 NES_CM_LISTENER_ACTIVE_STATE); 2286 NES_CM_LISTENER_ACTIVE_STATE);
2056 if (loopbackremotelistener == NULL) { 2287 if (loopbackremotelistener == NULL) {
2057 create_event(cm_node, NES_CM_EVENT_ABORTED); 2288 create_event(cm_node, NES_CM_EVENT_ABORTED);
2058 } else { 2289 } else {
@@ -2061,7 +2292,7 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2061 loopback_cm_info.rem_port = cm_info->loc_port; 2292 loopback_cm_info.rem_port = cm_info->loc_port;
2062 loopback_cm_info.cm_id = loopbackremotelistener->cm_id; 2293 loopback_cm_info.cm_id = loopbackremotelistener->cm_id;
2063 loopbackremotenode = make_cm_node(cm_core, nesvnic, 2294 loopbackremotenode = make_cm_node(cm_core, nesvnic,
2064 &loopback_cm_info, loopbackremotelistener); 2295 &loopback_cm_info, loopbackremotelistener);
2065 if (!loopbackremotenode) { 2296 if (!loopbackremotenode) {
2066 rem_ref_cm_node(cm_node->cm_core, cm_node); 2297 rem_ref_cm_node(cm_node->cm_core, cm_node);
2067 return NULL; 2298 return NULL;
@@ -2072,7 +2303,7 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2072 NES_CM_DEFAULT_RCV_WND_SCALE; 2303 NES_CM_DEFAULT_RCV_WND_SCALE;
2073 cm_node->loopbackpartner = loopbackremotenode; 2304 cm_node->loopbackpartner = loopbackremotenode;
2074 memcpy(loopbackremotenode->mpa_frame_buf, private_data, 2305 memcpy(loopbackremotenode->mpa_frame_buf, private_data,
2075 private_data_len); 2306 private_data_len);
2076 loopbackremotenode->mpa_frame_size = private_data_len; 2307 loopbackremotenode->mpa_frame_size = private_data_len;
2077 2308
2078 /* we are done handling this state. */ 2309 /* we are done handling this state. */
@@ -2100,12 +2331,10 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2100 return cm_node; 2331 return cm_node;
2101 } 2332 }
2102 2333
2103 /* set our node side to client (active) side */ 2334 start_buff = &cm_node->mpa_frame_buf[0] + sizeof(struct ietf_mpa_v2);
2104 cm_node->tcp_cntxt.client = 1; 2335 cm_node->mpa_frame_size = private_data_len;
2105 /* init our MPA frame ptr */
2106 memcpy(mpa_frame->priv_data, private_data, private_data_len);
2107 2336
2108 cm_node->mpa_frame_size = mpa_frame_size; 2337 memcpy(start_buff, private_data, private_data_len);
2109 2338
2110 /* send a syn and goto syn sent state */ 2339 /* send a syn and goto syn sent state */
2111 cm_node->state = NES_CM_STATE_SYN_SENT; 2340 cm_node->state = NES_CM_STATE_SYN_SENT;
@@ -2114,18 +2343,19 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2114 if (ret) { 2343 if (ret) {
2115 /* error in sending the syn free up the cm_node struct */ 2344 /* error in sending the syn free up the cm_node struct */
2116 nes_debug(NES_DBG_CM, "Api - connect() FAILED: dest " 2345 nes_debug(NES_DBG_CM, "Api - connect() FAILED: dest "
2117 "addr=0x%08X, port=0x%04x, cm_node=%p, cm_id = %p.\n", 2346 "addr=0x%08X, port=0x%04x, cm_node=%p, cm_id = %p.\n",
2118 cm_node->rem_addr, cm_node->rem_port, cm_node, 2347 cm_node->rem_addr, cm_node->rem_port, cm_node,
2119 cm_node->cm_id); 2348 cm_node->cm_id);
2120 rem_ref_cm_node(cm_node->cm_core, cm_node); 2349 rem_ref_cm_node(cm_node->cm_core, cm_node);
2121 cm_node = NULL; 2350 cm_node = NULL;
2122 } 2351 }
2123 2352
2124 if (cm_node) 2353 if (cm_node) {
2125 nes_debug(NES_DBG_CM, "Api - connect(): dest addr=0x%08X," 2354 nes_debug(NES_DBG_CM, "Api - connect(): dest addr=0x%08X,"
2126 "port=0x%04x, cm_node=%p, cm_id = %p.\n", 2355 "port=0x%04x, cm_node=%p, cm_id = %p.\n",
2127 cm_node->rem_addr, cm_node->rem_port, cm_node, 2356 cm_node->rem_addr, cm_node->rem_port, cm_node,
2128 cm_node->cm_id); 2357 cm_node->cm_id);
2358 }
2129 2359
2130 return cm_node; 2360 return cm_node;
2131} 2361}
@@ -2135,8 +2365,7 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
2135 * mini_cm_accept - accept a connection 2365 * mini_cm_accept - accept a connection
2136 * This function is never called 2366 * This function is never called
2137 */ 2367 */
2138static int mini_cm_accept(struct nes_cm_core *cm_core, 2368static int mini_cm_accept(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node)
2139 struct ietf_mpa_frame *mpa_frame, struct nes_cm_node *cm_node)
2140{ 2369{
2141 return 0; 2370 return 0;
2142} 2371}
@@ -2145,8 +2374,7 @@ static int mini_cm_accept(struct nes_cm_core *cm_core,
2145/** 2374/**
2146 * mini_cm_reject - reject and teardown a connection 2375 * mini_cm_reject - reject and teardown a connection
2147 */ 2376 */
2148static int mini_cm_reject(struct nes_cm_core *cm_core, 2377static int mini_cm_reject(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node)
2149 struct ietf_mpa_frame *mpa_frame, struct nes_cm_node *cm_node)
2150{ 2378{
2151 int ret = 0; 2379 int ret = 0;
2152 int err = 0; 2380 int err = 0;
@@ -2156,7 +2384,7 @@ static int mini_cm_reject(struct nes_cm_core *cm_core,
2156 struct nes_cm_node *loopback = cm_node->loopbackpartner; 2384 struct nes_cm_node *loopback = cm_node->loopbackpartner;
2157 2385
2158 nes_debug(NES_DBG_CM, "%s cm_node=%p type=%d state=%d\n", 2386 nes_debug(NES_DBG_CM, "%s cm_node=%p type=%d state=%d\n",
2159 __func__, cm_node, cm_node->tcp_cntxt.client, cm_node->state); 2387 __func__, cm_node, cm_node->tcp_cntxt.client, cm_node->state);
2160 2388
2161 if (cm_node->tcp_cntxt.client) 2389 if (cm_node->tcp_cntxt.client)
2162 return ret; 2390 return ret;
@@ -2177,8 +2405,9 @@ static int mini_cm_reject(struct nes_cm_core *cm_core,
2177 err = send_reset(cm_node, NULL); 2405 err = send_reset(cm_node, NULL);
2178 if (err) 2406 if (err)
2179 WARN_ON(1); 2407 WARN_ON(1);
2180 } else 2408 } else {
2181 cm_id->add_ref(cm_id); 2409 cm_id->add_ref(cm_id);
2410 }
2182 } 2411 }
2183 } 2412 }
2184 } else { 2413 } else {
@@ -2253,7 +2482,7 @@ static int mini_cm_close(struct nes_cm_core *cm_core, struct nes_cm_node *cm_nod
2253 case NES_CM_STATE_TSA: 2482 case NES_CM_STATE_TSA:
2254 if (cm_node->send_entry) 2483 if (cm_node->send_entry)
2255 printk(KERN_ERR "ERROR Close got called from STATE_TSA " 2484 printk(KERN_ERR "ERROR Close got called from STATE_TSA "
2256 "send_entry=%p\n", cm_node->send_entry); 2485 "send_entry=%p\n", cm_node->send_entry);
2257 ret = rem_ref_cm_node(cm_core, cm_node); 2486 ret = rem_ref_cm_node(cm_core, cm_node);
2258 break; 2487 break;
2259 } 2488 }
@@ -2266,7 +2495,7 @@ static int mini_cm_close(struct nes_cm_core *cm_core, struct nes_cm_node *cm_nod
2266 * node state machine 2495 * node state machine
2267 */ 2496 */
2268static int mini_cm_recv_pkt(struct nes_cm_core *cm_core, 2497static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
2269 struct nes_vnic *nesvnic, struct sk_buff *skb) 2498 struct nes_vnic *nesvnic, struct sk_buff *skb)
2270{ 2499{
2271 struct nes_cm_node *cm_node = NULL; 2500 struct nes_cm_node *cm_node = NULL;
2272 struct nes_cm_listener *listener = NULL; 2501 struct nes_cm_listener *listener = NULL;
@@ -2278,9 +2507,8 @@ static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
2278 2507
2279 if (!skb) 2508 if (!skb)
2280 return 0; 2509 return 0;
2281 if (skb->len < sizeof(struct iphdr) + sizeof(struct tcphdr)) { 2510 if (skb->len < sizeof(struct iphdr) + sizeof(struct tcphdr))
2282 return 0; 2511 return 0;
2283 }
2284 2512
2285 iph = (struct iphdr *)skb->data; 2513 iph = (struct iphdr *)skb->data;
2286 tcph = (struct tcphdr *)(skb->data + sizeof(struct iphdr)); 2514 tcph = (struct tcphdr *)(skb->data + sizeof(struct iphdr));
@@ -2298,8 +2526,8 @@ static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
2298 2526
2299 do { 2527 do {
2300 cm_node = find_node(cm_core, 2528 cm_node = find_node(cm_core,
2301 nfo.rem_port, nfo.rem_addr, 2529 nfo.rem_port, nfo.rem_addr,
2302 nfo.loc_port, nfo.loc_addr); 2530 nfo.loc_port, nfo.loc_addr);
2303 2531
2304 if (!cm_node) { 2532 if (!cm_node) {
2305 /* Only type of packet accepted are for */ 2533 /* Only type of packet accepted are for */
@@ -2309,8 +2537,8 @@ static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
2309 break; 2537 break;
2310 } 2538 }
2311 listener = find_listener(cm_core, nfo.loc_addr, 2539 listener = find_listener(cm_core, nfo.loc_addr,
2312 nfo.loc_port, 2540 nfo.loc_port,
2313 NES_CM_LISTENER_ACTIVE_STATE); 2541 NES_CM_LISTENER_ACTIVE_STATE);
2314 if (!listener) { 2542 if (!listener) {
2315 nfo.cm_id = NULL; 2543 nfo.cm_id = NULL;
2316 nfo.conn_type = 0; 2544 nfo.conn_type = 0;
@@ -2321,10 +2549,10 @@ static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
2321 nfo.cm_id = listener->cm_id; 2549 nfo.cm_id = listener->cm_id;
2322 nfo.conn_type = listener->conn_type; 2550 nfo.conn_type = listener->conn_type;
2323 cm_node = make_cm_node(cm_core, nesvnic, &nfo, 2551 cm_node = make_cm_node(cm_core, nesvnic, &nfo,
2324 listener); 2552 listener);
2325 if (!cm_node) { 2553 if (!cm_node) {
2326 nes_debug(NES_DBG_CM, "Unable to allocate " 2554 nes_debug(NES_DBG_CM, "Unable to allocate "
2327 "node\n"); 2555 "node\n");
2328 cm_packets_dropped++; 2556 cm_packets_dropped++;
2329 atomic_dec(&listener->ref_count); 2557 atomic_dec(&listener->ref_count);
2330 dev_kfree_skb_any(skb); 2558 dev_kfree_skb_any(skb);
@@ -2376,7 +2604,7 @@ static struct nes_cm_core *nes_cm_alloc_core(void)
2376 init_timer(&cm_core->tcp_timer); 2604 init_timer(&cm_core->tcp_timer);
2377 cm_core->tcp_timer.function = nes_cm_timer_tick; 2605 cm_core->tcp_timer.function = nes_cm_timer_tick;
2378 2606
2379 cm_core->mtu = NES_CM_DEFAULT_MTU; 2607 cm_core->mtu = NES_CM_DEFAULT_MTU;
2380 cm_core->state = NES_CM_STATE_INITED; 2608 cm_core->state = NES_CM_STATE_INITED;
2381 cm_core->free_tx_pkt_max = NES_CM_DEFAULT_FREE_PKTS; 2609 cm_core->free_tx_pkt_max = NES_CM_DEFAULT_FREE_PKTS;
2382 2610
@@ -2414,9 +2642,8 @@ static int mini_cm_dealloc_core(struct nes_cm_core *cm_core)
2414 2642
2415 barrier(); 2643 barrier();
2416 2644
2417 if (timer_pending(&cm_core->tcp_timer)) { 2645 if (timer_pending(&cm_core->tcp_timer))
2418 del_timer(&cm_core->tcp_timer); 2646 del_timer(&cm_core->tcp_timer);
2419 }
2420 2647
2421 destroy_workqueue(cm_core->event_wq); 2648 destroy_workqueue(cm_core->event_wq);
2422 destroy_workqueue(cm_core->disconn_wq); 2649 destroy_workqueue(cm_core->disconn_wq);
@@ -2471,8 +2698,8 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
2471 return -EINVAL; 2698 return -EINVAL;
2472 2699
2473 nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_IPV4 | 2700 nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_IPV4 |
2474 NES_QPCONTEXT_MISC_NO_NAGLE | NES_QPCONTEXT_MISC_DO_NOT_FRAG | 2701 NES_QPCONTEXT_MISC_NO_NAGLE | NES_QPCONTEXT_MISC_DO_NOT_FRAG |
2475 NES_QPCONTEXT_MISC_DROS); 2702 NES_QPCONTEXT_MISC_DROS);
2476 2703
2477 if (cm_node->tcp_cntxt.snd_wscale || cm_node->tcp_cntxt.rcv_wscale) 2704 if (cm_node->tcp_cntxt.snd_wscale || cm_node->tcp_cntxt.rcv_wscale)
2478 nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_WSCALE); 2705 nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_WSCALE);
@@ -2482,15 +2709,15 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
2482 nesqp->nesqp_context->mss |= cpu_to_le32(((u32)cm_node->tcp_cntxt.mss) << 16); 2709 nesqp->nesqp_context->mss |= cpu_to_le32(((u32)cm_node->tcp_cntxt.mss) << 16);
2483 2710
2484 nesqp->nesqp_context->tcp_state_flow_label |= cpu_to_le32( 2711 nesqp->nesqp_context->tcp_state_flow_label |= cpu_to_le32(
2485 (u32)NES_QPCONTEXT_TCPSTATE_EST << NES_QPCONTEXT_TCPFLOW_TCP_STATE_SHIFT); 2712 (u32)NES_QPCONTEXT_TCPSTATE_EST << NES_QPCONTEXT_TCPFLOW_TCP_STATE_SHIFT);
2486 2713
2487 nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32( 2714 nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32(
2488 (cm_node->tcp_cntxt.snd_wscale << NES_QPCONTEXT_PDWSCALE_SND_WSCALE_SHIFT) & 2715 (cm_node->tcp_cntxt.snd_wscale << NES_QPCONTEXT_PDWSCALE_SND_WSCALE_SHIFT) &
2489 NES_QPCONTEXT_PDWSCALE_SND_WSCALE_MASK); 2716 NES_QPCONTEXT_PDWSCALE_SND_WSCALE_MASK);
2490 2717
2491 nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32( 2718 nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32(
2492 (cm_node->tcp_cntxt.rcv_wscale << NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_SHIFT) & 2719 (cm_node->tcp_cntxt.rcv_wscale << NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_SHIFT) &
2493 NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_MASK); 2720 NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_MASK);
2494 2721
2495 nesqp->nesqp_context->keepalive = cpu_to_le32(0x80); 2722 nesqp->nesqp_context->keepalive = cpu_to_le32(0x80);
2496 nesqp->nesqp_context->ts_recent = 0; 2723 nesqp->nesqp_context->ts_recent = 0;
@@ -2499,24 +2726,24 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
2499 nesqp->nesqp_context->snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.snd_wnd); 2726 nesqp->nesqp_context->snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.snd_wnd);
2500 nesqp->nesqp_context->rcv_nxt = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt); 2727 nesqp->nesqp_context->rcv_nxt = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt);
2501 nesqp->nesqp_context->rcv_wnd = cpu_to_le32(cm_node->tcp_cntxt.rcv_wnd << 2728 nesqp->nesqp_context->rcv_wnd = cpu_to_le32(cm_node->tcp_cntxt.rcv_wnd <<
2502 cm_node->tcp_cntxt.rcv_wscale); 2729 cm_node->tcp_cntxt.rcv_wscale);
2503 nesqp->nesqp_context->snd_max = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num); 2730 nesqp->nesqp_context->snd_max = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
2504 nesqp->nesqp_context->snd_una = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num); 2731 nesqp->nesqp_context->snd_una = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
2505 nesqp->nesqp_context->srtt = 0; 2732 nesqp->nesqp_context->srtt = 0;
2506 nesqp->nesqp_context->rttvar = cpu_to_le32(0x6); 2733 nesqp->nesqp_context->rttvar = cpu_to_le32(0x6);
2507 nesqp->nesqp_context->ssthresh = cpu_to_le32(0x3FFFC000); 2734 nesqp->nesqp_context->ssthresh = cpu_to_le32(0x3FFFC000);
2508 nesqp->nesqp_context->cwnd = cpu_to_le32(2*cm_node->tcp_cntxt.mss); 2735 nesqp->nesqp_context->cwnd = cpu_to_le32(2 * cm_node->tcp_cntxt.mss);
2509 nesqp->nesqp_context->snd_wl1 = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt); 2736 nesqp->nesqp_context->snd_wl1 = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt);
2510 nesqp->nesqp_context->snd_wl2 = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num); 2737 nesqp->nesqp_context->snd_wl2 = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
2511 nesqp->nesqp_context->max_snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.max_snd_wnd); 2738 nesqp->nesqp_context->max_snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.max_snd_wnd);
2512 2739
2513 nes_debug(NES_DBG_CM, "QP%u: rcv_nxt = 0x%08X, snd_nxt = 0x%08X," 2740 nes_debug(NES_DBG_CM, "QP%u: rcv_nxt = 0x%08X, snd_nxt = 0x%08X,"
2514 " Setting MSS to %u, PDWscale = 0x%08X, rcv_wnd = %u, context misc = 0x%08X.\n", 2741 " Setting MSS to %u, PDWscale = 0x%08X, rcv_wnd = %u, context misc = 0x%08X.\n",
2515 nesqp->hwqp.qp_id, le32_to_cpu(nesqp->nesqp_context->rcv_nxt), 2742 nesqp->hwqp.qp_id, le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
2516 le32_to_cpu(nesqp->nesqp_context->snd_nxt), 2743 le32_to_cpu(nesqp->nesqp_context->snd_nxt),
2517 cm_node->tcp_cntxt.mss, le32_to_cpu(nesqp->nesqp_context->pd_index_wscale), 2744 cm_node->tcp_cntxt.mss, le32_to_cpu(nesqp->nesqp_context->pd_index_wscale),
2518 le32_to_cpu(nesqp->nesqp_context->rcv_wnd), 2745 le32_to_cpu(nesqp->nesqp_context->rcv_wnd),
2519 le32_to_cpu(nesqp->nesqp_context->misc)); 2746 le32_to_cpu(nesqp->nesqp_context->misc));
2520 nes_debug(NES_DBG_CM, " snd_wnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->snd_wnd)); 2747 nes_debug(NES_DBG_CM, " snd_wnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->snd_wnd));
2521 nes_debug(NES_DBG_CM, " snd_cwnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->cwnd)); 2748 nes_debug(NES_DBG_CM, " snd_cwnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->cwnd));
2522 nes_debug(NES_DBG_CM, " max_swnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->max_snd_wnd)); 2749 nes_debug(NES_DBG_CM, " max_swnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->max_snd_wnd));
@@ -2537,7 +2764,7 @@ int nes_cm_disconn(struct nes_qp *nesqp)
2537 2764
2538 work = kzalloc(sizeof *work, GFP_ATOMIC); 2765 work = kzalloc(sizeof *work, GFP_ATOMIC);
2539 if (!work) 2766 if (!work)
2540 return -ENOMEM; /* Timer will clean up */ 2767 return -ENOMEM; /* Timer will clean up */
2541 2768
2542 nes_add_ref(&nesqp->ibqp); 2769 nes_add_ref(&nesqp->ibqp);
2543 work->nesqp = nesqp; 2770 work->nesqp = nesqp;
@@ -2557,7 +2784,7 @@ static void nes_disconnect_worker(struct work_struct *work)
2557 2784
2558 kfree(dwork); 2785 kfree(dwork);
2559 nes_debug(NES_DBG_CM, "processing AEQE id 0x%04X for QP%u.\n", 2786 nes_debug(NES_DBG_CM, "processing AEQE id 0x%04X for QP%u.\n",
2560 nesqp->last_aeq, nesqp->hwqp.qp_id); 2787 nesqp->last_aeq, nesqp->hwqp.qp_id);
2561 nes_cm_disconn_true(nesqp); 2788 nes_cm_disconn_true(nesqp);
2562 nes_rem_ref(&nesqp->ibqp); 2789 nes_rem_ref(&nesqp->ibqp);
2563} 2790}
@@ -2593,7 +2820,7 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
2593 /* make sure we havent already closed this connection */ 2820 /* make sure we havent already closed this connection */
2594 if (!cm_id) { 2821 if (!cm_id) {
2595 nes_debug(NES_DBG_CM, "QP%u disconnect_worker cmid is NULL\n", 2822 nes_debug(NES_DBG_CM, "QP%u disconnect_worker cmid is NULL\n",
2596 nesqp->hwqp.qp_id); 2823 nesqp->hwqp.qp_id);
2597 spin_unlock_irqrestore(&nesqp->lock, flags); 2824 spin_unlock_irqrestore(&nesqp->lock, flags);
2598 return -1; 2825 return -1;
2599 } 2826 }
@@ -2602,7 +2829,7 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
2602 nes_debug(NES_DBG_CM, "Disconnecting QP%u\n", nesqp->hwqp.qp_id); 2829 nes_debug(NES_DBG_CM, "Disconnecting QP%u\n", nesqp->hwqp.qp_id);
2603 2830
2604 original_hw_tcp_state = nesqp->hw_tcp_state; 2831 original_hw_tcp_state = nesqp->hw_tcp_state;
2605 original_ibqp_state = nesqp->ibqp_state; 2832 original_ibqp_state = nesqp->ibqp_state;
2606 last_ae = nesqp->last_aeq; 2833 last_ae = nesqp->last_aeq;
2607 2834
2608 if (nesqp->term_flags) { 2835 if (nesqp->term_flags) {
@@ -2660,16 +2887,16 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
2660 cm_event.private_data_len = 0; 2887 cm_event.private_data_len = 0;
2661 2888
2662 nes_debug(NES_DBG_CM, "Generating a CM Disconnect Event" 2889 nes_debug(NES_DBG_CM, "Generating a CM Disconnect Event"
2663 " for QP%u, SQ Head = %u, SQ Tail = %u. " 2890 " for QP%u, SQ Head = %u, SQ Tail = %u. "
2664 "cm_id = %p, refcount = %u.\n", 2891 "cm_id = %p, refcount = %u.\n",
2665 nesqp->hwqp.qp_id, nesqp->hwqp.sq_head, 2892 nesqp->hwqp.qp_id, nesqp->hwqp.sq_head,
2666 nesqp->hwqp.sq_tail, cm_id, 2893 nesqp->hwqp.sq_tail, cm_id,
2667 atomic_read(&nesqp->refcount)); 2894 atomic_read(&nesqp->refcount));
2668 2895
2669 ret = cm_id->event_handler(cm_id, &cm_event); 2896 ret = cm_id->event_handler(cm_id, &cm_event);
2670 if (ret) 2897 if (ret)
2671 nes_debug(NES_DBG_CM, "OFA CM event_handler " 2898 nes_debug(NES_DBG_CM, "OFA CM event_handler "
2672 "returned, ret=%d\n", ret); 2899 "returned, ret=%d\n", ret);
2673 } 2900 }
2674 2901
2675 if (issue_close) { 2902 if (issue_close) {
@@ -2687,9 +2914,8 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
2687 cm_event.private_data_len = 0; 2914 cm_event.private_data_len = 0;
2688 2915
2689 ret = cm_id->event_handler(cm_id, &cm_event); 2916 ret = cm_id->event_handler(cm_id, &cm_event);
2690 if (ret) { 2917 if (ret)
2691 nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret); 2918 nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
2692 }
2693 2919
2694 cm_id->rem_ref(cm_id); 2920 cm_id->rem_ref(cm_id);
2695 } 2921 }
@@ -2729,8 +2955,8 @@ static int nes_disconnect(struct nes_qp *nesqp, int abrupt)
2729 if (nesqp->lsmm_mr) 2955 if (nesqp->lsmm_mr)
2730 nesibdev->ibdev.dereg_mr(nesqp->lsmm_mr); 2956 nesibdev->ibdev.dereg_mr(nesqp->lsmm_mr);
2731 pci_free_consistent(nesdev->pcidev, 2957 pci_free_consistent(nesdev->pcidev,
2732 nesqp->private_data_len+sizeof(struct ietf_mpa_frame), 2958 nesqp->private_data_len + nesqp->ietf_frame_size,
2733 nesqp->ietf_frame, nesqp->ietf_frame_pbase); 2959 nesqp->ietf_frame, nesqp->ietf_frame_pbase);
2734 } 2960 }
2735 } 2961 }
2736 2962
@@ -2769,6 +2995,12 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2769 struct ib_phys_buf ibphysbuf; 2995 struct ib_phys_buf ibphysbuf;
2770 struct nes_pd *nespd; 2996 struct nes_pd *nespd;
2771 u64 tagged_offset; 2997 u64 tagged_offset;
2998 u8 mpa_frame_offset = 0;
2999 struct ietf_mpa_v2 *mpa_v2_frame;
3000 u8 start_addr = 0;
3001 u8 *start_ptr = &start_addr;
3002 u8 **start_buff = &start_ptr;
3003 u16 buff_len = 0;
2772 3004
2773 ibqp = nes_get_qp(cm_id->device, conn_param->qpn); 3005 ibqp = nes_get_qp(cm_id->device, conn_param->qpn);
2774 if (!ibqp) 3006 if (!ibqp)
@@ -2809,53 +3041,49 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2809 nes_debug(NES_DBG_CM, "netdev refcnt = %u.\n", 3041 nes_debug(NES_DBG_CM, "netdev refcnt = %u.\n",
2810 netdev_refcnt_read(nesvnic->netdev)); 3042 netdev_refcnt_read(nesvnic->netdev));
2811 3043
3044 nesqp->ietf_frame_size = sizeof(struct ietf_mpa_v2);
2812 /* allocate the ietf frame and space for private data */ 3045 /* allocate the ietf frame and space for private data */
2813 nesqp->ietf_frame = pci_alloc_consistent(nesdev->pcidev, 3046 nesqp->ietf_frame = pci_alloc_consistent(nesdev->pcidev,
2814 sizeof(struct ietf_mpa_frame) + conn_param->private_data_len, 3047 nesqp->ietf_frame_size + conn_param->private_data_len,
2815 &nesqp->ietf_frame_pbase); 3048 &nesqp->ietf_frame_pbase);
2816 3049
2817 if (!nesqp->ietf_frame) { 3050 if (!nesqp->ietf_frame) {
2818 nes_debug(NES_DBG_CM, "Unable to allocate memory for private " 3051 nes_debug(NES_DBG_CM, "Unable to allocate memory for private data\n");
2819 "data\n");
2820 return -ENOMEM; 3052 return -ENOMEM;
2821 } 3053 }
3054 mpa_v2_frame = (struct ietf_mpa_v2 *)nesqp->ietf_frame;
2822 3055
3056 if (cm_node->mpa_frame_rev == IETF_MPA_V1)
3057 mpa_frame_offset = 4;
2823 3058
2824 /* setup the MPA frame */ 3059 memcpy(mpa_v2_frame->priv_data, conn_param->private_data,
2825 nesqp->private_data_len = conn_param->private_data_len; 3060 conn_param->private_data_len);
2826 memcpy(nesqp->ietf_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
2827
2828 memcpy(nesqp->ietf_frame->priv_data, conn_param->private_data,
2829 conn_param->private_data_len);
2830 3061
2831 nesqp->ietf_frame->priv_data_len = 3062 cm_build_mpa_frame(cm_node, start_buff, &buff_len, nesqp->ietf_frame, MPA_KEY_REPLY);
2832 cpu_to_be16(conn_param->private_data_len); 3063 nesqp->private_data_len = conn_param->private_data_len;
2833 nesqp->ietf_frame->rev = mpa_version;
2834 nesqp->ietf_frame->flags = IETF_MPA_FLAGS_CRC;
2835 3064
2836 /* setup our first outgoing iWarp send WQE (the IETF frame response) */ 3065 /* setup our first outgoing iWarp send WQE (the IETF frame response) */
2837 wqe = &nesqp->hwqp.sq_vbase[0]; 3066 wqe = &nesqp->hwqp.sq_vbase[0];
2838 3067
2839 if (cm_id->remote_addr.sin_addr.s_addr != 3068 if (cm_id->remote_addr.sin_addr.s_addr !=
2840 cm_id->local_addr.sin_addr.s_addr) { 3069 cm_id->local_addr.sin_addr.s_addr) {
2841 u64temp = (unsigned long)nesqp; 3070 u64temp = (unsigned long)nesqp;
2842 nesibdev = nesvnic->nesibdev; 3071 nesibdev = nesvnic->nesibdev;
2843 nespd = nesqp->nespd; 3072 nespd = nesqp->nespd;
2844 ibphysbuf.addr = nesqp->ietf_frame_pbase; 3073 ibphysbuf.addr = nesqp->ietf_frame_pbase + mpa_frame_offset;
2845 ibphysbuf.size = conn_param->private_data_len + 3074 ibphysbuf.size = buff_len;
2846 sizeof(struct ietf_mpa_frame); 3075 tagged_offset = (u64)(unsigned long)*start_buff;
2847 tagged_offset = (u64)(unsigned long)nesqp->ietf_frame;
2848 ibmr = nesibdev->ibdev.reg_phys_mr((struct ib_pd *)nespd, 3076 ibmr = nesibdev->ibdev.reg_phys_mr((struct ib_pd *)nespd,
2849 &ibphysbuf, 1, 3077 &ibphysbuf, 1,
2850 IB_ACCESS_LOCAL_WRITE, 3078 IB_ACCESS_LOCAL_WRITE,
2851 &tagged_offset); 3079 &tagged_offset);
2852 if (!ibmr) { 3080 if (!ibmr) {
2853 nes_debug(NES_DBG_CM, "Unable to register memory region" 3081 nes_debug(NES_DBG_CM, "Unable to register memory region"
2854 "for lSMM for cm_node = %p \n", 3082 "for lSMM for cm_node = %p \n",
2855 cm_node); 3083 cm_node);
2856 pci_free_consistent(nesdev->pcidev, 3084 pci_free_consistent(nesdev->pcidev,
2857 nesqp->private_data_len+sizeof(struct ietf_mpa_frame), 3085 nesqp->private_data_len + nesqp->ietf_frame_size,
2858 nesqp->ietf_frame, nesqp->ietf_frame_pbase); 3086 nesqp->ietf_frame, nesqp->ietf_frame_pbase);
2859 return -ENOMEM; 3087 return -ENOMEM;
2860 } 3088 }
2861 3089
@@ -2863,22 +3091,20 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2863 ibmr->device = nespd->ibpd.device; 3091 ibmr->device = nespd->ibpd.device;
2864 nesqp->lsmm_mr = ibmr; 3092 nesqp->lsmm_mr = ibmr;
2865 3093
2866 u64temp |= NES_SW_CONTEXT_ALIGN>>1; 3094 u64temp |= NES_SW_CONTEXT_ALIGN >> 1;
2867 set_wqe_64bit_value(wqe->wqe_words, 3095 set_wqe_64bit_value(wqe->wqe_words,
2868 NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX, 3096 NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX,
2869 u64temp); 3097 u64temp);
2870 wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] = 3098 wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
2871 cpu_to_le32(NES_IWARP_SQ_WQE_STREAMING | 3099 cpu_to_le32(NES_IWARP_SQ_WQE_STREAMING |
2872 NES_IWARP_SQ_WQE_WRPDU); 3100 NES_IWARP_SQ_WQE_WRPDU);
2873 wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] = 3101 wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] =
2874 cpu_to_le32(conn_param->private_data_len + 3102 cpu_to_le32(buff_len);
2875 sizeof(struct ietf_mpa_frame));
2876 set_wqe_64bit_value(wqe->wqe_words, 3103 set_wqe_64bit_value(wqe->wqe_words,
2877 NES_IWARP_SQ_WQE_FRAG0_LOW_IDX, 3104 NES_IWARP_SQ_WQE_FRAG0_LOW_IDX,
2878 (u64)(unsigned long)nesqp->ietf_frame); 3105 (u64)(unsigned long)(*start_buff));
2879 wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] = 3106 wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] =
2880 cpu_to_le32(conn_param->private_data_len + 3107 cpu_to_le32(buff_len);
2881 sizeof(struct ietf_mpa_frame));
2882 wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = ibmr->lkey; 3108 wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = ibmr->lkey;
2883 if (nesqp->sq_kmapped) { 3109 if (nesqp->sq_kmapped) {
2884 nesqp->sq_kmapped = 0; 3110 nesqp->sq_kmapped = 0;
@@ -2887,7 +3113,7 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2887 3113
2888 nesqp->nesqp_context->ird_ord_sizes |= 3114 nesqp->nesqp_context->ird_ord_sizes |=
2889 cpu_to_le32(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT | 3115 cpu_to_le32(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
2890 NES_QPCONTEXT_ORDIRD_WRPDU); 3116 NES_QPCONTEXT_ORDIRD_WRPDU);
2891 } else { 3117 } else {
2892 nesqp->nesqp_context->ird_ord_sizes |= 3118 nesqp->nesqp_context->ird_ord_sizes |=
2893 cpu_to_le32(NES_QPCONTEXT_ORDIRD_WRPDU); 3119 cpu_to_le32(NES_QPCONTEXT_ORDIRD_WRPDU);
@@ -2901,11 +3127,11 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2901 3127
2902 /* nesqp->cm_node = (void *)cm_id->provider_data; */ 3128 /* nesqp->cm_node = (void *)cm_id->provider_data; */
2903 cm_id->provider_data = nesqp; 3129 cm_id->provider_data = nesqp;
2904 nesqp->active_conn = 0; 3130 nesqp->active_conn = 0;
2905 3131
2906 if (cm_node->state == NES_CM_STATE_TSA) 3132 if (cm_node->state == NES_CM_STATE_TSA)
2907 nes_debug(NES_DBG_CM, "Already state = TSA for cm_node=%p\n", 3133 nes_debug(NES_DBG_CM, "Already state = TSA for cm_node=%p\n",
2908 cm_node); 3134 cm_node);
2909 3135
2910 nes_cm_init_tsa_conn(nesqp, cm_node); 3136 nes_cm_init_tsa_conn(nesqp, cm_node);
2911 3137
@@ -2922,13 +3148,13 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2922 cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr)); 3148 cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr));
2923 3149
2924 nesqp->nesqp_context->misc2 |= cpu_to_le32( 3150 nesqp->nesqp_context->misc2 |= cpu_to_le32(
2925 (u32)PCI_FUNC(nesdev->pcidev->devfn) << 3151 (u32)PCI_FUNC(nesdev->pcidev->devfn) <<
2926 NES_QPCONTEXT_MISC2_SRC_IP_SHIFT); 3152 NES_QPCONTEXT_MISC2_SRC_IP_SHIFT);
2927 3153
2928 nesqp->nesqp_context->arp_index_vlan |= 3154 nesqp->nesqp_context->arp_index_vlan |=
2929 cpu_to_le32(nes_arp_table(nesdev, 3155 cpu_to_le32(nes_arp_table(nesdev,
2930 le32_to_cpu(nesqp->nesqp_context->ip0), NULL, 3156 le32_to_cpu(nesqp->nesqp_context->ip0), NULL,
2931 NES_ARP_RESOLVE) << 16); 3157 NES_ARP_RESOLVE) << 16);
2932 3158
2933 nesqp->nesqp_context->ts_val_delta = cpu_to_le32( 3159 nesqp->nesqp_context->ts_val_delta = cpu_to_le32(
2934 jiffies - nes_read_indexed(nesdev, NES_IDX_TCP_NOW)); 3160 jiffies - nes_read_indexed(nesdev, NES_IDX_TCP_NOW));
@@ -2954,7 +3180,7 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2954 crc_value = get_crc_value(&nes_quad); 3180 crc_value = get_crc_value(&nes_quad);
2955 nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff); 3181 nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
2956 nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, CRC = 0x%08X\n", 3182 nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, CRC = 0x%08X\n",
2957 nesqp->hte_index, nesqp->hte_index & adapter->hte_index_mask); 3183 nesqp->hte_index, nesqp->hte_index & adapter->hte_index_mask);
2958 3184
2959 nesqp->hte_index &= adapter->hte_index_mask; 3185 nesqp->hte_index &= adapter->hte_index_mask;
2960 nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index); 3186 nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
@@ -2962,17 +3188,15 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2962 cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node); 3188 cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node);
2963 3189
2964 nes_debug(NES_DBG_CM, "QP%u, Destination IP = 0x%08X:0x%04X, local = " 3190 nes_debug(NES_DBG_CM, "QP%u, Destination IP = 0x%08X:0x%04X, local = "
2965 "0x%08X:0x%04X, rcv_nxt=0x%08X, snd_nxt=0x%08X, mpa + " 3191 "0x%08X:0x%04X, rcv_nxt=0x%08X, snd_nxt=0x%08X, mpa + "
2966 "private data length=%zu.\n", nesqp->hwqp.qp_id, 3192 "private data length=%u.\n", nesqp->hwqp.qp_id,
2967 ntohl(cm_id->remote_addr.sin_addr.s_addr), 3193 ntohl(cm_id->remote_addr.sin_addr.s_addr),
2968 ntohs(cm_id->remote_addr.sin_port), 3194 ntohs(cm_id->remote_addr.sin_port),
2969 ntohl(cm_id->local_addr.sin_addr.s_addr), 3195 ntohl(cm_id->local_addr.sin_addr.s_addr),
2970 ntohs(cm_id->local_addr.sin_port), 3196 ntohs(cm_id->local_addr.sin_port),
2971 le32_to_cpu(nesqp->nesqp_context->rcv_nxt), 3197 le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
2972 le32_to_cpu(nesqp->nesqp_context->snd_nxt), 3198 le32_to_cpu(nesqp->nesqp_context->snd_nxt),
2973 conn_param->private_data_len + 3199 buff_len);
2974 sizeof(struct ietf_mpa_frame));
2975
2976 3200
2977 /* notify OF layer that accept event was successful */ 3201 /* notify OF layer that accept event was successful */
2978 cm_id->add_ref(cm_id); 3202 cm_id->add_ref(cm_id);
@@ -2993,12 +3217,12 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2993 nesqp->private_data_len; 3217 nesqp->private_data_len;
2994 /* copy entire MPA frame to our cm_node's frame */ 3218 /* copy entire MPA frame to our cm_node's frame */
2995 memcpy(cm_node->loopbackpartner->mpa_frame_buf, 3219 memcpy(cm_node->loopbackpartner->mpa_frame_buf,
2996 nesqp->ietf_frame->priv_data, nesqp->private_data_len); 3220 conn_param->private_data, conn_param->private_data_len);
2997 create_event(cm_node->loopbackpartner, NES_CM_EVENT_CONNECTED); 3221 create_event(cm_node->loopbackpartner, NES_CM_EVENT_CONNECTED);
2998 } 3222 }
2999 if (ret) 3223 if (ret)
3000 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, " 3224 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
3001 "ret=%d\n", __func__, __LINE__, ret); 3225 "ret=%d\n", __func__, __LINE__, ret);
3002 3226
3003 return 0; 3227 return 0;
3004} 3228}
@@ -3011,34 +3235,28 @@ int nes_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3011{ 3235{
3012 struct nes_cm_node *cm_node; 3236 struct nes_cm_node *cm_node;
3013 struct nes_cm_node *loopback; 3237 struct nes_cm_node *loopback;
3014
3015 struct nes_cm_core *cm_core; 3238 struct nes_cm_core *cm_core;
3239 u8 *start_buff;
3016 3240
3017 atomic_inc(&cm_rejects); 3241 atomic_inc(&cm_rejects);
3018 cm_node = (struct nes_cm_node *) cm_id->provider_data; 3242 cm_node = (struct nes_cm_node *)cm_id->provider_data;
3019 loopback = cm_node->loopbackpartner; 3243 loopback = cm_node->loopbackpartner;
3020 cm_core = cm_node->cm_core; 3244 cm_core = cm_node->cm_core;
3021 cm_node->cm_id = cm_id; 3245 cm_node->cm_id = cm_id;
3022 cm_node->mpa_frame_size = sizeof(struct ietf_mpa_frame) + pdata_len;
3023 3246
3024 if (cm_node->mpa_frame_size > MAX_CM_BUFFER) 3247 if (pdata_len + sizeof(struct ietf_mpa_v2) > MAX_CM_BUFFER)
3025 return -EINVAL; 3248 return -EINVAL;
3026 3249
3027 memcpy(&cm_node->mpa_frame.key[0], IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
3028 if (loopback) { 3250 if (loopback) {
3029 memcpy(&loopback->mpa_frame.priv_data, pdata, pdata_len); 3251 memcpy(&loopback->mpa_frame.priv_data, pdata, pdata_len);
3030 loopback->mpa_frame.priv_data_len = pdata_len; 3252 loopback->mpa_frame.priv_data_len = pdata_len;
3031 loopback->mpa_frame_size = sizeof(struct ietf_mpa_frame) + 3253 loopback->mpa_frame_size = pdata_len;
3032 pdata_len;
3033 } else { 3254 } else {
3034 memcpy(&cm_node->mpa_frame.priv_data, pdata, pdata_len); 3255 start_buff = &cm_node->mpa_frame_buf[0] + sizeof(struct ietf_mpa_v2);
3035 cm_node->mpa_frame.priv_data_len = cpu_to_be16(pdata_len); 3256 cm_node->mpa_frame_size = pdata_len;
3257 memcpy(start_buff, pdata, pdata_len);
3036 } 3258 }
3037 3259 return cm_core->api->reject(cm_core, cm_node);
3038 cm_node->mpa_frame.rev = mpa_version;
3039 cm_node->mpa_frame.flags = IETF_MPA_FLAGS_CRC | IETF_MPA_FLAGS_REJECT;
3040
3041 return cm_core->api->reject(cm_core, &cm_node->mpa_frame, cm_node);
3042} 3260}
3043 3261
3044 3262
@@ -3065,7 +3283,7 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3065 nesvnic = to_nesvnic(nesqp->ibqp.device); 3283 nesvnic = to_nesvnic(nesqp->ibqp.device);
3066 if (!nesvnic) 3284 if (!nesvnic)
3067 return -EINVAL; 3285 return -EINVAL;
3068 nesdev = nesvnic->nesdev; 3286 nesdev = nesvnic->nesdev;
3069 if (!nesdev) 3287 if (!nesdev)
3070 return -EINVAL; 3288 return -EINVAL;
3071 3289
@@ -3073,12 +3291,12 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3073 return -EINVAL; 3291 return -EINVAL;
3074 3292
3075 nes_debug(NES_DBG_CM, "QP%u, current IP = 0x%08X, Destination IP = " 3293 nes_debug(NES_DBG_CM, "QP%u, current IP = 0x%08X, Destination IP = "
3076 "0x%08X:0x%04X, local = 0x%08X:0x%04X.\n", nesqp->hwqp.qp_id, 3294 "0x%08X:0x%04X, local = 0x%08X:0x%04X.\n", nesqp->hwqp.qp_id,
3077 ntohl(nesvnic->local_ipaddr), 3295 ntohl(nesvnic->local_ipaddr),
3078 ntohl(cm_id->remote_addr.sin_addr.s_addr), 3296 ntohl(cm_id->remote_addr.sin_addr.s_addr),
3079 ntohs(cm_id->remote_addr.sin_port), 3297 ntohs(cm_id->remote_addr.sin_port),
3080 ntohl(cm_id->local_addr.sin_addr.s_addr), 3298 ntohl(cm_id->local_addr.sin_addr.s_addr),
3081 ntohs(cm_id->local_addr.sin_port)); 3299 ntohs(cm_id->local_addr.sin_port));
3082 3300
3083 atomic_inc(&cm_connects); 3301 atomic_inc(&cm_connects);
3084 nesqp->active_conn = 1; 3302 nesqp->active_conn = 1;
@@ -3092,12 +3310,12 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3092 nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32((u32)conn_param->ord); 3310 nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32((u32)conn_param->ord);
3093 nes_debug(NES_DBG_CM, "requested ord = 0x%08X.\n", (u32)conn_param->ord); 3311 nes_debug(NES_DBG_CM, "requested ord = 0x%08X.\n", (u32)conn_param->ord);
3094 nes_debug(NES_DBG_CM, "mpa private data len =%u\n", 3312 nes_debug(NES_DBG_CM, "mpa private data len =%u\n",
3095 conn_param->private_data_len); 3313 conn_param->private_data_len);
3096 3314
3097 if (cm_id->local_addr.sin_addr.s_addr != 3315 if (cm_id->local_addr.sin_addr.s_addr !=
3098 cm_id->remote_addr.sin_addr.s_addr) { 3316 cm_id->remote_addr.sin_addr.s_addr) {
3099 nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port), 3317 nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port),
3100 PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD); 3318 PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD);
3101 apbvt_set = 1; 3319 apbvt_set = 1;
3102 } 3320 }
3103 3321
@@ -3113,13 +3331,13 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3113 3331
3114 /* create a connect CM node connection */ 3332 /* create a connect CM node connection */
3115 cm_node = g_cm_core->api->connect(g_cm_core, nesvnic, 3333 cm_node = g_cm_core->api->connect(g_cm_core, nesvnic,
3116 conn_param->private_data_len, (void *)conn_param->private_data, 3334 conn_param->private_data_len, (void *)conn_param->private_data,
3117 &cm_info); 3335 &cm_info);
3118 if (!cm_node) { 3336 if (!cm_node) {
3119 if (apbvt_set) 3337 if (apbvt_set)
3120 nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port), 3338 nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port),
3121 PCI_FUNC(nesdev->pcidev->devfn), 3339 PCI_FUNC(nesdev->pcidev->devfn),
3122 NES_MANAGE_APBVT_DEL); 3340 NES_MANAGE_APBVT_DEL);
3123 3341
3124 cm_id->rem_ref(cm_id); 3342 cm_id->rem_ref(cm_id);
3125 return -ENOMEM; 3343 return -ENOMEM;
@@ -3169,7 +3387,7 @@ int nes_create_listen(struct iw_cm_id *cm_id, int backlog)
3169 cm_node = g_cm_core->api->listen(g_cm_core, nesvnic, &cm_info); 3387 cm_node = g_cm_core->api->listen(g_cm_core, nesvnic, &cm_info);
3170 if (!cm_node) { 3388 if (!cm_node) {
3171 printk(KERN_ERR "%s[%u] Error returned from listen API call\n", 3389 printk(KERN_ERR "%s[%u] Error returned from listen API call\n",
3172 __func__, __LINE__); 3390 __func__, __LINE__);
3173 return -ENOMEM; 3391 return -ENOMEM;
3174 } 3392 }
3175 3393
@@ -3177,12 +3395,12 @@ int nes_create_listen(struct iw_cm_id *cm_id, int backlog)
3177 3395
3178 if (!cm_node->reused_node) { 3396 if (!cm_node->reused_node) {
3179 err = nes_manage_apbvt(nesvnic, 3397 err = nes_manage_apbvt(nesvnic,
3180 ntohs(cm_id->local_addr.sin_port), 3398 ntohs(cm_id->local_addr.sin_port),
3181 PCI_FUNC(nesvnic->nesdev->pcidev->devfn), 3399 PCI_FUNC(nesvnic->nesdev->pcidev->devfn),
3182 NES_MANAGE_APBVT_ADD); 3400 NES_MANAGE_APBVT_ADD);
3183 if (err) { 3401 if (err) {
3184 printk(KERN_ERR "nes_manage_apbvt call returned %d.\n", 3402 printk(KERN_ERR "nes_manage_apbvt call returned %d.\n",
3185 err); 3403 err);
3186 g_cm_core->api->stop_listener(g_cm_core, (void *)cm_node); 3404 g_cm_core->api->stop_listener(g_cm_core, (void *)cm_node);
3187 return err; 3405 return err;
3188 } 3406 }
@@ -3219,13 +3437,13 @@ int nes_destroy_listen(struct iw_cm_id *cm_id)
3219int nes_cm_recv(struct sk_buff *skb, struct net_device *netdevice) 3437int nes_cm_recv(struct sk_buff *skb, struct net_device *netdevice)
3220{ 3438{
3221 int rc = 0; 3439 int rc = 0;
3440
3222 cm_packets_received++; 3441 cm_packets_received++;
3223 if ((g_cm_core) && (g_cm_core->api)) { 3442 if ((g_cm_core) && (g_cm_core->api))
3224 rc = g_cm_core->api->recv_pkt(g_cm_core, netdev_priv(netdevice), skb); 3443 rc = g_cm_core->api->recv_pkt(g_cm_core, netdev_priv(netdevice), skb);
3225 } else { 3444 else
3226 nes_debug(NES_DBG_CM, "Unable to process packet for CM," 3445 nes_debug(NES_DBG_CM, "Unable to process packet for CM,"
3227 " cm is not setup properly.\n"); 3446 " cm is not setup properly.\n");
3228 }
3229 3447
3230 return rc; 3448 return rc;
3231} 3449}
@@ -3240,11 +3458,10 @@ int nes_cm_start(void)
3240 nes_debug(NES_DBG_CM, "\n"); 3458 nes_debug(NES_DBG_CM, "\n");
3241 /* create the primary CM core, pass this handle to subsequent core inits */ 3459 /* create the primary CM core, pass this handle to subsequent core inits */
3242 g_cm_core = nes_cm_alloc_core(); 3460 g_cm_core = nes_cm_alloc_core();
3243 if (g_cm_core) { 3461 if (g_cm_core)
3244 return 0; 3462 return 0;
3245 } else { 3463 else
3246 return -ENOMEM; 3464 return -ENOMEM;
3247 }
3248} 3465}
3249 3466
3250 3467
@@ -3265,7 +3482,6 @@ int nes_cm_stop(void)
3265 */ 3482 */
3266static void cm_event_connected(struct nes_cm_event *event) 3483static void cm_event_connected(struct nes_cm_event *event)
3267{ 3484{
3268 u64 u64temp;
3269 struct nes_qp *nesqp; 3485 struct nes_qp *nesqp;
3270 struct nes_vnic *nesvnic; 3486 struct nes_vnic *nesvnic;
3271 struct nes_device *nesdev; 3487 struct nes_device *nesdev;
@@ -3274,7 +3490,6 @@ static void cm_event_connected(struct nes_cm_event *event)
3274 struct ib_qp_attr attr; 3490 struct ib_qp_attr attr;
3275 struct iw_cm_id *cm_id; 3491 struct iw_cm_id *cm_id;
3276 struct iw_cm_event cm_event; 3492 struct iw_cm_event cm_event;
3277 struct nes_hw_qp_wqe *wqe;
3278 struct nes_v4_quad nes_quad; 3493 struct nes_v4_quad nes_quad;
3279 u32 crc_value; 3494 u32 crc_value;
3280 int ret; 3495 int ret;
@@ -3288,17 +3503,16 @@ static void cm_event_connected(struct nes_cm_event *event)
3288 nesdev = nesvnic->nesdev; 3503 nesdev = nesvnic->nesdev;
3289 nesadapter = nesdev->nesadapter; 3504 nesadapter = nesdev->nesadapter;
3290 3505
3291 if (nesqp->destroyed) { 3506 if (nesqp->destroyed)
3292 return; 3507 return;
3293 }
3294 atomic_inc(&cm_connecteds); 3508 atomic_inc(&cm_connecteds);
3295 nes_debug(NES_DBG_CM, "QP%u attempting to connect to 0x%08X:0x%04X on" 3509 nes_debug(NES_DBG_CM, "QP%u attempting to connect to 0x%08X:0x%04X on"
3296 " local port 0x%04X. jiffies = %lu.\n", 3510 " local port 0x%04X. jiffies = %lu.\n",
3297 nesqp->hwqp.qp_id, 3511 nesqp->hwqp.qp_id,
3298 ntohl(cm_id->remote_addr.sin_addr.s_addr), 3512 ntohl(cm_id->remote_addr.sin_addr.s_addr),
3299 ntohs(cm_id->remote_addr.sin_port), 3513 ntohs(cm_id->remote_addr.sin_port),
3300 ntohs(cm_id->local_addr.sin_port), 3514 ntohs(cm_id->local_addr.sin_port),
3301 jiffies); 3515 jiffies);
3302 3516
3303 nes_cm_init_tsa_conn(nesqp, cm_node); 3517 nes_cm_init_tsa_conn(nesqp, cm_node);
3304 3518
@@ -3329,40 +3543,12 @@ static void cm_event_connected(struct nes_cm_event *event)
3329 NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT); 3543 NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT);
3330 3544
3331 /* Adjust tail for not having a LSMM */ 3545 /* Adjust tail for not having a LSMM */
3332 nesqp->hwqp.sq_tail = 1; 3546 /*nesqp->hwqp.sq_tail = 1;*/
3333
3334#if defined(NES_SEND_FIRST_WRITE)
3335 if (cm_node->send_write0) {
3336 nes_debug(NES_DBG_CM, "Sending first write.\n");
3337 wqe = &nesqp->hwqp.sq_vbase[0];
3338 u64temp = (unsigned long)nesqp;
3339 u64temp |= NES_SW_CONTEXT_ALIGN>>1;
3340 set_wqe_64bit_value(wqe->wqe_words,
3341 NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX, u64temp);
3342 wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
3343 cpu_to_le32(NES_IWARP_SQ_OP_RDMAW);
3344 wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] = 0;
3345 wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_LOW_IDX] = 0;
3346 wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_HIGH_IDX] = 0;
3347 wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] = 0;
3348 wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 0;
3349 3547
3350 if (nesqp->sq_kmapped) { 3548 build_rdma0_msg(cm_node, &nesqp);
3351 nesqp->sq_kmapped = 0;
3352 kunmap(nesqp->page);
3353 }
3354 3549
3355 /* use the reserved spot on the WQ for the extra first WQE */ 3550 nes_write32(nesdev->regs + NES_WQE_ALLOC,
3356 nesqp->nesqp_context->ird_ord_sizes &= 3551 (1 << 24) | 0x00800000 | nesqp->hwqp.qp_id);
3357 cpu_to_le32(~(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
3358 NES_QPCONTEXT_ORDIRD_WRPDU |
3359 NES_QPCONTEXT_ORDIRD_ALSMM));
3360 nesqp->skip_lsmm = 1;
3361 nesqp->hwqp.sq_tail = 0;
3362 nes_write32(nesdev->regs + NES_WQE_ALLOC,
3363 (1 << 24) | 0x00800000 | nesqp->hwqp.qp_id);
3364 }
3365#endif
3366 3552
3367 memset(&nes_quad, 0, sizeof(nes_quad)); 3553 memset(&nes_quad, 0, sizeof(nes_quad));
3368 3554
@@ -3379,13 +3565,13 @@ static void cm_event_connected(struct nes_cm_event *event)
3379 crc_value = get_crc_value(&nes_quad); 3565 crc_value = get_crc_value(&nes_quad);
3380 nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff); 3566 nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
3381 nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, After CRC = 0x%08X\n", 3567 nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, After CRC = 0x%08X\n",
3382 nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask); 3568 nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask);
3383 3569
3384 nesqp->hte_index &= nesadapter->hte_index_mask; 3570 nesqp->hte_index &= nesadapter->hte_index_mask;
3385 nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index); 3571 nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
3386 3572
3387 nesqp->ietf_frame = &cm_node->mpa_frame; 3573 nesqp->ietf_frame = &cm_node->mpa_frame;
3388 nesqp->private_data_len = (u8) cm_node->mpa_frame_size; 3574 nesqp->private_data_len = (u8)cm_node->mpa_frame_size;
3389 cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node); 3575 cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node);
3390 3576
3391 /* notify OF layer we successfully created the requested connection */ 3577 /* notify OF layer we successfully created the requested connection */
@@ -3397,7 +3583,9 @@ static void cm_event_connected(struct nes_cm_event *event)
3397 cm_event.remote_addr = cm_id->remote_addr; 3583 cm_event.remote_addr = cm_id->remote_addr;
3398 3584
3399 cm_event.private_data = (void *)event->cm_node->mpa_frame_buf; 3585 cm_event.private_data = (void *)event->cm_node->mpa_frame_buf;
3400 cm_event.private_data_len = (u8) event->cm_node->mpa_frame_size; 3586 cm_event.private_data_len = (u8)event->cm_node->mpa_frame_size;
3587 cm_event.ird = cm_node->ird_size;
3588 cm_event.ord = cm_node->ord_size;
3401 3589
3402 cm_event.local_addr.sin_addr.s_addr = event->cm_info.rem_addr; 3590 cm_event.local_addr.sin_addr.s_addr = event->cm_info.rem_addr;
3403 ret = cm_id->event_handler(cm_id, &cm_event); 3591 ret = cm_id->event_handler(cm_id, &cm_event);
@@ -3405,12 +3593,12 @@ static void cm_event_connected(struct nes_cm_event *event)
3405 3593
3406 if (ret) 3594 if (ret)
3407 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, " 3595 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
3408 "ret=%d\n", __func__, __LINE__, ret); 3596 "ret=%d\n", __func__, __LINE__, ret);
3409 attr.qp_state = IB_QPS_RTS; 3597 attr.qp_state = IB_QPS_RTS;
3410 nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL); 3598 nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL);
3411 3599
3412 nes_debug(NES_DBG_CM, "Exiting connect thread for QP%u. jiffies = " 3600 nes_debug(NES_DBG_CM, "Exiting connect thread for QP%u. jiffies = "
3413 "%lu\n", nesqp->hwqp.qp_id, jiffies); 3601 "%lu\n", nesqp->hwqp.qp_id, jiffies);
3414 3602
3415 return; 3603 return;
3416} 3604}
@@ -3431,16 +3619,14 @@ static void cm_event_connect_error(struct nes_cm_event *event)
3431 return; 3619 return;
3432 3620
3433 cm_id = event->cm_node->cm_id; 3621 cm_id = event->cm_node->cm_id;
3434 if (!cm_id) { 3622 if (!cm_id)
3435 return; 3623 return;
3436 }
3437 3624
3438 nes_debug(NES_DBG_CM, "cm_node=%p, cm_id=%p\n", event->cm_node, cm_id); 3625 nes_debug(NES_DBG_CM, "cm_node=%p, cm_id=%p\n", event->cm_node, cm_id);
3439 nesqp = cm_id->provider_data; 3626 nesqp = cm_id->provider_data;
3440 3627
3441 if (!nesqp) { 3628 if (!nesqp)
3442 return; 3629 return;
3443 }
3444 3630
3445 /* notify OF layer about this connection error event */ 3631 /* notify OF layer about this connection error event */
3446 /* cm_id->rem_ref(cm_id); */ 3632 /* cm_id->rem_ref(cm_id); */
@@ -3455,14 +3641,14 @@ static void cm_event_connect_error(struct nes_cm_event *event)
3455 cm_event.private_data_len = 0; 3641 cm_event.private_data_len = 0;
3456 3642
3457 nes_debug(NES_DBG_CM, "call CM_EVENT REJECTED, local_addr=%08x, " 3643 nes_debug(NES_DBG_CM, "call CM_EVENT REJECTED, local_addr=%08x, "
3458 "remove_addr=%08x\n", cm_event.local_addr.sin_addr.s_addr, 3644 "remove_addr=%08x\n", cm_event.local_addr.sin_addr.s_addr,
3459 cm_event.remote_addr.sin_addr.s_addr); 3645 cm_event.remote_addr.sin_addr.s_addr);
3460 3646
3461 ret = cm_id->event_handler(cm_id, &cm_event); 3647 ret = cm_id->event_handler(cm_id, &cm_event);
3462 nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret); 3648 nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
3463 if (ret) 3649 if (ret)
3464 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, " 3650 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
3465 "ret=%d\n", __func__, __LINE__, ret); 3651 "ret=%d\n", __func__, __LINE__, ret);
3466 cm_id->rem_ref(cm_id); 3652 cm_id->rem_ref(cm_id);
3467 3653
3468 rem_ref_cm_node(event->cm_node->cm_core, event->cm_node); 3654 rem_ref_cm_node(event->cm_node->cm_core, event->cm_node);
@@ -3532,7 +3718,7 @@ static void cm_event_reset(struct nes_cm_event *event)
3532 */ 3718 */
3533static void cm_event_mpa_req(struct nes_cm_event *event) 3719static void cm_event_mpa_req(struct nes_cm_event *event)
3534{ 3720{
3535 struct iw_cm_id *cm_id; 3721 struct iw_cm_id *cm_id;
3536 struct iw_cm_event cm_event; 3722 struct iw_cm_event cm_event;
3537 int ret; 3723 int ret;
3538 struct nes_cm_node *cm_node; 3724 struct nes_cm_node *cm_node;
@@ -3544,7 +3730,7 @@ static void cm_event_mpa_req(struct nes_cm_event *event)
3544 3730
3545 atomic_inc(&cm_connect_reqs); 3731 atomic_inc(&cm_connect_reqs);
3546 nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n", 3732 nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n",
3547 cm_node, cm_id, jiffies); 3733 cm_node, cm_id, jiffies);
3548 3734
3549 cm_event.event = IW_CM_EVENT_CONNECT_REQUEST; 3735 cm_event.event = IW_CM_EVENT_CONNECT_REQUEST;
3550 cm_event.status = 0; 3736 cm_event.status = 0;
@@ -3558,19 +3744,21 @@ static void cm_event_mpa_req(struct nes_cm_event *event)
3558 cm_event.remote_addr.sin_port = htons(event->cm_info.rem_port); 3744 cm_event.remote_addr.sin_port = htons(event->cm_info.rem_port);
3559 cm_event.remote_addr.sin_addr.s_addr = htonl(event->cm_info.rem_addr); 3745 cm_event.remote_addr.sin_addr.s_addr = htonl(event->cm_info.rem_addr);
3560 cm_event.private_data = cm_node->mpa_frame_buf; 3746 cm_event.private_data = cm_node->mpa_frame_buf;
3561 cm_event.private_data_len = (u8) cm_node->mpa_frame_size; 3747 cm_event.private_data_len = (u8)cm_node->mpa_frame_size;
3748 cm_event.ird = cm_node->ird_size;
3749 cm_event.ord = cm_node->ord_size;
3562 3750
3563 ret = cm_id->event_handler(cm_id, &cm_event); 3751 ret = cm_id->event_handler(cm_id, &cm_event);
3564 if (ret) 3752 if (ret)
3565 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n", 3753 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n",
3566 __func__, __LINE__, ret); 3754 __func__, __LINE__, ret);
3567 return; 3755 return;
3568} 3756}
3569 3757
3570 3758
3571static void cm_event_mpa_reject(struct nes_cm_event *event) 3759static void cm_event_mpa_reject(struct nes_cm_event *event)
3572{ 3760{
3573 struct iw_cm_id *cm_id; 3761 struct iw_cm_id *cm_id;
3574 struct iw_cm_event cm_event; 3762 struct iw_cm_event cm_event;
3575 struct nes_cm_node *cm_node; 3763 struct nes_cm_node *cm_node;
3576 int ret; 3764 int ret;
@@ -3582,7 +3770,7 @@ static void cm_event_mpa_reject(struct nes_cm_event *event)
3582 3770
3583 atomic_inc(&cm_connect_reqs); 3771 atomic_inc(&cm_connect_reqs);
3584 nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n", 3772 nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n",
3585 cm_node, cm_id, jiffies); 3773 cm_node, cm_id, jiffies);
3586 3774
3587 cm_event.event = IW_CM_EVENT_CONNECT_REPLY; 3775 cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
3588 cm_event.status = -ECONNREFUSED; 3776 cm_event.status = -ECONNREFUSED;
@@ -3597,17 +3785,17 @@ static void cm_event_mpa_reject(struct nes_cm_event *event)
3597 cm_event.remote_addr.sin_addr.s_addr = htonl(event->cm_info.rem_addr); 3785 cm_event.remote_addr.sin_addr.s_addr = htonl(event->cm_info.rem_addr);
3598 3786
3599 cm_event.private_data = cm_node->mpa_frame_buf; 3787 cm_event.private_data = cm_node->mpa_frame_buf;
3600 cm_event.private_data_len = (u8) cm_node->mpa_frame_size; 3788 cm_event.private_data_len = (u8)cm_node->mpa_frame_size;
3601 3789
3602 nes_debug(NES_DBG_CM, "call CM_EVENT_MPA_REJECTED, local_addr=%08x, " 3790 nes_debug(NES_DBG_CM, "call CM_EVENT_MPA_REJECTED, local_addr=%08x, "
3603 "remove_addr=%08x\n", 3791 "remove_addr=%08x\n",
3604 cm_event.local_addr.sin_addr.s_addr, 3792 cm_event.local_addr.sin_addr.s_addr,
3605 cm_event.remote_addr.sin_addr.s_addr); 3793 cm_event.remote_addr.sin_addr.s_addr);
3606 3794
3607 ret = cm_id->event_handler(cm_id, &cm_event); 3795 ret = cm_id->event_handler(cm_id, &cm_event);
3608 if (ret) 3796 if (ret)
3609 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n", 3797 printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n",
3610 __func__, __LINE__, ret); 3798 __func__, __LINE__, ret);
3611 3799
3612 return; 3800 return;
3613} 3801}
@@ -3626,7 +3814,7 @@ static int nes_cm_post_event(struct nes_cm_event *event)
3626 event->cm_info.cm_id->add_ref(event->cm_info.cm_id); 3814 event->cm_info.cm_id->add_ref(event->cm_info.cm_id);
3627 INIT_WORK(&event->event_work, nes_cm_event_handler); 3815 INIT_WORK(&event->event_work, nes_cm_event_handler);
3628 nes_debug(NES_DBG_CM, "cm_node=%p queue_work, event=%p\n", 3816 nes_debug(NES_DBG_CM, "cm_node=%p queue_work, event=%p\n",
3629 event->cm_node, event); 3817 event->cm_node, event);
3630 3818
3631 queue_work(event->cm_node->cm_core->event_wq, &event->event_work); 3819 queue_work(event->cm_node->cm_core->event_wq, &event->event_work);
3632 3820
@@ -3643,7 +3831,7 @@ static int nes_cm_post_event(struct nes_cm_event *event)
3643static void nes_cm_event_handler(struct work_struct *work) 3831static void nes_cm_event_handler(struct work_struct *work)
3644{ 3832{
3645 struct nes_cm_event *event = container_of(work, struct nes_cm_event, 3833 struct nes_cm_event *event = container_of(work, struct nes_cm_event,
3646 event_work); 3834 event_work);
3647 struct nes_cm_core *cm_core; 3835 struct nes_cm_core *cm_core;
3648 3836
3649 if ((!event) || (!event->cm_node) || (!event->cm_node->cm_core)) 3837 if ((!event) || (!event->cm_node) || (!event->cm_node->cm_core))
@@ -3651,29 +3839,29 @@ static void nes_cm_event_handler(struct work_struct *work)
3651 3839
3652 cm_core = event->cm_node->cm_core; 3840 cm_core = event->cm_node->cm_core;
3653 nes_debug(NES_DBG_CM, "event=%p, event->type=%u, events posted=%u\n", 3841 nes_debug(NES_DBG_CM, "event=%p, event->type=%u, events posted=%u\n",
3654 event, event->type, atomic_read(&cm_core->events_posted)); 3842 event, event->type, atomic_read(&cm_core->events_posted));
3655 3843
3656 switch (event->type) { 3844 switch (event->type) {
3657 case NES_CM_EVENT_MPA_REQ: 3845 case NES_CM_EVENT_MPA_REQ:
3658 cm_event_mpa_req(event); 3846 cm_event_mpa_req(event);
3659 nes_debug(NES_DBG_CM, "cm_node=%p CM Event: MPA REQUEST\n", 3847 nes_debug(NES_DBG_CM, "cm_node=%p CM Event: MPA REQUEST\n",
3660 event->cm_node); 3848 event->cm_node);
3661 break; 3849 break;
3662 case NES_CM_EVENT_RESET: 3850 case NES_CM_EVENT_RESET:
3663 nes_debug(NES_DBG_CM, "cm_node = %p CM Event: RESET\n", 3851 nes_debug(NES_DBG_CM, "cm_node = %p CM Event: RESET\n",
3664 event->cm_node); 3852 event->cm_node);
3665 cm_event_reset(event); 3853 cm_event_reset(event);
3666 break; 3854 break;
3667 case NES_CM_EVENT_CONNECTED: 3855 case NES_CM_EVENT_CONNECTED:
3668 if ((!event->cm_node->cm_id) || 3856 if ((!event->cm_node->cm_id) ||
3669 (event->cm_node->state != NES_CM_STATE_TSA)) 3857 (event->cm_node->state != NES_CM_STATE_TSA))
3670 break; 3858 break;
3671 cm_event_connected(event); 3859 cm_event_connected(event);
3672 nes_debug(NES_DBG_CM, "CM Event: CONNECTED\n"); 3860 nes_debug(NES_DBG_CM, "CM Event: CONNECTED\n");
3673 break; 3861 break;
3674 case NES_CM_EVENT_MPA_REJECT: 3862 case NES_CM_EVENT_MPA_REJECT:
3675 if ((!event->cm_node->cm_id) || 3863 if ((!event->cm_node->cm_id) ||
3676 (event->cm_node->state == NES_CM_STATE_TSA)) 3864 (event->cm_node->state == NES_CM_STATE_TSA))
3677 break; 3865 break;
3678 cm_event_mpa_reject(event); 3866 cm_event_mpa_reject(event);
3679 nes_debug(NES_DBG_CM, "CM Event: REJECT\n"); 3867 nes_debug(NES_DBG_CM, "CM Event: REJECT\n");
@@ -3681,7 +3869,7 @@ static void nes_cm_event_handler(struct work_struct *work)
3681 3869
3682 case NES_CM_EVENT_ABORTED: 3870 case NES_CM_EVENT_ABORTED:
3683 if ((!event->cm_node->cm_id) || 3871 if ((!event->cm_node->cm_id) ||
3684 (event->cm_node->state == NES_CM_STATE_TSA)) 3872 (event->cm_node->state == NES_CM_STATE_TSA))
3685 break; 3873 break;
3686 cm_event_connect_error(event); 3874 cm_event_connect_error(event);
3687 nes_debug(NES_DBG_CM, "CM Event: ABORTED\n"); 3875 nes_debug(NES_DBG_CM, "CM Event: ABORTED\n");
diff --git a/drivers/infiniband/hw/nes/nes_cm.h b/drivers/infiniband/hw/nes/nes_cm.h
index 130c185cde0d..bdfa1fbb35fc 100644
--- a/drivers/infiniband/hw/nes/nes_cm.h
+++ b/drivers/infiniband/hw/nes/nes_cm.h
@@ -48,7 +48,16 @@
48#define IETF_MPA_KEY_SIZE 16 48#define IETF_MPA_KEY_SIZE 16
49#define IETF_MPA_VERSION 1 49#define IETF_MPA_VERSION 1
50#define IETF_MAX_PRIV_DATA_LEN 512 50#define IETF_MAX_PRIV_DATA_LEN 512
51#define IETF_MPA_FRAME_SIZE 20 51#define IETF_MPA_FRAME_SIZE 20
52#define IETF_RTR_MSG_SIZE 4
53#define IETF_MPA_V2_FLAG 0x10
54
55/* IETF RTR MSG Fields */
56#define IETF_PEER_TO_PEER 0x8000
57#define IETF_FLPDU_ZERO_LEN 0x4000
58#define IETF_RDMA0_WRITE 0x8000
59#define IETF_RDMA0_READ 0x4000
60#define IETF_NO_IRD_ORD 0x3FFF
52 61
53enum ietf_mpa_flags { 62enum ietf_mpa_flags {
54 IETF_MPA_FLAGS_MARKERS = 0x80, /* receive Markers */ 63 IETF_MPA_FLAGS_MARKERS = 0x80, /* receive Markers */
@@ -56,7 +65,7 @@ enum ietf_mpa_flags {
56 IETF_MPA_FLAGS_REJECT = 0x20, /* Reject */ 65 IETF_MPA_FLAGS_REJECT = 0x20, /* Reject */
57}; 66};
58 67
59struct ietf_mpa_frame { 68struct ietf_mpa_v1 {
60 u8 key[IETF_MPA_KEY_SIZE]; 69 u8 key[IETF_MPA_KEY_SIZE];
61 u8 flags; 70 u8 flags;
62 u8 rev; 71 u8 rev;
@@ -66,6 +75,20 @@ struct ietf_mpa_frame {
66 75
67#define ietf_mpa_req_resp_frame ietf_mpa_frame 76#define ietf_mpa_req_resp_frame ietf_mpa_frame
68 77
78struct ietf_rtr_msg {
79 __be16 ctrl_ird;
80 __be16 ctrl_ord;
81};
82
83struct ietf_mpa_v2 {
84 u8 key[IETF_MPA_KEY_SIZE];
85 u8 flags;
86 u8 rev;
87 __be16 priv_data_len;
88 struct ietf_rtr_msg rtr_msg;
89 u8 priv_data[0];
90};
91
69struct nes_v4_quad { 92struct nes_v4_quad {
70 u32 rsvd0; 93 u32 rsvd0;
71 __le32 DstIpAdrIndex; /* Only most significant 5 bits are valid */ 94 __le32 DstIpAdrIndex; /* Only most significant 5 bits are valid */
@@ -171,8 +194,7 @@ struct nes_timer_entry {
171 194
172#define NES_CM_DEF_SEQ2 0x18ed5740 195#define NES_CM_DEF_SEQ2 0x18ed5740
173#define NES_CM_DEF_LOCAL_ID2 0xb807 196#define NES_CM_DEF_LOCAL_ID2 0xb807
174#define MAX_CM_BUFFER (IETF_MPA_FRAME_SIZE + IETF_MAX_PRIV_DATA_LEN) 197#define MAX_CM_BUFFER (IETF_MPA_FRAME_SIZE + IETF_RTR_MSG_SIZE + IETF_MAX_PRIV_DATA_LEN)
175
176 198
177typedef u32 nes_addr_t; 199typedef u32 nes_addr_t;
178 200
@@ -204,6 +226,21 @@ enum nes_cm_node_state {
204 NES_CM_STATE_CLOSED 226 NES_CM_STATE_CLOSED
205}; 227};
206 228
229enum mpa_frame_version {
230 IETF_MPA_V1 = 1,
231 IETF_MPA_V2 = 2
232};
233
234enum mpa_frame_key {
235 MPA_KEY_REQUEST,
236 MPA_KEY_REPLY
237};
238
239enum send_rdma0 {
240 SEND_RDMA_READ_ZERO = 1,
241 SEND_RDMA_WRITE_ZERO = 2
242};
243
207enum nes_tcpip_pkt_type { 244enum nes_tcpip_pkt_type {
208 NES_PKT_TYPE_UNKNOWN, 245 NES_PKT_TYPE_UNKNOWN,
209 NES_PKT_TYPE_SYN, 246 NES_PKT_TYPE_SYN,
@@ -245,9 +282,9 @@ struct nes_cm_tcp_context {
245 282
246 283
247enum nes_cm_listener_state { 284enum nes_cm_listener_state {
248 NES_CM_LISTENER_PASSIVE_STATE=1, 285 NES_CM_LISTENER_PASSIVE_STATE = 1,
249 NES_CM_LISTENER_ACTIVE_STATE=2, 286 NES_CM_LISTENER_ACTIVE_STATE = 2,
250 NES_CM_LISTENER_EITHER_STATE=3 287 NES_CM_LISTENER_EITHER_STATE = 3
251}; 288};
252 289
253struct nes_cm_listener { 290struct nes_cm_listener {
@@ -283,16 +320,20 @@ struct nes_cm_node {
283 320
284 struct nes_cm_node *loopbackpartner; 321 struct nes_cm_node *loopbackpartner;
285 322
286 struct nes_timer_entry *send_entry; 323 struct nes_timer_entry *send_entry;
287 324 struct nes_timer_entry *recv_entry;
288 spinlock_t retrans_list_lock; 325 spinlock_t retrans_list_lock;
289 struct nes_timer_entry *recv_entry; 326 enum send_rdma0 send_rdma0_op;
290 327
291 int send_write0;
292 union { 328 union {
293 struct ietf_mpa_frame mpa_frame; 329 struct ietf_mpa_v1 mpa_frame;
294 u8 mpa_frame_buf[MAX_CM_BUFFER]; 330 struct ietf_mpa_v2 mpa_v2_frame;
331 u8 mpa_frame_buf[MAX_CM_BUFFER];
295 }; 332 };
333 enum mpa_frame_version mpa_frame_rev;
334 u16 ird_size;
335 u16 ord_size;
336
296 u16 mpa_frame_size; 337 u16 mpa_frame_size;
297 struct iw_cm_id *cm_id; 338 struct iw_cm_id *cm_id;
298 struct list_head list; 339 struct list_head list;
@@ -399,10 +440,8 @@ struct nes_cm_ops {
399 struct nes_vnic *, u16, void *, 440 struct nes_vnic *, u16, void *,
400 struct nes_cm_info *); 441 struct nes_cm_info *);
401 int (*close)(struct nes_cm_core *, struct nes_cm_node *); 442 int (*close)(struct nes_cm_core *, struct nes_cm_node *);
402 int (*accept)(struct nes_cm_core *, struct ietf_mpa_frame *, 443 int (*accept)(struct nes_cm_core *, struct nes_cm_node *);
403 struct nes_cm_node *); 444 int (*reject)(struct nes_cm_core *, struct nes_cm_node *);
404 int (*reject)(struct nes_cm_core *, struct ietf_mpa_frame *,
405 struct nes_cm_node *);
406 int (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *, 445 int (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *,
407 struct sk_buff *); 446 struct sk_buff *);
408 int (*destroy_cm_core)(struct nes_cm_core *); 447 int (*destroy_cm_core)(struct nes_cm_core *);
diff --git a/drivers/infiniband/hw/nes/nes_verbs.h b/drivers/infiniband/hw/nes/nes_verbs.h
index 280057353343..fe6b6e92fa90 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.h
+++ b/drivers/infiniband/hw/nes/nes_verbs.h
@@ -139,7 +139,8 @@ struct nes_qp {
139 struct nes_cq *nesrcq; 139 struct nes_cq *nesrcq;
140 struct nes_pd *nespd; 140 struct nes_pd *nespd;
141 void *cm_node; /* handle of the node this QP is associated with */ 141 void *cm_node; /* handle of the node this QP is associated with */
142 struct ietf_mpa_frame *ietf_frame; 142 void *ietf_frame;
143 u8 ietf_frame_size;
143 dma_addr_t ietf_frame_pbase; 144 dma_addr_t ietf_frame_pbase;
144 struct ib_mr *lsmm_mr; 145 struct ib_mr *lsmm_mr;
145 struct nes_hw_qp hwqp; 146 struct nes_hw_qp hwqp;
diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
index c9624ea87209..b881bdc401f5 100644
--- a/drivers/infiniband/hw/qib/qib.h
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -171,7 +171,9 @@ struct qib_ctxtdata {
171 /* how many alloc_pages() chunks in rcvegrbuf_pages */ 171 /* how many alloc_pages() chunks in rcvegrbuf_pages */
172 u32 rcvegrbuf_chunks; 172 u32 rcvegrbuf_chunks;
173 /* how many egrbufs per chunk */ 173 /* how many egrbufs per chunk */
174 u32 rcvegrbufs_perchunk; 174 u16 rcvegrbufs_perchunk;
175 /* ilog2 of above */
176 u16 rcvegrbufs_perchunk_shift;
175 /* order for rcvegrbuf_pages */ 177 /* order for rcvegrbuf_pages */
176 size_t rcvegrbuf_size; 178 size_t rcvegrbuf_size;
177 /* rcvhdrq size (for freeing) */ 179 /* rcvhdrq size (for freeing) */
@@ -221,6 +223,9 @@ struct qib_ctxtdata {
221 /* ctxt rcvhdrq head offset */ 223 /* ctxt rcvhdrq head offset */
222 u32 head; 224 u32 head;
223 u32 pkt_count; 225 u32 pkt_count;
226 /* lookaside fields */
227 struct qib_qp *lookaside_qp;
228 u32 lookaside_qpn;
224 /* QPs waiting for context processing */ 229 /* QPs waiting for context processing */
225 struct list_head qp_wait_list; 230 struct list_head qp_wait_list;
226}; 231};
@@ -807,6 +812,10 @@ struct qib_devdata {
807 * supports, less gives more pio bufs/ctxt, etc. 812 * supports, less gives more pio bufs/ctxt, etc.
808 */ 813 */
809 u32 cfgctxts; 814 u32 cfgctxts;
815 /*
816 * number of ctxts available for PSM open
817 */
818 u32 freectxts;
810 819
811 /* 820 /*
812 * hint that we should update pioavailshadow before 821 * hint that we should update pioavailshadow before
@@ -936,7 +945,9 @@ struct qib_devdata {
936 /* chip address space used by 4k pio buffers */ 945 /* chip address space used by 4k pio buffers */
937 u32 align4k; 946 u32 align4k;
938 /* size of each rcvegrbuffer */ 947 /* size of each rcvegrbuffer */
939 u32 rcvegrbufsize; 948 u16 rcvegrbufsize;
949 /* log2 of above */
950 u16 rcvegrbufsize_shift;
940 /* localbus width (1, 2,4,8,16,32) from config space */ 951 /* localbus width (1, 2,4,8,16,32) from config space */
941 u32 lbus_width; 952 u32 lbus_width;
942 /* localbus speed in MHz */ 953 /* localbus speed in MHz */
diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
index 23e584f4c36c..9a9047f385ae 100644
--- a/drivers/infiniband/hw/qib/qib_driver.c
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -279,10 +279,10 @@ bail:
279 */ 279 */
280static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail) 280static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
281{ 281{
282 const u32 chunk = etail / rcd->rcvegrbufs_perchunk; 282 const u32 chunk = etail >> rcd->rcvegrbufs_perchunk_shift;
283 const u32 idx = etail % rcd->rcvegrbufs_perchunk; 283 const u32 idx = etail & ((u32)rcd->rcvegrbufs_perchunk - 1);
284 284
285 return rcd->rcvegrbuf[chunk] + idx * rcd->dd->rcvegrbufsize; 285 return rcd->rcvegrbuf[chunk] + (idx << rcd->dd->rcvegrbufsize_shift);
286} 286}
287 287
288/* 288/*
@@ -310,7 +310,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
310 u32 opcode; 310 u32 opcode;
311 u32 psn; 311 u32 psn;
312 int diff; 312 int diff;
313 unsigned long flags;
314 313
315 /* Sanity check packet */ 314 /* Sanity check packet */
316 if (tlen < 24) 315 if (tlen < 24)
@@ -365,7 +364,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
365 364
366 switch (qp->ibqp.qp_type) { 365 switch (qp->ibqp.qp_type) {
367 case IB_QPT_RC: 366 case IB_QPT_RC:
368 spin_lock_irqsave(&qp->s_lock, flags);
369 ruc_res = 367 ruc_res =
370 qib_ruc_check_hdr( 368 qib_ruc_check_hdr(
371 ibp, hdr, 369 ibp, hdr,
@@ -373,11 +371,8 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
373 qp, 371 qp,
374 be32_to_cpu(ohdr->bth[0])); 372 be32_to_cpu(ohdr->bth[0]));
375 if (ruc_res) { 373 if (ruc_res) {
376 spin_unlock_irqrestore(&qp->s_lock,
377 flags);
378 goto unlock; 374 goto unlock;
379 } 375 }
380 spin_unlock_irqrestore(&qp->s_lock, flags);
381 376
382 /* Only deal with RDMA Writes for now */ 377 /* Only deal with RDMA Writes for now */
383 if (opcode < 378 if (opcode <
@@ -547,6 +542,15 @@ move_along:
547 updegr = 0; 542 updegr = 0;
548 } 543 }
549 } 544 }
545 /*
546 * Notify qib_destroy_qp() if it is waiting
547 * for lookaside_qp to finish.
548 */
549 if (rcd->lookaside_qp) {
550 if (atomic_dec_and_test(&rcd->lookaside_qp->refcount))
551 wake_up(&rcd->lookaside_qp->wait);
552 rcd->lookaside_qp = NULL;
553 }
550 554
551 rcd->head = l; 555 rcd->head = l;
552 rcd->pkt_count += i; 556 rcd->pkt_count += i;
diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
index 26253039d2c7..77633666f81c 100644
--- a/drivers/infiniband/hw/qib/qib_file_ops.c
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -1284,6 +1284,7 @@ static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
1284 strlcpy(rcd->comm, current->comm, sizeof(rcd->comm)); 1284 strlcpy(rcd->comm, current->comm, sizeof(rcd->comm));
1285 ctxt_fp(fp) = rcd; 1285 ctxt_fp(fp) = rcd;
1286 qib_stats.sps_ctxts++; 1286 qib_stats.sps_ctxts++;
1287 dd->freectxts++;
1287 ret = 0; 1288 ret = 0;
1288 goto bail; 1289 goto bail;
1289 1290
@@ -1792,6 +1793,7 @@ static int qib_close(struct inode *in, struct file *fp)
1792 if (dd->pageshadow) 1793 if (dd->pageshadow)
1793 unlock_expected_tids(rcd); 1794 unlock_expected_tids(rcd);
1794 qib_stats.sps_ctxts--; 1795 qib_stats.sps_ctxts--;
1796 dd->freectxts--;
1795 } 1797 }
1796 1798
1797 mutex_unlock(&qib_mutex); 1799 mutex_unlock(&qib_mutex);
diff --git a/drivers/infiniband/hw/qib/qib_iba6120.c b/drivers/infiniband/hw/qib/qib_iba6120.c
index d8ca0a0b970d..781a802a321f 100644
--- a/drivers/infiniband/hw/qib/qib_iba6120.c
+++ b/drivers/infiniband/hw/qib/qib_iba6120.c
@@ -3273,6 +3273,8 @@ static int init_6120_variables(struct qib_devdata *dd)
3273 /* we always allocate at least 2048 bytes for eager buffers */ 3273 /* we always allocate at least 2048 bytes for eager buffers */
3274 ret = ib_mtu_enum_to_int(qib_ibmtu); 3274 ret = ib_mtu_enum_to_int(qib_ibmtu);
3275 dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU; 3275 dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
3276 BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
3277 dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
3276 3278
3277 qib_6120_tidtemplate(dd); 3279 qib_6120_tidtemplate(dd);
3278 3280
diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c
index e1f947446c2a..3f1d562ba898 100644
--- a/drivers/infiniband/hw/qib/qib_iba7220.c
+++ b/drivers/infiniband/hw/qib/qib_iba7220.c
@@ -4085,6 +4085,8 @@ static int qib_init_7220_variables(struct qib_devdata *dd)
4085 /* we always allocate at least 2048 bytes for eager buffers */ 4085 /* we always allocate at least 2048 bytes for eager buffers */
4086 ret = ib_mtu_enum_to_int(qib_ibmtu); 4086 ret = ib_mtu_enum_to_int(qib_ibmtu);
4087 dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU; 4087 dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
4088 BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
4089 dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
4088 4090
4089 qib_7220_tidtemplate(dd); 4091 qib_7220_tidtemplate(dd);
4090 4092
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
index 5ea9ece23b33..efd0a110091f 100644
--- a/drivers/infiniband/hw/qib/qib_iba7322.c
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -2310,12 +2310,15 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
2310 val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE << 2310 val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE <<
2311 QLOGIC_IB_IBCC_LINKINITCMD_SHIFT); 2311 QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
2312 2312
2313 ppd->cpspec->ibcctrl_a = val;
2313 /* 2314 /*
2314 * Reset the PCS interface to the serdes (and also ibc, which is still 2315 * Reset the PCS interface to the serdes (and also ibc, which is still
2315 * in reset from above). Writes new value of ibcctrl_a as last step. 2316 * in reset from above). Writes new value of ibcctrl_a as last step.
2316 */ 2317 */
2317 qib_7322_mini_pcs_reset(ppd); 2318 qib_7322_mini_pcs_reset(ppd);
2318 qib_write_kreg(dd, kr_scratch, 0ULL); 2319 qib_write_kreg(dd, kr_scratch, 0ULL);
2320 /* clear the linkinit cmds */
2321 ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0, LinkInitCmd);
2319 2322
2320 if (!ppd->cpspec->ibcctrl_b) { 2323 if (!ppd->cpspec->ibcctrl_b) {
2321 unsigned lse = ppd->link_speed_enabled; 2324 unsigned lse = ppd->link_speed_enabled;
@@ -2387,11 +2390,6 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
2387 qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl); 2390 qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl);
2388 spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags); 2391 spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
2389 2392
2390 /* Hold the link state machine for mezz boards */
2391 if (IS_QMH(dd) || IS_QME(dd))
2392 qib_set_ib_7322_lstate(ppd, 0,
2393 QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
2394
2395 /* Also enable IBSTATUSCHG interrupt. */ 2393 /* Also enable IBSTATUSCHG interrupt. */
2396 val = qib_read_kreg_port(ppd, krp_errmask); 2394 val = qib_read_kreg_port(ppd, krp_errmask);
2397 qib_write_kreg_port(ppd, krp_errmask, 2395 qib_write_kreg_port(ppd, krp_errmask,
@@ -2853,9 +2851,8 @@ static irqreturn_t qib_7322intr(int irq, void *data)
2853 for (i = 0; i < dd->first_user_ctxt; i++) { 2851 for (i = 0; i < dd->first_user_ctxt; i++) {
2854 if (ctxtrbits & rmask) { 2852 if (ctxtrbits & rmask) {
2855 ctxtrbits &= ~rmask; 2853 ctxtrbits &= ~rmask;
2856 if (dd->rcd[i]) { 2854 if (dd->rcd[i])
2857 qib_kreceive(dd->rcd[i], NULL, &npkts); 2855 qib_kreceive(dd->rcd[i], NULL, &npkts);
2858 }
2859 } 2856 }
2860 rmask <<= 1; 2857 rmask <<= 1;
2861 } 2858 }
@@ -5230,6 +5227,8 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
5230 QIBL_IB_AUTONEG_INPROG))) 5227 QIBL_IB_AUTONEG_INPROG)))
5231 set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled); 5228 set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled);
5232 if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) { 5229 if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
5230 struct qib_qsfp_data *qd =
5231 &ppd->cpspec->qsfp_data;
5233 /* unlock the Tx settings, speed may change */ 5232 /* unlock the Tx settings, speed may change */
5234 qib_write_kreg_port(ppd, krp_tx_deemph_override, 5233 qib_write_kreg_port(ppd, krp_tx_deemph_override,
5235 SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, 5234 SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
@@ -5237,6 +5236,12 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
5237 qib_cancel_sends(ppd); 5236 qib_cancel_sends(ppd);
5238 /* on link down, ensure sane pcs state */ 5237 /* on link down, ensure sane pcs state */
5239 qib_7322_mini_pcs_reset(ppd); 5238 qib_7322_mini_pcs_reset(ppd);
5239 /* schedule the qsfp refresh which should turn the link
5240 off */
5241 if (ppd->dd->flags & QIB_HAS_QSFP) {
5242 qd->t_insert = get_jiffies_64();
5243 schedule_work(&qd->work);
5244 }
5240 spin_lock_irqsave(&ppd->sdma_lock, flags); 5245 spin_lock_irqsave(&ppd->sdma_lock, flags);
5241 if (__qib_sdma_running(ppd)) 5246 if (__qib_sdma_running(ppd))
5242 __qib_sdma_process_event(ppd, 5247 __qib_sdma_process_event(ppd,
@@ -5587,43 +5592,79 @@ static void qsfp_7322_event(struct work_struct *work)
5587 struct qib_qsfp_data *qd; 5592 struct qib_qsfp_data *qd;
5588 struct qib_pportdata *ppd; 5593 struct qib_pportdata *ppd;
5589 u64 pwrup; 5594 u64 pwrup;
5595 unsigned long flags;
5590 int ret; 5596 int ret;
5591 u32 le2; 5597 u32 le2;
5592 5598
5593 qd = container_of(work, struct qib_qsfp_data, work); 5599 qd = container_of(work, struct qib_qsfp_data, work);
5594 ppd = qd->ppd; 5600 ppd = qd->ppd;
5595 pwrup = qd->t_insert + msecs_to_jiffies(QSFP_PWR_LAG_MSEC); 5601 pwrup = qd->t_insert +
5602 msecs_to_jiffies(QSFP_PWR_LAG_MSEC - QSFP_MODPRS_LAG_MSEC);
5596 5603
5597 /* 5604 /* Delay for 20 msecs to allow ModPrs resistor to setup */
5598 * Some QSFP's not only do not respond until the full power-up 5605 mdelay(QSFP_MODPRS_LAG_MSEC);
5599 * time, but may behave badly if we try. So hold off responding 5606
5600 * to insertion. 5607 if (!qib_qsfp_mod_present(ppd)) {
5601 */ 5608 ppd->cpspec->qsfp_data.modpresent = 0;
5602 while (1) { 5609 /* Set the physical link to disabled */
5603 u64 now = get_jiffies_64(); 5610 qib_set_ib_7322_lstate(ppd, 0,
5604 if (time_after64(now, pwrup)) 5611 QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
5605 break; 5612 spin_lock_irqsave(&ppd->lflags_lock, flags);
5606 msleep(20); 5613 ppd->lflags &= ~QIBL_LINKV;
5607 } 5614 spin_unlock_irqrestore(&ppd->lflags_lock, flags);
5608 ret = qib_refresh_qsfp_cache(ppd, &qd->cache); 5615 } else {
5609 /* 5616 /*
5610 * Need to change LE2 back to defaults if we couldn't 5617 * Some QSFP's not only do not respond until the full power-up
5611 * read the cable type (to handle cable swaps), so do this 5618 * time, but may behave badly if we try. So hold off responding
5612 * even on failure to read cable information. We don't 5619 * to insertion.
5613 * get here for QME, so IS_QME check not needed here. 5620 */
5614 */ 5621 while (1) {
5615 if (!ret && !ppd->dd->cspec->r1) { 5622 u64 now = get_jiffies_64();
5616 if (QSFP_IS_ACTIVE_FAR(qd->cache.tech)) 5623 if (time_after64(now, pwrup))
5617 le2 = LE2_QME; 5624 break;
5618 else if (qd->cache.atten[1] >= qib_long_atten && 5625 msleep(20);
5619 QSFP_IS_CU(qd->cache.tech)) 5626 }
5620 le2 = LE2_5m; 5627
5621 else 5628 ret = qib_refresh_qsfp_cache(ppd, &qd->cache);
5629
5630 /*
5631 * Need to change LE2 back to defaults if we couldn't
5632 * read the cable type (to handle cable swaps), so do this
5633 * even on failure to read cable information. We don't
5634 * get here for QME, so IS_QME check not needed here.
5635 */
5636 if (!ret && !ppd->dd->cspec->r1) {
5637 if (QSFP_IS_ACTIVE_FAR(qd->cache.tech))
5638 le2 = LE2_QME;
5639 else if (qd->cache.atten[1] >= qib_long_atten &&
5640 QSFP_IS_CU(qd->cache.tech))
5641 le2 = LE2_5m;
5642 else
5643 le2 = LE2_DEFAULT;
5644 } else
5622 le2 = LE2_DEFAULT; 5645 le2 = LE2_DEFAULT;
5623 } else 5646 ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7));
5624 le2 = LE2_DEFAULT; 5647 /*
5625 ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7)); 5648 * We always change parameteters, since we can choose
5626 init_txdds_table(ppd, 0); 5649 * values for cables without eeproms, and the cable may have
5650 * changed from a cable with full or partial eeprom content
5651 * to one with partial or no content.
5652 */
5653 init_txdds_table(ppd, 0);
5654 /* The physical link is being re-enabled only when the
5655 * previous state was DISABLED and the VALID bit is not
5656 * set. This should only happen when the cable has been
5657 * physically pulled. */
5658 if (!ppd->cpspec->qsfp_data.modpresent &&
5659 (ppd->lflags & (QIBL_LINKV | QIBL_IB_LINK_DISABLED))) {
5660 ppd->cpspec->qsfp_data.modpresent = 1;
5661 qib_set_ib_7322_lstate(ppd, 0,
5662 QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
5663 spin_lock_irqsave(&ppd->lflags_lock, flags);
5664 ppd->lflags |= QIBL_LINKV;
5665 spin_unlock_irqrestore(&ppd->lflags_lock, flags);
5666 }
5667 }
5627} 5668}
5628 5669
5629/* 5670/*
@@ -5727,7 +5768,8 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
5727 /* now change the IBC and serdes, overriding generic */ 5768 /* now change the IBC and serdes, overriding generic */
5728 init_txdds_table(ppd, 1); 5769 init_txdds_table(ppd, 1);
5729 /* Re-enable the physical state machine on mezz boards 5770 /* Re-enable the physical state machine on mezz boards
5730 * now that the correct settings have been set. */ 5771 * now that the correct settings have been set.
5772 * QSFP boards are handles by the QSFP event handler */
5731 if (IS_QMH(dd) || IS_QME(dd)) 5773 if (IS_QMH(dd) || IS_QME(dd))
5732 qib_set_ib_7322_lstate(ppd, 0, 5774 qib_set_ib_7322_lstate(ppd, 0,
5733 QLOGIC_IB_IBCC_LINKINITCMD_SLEEP); 5775 QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
@@ -6205,6 +6247,8 @@ static int qib_init_7322_variables(struct qib_devdata *dd)
6205 6247
6206 /* we always allocate at least 2048 bytes for eager buffers */ 6248 /* we always allocate at least 2048 bytes for eager buffers */
6207 dd->rcvegrbufsize = max(mtu, 2048); 6249 dd->rcvegrbufsize = max(mtu, 2048);
6250 BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
6251 dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
6208 6252
6209 qib_7322_tidtemplate(dd); 6253 qib_7322_tidtemplate(dd);
6210 6254
@@ -7147,7 +7191,8 @@ static void find_best_ent(struct qib_pportdata *ppd,
7147 } 7191 }
7148 } 7192 }
7149 7193
7150 /* Lookup serdes setting by cable type and attenuation */ 7194 /* Active cables don't have attenuation so we only set SERDES
7195 * settings to account for the attenuation of the board traces. */
7151 if (!override && QSFP_IS_ACTIVE(qd->tech)) { 7196 if (!override && QSFP_IS_ACTIVE(qd->tech)) {
7152 *sdr_dds = txdds_sdr + ppd->dd->board_atten; 7197 *sdr_dds = txdds_sdr + ppd->dd->board_atten;
7153 *ddr_dds = txdds_ddr + ppd->dd->board_atten; 7198 *ddr_dds = txdds_ddr + ppd->dd->board_atten;
@@ -7464,12 +7509,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
7464 u32 le_val, rxcaldone; 7509 u32 le_val, rxcaldone;
7465 int chan, chan_done = (1 << SERDES_CHANS) - 1; 7510 int chan, chan_done = (1 << SERDES_CHANS) - 1;
7466 7511
7467 /*
7468 * Initialize the Tx DDS tables. Also done every QSFP event,
7469 * for adapters with QSFP
7470 */
7471 init_txdds_table(ppd, 0);
7472
7473 /* Clear cmode-override, may be set from older driver */ 7512 /* Clear cmode-override, may be set from older driver */
7474 ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14); 7513 ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14);
7475 7514
@@ -7655,6 +7694,12 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
7655 /* VGA output common mode */ 7694 /* VGA output common mode */
7656 ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2)); 7695 ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
7657 7696
7697 /*
7698 * Initialize the Tx DDS tables. Also done every QSFP event,
7699 * for adapters with QSFP
7700 */
7701 init_txdds_table(ppd, 0);
7702
7658 return 0; 7703 return 0;
7659} 7704}
7660 7705
diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c
index a01f3fce8eb3..b093a0b53b2f 100644
--- a/drivers/infiniband/hw/qib/qib_init.c
+++ b/drivers/infiniband/hw/qib/qib_init.c
@@ -183,6 +183,9 @@ struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt)
183 rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt + 183 rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt +
184 rcd->rcvegrbufs_perchunk - 1) / 184 rcd->rcvegrbufs_perchunk - 1) /
185 rcd->rcvegrbufs_perchunk; 185 rcd->rcvegrbufs_perchunk;
186 BUG_ON(!is_power_of_2(rcd->rcvegrbufs_perchunk));
187 rcd->rcvegrbufs_perchunk_shift =
188 ilog2(rcd->rcvegrbufs_perchunk);
186 } 189 }
187 return rcd; 190 return rcd;
188} 191}
@@ -398,6 +401,7 @@ static void enable_chip(struct qib_devdata *dd)
398 if (rcd) 401 if (rcd)
399 dd->f_rcvctrl(rcd->ppd, rcvmask, i); 402 dd->f_rcvctrl(rcd->ppd, rcvmask, i);
400 } 403 }
404 dd->freectxts = dd->cfgctxts - dd->first_user_ctxt;
401} 405}
402 406
403static void verify_interrupt(unsigned long opaque) 407static void verify_interrupt(unsigned long opaque)
@@ -581,10 +585,6 @@ int qib_init(struct qib_devdata *dd, int reinit)
581 continue; 585 continue;
582 } 586 }
583 587
584 /* let link come up, and enable IBC */
585 spin_lock_irqsave(&ppd->lflags_lock, flags);
586 ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
587 spin_unlock_irqrestore(&ppd->lflags_lock, flags);
588 portok++; 588 portok++;
589 } 589 }
590 590
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index e16751f8639e..7e7e16fbee99 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -34,6 +34,7 @@
34 34
35#include <linux/err.h> 35#include <linux/err.h>
36#include <linux/vmalloc.h> 36#include <linux/vmalloc.h>
37#include <linux/jhash.h>
37 38
38#include "qib.h" 39#include "qib.h"
39 40
@@ -204,6 +205,13 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
204 clear_bit(qpn & BITS_PER_PAGE_MASK, map->page); 205 clear_bit(qpn & BITS_PER_PAGE_MASK, map->page);
205} 206}
206 207
208static inline unsigned qpn_hash(struct qib_ibdev *dev, u32 qpn)
209{
210 return jhash_1word(qpn, dev->qp_rnd) &
211 (dev->qp_table_size - 1);
212}
213
214
207/* 215/*
208 * Put the QP into the hash table. 216 * Put the QP into the hash table.
209 * The hash table holds a reference to the QP. 217 * The hash table holds a reference to the QP.
@@ -211,22 +219,23 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
211static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp) 219static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
212{ 220{
213 struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); 221 struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
214 unsigned n = qp->ibqp.qp_num % dev->qp_table_size;
215 unsigned long flags; 222 unsigned long flags;
223 unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
216 224
217 spin_lock_irqsave(&dev->qpt_lock, flags); 225 spin_lock_irqsave(&dev->qpt_lock, flags);
226 atomic_inc(&qp->refcount);
218 227
219 if (qp->ibqp.qp_num == 0) 228 if (qp->ibqp.qp_num == 0)
220 ibp->qp0 = qp; 229 rcu_assign_pointer(ibp->qp0, qp);
221 else if (qp->ibqp.qp_num == 1) 230 else if (qp->ibqp.qp_num == 1)
222 ibp->qp1 = qp; 231 rcu_assign_pointer(ibp->qp1, qp);
223 else { 232 else {
224 qp->next = dev->qp_table[n]; 233 qp->next = dev->qp_table[n];
225 dev->qp_table[n] = qp; 234 rcu_assign_pointer(dev->qp_table[n], qp);
226 } 235 }
227 atomic_inc(&qp->refcount);
228 236
229 spin_unlock_irqrestore(&dev->qpt_lock, flags); 237 spin_unlock_irqrestore(&dev->qpt_lock, flags);
238 synchronize_rcu();
230} 239}
231 240
232/* 241/*
@@ -236,29 +245,32 @@ static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
236static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp) 245static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp)
237{ 246{
238 struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); 247 struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
239 struct qib_qp *q, **qpp; 248 unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
240 unsigned long flags; 249 unsigned long flags;
241 250
242 qpp = &dev->qp_table[qp->ibqp.qp_num % dev->qp_table_size];
243
244 spin_lock_irqsave(&dev->qpt_lock, flags); 251 spin_lock_irqsave(&dev->qpt_lock, flags);
245 252
246 if (ibp->qp0 == qp) { 253 if (ibp->qp0 == qp) {
247 ibp->qp0 = NULL;
248 atomic_dec(&qp->refcount); 254 atomic_dec(&qp->refcount);
255 rcu_assign_pointer(ibp->qp0, NULL);
249 } else if (ibp->qp1 == qp) { 256 } else if (ibp->qp1 == qp) {
250 ibp->qp1 = NULL;
251 atomic_dec(&qp->refcount); 257 atomic_dec(&qp->refcount);
252 } else 258 rcu_assign_pointer(ibp->qp1, NULL);
259 } else {
260 struct qib_qp *q, **qpp;
261
262 qpp = &dev->qp_table[n];
253 for (; (q = *qpp) != NULL; qpp = &q->next) 263 for (; (q = *qpp) != NULL; qpp = &q->next)
254 if (q == qp) { 264 if (q == qp) {
255 *qpp = qp->next;
256 qp->next = NULL;
257 atomic_dec(&qp->refcount); 265 atomic_dec(&qp->refcount);
266 rcu_assign_pointer(*qpp, qp->next);
267 qp->next = NULL;
258 break; 268 break;
259 } 269 }
270 }
260 271
261 spin_unlock_irqrestore(&dev->qpt_lock, flags); 272 spin_unlock_irqrestore(&dev->qpt_lock, flags);
273 synchronize_rcu();
262} 274}
263 275
264/** 276/**
@@ -280,21 +292,24 @@ unsigned qib_free_all_qps(struct qib_devdata *dd)
280 292
281 if (!qib_mcast_tree_empty(ibp)) 293 if (!qib_mcast_tree_empty(ibp))
282 qp_inuse++; 294 qp_inuse++;
283 if (ibp->qp0) 295 rcu_read_lock();
296 if (rcu_dereference(ibp->qp0))
284 qp_inuse++; 297 qp_inuse++;
285 if (ibp->qp1) 298 if (rcu_dereference(ibp->qp1))
286 qp_inuse++; 299 qp_inuse++;
300 rcu_read_unlock();
287 } 301 }
288 302
289 spin_lock_irqsave(&dev->qpt_lock, flags); 303 spin_lock_irqsave(&dev->qpt_lock, flags);
290 for (n = 0; n < dev->qp_table_size; n++) { 304 for (n = 0; n < dev->qp_table_size; n++) {
291 qp = dev->qp_table[n]; 305 qp = dev->qp_table[n];
292 dev->qp_table[n] = NULL; 306 rcu_assign_pointer(dev->qp_table[n], NULL);
293 307
294 for (; qp; qp = qp->next) 308 for (; qp; qp = qp->next)
295 qp_inuse++; 309 qp_inuse++;
296 } 310 }
297 spin_unlock_irqrestore(&dev->qpt_lock, flags); 311 spin_unlock_irqrestore(&dev->qpt_lock, flags);
312 synchronize_rcu();
298 313
299 return qp_inuse; 314 return qp_inuse;
300} 315}
@@ -309,25 +324,28 @@ unsigned qib_free_all_qps(struct qib_devdata *dd)
309 */ 324 */
310struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn) 325struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn)
311{ 326{
312 struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev; 327 struct qib_qp *qp = NULL;
313 unsigned long flags;
314 struct qib_qp *qp;
315 328
316 spin_lock_irqsave(&dev->qpt_lock, flags); 329 if (unlikely(qpn <= 1)) {
330 rcu_read_lock();
331 if (qpn == 0)
332 qp = rcu_dereference(ibp->qp0);
333 else
334 qp = rcu_dereference(ibp->qp1);
335 } else {
336 struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev;
337 unsigned n = qpn_hash(dev, qpn);
317 338
318 if (qpn == 0) 339 rcu_read_lock();
319 qp = ibp->qp0; 340 for (qp = dev->qp_table[n]; rcu_dereference(qp); qp = qp->next)
320 else if (qpn == 1)
321 qp = ibp->qp1;
322 else
323 for (qp = dev->qp_table[qpn % dev->qp_table_size]; qp;
324 qp = qp->next)
325 if (qp->ibqp.qp_num == qpn) 341 if (qp->ibqp.qp_num == qpn)
326 break; 342 break;
343 }
327 if (qp) 344 if (qp)
328 atomic_inc(&qp->refcount); 345 if (unlikely(!atomic_inc_not_zero(&qp->refcount)))
346 qp = NULL;
329 347
330 spin_unlock_irqrestore(&dev->qpt_lock, flags); 348 rcu_read_unlock();
331 return qp; 349 return qp;
332} 350}
333 351
@@ -765,8 +783,10 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
765 } 783 }
766 } 784 }
767 785
768 if (attr_mask & IB_QP_PATH_MTU) 786 if (attr_mask & IB_QP_PATH_MTU) {
769 qp->path_mtu = pmtu; 787 qp->path_mtu = pmtu;
788 qp->pmtu = ib_mtu_enum_to_int(pmtu);
789 }
770 790
771 if (attr_mask & IB_QP_RETRY_CNT) { 791 if (attr_mask & IB_QP_RETRY_CNT) {
772 qp->s_retry_cnt = attr->retry_cnt; 792 qp->s_retry_cnt = attr->retry_cnt;
@@ -781,8 +801,12 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
781 if (attr_mask & IB_QP_MIN_RNR_TIMER) 801 if (attr_mask & IB_QP_MIN_RNR_TIMER)
782 qp->r_min_rnr_timer = attr->min_rnr_timer; 802 qp->r_min_rnr_timer = attr->min_rnr_timer;
783 803
784 if (attr_mask & IB_QP_TIMEOUT) 804 if (attr_mask & IB_QP_TIMEOUT) {
785 qp->timeout = attr->timeout; 805 qp->timeout = attr->timeout;
806 qp->timeout_jiffies =
807 usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
808 1000UL);
809 }
786 810
787 if (attr_mask & IB_QP_QKEY) 811 if (attr_mask & IB_QP_QKEY)
788 qp->qkey = attr->qkey; 812 qp->qkey = attr->qkey;
@@ -1013,6 +1037,10 @@ struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
1013 ret = ERR_PTR(-ENOMEM); 1037 ret = ERR_PTR(-ENOMEM);
1014 goto bail_swq; 1038 goto bail_swq;
1015 } 1039 }
1040 RCU_INIT_POINTER(qp->next, NULL);
1041 qp->timeout_jiffies =
1042 usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
1043 1000UL);
1016 if (init_attr->srq) 1044 if (init_attr->srq)
1017 sz = 0; 1045 sz = 0;
1018 else { 1046 else {
diff --git a/drivers/infiniband/hw/qib/qib_qsfp.c b/drivers/infiniband/hw/qib/qib_qsfp.c
index 3374a52232c1..e06c4ed383f1 100644
--- a/drivers/infiniband/hw/qib/qib_qsfp.c
+++ b/drivers/infiniband/hw/qib/qib_qsfp.c
@@ -273,18 +273,12 @@ int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp)
273 int ret; 273 int ret;
274 int idx; 274 int idx;
275 u16 cks; 275 u16 cks;
276 u32 mask;
277 u8 peek[4]; 276 u8 peek[4];
278 277
279 /* ensure sane contents on invalid reads, for cable swaps */ 278 /* ensure sane contents on invalid reads, for cable swaps */
280 memset(cp, 0, sizeof(*cp)); 279 memset(cp, 0, sizeof(*cp));
281 280
282 mask = QSFP_GPIO_MOD_PRS_N; 281 if (!qib_qsfp_mod_present(ppd)) {
283 if (ppd->hw_pidx)
284 mask <<= QSFP_GPIO_PORT2_SHIFT;
285
286 ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
287 if (ret & mask) {
288 ret = -ENODEV; 282 ret = -ENODEV;
289 goto bail; 283 goto bail;
290 } 284 }
@@ -444,6 +438,19 @@ const char * const qib_qsfp_devtech[16] = {
444 438
445static const char *pwr_codes = "1.5W2.0W2.5W3.5W"; 439static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
446 440
441int qib_qsfp_mod_present(struct qib_pportdata *ppd)
442{
443 u32 mask;
444 int ret;
445
446 mask = QSFP_GPIO_MOD_PRS_N <<
447 (ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT);
448 ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
449
450 return !((ret & mask) >>
451 ((ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT) + 3));
452}
453
447/* 454/*
448 * Initialize structures that control access to QSFP. Called once per port 455 * Initialize structures that control access to QSFP. Called once per port
449 * on cards that support QSFP. 456 * on cards that support QSFP.
@@ -452,7 +459,6 @@ void qib_qsfp_init(struct qib_qsfp_data *qd,
452 void (*fevent)(struct work_struct *)) 459 void (*fevent)(struct work_struct *))
453{ 460{
454 u32 mask, highs; 461 u32 mask, highs;
455 int pins;
456 462
457 struct qib_devdata *dd = qd->ppd->dd; 463 struct qib_devdata *dd = qd->ppd->dd;
458 464
@@ -480,8 +486,7 @@ void qib_qsfp_init(struct qib_qsfp_data *qd,
480 mask <<= QSFP_GPIO_PORT2_SHIFT; 486 mask <<= QSFP_GPIO_PORT2_SHIFT;
481 487
482 /* Do not try to wait here. Better to let event handle it */ 488 /* Do not try to wait here. Better to let event handle it */
483 pins = dd->f_gpio_mod(dd, 0, 0, 0); 489 if (!qib_qsfp_mod_present(qd->ppd))
484 if (pins & mask)
485 goto bail; 490 goto bail;
486 /* We see a module, but it may be unwise to look yet. Just schedule */ 491 /* We see a module, but it may be unwise to look yet. Just schedule */
487 qd->t_insert = get_jiffies_64(); 492 qd->t_insert = get_jiffies_64();
diff --git a/drivers/infiniband/hw/qib/qib_qsfp.h b/drivers/infiniband/hw/qib/qib_qsfp.h
index c109bbdc90ac..46002a9417c0 100644
--- a/drivers/infiniband/hw/qib/qib_qsfp.h
+++ b/drivers/infiniband/hw/qib/qib_qsfp.h
@@ -34,6 +34,7 @@
34 34
35#define QSFP_DEV 0xA0 35#define QSFP_DEV 0xA0
36#define QSFP_PWR_LAG_MSEC 2000 36#define QSFP_PWR_LAG_MSEC 2000
37#define QSFP_MODPRS_LAG_MSEC 20
37 38
38/* 39/*
39 * Below are masks for various QSFP signals, for Port 1. 40 * Below are masks for various QSFP signals, for Port 1.
@@ -177,10 +178,12 @@ struct qib_qsfp_data {
177 struct work_struct work; 178 struct work_struct work;
178 struct qib_qsfp_cache cache; 179 struct qib_qsfp_cache cache;
179 u64 t_insert; 180 u64 t_insert;
181 u8 modpresent;
180}; 182};
181 183
182extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, 184extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd,
183 struct qib_qsfp_cache *cp); 185 struct qib_qsfp_cache *cp);
186extern int qib_qsfp_mod_present(struct qib_pportdata *ppd);
184extern void qib_qsfp_init(struct qib_qsfp_data *qd, 187extern void qib_qsfp_init(struct qib_qsfp_data *qd,
185 void (*fevent)(struct work_struct *)); 188 void (*fevent)(struct work_struct *));
186extern void qib_qsfp_deinit(struct qib_qsfp_data *qd); 189extern void qib_qsfp_deinit(struct qib_qsfp_data *qd);
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index eca0c41f1226..afaf4ac79f42 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -59,8 +59,7 @@ static void start_timer(struct qib_qp *qp)
59 qp->s_flags |= QIB_S_TIMER; 59 qp->s_flags |= QIB_S_TIMER;
60 qp->s_timer.function = rc_timeout; 60 qp->s_timer.function = rc_timeout;
61 /* 4.096 usec. * (1 << qp->timeout) */ 61 /* 4.096 usec. * (1 << qp->timeout) */
62 qp->s_timer.expires = jiffies + 62 qp->s_timer.expires = jiffies + qp->timeout_jiffies;
63 usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / 1000UL);
64 add_timer(&qp->s_timer); 63 add_timer(&qp->s_timer);
65} 64}
66 65
@@ -239,7 +238,7 @@ int qib_make_rc_req(struct qib_qp *qp)
239 u32 len; 238 u32 len;
240 u32 bth0; 239 u32 bth0;
241 u32 bth2; 240 u32 bth2;
242 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); 241 u32 pmtu = qp->pmtu;
243 char newreq; 242 char newreq;
244 unsigned long flags; 243 unsigned long flags;
245 int ret = 0; 244 int ret = 0;
@@ -1519,9 +1518,7 @@ read_middle:
1519 * 4.096 usec. * (1 << qp->timeout) 1518 * 4.096 usec. * (1 << qp->timeout)
1520 */ 1519 */
1521 qp->s_flags |= QIB_S_TIMER; 1520 qp->s_flags |= QIB_S_TIMER;
1522 mod_timer(&qp->s_timer, jiffies + 1521 mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies);
1523 usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
1524 1000UL));
1525 if (qp->s_flags & QIB_S_WAIT_ACK) { 1522 if (qp->s_flags & QIB_S_WAIT_ACK) {
1526 qp->s_flags &= ~QIB_S_WAIT_ACK; 1523 qp->s_flags &= ~QIB_S_WAIT_ACK;
1527 qib_schedule_send(qp); 1524 qib_schedule_send(qp);
@@ -1732,7 +1729,7 @@ static int qib_rc_rcv_error(struct qib_other_headers *ohdr,
1732 * same request. 1729 * same request.
1733 */ 1730 */
1734 offset = ((psn - e->psn) & QIB_PSN_MASK) * 1731 offset = ((psn - e->psn) & QIB_PSN_MASK) *
1735 ib_mtu_enum_to_int(qp->path_mtu); 1732 qp->pmtu;
1736 len = be32_to_cpu(reth->length); 1733 len = be32_to_cpu(reth->length);
1737 if (unlikely(offset + len != e->rdma_sge.sge_length)) 1734 if (unlikely(offset + len != e->rdma_sge.sge_length))
1738 goto unlock_done; 1735 goto unlock_done;
@@ -1876,7 +1873,7 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
1876 u32 psn; 1873 u32 psn;
1877 u32 pad; 1874 u32 pad;
1878 struct ib_wc wc; 1875 struct ib_wc wc;
1879 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); 1876 u32 pmtu = qp->pmtu;
1880 int diff; 1877 int diff;
1881 struct ib_reth *reth; 1878 struct ib_reth *reth;
1882 unsigned long flags; 1879 unsigned long flags;
@@ -1892,10 +1889,8 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
1892 } 1889 }
1893 1890
1894 opcode = be32_to_cpu(ohdr->bth[0]); 1891 opcode = be32_to_cpu(ohdr->bth[0]);
1895 spin_lock_irqsave(&qp->s_lock, flags);
1896 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode)) 1892 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
1897 goto sunlock; 1893 return;
1898 spin_unlock_irqrestore(&qp->s_lock, flags);
1899 1894
1900 psn = be32_to_cpu(ohdr->bth[2]); 1895 psn = be32_to_cpu(ohdr->bth[2]);
1901 opcode >>= 24; 1896 opcode >>= 24;
@@ -1955,8 +1950,6 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
1955 break; 1950 break;
1956 } 1951 }
1957 1952
1958 memset(&wc, 0, sizeof wc);
1959
1960 if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) { 1953 if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
1961 qp->r_flags |= QIB_R_COMM_EST; 1954 qp->r_flags |= QIB_R_COMM_EST;
1962 if (qp->ibqp.event_handler) { 1955 if (qp->ibqp.event_handler) {
@@ -2009,16 +2002,19 @@ send_middle:
2009 goto rnr_nak; 2002 goto rnr_nak;
2010 qp->r_rcv_len = 0; 2003 qp->r_rcv_len = 0;
2011 if (opcode == OP(SEND_ONLY)) 2004 if (opcode == OP(SEND_ONLY))
2012 goto send_last; 2005 goto no_immediate_data;
2013 /* FALLTHROUGH */ 2006 /* FALLTHROUGH for SEND_ONLY_WITH_IMMEDIATE */
2014 case OP(SEND_LAST_WITH_IMMEDIATE): 2007 case OP(SEND_LAST_WITH_IMMEDIATE):
2015send_last_imm: 2008send_last_imm:
2016 wc.ex.imm_data = ohdr->u.imm_data; 2009 wc.ex.imm_data = ohdr->u.imm_data;
2017 hdrsize += 4; 2010 hdrsize += 4;
2018 wc.wc_flags = IB_WC_WITH_IMM; 2011 wc.wc_flags = IB_WC_WITH_IMM;
2019 /* FALLTHROUGH */ 2012 goto send_last;
2020 case OP(SEND_LAST): 2013 case OP(SEND_LAST):
2021 case OP(RDMA_WRITE_LAST): 2014 case OP(RDMA_WRITE_LAST):
2015no_immediate_data:
2016 wc.wc_flags = 0;
2017 wc.ex.imm_data = 0;
2022send_last: 2018send_last:
2023 /* Get the number of bytes the message was padded by. */ 2019 /* Get the number of bytes the message was padded by. */
2024 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3; 2020 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
@@ -2051,6 +2047,12 @@ send_last:
2051 wc.src_qp = qp->remote_qpn; 2047 wc.src_qp = qp->remote_qpn;
2052 wc.slid = qp->remote_ah_attr.dlid; 2048 wc.slid = qp->remote_ah_attr.dlid;
2053 wc.sl = qp->remote_ah_attr.sl; 2049 wc.sl = qp->remote_ah_attr.sl;
2050 /* zero fields that are N/A */
2051 wc.vendor_err = 0;
2052 wc.pkey_index = 0;
2053 wc.dlid_path_bits = 0;
2054 wc.port_num = 0;
2055 wc.csum_ok = 0;
2054 /* Signal completion event if the solicited bit is set. */ 2056 /* Signal completion event if the solicited bit is set. */
2055 qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 2057 qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
2056 (ohdr->bth[0] & 2058 (ohdr->bth[0] &
@@ -2089,7 +2091,7 @@ send_last:
2089 if (opcode == OP(RDMA_WRITE_FIRST)) 2091 if (opcode == OP(RDMA_WRITE_FIRST))
2090 goto send_middle; 2092 goto send_middle;
2091 else if (opcode == OP(RDMA_WRITE_ONLY)) 2093 else if (opcode == OP(RDMA_WRITE_ONLY))
2092 goto send_last; 2094 goto no_immediate_data;
2093 ret = qib_get_rwqe(qp, 1); 2095 ret = qib_get_rwqe(qp, 1);
2094 if (ret < 0) 2096 if (ret < 0)
2095 goto nack_op_err; 2097 goto nack_op_err;
diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c
index eb78d9367f06..b4b37e47321a 100644
--- a/drivers/infiniband/hw/qib/qib_ruc.c
+++ b/drivers/infiniband/hw/qib/qib_ruc.c
@@ -260,12 +260,15 @@ static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id)
260 260
261/* 261/*
262 * 262 *
263 * This should be called with the QP s_lock held. 263 * This should be called with the QP r_lock held.
264 *
265 * The s_lock will be acquired around the qib_migrate_qp() call.
264 */ 266 */
265int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr, 267int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
266 int has_grh, struct qib_qp *qp, u32 bth0) 268 int has_grh, struct qib_qp *qp, u32 bth0)
267{ 269{
268 __be64 guid; 270 __be64 guid;
271 unsigned long flags;
269 272
270 if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) { 273 if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
271 if (!has_grh) { 274 if (!has_grh) {
@@ -295,7 +298,9 @@ int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
295 if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid || 298 if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid ||
296 ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num) 299 ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num)
297 goto err; 300 goto err;
301 spin_lock_irqsave(&qp->s_lock, flags);
298 qib_migrate_qp(qp); 302 qib_migrate_qp(qp);
303 spin_unlock_irqrestore(&qp->s_lock, flags);
299 } else { 304 } else {
300 if (!has_grh) { 305 if (!has_grh) {
301 if (qp->remote_ah_attr.ah_flags & IB_AH_GRH) 306 if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
diff --git a/drivers/infiniband/hw/qib/qib_srq.c b/drivers/infiniband/hw/qib/qib_srq.c
index c3ec8efc2ed8..d6235931a1ba 100644
--- a/drivers/infiniband/hw/qib/qib_srq.c
+++ b/drivers/infiniband/hw/qib/qib_srq.c
@@ -107,6 +107,11 @@ struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
107 u32 sz; 107 u32 sz;
108 struct ib_srq *ret; 108 struct ib_srq *ret;
109 109
110 if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
111 ret = ERR_PTR(-ENOSYS);
112 goto done;
113 }
114
110 if (srq_init_attr->attr.max_sge == 0 || 115 if (srq_init_attr->attr.max_sge == 0 ||
111 srq_init_attr->attr.max_sge > ib_qib_max_srq_sges || 116 srq_init_attr->attr.max_sge > ib_qib_max_srq_sges ||
112 srq_init_attr->attr.max_wr == 0 || 117 srq_init_attr->attr.max_wr == 0 ||
diff --git a/drivers/infiniband/hw/qib/qib_sysfs.c b/drivers/infiniband/hw/qib/qib_sysfs.c
index 14d129de4320..78fbd56879d4 100644
--- a/drivers/infiniband/hw/qib/qib_sysfs.c
+++ b/drivers/infiniband/hw/qib/qib_sysfs.c
@@ -515,8 +515,7 @@ static ssize_t show_nfreectxts(struct device *device,
515 struct qib_devdata *dd = dd_from_dev(dev); 515 struct qib_devdata *dd = dd_from_dev(dev);
516 516
517 /* Return the number of free user ports (contexts) available. */ 517 /* Return the number of free user ports (contexts) available. */
518 return scnprintf(buf, PAGE_SIZE, "%u\n", dd->cfgctxts - 518 return scnprintf(buf, PAGE_SIZE, "%u\n", dd->freectxts);
519 dd->first_user_ctxt - (u32)qib_stats.sps_ctxts);
520} 519}
521 520
522static ssize_t show_serial(struct device *device, 521static ssize_t show_serial(struct device *device,
diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c
index 32ccf3c824ca..847e7afdfd94 100644
--- a/drivers/infiniband/hw/qib/qib_uc.c
+++ b/drivers/infiniband/hw/qib/qib_uc.c
@@ -51,7 +51,7 @@ int qib_make_uc_req(struct qib_qp *qp)
51 u32 hwords; 51 u32 hwords;
52 u32 bth0; 52 u32 bth0;
53 u32 len; 53 u32 len;
54 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); 54 u32 pmtu = qp->pmtu;
55 int ret = 0; 55 int ret = 0;
56 56
57 spin_lock_irqsave(&qp->s_lock, flags); 57 spin_lock_irqsave(&qp->s_lock, flags);
@@ -243,13 +243,12 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
243 int has_grh, void *data, u32 tlen, struct qib_qp *qp) 243 int has_grh, void *data, u32 tlen, struct qib_qp *qp)
244{ 244{
245 struct qib_other_headers *ohdr; 245 struct qib_other_headers *ohdr;
246 unsigned long flags;
247 u32 opcode; 246 u32 opcode;
248 u32 hdrsize; 247 u32 hdrsize;
249 u32 psn; 248 u32 psn;
250 u32 pad; 249 u32 pad;
251 struct ib_wc wc; 250 struct ib_wc wc;
252 u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu); 251 u32 pmtu = qp->pmtu;
253 struct ib_reth *reth; 252 struct ib_reth *reth;
254 int ret; 253 int ret;
255 254
@@ -263,14 +262,11 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
263 } 262 }
264 263
265 opcode = be32_to_cpu(ohdr->bth[0]); 264 opcode = be32_to_cpu(ohdr->bth[0]);
266 spin_lock_irqsave(&qp->s_lock, flags);
267 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode)) 265 if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
268 goto sunlock; 266 return;
269 spin_unlock_irqrestore(&qp->s_lock, flags);
270 267
271 psn = be32_to_cpu(ohdr->bth[2]); 268 psn = be32_to_cpu(ohdr->bth[2]);
272 opcode >>= 24; 269 opcode >>= 24;
273 memset(&wc, 0, sizeof wc);
274 270
275 /* Compare the PSN verses the expected PSN. */ 271 /* Compare the PSN verses the expected PSN. */
276 if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) { 272 if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
@@ -370,7 +366,7 @@ send_first:
370 } 366 }
371 qp->r_rcv_len = 0; 367 qp->r_rcv_len = 0;
372 if (opcode == OP(SEND_ONLY)) 368 if (opcode == OP(SEND_ONLY))
373 goto send_last; 369 goto no_immediate_data;
374 else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE)) 370 else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
375 goto send_last_imm; 371 goto send_last_imm;
376 /* FALLTHROUGH */ 372 /* FALLTHROUGH */
@@ -389,8 +385,11 @@ send_last_imm:
389 wc.ex.imm_data = ohdr->u.imm_data; 385 wc.ex.imm_data = ohdr->u.imm_data;
390 hdrsize += 4; 386 hdrsize += 4;
391 wc.wc_flags = IB_WC_WITH_IMM; 387 wc.wc_flags = IB_WC_WITH_IMM;
392 /* FALLTHROUGH */ 388 goto send_last;
393 case OP(SEND_LAST): 389 case OP(SEND_LAST):
390no_immediate_data:
391 wc.ex.imm_data = 0;
392 wc.wc_flags = 0;
394send_last: 393send_last:
395 /* Get the number of bytes the message was padded by. */ 394 /* Get the number of bytes the message was padded by. */
396 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3; 395 pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
@@ -418,6 +417,12 @@ last_imm:
418 wc.src_qp = qp->remote_qpn; 417 wc.src_qp = qp->remote_qpn;
419 wc.slid = qp->remote_ah_attr.dlid; 418 wc.slid = qp->remote_ah_attr.dlid;
420 wc.sl = qp->remote_ah_attr.sl; 419 wc.sl = qp->remote_ah_attr.sl;
420 /* zero fields that are N/A */
421 wc.vendor_err = 0;
422 wc.pkey_index = 0;
423 wc.dlid_path_bits = 0;
424 wc.port_num = 0;
425 wc.csum_ok = 0;
421 /* Signal completion event if the solicited bit is set. */ 426 /* Signal completion event if the solicited bit is set. */
422 qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 427 qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
423 (ohdr->bth[0] & 428 (ohdr->bth[0] &
@@ -546,6 +551,4 @@ op_err:
546 qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR); 551 qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
547 return; 552 return;
548 553
549sunlock:
550 spin_unlock_irqrestore(&qp->s_lock, flags);
551} 554}
diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c
index 9fab40488850..9627cb737125 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.c
+++ b/drivers/infiniband/hw/qib/qib_verbs.c
@@ -38,11 +38,12 @@
38#include <linux/utsname.h> 38#include <linux/utsname.h>
39#include <linux/rculist.h> 39#include <linux/rculist.h>
40#include <linux/mm.h> 40#include <linux/mm.h>
41#include <linux/random.h>
41 42
42#include "qib.h" 43#include "qib.h"
43#include "qib_common.h" 44#include "qib_common.h"
44 45
45static unsigned int ib_qib_qp_table_size = 251; 46static unsigned int ib_qib_qp_table_size = 256;
46module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO); 47module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO);
47MODULE_PARM_DESC(qp_table_size, "QP table size"); 48MODULE_PARM_DESC(qp_table_size, "QP table size");
48 49
@@ -659,17 +660,25 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
659 if (atomic_dec_return(&mcast->refcount) <= 1) 660 if (atomic_dec_return(&mcast->refcount) <= 1)
660 wake_up(&mcast->wait); 661 wake_up(&mcast->wait);
661 } else { 662 } else {
662 qp = qib_lookup_qpn(ibp, qp_num); 663 if (rcd->lookaside_qp) {
663 if (!qp) 664 if (rcd->lookaside_qpn != qp_num) {
664 goto drop; 665 if (atomic_dec_and_test(
666 &rcd->lookaside_qp->refcount))
667 wake_up(
668 &rcd->lookaside_qp->wait);
669 rcd->lookaside_qp = NULL;
670 }
671 }
672 if (!rcd->lookaside_qp) {
673 qp = qib_lookup_qpn(ibp, qp_num);
674 if (!qp)
675 goto drop;
676 rcd->lookaside_qp = qp;
677 rcd->lookaside_qpn = qp_num;
678 } else
679 qp = rcd->lookaside_qp;
665 ibp->n_unicast_rcv++; 680 ibp->n_unicast_rcv++;
666 qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp); 681 qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp);
667 /*
668 * Notify qib_destroy_qp() if it is waiting
669 * for us to finish.
670 */
671 if (atomic_dec_and_test(&qp->refcount))
672 wake_up(&qp->wait);
673 } 682 }
674 return; 683 return;
675 684
@@ -1974,6 +1983,8 @@ static void init_ibport(struct qib_pportdata *ppd)
1974 ibp->z_excessive_buffer_overrun_errors = 1983 ibp->z_excessive_buffer_overrun_errors =
1975 cntrs.excessive_buffer_overrun_errors; 1984 cntrs.excessive_buffer_overrun_errors;
1976 ibp->z_vl15_dropped = cntrs.vl15_dropped; 1985 ibp->z_vl15_dropped = cntrs.vl15_dropped;
1986 RCU_INIT_POINTER(ibp->qp0, NULL);
1987 RCU_INIT_POINTER(ibp->qp1, NULL);
1977} 1988}
1978 1989
1979/** 1990/**
@@ -1990,12 +2001,15 @@ int qib_register_ib_device(struct qib_devdata *dd)
1990 int ret; 2001 int ret;
1991 2002
1992 dev->qp_table_size = ib_qib_qp_table_size; 2003 dev->qp_table_size = ib_qib_qp_table_size;
1993 dev->qp_table = kzalloc(dev->qp_table_size * sizeof *dev->qp_table, 2004 get_random_bytes(&dev->qp_rnd, sizeof(dev->qp_rnd));
2005 dev->qp_table = kmalloc(dev->qp_table_size * sizeof *dev->qp_table,
1994 GFP_KERNEL); 2006 GFP_KERNEL);
1995 if (!dev->qp_table) { 2007 if (!dev->qp_table) {
1996 ret = -ENOMEM; 2008 ret = -ENOMEM;
1997 goto err_qpt; 2009 goto err_qpt;
1998 } 2010 }
2011 for (i = 0; i < dev->qp_table_size; i++)
2012 RCU_INIT_POINTER(dev->qp_table[i], NULL);
1999 2013
2000 for (i = 0; i < dd->num_pports; i++) 2014 for (i = 0; i < dd->num_pports; i++)
2001 init_ibport(ppd + i); 2015 init_ibport(ppd + i);
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
index 95e5b47223b3..0c19ef0c4123 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.h
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -485,6 +485,7 @@ struct qib_qp {
485 u8 alt_timeout; /* Alternate path timeout for this QP */ 485 u8 alt_timeout; /* Alternate path timeout for this QP */
486 u8 port_num; 486 u8 port_num;
487 enum ib_mtu path_mtu; 487 enum ib_mtu path_mtu;
488 u32 pmtu; /* decoded from path_mtu */
488 u32 remote_qpn; 489 u32 remote_qpn;
489 u32 qkey; /* QKEY for this QP (for UD or RD) */ 490 u32 qkey; /* QKEY for this QP (for UD or RD) */
490 u32 s_size; /* send work queue size */ 491 u32 s_size; /* send work queue size */
@@ -495,6 +496,7 @@ struct qib_qp {
495 u32 s_last; /* last completed entry */ 496 u32 s_last; /* last completed entry */
496 u32 s_ssn; /* SSN of tail entry */ 497 u32 s_ssn; /* SSN of tail entry */
497 u32 s_lsn; /* limit sequence number (credit) */ 498 u32 s_lsn; /* limit sequence number (credit) */
499 unsigned long timeout_jiffies; /* computed from timeout */
498 struct qib_swqe *s_wq; /* send work queue */ 500 struct qib_swqe *s_wq; /* send work queue */
499 struct qib_swqe *s_wqe; 501 struct qib_swqe *s_wqe;
500 struct qib_rq r_rq; /* receive work queue */ 502 struct qib_rq r_rq; /* receive work queue */
@@ -723,7 +725,8 @@ struct qib_ibdev {
723 dma_addr_t pio_hdrs_phys; 725 dma_addr_t pio_hdrs_phys;
724 /* list of QPs waiting for RNR timer */ 726 /* list of QPs waiting for RNR timer */
725 spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */ 727 spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
726 unsigned qp_table_size; /* size of the hash table */ 728 u32 qp_table_size; /* size of the hash table */
729 u32 qp_rnd; /* random bytes for hash */
727 spinlock_t qpt_lock; 730 spinlock_t qpt_lock;
728 731
729 u32 n_piowait; 732 u32 n_piowait;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index 39913a065f99..fe48677fd748 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -84,7 +84,7 @@ static void ipoib_cm_dma_unmap_rx(struct ipoib_dev_priv *priv, int frags,
84 ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE); 84 ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE);
85 85
86 for (i = 0; i < frags; ++i) 86 for (i = 0; i < frags; ++i)
87 ib_dma_unmap_single(priv->ca, mapping[i + 1], PAGE_SIZE, DMA_FROM_DEVICE); 87 ib_dma_unmap_page(priv->ca, mapping[i + 1], PAGE_SIZE, DMA_FROM_DEVICE);
88} 88}
89 89
90static int ipoib_cm_post_receive_srq(struct net_device *dev, int id) 90static int ipoib_cm_post_receive_srq(struct net_device *dev, int id)
@@ -183,7 +183,7 @@ partial_error:
183 ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE); 183 ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE);
184 184
185 for (; i > 0; --i) 185 for (; i > 0; --i)
186 ib_dma_unmap_single(priv->ca, mapping[i], PAGE_SIZE, DMA_FROM_DEVICE); 186 ib_dma_unmap_page(priv->ca, mapping[i], PAGE_SIZE, DMA_FROM_DEVICE);
187 187
188 dev_kfree_skb_any(skb); 188 dev_kfree_skb_any(skb);
189 return NULL; 189 return NULL;
@@ -1496,6 +1496,7 @@ static void ipoib_cm_create_srq(struct net_device *dev, int max_sge)
1496{ 1496{
1497 struct ipoib_dev_priv *priv = netdev_priv(dev); 1497 struct ipoib_dev_priv *priv = netdev_priv(dev);
1498 struct ib_srq_init_attr srq_init_attr = { 1498 struct ib_srq_init_attr srq_init_attr = {
1499 .srq_type = IB_SRQT_BASIC,
1499 .attr = { 1500 .attr = {
1500 .max_wr = ipoib_recvq_size, 1501 .max_wr = ipoib_recvq_size,
1501 .max_sge = max_sge 1502 .max_sge = max_sge
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_fs.c b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
index 86eae229dc49..0e2fe4631ba8 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_fs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
@@ -212,16 +212,15 @@ static int ipoib_path_seq_show(struct seq_file *file, void *iter_ptr)
212 gid_buf, path.pathrec.dlid ? "yes" : "no"); 212 gid_buf, path.pathrec.dlid ? "yes" : "no");
213 213
214 if (path.pathrec.dlid) { 214 if (path.pathrec.dlid) {
215 rate = ib_rate_to_mult(path.pathrec.rate) * 25; 215 rate = ib_rate_to_mbps(path.pathrec.rate);
216 216
217 seq_printf(file, 217 seq_printf(file,
218 " DLID: 0x%04x\n" 218 " DLID: 0x%04x\n"
219 " SL: %12d\n" 219 " SL: %12d\n"
220 " rate: %*d%s Gb/sec\n", 220 " rate: %8d.%d Gb/sec\n",
221 be16_to_cpu(path.pathrec.dlid), 221 be16_to_cpu(path.pathrec.dlid),
222 path.pathrec.sl, 222 path.pathrec.sl,
223 10 - ((rate % 10) ? 2 : 0), 223 rate / 1000, rate % 1000);
224 rate / 10, rate % 10 ? ".5" : "");
225 } 224 }
226 225
227 seq_putc(file, '\n'); 226 seq_putc(file, '\n');
diff --git a/drivers/net/mlx4/eq.c b/drivers/net/mlx4/eq.c
index 1ad1f6029af8..869a2c220a7b 100644
--- a/drivers/net/mlx4/eq.c
+++ b/drivers/net/mlx4/eq.c
@@ -484,7 +484,7 @@ static void mlx4_free_eq(struct mlx4_dev *dev,
484 484
485 mlx4_mtt_cleanup(dev, &eq->mtt); 485 mlx4_mtt_cleanup(dev, &eq->mtt);
486 for (i = 0; i < npages; ++i) 486 for (i = 0; i < npages; ++i)
487 pci_free_consistent(dev->pdev, PAGE_SIZE, 487 dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
488 eq->page_list[i].buf, 488 eq->page_list[i].buf,
489 eq->page_list[i].map); 489 eq->page_list[i].map);
490 490
diff --git a/drivers/net/mlx4/fw.c b/drivers/net/mlx4/fw.c
index 7eb8ba822e97..875838b8799c 100644
--- a/drivers/net/mlx4/fw.c
+++ b/drivers/net/mlx4/fw.c
@@ -204,6 +204,8 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
204#define QUERY_DEV_CAP_MAX_MCG_OFFSET 0x63 204#define QUERY_DEV_CAP_MAX_MCG_OFFSET 0x63
205#define QUERY_DEV_CAP_RSVD_PD_OFFSET 0x64 205#define QUERY_DEV_CAP_RSVD_PD_OFFSET 0x64
206#define QUERY_DEV_CAP_MAX_PD_OFFSET 0x65 206#define QUERY_DEV_CAP_MAX_PD_OFFSET 0x65
207#define QUERY_DEV_CAP_RSVD_XRC_OFFSET 0x66
208#define QUERY_DEV_CAP_MAX_XRC_OFFSET 0x67
207#define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET 0x68 209#define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET 0x68
208#define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET 0x80 210#define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET 0x80
209#define QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET 0x82 211#define QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET 0x82
@@ -318,6 +320,10 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
318 dev_cap->reserved_pds = field >> 4; 320 dev_cap->reserved_pds = field >> 4;
319 MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET); 321 MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
320 dev_cap->max_pds = 1 << (field & 0x3f); 322 dev_cap->max_pds = 1 << (field & 0x3f);
323 MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_XRC_OFFSET);
324 dev_cap->reserved_xrcds = field >> 4;
325 MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
326 dev_cap->max_xrcds = 1 << (field & 0x1f);
321 327
322 MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET); 328 MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET);
323 dev_cap->rdmarc_entry_sz = size; 329 dev_cap->rdmarc_entry_sz = size;
diff --git a/drivers/net/mlx4/fw.h b/drivers/net/mlx4/fw.h
index 1e8ecc3708e2..bf5ec2286528 100644
--- a/drivers/net/mlx4/fw.h
+++ b/drivers/net/mlx4/fw.h
@@ -93,6 +93,8 @@ struct mlx4_dev_cap {
93 int max_mcgs; 93 int max_mcgs;
94 int reserved_pds; 94 int reserved_pds;
95 int max_pds; 95 int max_pds;
96 int reserved_xrcds;
97 int max_xrcds;
96 int qpc_entry_sz; 98 int qpc_entry_sz;
97 int rdmarc_entry_sz; 99 int rdmarc_entry_sz;
98 int altc_entry_sz; 100 int altc_entry_sz;
diff --git a/drivers/net/mlx4/main.c b/drivers/net/mlx4/main.c
index f0ee35df4dd7..94bbc85a532d 100644
--- a/drivers/net/mlx4/main.c
+++ b/drivers/net/mlx4/main.c
@@ -96,6 +96,8 @@ MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
96static int log_num_vlan; 96static int log_num_vlan;
97module_param_named(log_num_vlan, log_num_vlan, int, 0444); 97module_param_named(log_num_vlan, log_num_vlan, int, 0444);
98MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)"); 98MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
99/* Log2 max number of VLANs per ETH port (0-7) */
100#define MLX4_LOG_NUM_VLANS 7
99 101
100static int use_prio; 102static int use_prio;
101module_param_named(use_prio, use_prio, bool, 0444); 103module_param_named(use_prio, use_prio, bool, 0444);
@@ -220,6 +222,10 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
220 dev->caps.reserved_mrws = dev_cap->reserved_mrws; 222 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
221 dev->caps.reserved_uars = dev_cap->reserved_uars; 223 dev->caps.reserved_uars = dev_cap->reserved_uars;
222 dev->caps.reserved_pds = dev_cap->reserved_pds; 224 dev->caps.reserved_pds = dev_cap->reserved_pds;
225 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
226 dev_cap->reserved_xrcds : 0;
227 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
228 dev_cap->max_xrcds : 0;
223 dev->caps.mtt_entry_sz = dev->caps.mtts_per_seg * dev_cap->mtt_entry_sz; 229 dev->caps.mtt_entry_sz = dev->caps.mtts_per_seg * dev_cap->mtt_entry_sz;
224 dev->caps.max_msg_sz = dev_cap->max_msg_sz; 230 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
225 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1); 231 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
@@ -230,7 +236,7 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
230 dev->caps.max_gso_sz = dev_cap->max_gso_sz; 236 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
231 237
232 dev->caps.log_num_macs = log_num_mac; 238 dev->caps.log_num_macs = log_num_mac;
233 dev->caps.log_num_vlans = log_num_vlan; 239 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
234 dev->caps.log_num_prios = use_prio ? 3 : 0; 240 dev->caps.log_num_prios = use_prio ? 3 : 0;
235 241
236 for (i = 1; i <= dev->caps.num_ports; ++i) { 242 for (i = 1; i <= dev->caps.num_ports; ++i) {
@@ -912,11 +918,18 @@ static int mlx4_setup_hca(struct mlx4_dev *dev)
912 goto err_kar_unmap; 918 goto err_kar_unmap;
913 } 919 }
914 920
921 err = mlx4_init_xrcd_table(dev);
922 if (err) {
923 mlx4_err(dev, "Failed to initialize "
924 "reliable connection domain table, aborting.\n");
925 goto err_pd_table_free;
926 }
927
915 err = mlx4_init_mr_table(dev); 928 err = mlx4_init_mr_table(dev);
916 if (err) { 929 if (err) {
917 mlx4_err(dev, "Failed to initialize " 930 mlx4_err(dev, "Failed to initialize "
918 "memory region table, aborting.\n"); 931 "memory region table, aborting.\n");
919 goto err_pd_table_free; 932 goto err_xrcd_table_free;
920 } 933 }
921 934
922 err = mlx4_init_eq_table(dev); 935 err = mlx4_init_eq_table(dev);
@@ -998,6 +1011,13 @@ static int mlx4_setup_hca(struct mlx4_dev *dev)
998 "ib capabilities (%d). Continuing with " 1011 "ib capabilities (%d). Continuing with "
999 "caps = 0\n", port, err); 1012 "caps = 0\n", port, err);
1000 dev->caps.ib_port_def_cap[port] = ib_port_default_caps; 1013 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1014
1015 err = mlx4_check_ext_port_caps(dev, port);
1016 if (err)
1017 mlx4_warn(dev, "failed to get port %d extended "
1018 "port capabilities support info (%d)."
1019 " Assuming not supported\n", port, err);
1020
1001 err = mlx4_SET_PORT(dev, port); 1021 err = mlx4_SET_PORT(dev, port);
1002 if (err) { 1022 if (err) {
1003 mlx4_err(dev, "Failed to set port %d, aborting\n", 1023 mlx4_err(dev, "Failed to set port %d, aborting\n",
@@ -1033,6 +1053,9 @@ err_eq_table_free:
1033err_mr_table_free: 1053err_mr_table_free:
1034 mlx4_cleanup_mr_table(dev); 1054 mlx4_cleanup_mr_table(dev);
1035 1055
1056err_xrcd_table_free:
1057 mlx4_cleanup_xrcd_table(dev);
1058
1036err_pd_table_free: 1059err_pd_table_free:
1037 mlx4_cleanup_pd_table(dev); 1060 mlx4_cleanup_pd_table(dev);
1038 1061
@@ -1355,6 +1378,7 @@ err_port:
1355 mlx4_cmd_use_polling(dev); 1378 mlx4_cmd_use_polling(dev);
1356 mlx4_cleanup_eq_table(dev); 1379 mlx4_cleanup_eq_table(dev);
1357 mlx4_cleanup_mr_table(dev); 1380 mlx4_cleanup_mr_table(dev);
1381 mlx4_cleanup_xrcd_table(dev);
1358 mlx4_cleanup_pd_table(dev); 1382 mlx4_cleanup_pd_table(dev);
1359 mlx4_cleanup_uar_table(dev); 1383 mlx4_cleanup_uar_table(dev);
1360 1384
@@ -1416,6 +1440,7 @@ static void mlx4_remove_one(struct pci_dev *pdev)
1416 mlx4_cmd_use_polling(dev); 1440 mlx4_cmd_use_polling(dev);
1417 mlx4_cleanup_eq_table(dev); 1441 mlx4_cleanup_eq_table(dev);
1418 mlx4_cleanup_mr_table(dev); 1442 mlx4_cleanup_mr_table(dev);
1443 mlx4_cleanup_xrcd_table(dev);
1419 mlx4_cleanup_pd_table(dev); 1444 mlx4_cleanup_pd_table(dev);
1420 1445
1421 iounmap(priv->kar); 1446 iounmap(priv->kar);
@@ -1489,10 +1514,9 @@ static int __init mlx4_verify_params(void)
1489 return -1; 1514 return -1;
1490 } 1515 }
1491 1516
1492 if ((log_num_vlan < 0) || (log_num_vlan > 7)) { 1517 if (log_num_vlan != 0)
1493 pr_warning("mlx4_core: bad num_vlan: %d\n", log_num_vlan); 1518 pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
1494 return -1; 1519 MLX4_LOG_NUM_VLANS);
1495 }
1496 1520
1497 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) { 1521 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
1498 pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg); 1522 pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
diff --git a/drivers/net/mlx4/mlx4.h b/drivers/net/mlx4/mlx4.h
index a2fcd8402d37..5dfa68ffc11c 100644
--- a/drivers/net/mlx4/mlx4.h
+++ b/drivers/net/mlx4/mlx4.h
@@ -335,6 +335,7 @@ struct mlx4_priv {
335 struct mlx4_cmd cmd; 335 struct mlx4_cmd cmd;
336 336
337 struct mlx4_bitmap pd_bitmap; 337 struct mlx4_bitmap pd_bitmap;
338 struct mlx4_bitmap xrcd_bitmap;
338 struct mlx4_uar_table uar_table; 339 struct mlx4_uar_table uar_table;
339 struct mlx4_mr_table mr_table; 340 struct mlx4_mr_table mr_table;
340 struct mlx4_cq_table cq_table; 341 struct mlx4_cq_table cq_table;
@@ -384,6 +385,7 @@ int mlx4_alloc_eq_table(struct mlx4_dev *dev);
384void mlx4_free_eq_table(struct mlx4_dev *dev); 385void mlx4_free_eq_table(struct mlx4_dev *dev);
385 386
386int mlx4_init_pd_table(struct mlx4_dev *dev); 387int mlx4_init_pd_table(struct mlx4_dev *dev);
388int mlx4_init_xrcd_table(struct mlx4_dev *dev);
387int mlx4_init_uar_table(struct mlx4_dev *dev); 389int mlx4_init_uar_table(struct mlx4_dev *dev);
388int mlx4_init_mr_table(struct mlx4_dev *dev); 390int mlx4_init_mr_table(struct mlx4_dev *dev);
389int mlx4_init_eq_table(struct mlx4_dev *dev); 391int mlx4_init_eq_table(struct mlx4_dev *dev);
@@ -393,6 +395,7 @@ int mlx4_init_srq_table(struct mlx4_dev *dev);
393int mlx4_init_mcg_table(struct mlx4_dev *dev); 395int mlx4_init_mcg_table(struct mlx4_dev *dev);
394 396
395void mlx4_cleanup_pd_table(struct mlx4_dev *dev); 397void mlx4_cleanup_pd_table(struct mlx4_dev *dev);
398void mlx4_cleanup_xrcd_table(struct mlx4_dev *dev);
396void mlx4_cleanup_uar_table(struct mlx4_dev *dev); 399void mlx4_cleanup_uar_table(struct mlx4_dev *dev);
397void mlx4_cleanup_mr_table(struct mlx4_dev *dev); 400void mlx4_cleanup_mr_table(struct mlx4_dev *dev);
398void mlx4_cleanup_eq_table(struct mlx4_dev *dev); 401void mlx4_cleanup_eq_table(struct mlx4_dev *dev);
@@ -450,6 +453,7 @@ void mlx4_init_vlan_table(struct mlx4_dev *dev, struct mlx4_vlan_table *table);
450 453
451int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port); 454int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port);
452int mlx4_get_port_ib_caps(struct mlx4_dev *dev, u8 port, __be32 *caps); 455int mlx4_get_port_ib_caps(struct mlx4_dev *dev, u8 port, __be32 *caps);
456int mlx4_check_ext_port_caps(struct mlx4_dev *dev, u8 port);
453 457
454int mlx4_qp_detach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16], 458int mlx4_qp_detach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
455 enum mlx4_protocol prot, enum mlx4_steer_type steer); 459 enum mlx4_protocol prot, enum mlx4_steer_type steer);
diff --git a/drivers/net/mlx4/mr.c b/drivers/net/mlx4/mr.c
index 9c188bdd7f4f..ab639cfef78e 100644
--- a/drivers/net/mlx4/mr.c
+++ b/drivers/net/mlx4/mr.c
@@ -139,7 +139,7 @@ static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
139 139
140 buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *), 140 buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
141 GFP_KERNEL); 141 GFP_KERNEL);
142 buddy->num_free = kzalloc((buddy->max_order + 1) * sizeof (int *), 142 buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
143 GFP_KERNEL); 143 GFP_KERNEL);
144 if (!buddy->bits || !buddy->num_free) 144 if (!buddy->bits || !buddy->num_free)
145 goto err_out; 145 goto err_out;
diff --git a/drivers/net/mlx4/pd.c b/drivers/net/mlx4/pd.c
index 1286b886dcea..3736163e30e9 100644
--- a/drivers/net/mlx4/pd.c
+++ b/drivers/net/mlx4/pd.c
@@ -61,6 +61,24 @@ void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn)
61} 61}
62EXPORT_SYMBOL_GPL(mlx4_pd_free); 62EXPORT_SYMBOL_GPL(mlx4_pd_free);
63 63
64int mlx4_xrcd_alloc(struct mlx4_dev *dev, u32 *xrcdn)
65{
66 struct mlx4_priv *priv = mlx4_priv(dev);
67
68 *xrcdn = mlx4_bitmap_alloc(&priv->xrcd_bitmap);
69 if (*xrcdn == -1)
70 return -ENOMEM;
71
72 return 0;
73}
74EXPORT_SYMBOL_GPL(mlx4_xrcd_alloc);
75
76void mlx4_xrcd_free(struct mlx4_dev *dev, u32 xrcdn)
77{
78 mlx4_bitmap_free(&mlx4_priv(dev)->xrcd_bitmap, xrcdn);
79}
80EXPORT_SYMBOL_GPL(mlx4_xrcd_free);
81
64int mlx4_init_pd_table(struct mlx4_dev *dev) 82int mlx4_init_pd_table(struct mlx4_dev *dev)
65{ 83{
66 struct mlx4_priv *priv = mlx4_priv(dev); 84 struct mlx4_priv *priv = mlx4_priv(dev);
@@ -74,6 +92,18 @@ void mlx4_cleanup_pd_table(struct mlx4_dev *dev)
74 mlx4_bitmap_cleanup(&mlx4_priv(dev)->pd_bitmap); 92 mlx4_bitmap_cleanup(&mlx4_priv(dev)->pd_bitmap);
75} 93}
76 94
95int mlx4_init_xrcd_table(struct mlx4_dev *dev)
96{
97 struct mlx4_priv *priv = mlx4_priv(dev);
98
99 return mlx4_bitmap_init(&priv->xrcd_bitmap, (1 << 16),
100 (1 << 16) - 1, dev->caps.reserved_xrcds + 1, 0);
101}
102
103void mlx4_cleanup_xrcd_table(struct mlx4_dev *dev)
104{
105 mlx4_bitmap_cleanup(&mlx4_priv(dev)->xrcd_bitmap);
106}
77 107
78int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar) 108int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar)
79{ 109{
diff --git a/drivers/net/mlx4/port.c b/drivers/net/mlx4/port.c
index 609e0ec14cee..881592eec614 100644
--- a/drivers/net/mlx4/port.c
+++ b/drivers/net/mlx4/port.c
@@ -148,22 +148,26 @@ int mlx4_register_mac(struct mlx4_dev *dev, u8 port, u64 mac, int *qpn, u8 wrap)
148 148
149 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER) { 149 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER) {
150 err = mlx4_uc_steer_add(dev, port, mac, qpn, 1); 150 err = mlx4_uc_steer_add(dev, port, mac, qpn, 1);
151 if (!err) { 151 if (err)
152 entry = kmalloc(sizeof *entry, GFP_KERNEL);
153 if (!entry) {
154 mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
155 return -ENOMEM;
156 }
157 entry->mac = mac;
158 err = radix_tree_insert(&info->mac_tree, *qpn, entry);
159 if (err) {
160 mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
161 return err;
162 }
163 } else
164 return err; 152 return err;
153
154 entry = kmalloc(sizeof *entry, GFP_KERNEL);
155 if (!entry) {
156 mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
157 return -ENOMEM;
158 }
159
160 entry->mac = mac;
161 err = radix_tree_insert(&info->mac_tree, *qpn, entry);
162 if (err) {
163 kfree(entry);
164 mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
165 return err;
166 }
165 } 167 }
168
166 mlx4_dbg(dev, "Registering MAC: 0x%llx\n", (unsigned long long) mac); 169 mlx4_dbg(dev, "Registering MAC: 0x%llx\n", (unsigned long long) mac);
170
167 mutex_lock(&table->mutex); 171 mutex_lock(&table->mutex);
168 for (i = 0; i < MLX4_MAX_MAC_NUM - 1; i++) { 172 for (i = 0; i < MLX4_MAX_MAC_NUM - 1; i++) {
169 if (free < 0 && !table->refs[i]) { 173 if (free < 0 && !table->refs[i]) {
@@ -464,6 +468,48 @@ int mlx4_get_port_ib_caps(struct mlx4_dev *dev, u8 port, __be32 *caps)
464 return err; 468 return err;
465} 469}
466 470
471int mlx4_check_ext_port_caps(struct mlx4_dev *dev, u8 port)
472{
473 struct mlx4_cmd_mailbox *inmailbox, *outmailbox;
474 u8 *inbuf, *outbuf;
475 int err, packet_error;
476
477 inmailbox = mlx4_alloc_cmd_mailbox(dev);
478 if (IS_ERR(inmailbox))
479 return PTR_ERR(inmailbox);
480
481 outmailbox = mlx4_alloc_cmd_mailbox(dev);
482 if (IS_ERR(outmailbox)) {
483 mlx4_free_cmd_mailbox(dev, inmailbox);
484 return PTR_ERR(outmailbox);
485 }
486
487 inbuf = inmailbox->buf;
488 outbuf = outmailbox->buf;
489 memset(inbuf, 0, 256);
490 memset(outbuf, 0, 256);
491 inbuf[0] = 1;
492 inbuf[1] = 1;
493 inbuf[2] = 1;
494 inbuf[3] = 1;
495
496 *(__be16 *) (&inbuf[16]) = MLX4_ATTR_EXTENDED_PORT_INFO;
497 *(__be32 *) (&inbuf[20]) = cpu_to_be32(port);
498
499 err = mlx4_cmd_box(dev, inmailbox->dma, outmailbox->dma, port, 3,
500 MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C);
501
502 packet_error = be16_to_cpu(*(__be16 *) (outbuf + 4));
503
504 dev->caps.ext_port_cap[port] = (!err && !packet_error) ?
505 MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO
506 : 0;
507
508 mlx4_free_cmd_mailbox(dev, inmailbox);
509 mlx4_free_cmd_mailbox(dev, outmailbox);
510 return err;
511}
512
467int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port) 513int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port)
468{ 514{
469 struct mlx4_cmd_mailbox *mailbox; 515 struct mlx4_cmd_mailbox *mailbox;
diff --git a/drivers/net/mlx4/qp.c b/drivers/net/mlx4/qp.c
index ec9350e5f21a..51c53898c35f 100644
--- a/drivers/net/mlx4/qp.c
+++ b/drivers/net/mlx4/qp.c
@@ -280,6 +280,9 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
280 * We reserve 2 extra QPs per port for the special QPs. The 280 * We reserve 2 extra QPs per port for the special QPs. The
281 * block of special QPs must be aligned to a multiple of 8, so 281 * block of special QPs must be aligned to a multiple of 8, so
282 * round up. 282 * round up.
283 *
284 * We also reserve the MSB of the 24-bit QP number to indicate
285 * that a QP is an XRC QP.
283 */ 286 */
284 dev->caps.sqp_start = 287 dev->caps.sqp_start =
285 ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8); 288 ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);
diff --git a/drivers/net/mlx4/srq.c b/drivers/net/mlx4/srq.c
index 3b07b80a0456..a20b141dbb5c 100644
--- a/drivers/net/mlx4/srq.c
+++ b/drivers/net/mlx4/srq.c
@@ -40,20 +40,20 @@
40struct mlx4_srq_context { 40struct mlx4_srq_context {
41 __be32 state_logsize_srqn; 41 __be32 state_logsize_srqn;
42 u8 logstride; 42 u8 logstride;
43 u8 reserved1[3]; 43 u8 reserved1;
44 u8 pg_offset; 44 __be16 xrcd;
45 u8 reserved2[3]; 45 __be32 pg_offset_cqn;
46 u32 reserved3; 46 u32 reserved2;
47 u8 log_page_size; 47 u8 log_page_size;
48 u8 reserved4[2]; 48 u8 reserved3[2];
49 u8 mtt_base_addr_h; 49 u8 mtt_base_addr_h;
50 __be32 mtt_base_addr_l; 50 __be32 mtt_base_addr_l;
51 __be32 pd; 51 __be32 pd;
52 __be16 limit_watermark; 52 __be16 limit_watermark;
53 __be16 wqe_cnt; 53 __be16 wqe_cnt;
54 u16 reserved5; 54 u16 reserved4;
55 __be16 wqe_counter; 55 __be16 wqe_counter;
56 u32 reserved6; 56 u32 reserved5;
57 __be64 db_rec_addr; 57 __be64 db_rec_addr;
58}; 58};
59 59
@@ -109,8 +109,8 @@ static int mlx4_QUERY_SRQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox
109 MLX4_CMD_TIME_CLASS_A); 109 MLX4_CMD_TIME_CLASS_A);
110} 110}
111 111
112int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt, 112int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcd,
113 u64 db_rec, struct mlx4_srq *srq) 113 struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq)
114{ 114{
115 struct mlx4_srq_table *srq_table = &mlx4_priv(dev)->srq_table; 115 struct mlx4_srq_table *srq_table = &mlx4_priv(dev)->srq_table;
116 struct mlx4_cmd_mailbox *mailbox; 116 struct mlx4_cmd_mailbox *mailbox;
@@ -148,6 +148,8 @@ int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
148 srq_context->state_logsize_srqn = cpu_to_be32((ilog2(srq->max) << 24) | 148 srq_context->state_logsize_srqn = cpu_to_be32((ilog2(srq->max) << 24) |
149 srq->srqn); 149 srq->srqn);
150 srq_context->logstride = srq->wqe_shift - 4; 150 srq_context->logstride = srq->wqe_shift - 4;
151 srq_context->xrcd = cpu_to_be16(xrcd);
152 srq_context->pg_offset_cqn = cpu_to_be32(cqn & 0xffffff);
151 srq_context->log_page_size = mtt->page_shift - MLX4_ICM_PAGE_SHIFT; 153 srq_context->log_page_size = mtt->page_shift - MLX4_ICM_PAGE_SHIFT;
152 154
153 mtt_addr = mlx4_mtt_addr(dev, mtt); 155 mtt_addr = mlx4_mtt_addr(dev, mtt);
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 53ef894bfa05..ff3ccd5c44d6 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -61,6 +61,7 @@ enum {
61 MLX4_DEV_CAP_FLAG_RC = 1LL << 0, 61 MLX4_DEV_CAP_FLAG_RC = 1LL << 0,
62 MLX4_DEV_CAP_FLAG_UC = 1LL << 1, 62 MLX4_DEV_CAP_FLAG_UC = 1LL << 1,
63 MLX4_DEV_CAP_FLAG_UD = 1LL << 2, 63 MLX4_DEV_CAP_FLAG_UD = 1LL << 2,
64 MLX4_DEV_CAP_FLAG_XRC = 1LL << 3,
64 MLX4_DEV_CAP_FLAG_SRQ = 1LL << 6, 65 MLX4_DEV_CAP_FLAG_SRQ = 1LL << 6,
65 MLX4_DEV_CAP_FLAG_IPOIB_CSUM = 1LL << 7, 66 MLX4_DEV_CAP_FLAG_IPOIB_CSUM = 1LL << 7,
66 MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR = 1LL << 8, 67 MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR = 1LL << 8,
@@ -82,6 +83,12 @@ enum {
82 MLX4_DEV_CAP_FLAG_COUNTERS = 1LL << 48 83 MLX4_DEV_CAP_FLAG_COUNTERS = 1LL << 48
83}; 84};
84 85
86#define MLX4_ATTR_EXTENDED_PORT_INFO cpu_to_be16(0xff90)
87
88enum {
89 MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO = 1 << 0
90};
91
85enum { 92enum {
86 MLX4_BMME_FLAG_LOCAL_INV = 1 << 6, 93 MLX4_BMME_FLAG_LOCAL_INV = 1 << 6,
87 MLX4_BMME_FLAG_REMOTE_INV = 1 << 7, 94 MLX4_BMME_FLAG_REMOTE_INV = 1 << 7,
@@ -256,6 +263,8 @@ struct mlx4_caps {
256 int num_qp_per_mgm; 263 int num_qp_per_mgm;
257 int num_pds; 264 int num_pds;
258 int reserved_pds; 265 int reserved_pds;
266 int max_xrcds;
267 int reserved_xrcds;
259 int mtt_entry_sz; 268 int mtt_entry_sz;
260 u32 max_msg_sz; 269 u32 max_msg_sz;
261 u32 page_size_cap; 270 u32 page_size_cap;
@@ -276,6 +285,7 @@ struct mlx4_caps {
276 u32 port_mask; 285 u32 port_mask;
277 enum mlx4_port_type possible_type[MLX4_MAX_PORTS + 1]; 286 enum mlx4_port_type possible_type[MLX4_MAX_PORTS + 1];
278 u32 max_counters; 287 u32 max_counters;
288 u8 ext_port_cap[MLX4_MAX_PORTS + 1];
279}; 289};
280 290
281struct mlx4_buf_list { 291struct mlx4_buf_list {
@@ -499,6 +509,8 @@ static inline void *mlx4_buf_offset(struct mlx4_buf *buf, int offset)
499 509
500int mlx4_pd_alloc(struct mlx4_dev *dev, u32 *pdn); 510int mlx4_pd_alloc(struct mlx4_dev *dev, u32 *pdn);
501void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn); 511void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn);
512int mlx4_xrcd_alloc(struct mlx4_dev *dev, u32 *xrcdn);
513void mlx4_xrcd_free(struct mlx4_dev *dev, u32 xrcdn);
502 514
503int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar); 515int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar);
504void mlx4_uar_free(struct mlx4_dev *dev, struct mlx4_uar *uar); 516void mlx4_uar_free(struct mlx4_dev *dev, struct mlx4_uar *uar);
@@ -538,8 +550,8 @@ void mlx4_qp_release_range(struct mlx4_dev *dev, int base_qpn, int cnt);
538int mlx4_qp_alloc(struct mlx4_dev *dev, int qpn, struct mlx4_qp *qp); 550int mlx4_qp_alloc(struct mlx4_dev *dev, int qpn, struct mlx4_qp *qp);
539void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp); 551void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp);
540 552
541int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt, 553int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcdn,
542 u64 db_rec, struct mlx4_srq *srq); 554 struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq);
543void mlx4_srq_free(struct mlx4_dev *dev, struct mlx4_srq *srq); 555void mlx4_srq_free(struct mlx4_dev *dev, struct mlx4_srq *srq);
544int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark); 556int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark);
545int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark); 557int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark);
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 4001c8249dbb..48cc4cb97858 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -75,6 +75,7 @@ enum {
75 MLX4_QP_ST_UC = 0x1, 75 MLX4_QP_ST_UC = 0x1,
76 MLX4_QP_ST_RD = 0x2, 76 MLX4_QP_ST_RD = 0x2,
77 MLX4_QP_ST_UD = 0x3, 77 MLX4_QP_ST_UD = 0x3,
78 MLX4_QP_ST_XRC = 0x6,
78 MLX4_QP_ST_MLX = 0x7 79 MLX4_QP_ST_MLX = 0x7
79}; 80};
80 81
@@ -137,7 +138,7 @@ struct mlx4_qp_context {
137 __be32 ssn; 138 __be32 ssn;
138 __be32 params2; 139 __be32 params2;
139 __be32 rnr_nextrecvpsn; 140 __be32 rnr_nextrecvpsn;
140 __be32 srcd; 141 __be32 xrcd;
141 __be32 cqn_recv; 142 __be32 cqn_recv;
142 __be64 db_rec_addr; 143 __be64 db_rec_addr;
143 __be32 qkey; 144 __be32 qkey;
diff --git a/include/rdma/ib_user_verbs.h b/include/rdma/ib_user_verbs.h
index fe5b05177a2c..81aba3a73aa3 100644
--- a/include/rdma/ib_user_verbs.h
+++ b/include/rdma/ib_user_verbs.h
@@ -81,7 +81,11 @@ enum {
81 IB_USER_VERBS_CMD_MODIFY_SRQ, 81 IB_USER_VERBS_CMD_MODIFY_SRQ,
82 IB_USER_VERBS_CMD_QUERY_SRQ, 82 IB_USER_VERBS_CMD_QUERY_SRQ,
83 IB_USER_VERBS_CMD_DESTROY_SRQ, 83 IB_USER_VERBS_CMD_DESTROY_SRQ,
84 IB_USER_VERBS_CMD_POST_SRQ_RECV 84 IB_USER_VERBS_CMD_POST_SRQ_RECV,
85 IB_USER_VERBS_CMD_OPEN_XRCD,
86 IB_USER_VERBS_CMD_CLOSE_XRCD,
87 IB_USER_VERBS_CMD_CREATE_XSRQ,
88 IB_USER_VERBS_CMD_OPEN_QP
85}; 89};
86 90
87/* 91/*
@@ -222,6 +226,21 @@ struct ib_uverbs_dealloc_pd {
222 __u32 pd_handle; 226 __u32 pd_handle;
223}; 227};
224 228
229struct ib_uverbs_open_xrcd {
230 __u64 response;
231 __u32 fd;
232 __u32 oflags;
233 __u64 driver_data[0];
234};
235
236struct ib_uverbs_open_xrcd_resp {
237 __u32 xrcd_handle;
238};
239
240struct ib_uverbs_close_xrcd {
241 __u32 xrcd_handle;
242};
243
225struct ib_uverbs_reg_mr { 244struct ib_uverbs_reg_mr {
226 __u64 response; 245 __u64 response;
227 __u64 start; 246 __u64 start;
@@ -404,6 +423,17 @@ struct ib_uverbs_create_qp {
404 __u64 driver_data[0]; 423 __u64 driver_data[0];
405}; 424};
406 425
426struct ib_uverbs_open_qp {
427 __u64 response;
428 __u64 user_handle;
429 __u32 pd_handle;
430 __u32 qpn;
431 __u8 qp_type;
432 __u8 reserved[7];
433 __u64 driver_data[0];
434};
435
436/* also used for open response */
407struct ib_uverbs_create_qp_resp { 437struct ib_uverbs_create_qp_resp {
408 __u32 qp_handle; 438 __u32 qp_handle;
409 __u32 qpn; 439 __u32 qpn;
@@ -648,11 +678,25 @@ struct ib_uverbs_create_srq {
648 __u64 driver_data[0]; 678 __u64 driver_data[0];
649}; 679};
650 680
681struct ib_uverbs_create_xsrq {
682 __u64 response;
683 __u64 user_handle;
684 __u32 srq_type;
685 __u32 pd_handle;
686 __u32 max_wr;
687 __u32 max_sge;
688 __u32 srq_limit;
689 __u32 reserved;
690 __u32 xrcd_handle;
691 __u32 cq_handle;
692 __u64 driver_data[0];
693};
694
651struct ib_uverbs_create_srq_resp { 695struct ib_uverbs_create_srq_resp {
652 __u32 srq_handle; 696 __u32 srq_handle;
653 __u32 max_wr; 697 __u32 max_wr;
654 __u32 max_sge; 698 __u32 max_sge;
655 __u32 reserved; 699 __u32 srqn;
656}; 700};
657 701
658struct ib_uverbs_modify_srq { 702struct ib_uverbs_modify_srq {
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 228be3e220d9..bf5daafe8ecc 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -112,6 +112,7 @@ enum ib_device_cap_flags {
112 */ 112 */
113 IB_DEVICE_UD_IP_CSUM = (1<<18), 113 IB_DEVICE_UD_IP_CSUM = (1<<18),
114 IB_DEVICE_UD_TSO = (1<<19), 114 IB_DEVICE_UD_TSO = (1<<19),
115 IB_DEVICE_XRC = (1<<20),
115 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21), 116 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
116 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22), 117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
117}; 118};
@@ -207,6 +208,7 @@ enum ib_port_cap_flags {
207 IB_PORT_SM_DISABLED = 1 << 10, 208 IB_PORT_SM_DISABLED = 1 << 10,
208 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 209 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
209 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 210 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
211 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
210 IB_PORT_CM_SUP = 1 << 16, 212 IB_PORT_CM_SUP = 1 << 16,
211 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 213 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
212 IB_PORT_REINIT_SUP = 1 << 18, 214 IB_PORT_REINIT_SUP = 1 << 18,
@@ -415,7 +417,15 @@ enum ib_rate {
415 IB_RATE_40_GBPS = 7, 417 IB_RATE_40_GBPS = 7,
416 IB_RATE_60_GBPS = 8, 418 IB_RATE_60_GBPS = 8,
417 IB_RATE_80_GBPS = 9, 419 IB_RATE_80_GBPS = 9,
418 IB_RATE_120_GBPS = 10 420 IB_RATE_120_GBPS = 10,
421 IB_RATE_14_GBPS = 11,
422 IB_RATE_56_GBPS = 12,
423 IB_RATE_112_GBPS = 13,
424 IB_RATE_168_GBPS = 14,
425 IB_RATE_25_GBPS = 15,
426 IB_RATE_100_GBPS = 16,
427 IB_RATE_200_GBPS = 17,
428 IB_RATE_300_GBPS = 18
419}; 429};
420 430
421/** 431/**
@@ -427,6 +437,13 @@ enum ib_rate {
427int ib_rate_to_mult(enum ib_rate rate) __attribute_const__; 437int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
428 438
429/** 439/**
440 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
441 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
442 * @rate: rate to convert.
443 */
444int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
445
446/**
430 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 447 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
431 * enum. 448 * enum.
432 * @mult: multiple to convert. 449 * @mult: multiple to convert.
@@ -522,6 +539,11 @@ enum ib_cq_notify_flags {
522 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 539 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
523}; 540};
524 541
542enum ib_srq_type {
543 IB_SRQT_BASIC,
544 IB_SRQT_XRC
545};
546
525enum ib_srq_attr_mask { 547enum ib_srq_attr_mask {
526 IB_SRQ_MAX_WR = 1 << 0, 548 IB_SRQ_MAX_WR = 1 << 0,
527 IB_SRQ_LIMIT = 1 << 1, 549 IB_SRQ_LIMIT = 1 << 1,
@@ -537,6 +559,14 @@ struct ib_srq_init_attr {
537 void (*event_handler)(struct ib_event *, void *); 559 void (*event_handler)(struct ib_event *, void *);
538 void *srq_context; 560 void *srq_context;
539 struct ib_srq_attr attr; 561 struct ib_srq_attr attr;
562 enum ib_srq_type srq_type;
563
564 union {
565 struct {
566 struct ib_xrcd *xrcd;
567 struct ib_cq *cq;
568 } xrc;
569 } ext;
540}; 570};
541 571
542struct ib_qp_cap { 572struct ib_qp_cap {
@@ -565,7 +595,11 @@ enum ib_qp_type {
565 IB_QPT_UC, 595 IB_QPT_UC,
566 IB_QPT_UD, 596 IB_QPT_UD,
567 IB_QPT_RAW_IPV6, 597 IB_QPT_RAW_IPV6,
568 IB_QPT_RAW_ETHERTYPE 598 IB_QPT_RAW_ETHERTYPE,
599 /* Save 8 for RAW_PACKET */
600 IB_QPT_XRC_INI = 9,
601 IB_QPT_XRC_TGT,
602 IB_QPT_MAX
569}; 603};
570 604
571enum ib_qp_create_flags { 605enum ib_qp_create_flags {
@@ -579,6 +613,7 @@ struct ib_qp_init_attr {
579 struct ib_cq *send_cq; 613 struct ib_cq *send_cq;
580 struct ib_cq *recv_cq; 614 struct ib_cq *recv_cq;
581 struct ib_srq *srq; 615 struct ib_srq *srq;
616 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
582 struct ib_qp_cap cap; 617 struct ib_qp_cap cap;
583 enum ib_sig_type sq_sig_type; 618 enum ib_sig_type sq_sig_type;
584 enum ib_qp_type qp_type; 619 enum ib_qp_type qp_type;
@@ -586,6 +621,13 @@ struct ib_qp_init_attr {
586 u8 port_num; /* special QP types only */ 621 u8 port_num; /* special QP types only */
587}; 622};
588 623
624struct ib_qp_open_attr {
625 void (*event_handler)(struct ib_event *, void *);
626 void *qp_context;
627 u32 qp_num;
628 enum ib_qp_type qp_type;
629};
630
589enum ib_rnr_timeout { 631enum ib_rnr_timeout {
590 IB_RNR_TIMER_655_36 = 0, 632 IB_RNR_TIMER_655_36 = 0,
591 IB_RNR_TIMER_000_01 = 1, 633 IB_RNR_TIMER_000_01 = 1,
@@ -770,6 +812,7 @@ struct ib_send_wr {
770 u32 rkey; 812 u32 rkey;
771 } fast_reg; 813 } fast_reg;
772 } wr; 814 } wr;
815 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
773}; 816};
774 817
775struct ib_recv_wr { 818struct ib_recv_wr {
@@ -831,6 +874,7 @@ struct ib_ucontext {
831 struct list_head qp_list; 874 struct list_head qp_list;
832 struct list_head srq_list; 875 struct list_head srq_list;
833 struct list_head ah_list; 876 struct list_head ah_list;
877 struct list_head xrcd_list;
834 int closing; 878 int closing;
835}; 879};
836 880
@@ -858,6 +902,15 @@ struct ib_pd {
858 atomic_t usecnt; /* count all resources */ 902 atomic_t usecnt; /* count all resources */
859}; 903};
860 904
905struct ib_xrcd {
906 struct ib_device *device;
907 atomic_t usecnt; /* count all exposed resources */
908 struct inode *inode;
909
910 struct mutex tgt_qp_mutex;
911 struct list_head tgt_qp_list;
912};
913
861struct ib_ah { 914struct ib_ah {
862 struct ib_device *device; 915 struct ib_device *device;
863 struct ib_pd *pd; 916 struct ib_pd *pd;
@@ -882,7 +935,16 @@ struct ib_srq {
882 struct ib_uobject *uobject; 935 struct ib_uobject *uobject;
883 void (*event_handler)(struct ib_event *, void *); 936 void (*event_handler)(struct ib_event *, void *);
884 void *srq_context; 937 void *srq_context;
938 enum ib_srq_type srq_type;
885 atomic_t usecnt; 939 atomic_t usecnt;
940
941 union {
942 struct {
943 struct ib_xrcd *xrcd;
944 struct ib_cq *cq;
945 u32 srq_num;
946 } xrc;
947 } ext;
886}; 948};
887 949
888struct ib_qp { 950struct ib_qp {
@@ -891,6 +953,11 @@ struct ib_qp {
891 struct ib_cq *send_cq; 953 struct ib_cq *send_cq;
892 struct ib_cq *recv_cq; 954 struct ib_cq *recv_cq;
893 struct ib_srq *srq; 955 struct ib_srq *srq;
956 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
957 struct list_head xrcd_list;
958 atomic_t usecnt; /* count times opened */
959 struct list_head open_list;
960 struct ib_qp *real_qp;
894 struct ib_uobject *uobject; 961 struct ib_uobject *uobject;
895 void (*event_handler)(struct ib_event *, void *); 962 void (*event_handler)(struct ib_event *, void *);
896 void *qp_context; 963 void *qp_context;
@@ -1149,6 +1216,10 @@ struct ib_device {
1149 struct ib_grh *in_grh, 1216 struct ib_grh *in_grh,
1150 struct ib_mad *in_mad, 1217 struct ib_mad *in_mad,
1151 struct ib_mad *out_mad); 1218 struct ib_mad *out_mad);
1219 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1220 struct ib_ucontext *ucontext,
1221 struct ib_udata *udata);
1222 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1152 1223
1153 struct ib_dma_mapping_ops *dma_ops; 1224 struct ib_dma_mapping_ops *dma_ops;
1154 1225
@@ -1443,6 +1514,25 @@ int ib_query_qp(struct ib_qp *qp,
1443int ib_destroy_qp(struct ib_qp *qp); 1514int ib_destroy_qp(struct ib_qp *qp);
1444 1515
1445/** 1516/**
1517 * ib_open_qp - Obtain a reference to an existing sharable QP.
1518 * @xrcd - XRC domain
1519 * @qp_open_attr: Attributes identifying the QP to open.
1520 *
1521 * Returns a reference to a sharable QP.
1522 */
1523struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1524 struct ib_qp_open_attr *qp_open_attr);
1525
1526/**
1527 * ib_close_qp - Release an external reference to a QP.
1528 * @qp: The QP handle to release
1529 *
1530 * The opened QP handle is released by the caller. The underlying
1531 * shared QP is not destroyed until all internal references are released.
1532 */
1533int ib_close_qp(struct ib_qp *qp);
1534
1535/**
1446 * ib_post_send - Posts a list of work requests to the send queue of 1536 * ib_post_send - Posts a list of work requests to the send queue of
1447 * the specified QP. 1537 * the specified QP.
1448 * @qp: The QP to post the work request on. 1538 * @qp: The QP to post the work request on.
@@ -2060,4 +2150,16 @@ int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2060 */ 2150 */
2061int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2151int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2062 2152
2153/**
2154 * ib_alloc_xrcd - Allocates an XRC domain.
2155 * @device: The device on which to allocate the XRC domain.
2156 */
2157struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2158
2159/**
2160 * ib_dealloc_xrcd - Deallocates an XRC domain.
2161 * @xrcd: The XRC domain to deallocate.
2162 */
2163int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2164
2063#endif /* IB_VERBS_H */ 2165#endif /* IB_VERBS_H */
diff --git a/include/rdma/iw_cm.h b/include/rdma/iw_cm.h
index 2d0191c90f9e..1a046b1595cc 100644
--- a/include/rdma/iw_cm.h
+++ b/include/rdma/iw_cm.h
@@ -52,8 +52,10 @@ struct iw_cm_event {
52 struct sockaddr_in local_addr; 52 struct sockaddr_in local_addr;
53 struct sockaddr_in remote_addr; 53 struct sockaddr_in remote_addr;
54 void *private_data; 54 void *private_data;
55 u8 private_data_len;
56 void *provider_data; 55 void *provider_data;
56 u8 private_data_len;
57 u8 ord;
58 u8 ird;
57}; 59};
58 60
59/** 61/**
diff --git a/include/rdma/rdma_cm.h b/include/rdma/rdma_cm.h
index 26977c149c41..51988f808181 100644
--- a/include/rdma/rdma_cm.h
+++ b/include/rdma/rdma_cm.h
@@ -65,6 +65,7 @@ enum rdma_cm_event_type {
65enum rdma_port_space { 65enum rdma_port_space {
66 RDMA_PS_SDP = 0x0001, 66 RDMA_PS_SDP = 0x0001,
67 RDMA_PS_IPOIB = 0x0002, 67 RDMA_PS_IPOIB = 0x0002,
68 RDMA_PS_IB = 0x013F,
68 RDMA_PS_TCP = 0x0106, 69 RDMA_PS_TCP = 0x0106,
69 RDMA_PS_UDP = 0x0111, 70 RDMA_PS_UDP = 0x0111,
70}; 71};
diff --git a/include/rdma/rdma_user_cm.h b/include/rdma/rdma_user_cm.h
index fc82c1896f75..5348a000c8f3 100644
--- a/include/rdma/rdma_user_cm.h
+++ b/include/rdma/rdma_user_cm.h
@@ -77,7 +77,8 @@ struct rdma_ucm_create_id {
77 __u64 uid; 77 __u64 uid;
78 __u64 response; 78 __u64 response;
79 __u16 ps; 79 __u16 ps;
80 __u8 reserved[6]; 80 __u8 qp_type;
81 __u8 reserved[5];
81}; 82};
82 83
83struct rdma_ucm_create_id_resp { 84struct rdma_ucm_create_id_resp {
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-28 21:38:16 -0500