aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2013-02-26 14:41:08 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2013-02-26 14:41:08 -0500
commit70a3a06d01ed9ca887316a881813cdefb8a20170 (patch)
treefbdb7982040ba77818e4b738d76eef8bb06fb47f
parentf6c0ffa8f0b0781f4954cb06f0a81d6c10c1b434 (diff)
parentef4e359d9b9e2dc022f79840fd207796b524a893 (diff)
Merge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband
Pull infiniband update from Roland Dreier: "Main batch of InfiniBand/RDMA changes for 3.9: - SRP error handling fixes from Bart Van Assche - Implementation of memory windows for mlx4 from Shani Michaeli - Lots of cxgb4 HW driver fixes from Vipul Pandya - Make iSER work for virtual functions, other fixes from Or Gerlitz - Fix for bug in qib HW driver from Mike Marciniszyn - IPoIB fixes from me, Itai Garbi, Shlomo Pongratz, Yan Burman - Various cleanups and warning fixes from Julia Lawall, Paul Bolle, Wei Yongjun" * tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (41 commits) IB/mlx4: Advertise MW support IB/mlx4: Support memory window binding mlx4: Implement memory windows allocation and deallocation mlx4_core: Enable memory windows in {INIT, QUERY}_HCA mlx4_core: Disable memory windows for virtual functions IPoIB: Free ipoib neigh on path record failure so path rec queries are retried IB/srp: Fail I/O requests if the transport is offline IB/srp: Avoid endless SCSI error handling loop IB/srp: Avoid sending a task management function needlessly IB/srp: Track connection state properly IB/mlx4: Remove redundant NULL check before kfree IB/mlx4: Fix compiler warning about uninitialized 'vlan' variable IB/mlx4: Convert is_xxx variables in build_mlx_header() to bool IB/iser: Enable iser when FMRs are not supported IB/iser: Avoid error prints on EAGAIN registration failures IB/iser: Use proper define for the commands per LUN value advertised to SCSI ML IB/uverbs: Implement memory windows support in uverbs IB/core: Add "type 2" memory windows support mlx4_core: Propagate MR deregistration failures to caller mlx4_core: Rename MPT-related functions to have mpt_ prefix ...
Diffstat
-rw-r--r--drivers/infiniband/core/uverbs.h2
-rw-r--r--drivers/infiniband/core/uverbs_cmd.c121
-rw-r--r--drivers/infiniband/core/uverbs_main.c13
-rw-r--r--drivers/infiniband/core/verbs.c5
-rw-r--r--drivers/infiniband/hw/amso1100/c2.c13
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_provider.c5
-rw-r--r--drivers/infiniband/hw/cxgb3/iwch_qp.c15
-rw-r--r--drivers/infiniband/hw/cxgb4/cm.c170
-rw-r--r--drivers/infiniband/hw/cxgb4/device.c5
-rw-r--r--drivers/infiniband/hw/cxgb4/ev.c8
-rw-r--r--drivers/infiniband/hw/cxgb4/iw_cxgb4.h4
-rw-r--r--drivers/infiniband/hw/cxgb4/mem.c5
-rw-r--r--drivers/infiniband/hw/cxgb4/qp.c1
-rw-r--r--drivers/infiniband/hw/ehca/ehca_iverbs.h2
-rw-r--r--drivers/infiniband/hw/ehca/ehca_mrmw.c5
-rw-r--r--drivers/infiniband/hw/mlx4/mad.c7
-rw-r--r--drivers/infiniband/hw/mlx4/main.c22
-rw-r--r--drivers/infiniband/hw/mlx4/mlx4_ib.h18
-rw-r--r--drivers/infiniband/hw/mlx4/mr.c87
-rw-r--r--drivers/infiniband/hw/mlx4/qp.c49
-rw-r--r--drivers/infiniband/hw/mlx4/sysfs.c2
-rw-r--r--drivers/infiniband/hw/nes/nes_verbs.c19
-rw-r--r--drivers/infiniband/hw/qib/qib_qp.c5
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib.h4
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_ethtool.c19
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_main.c14
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.h2
-rw-r--r--drivers/infiniband/ulp/iser/iser_memory.c9
-rw-r--r--drivers/infiniband/ulp/iser/iser_verbs.c8
-rw-r--r--drivers/infiniband/ulp/srp/ib_srp.c42
-rw-r--r--drivers/infiniband/ulp/srp/ib_srp.h1
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/en_main.c4
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/fw.c14
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/fw.h1
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/main.c4
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/mlx4.h34
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/mr.c186
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/resource_tracker.c63
-rw-r--r--include/linux/mlx4/device.h22
-rw-r--r--include/linux/mlx4/qp.h19
-rw-r--r--include/rdma/ib_verbs.h73
-rw-r--r--include/uapi/rdma/ib_user_verbs.h16
-rw-r--r--net/sunrpc/xprtrdma/verbs.c20
43 files changed, 885 insertions, 253 deletions
diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h
index 5bcb2afd3dcb..0fcd7aa26fa2 100644
--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h
@@ -188,6 +188,8 @@ IB_UVERBS_DECLARE_CMD(alloc_pd);
188IB_UVERBS_DECLARE_CMD(dealloc_pd); 188IB_UVERBS_DECLARE_CMD(dealloc_pd);
189IB_UVERBS_DECLARE_CMD(reg_mr); 189IB_UVERBS_DECLARE_CMD(reg_mr);
190IB_UVERBS_DECLARE_CMD(dereg_mr); 190IB_UVERBS_DECLARE_CMD(dereg_mr);
191IB_UVERBS_DECLARE_CMD(alloc_mw);
192IB_UVERBS_DECLARE_CMD(dealloc_mw);
191IB_UVERBS_DECLARE_CMD(create_comp_channel); 193IB_UVERBS_DECLARE_CMD(create_comp_channel);
192IB_UVERBS_DECLARE_CMD(create_cq); 194IB_UVERBS_DECLARE_CMD(create_cq);
193IB_UVERBS_DECLARE_CMD(resize_cq); 195IB_UVERBS_DECLARE_CMD(resize_cq);
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 0cb0007724a2..3983a0552775 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -48,6 +48,7 @@ struct uverbs_lock_class {
48 48
49static struct uverbs_lock_class pd_lock_class = { .name = "PD-uobj" }; 49static struct uverbs_lock_class pd_lock_class = { .name = "PD-uobj" };
50static struct uverbs_lock_class mr_lock_class = { .name = "MR-uobj" }; 50static struct uverbs_lock_class mr_lock_class = { .name = "MR-uobj" };
51static struct uverbs_lock_class mw_lock_class = { .name = "MW-uobj" };
51static struct uverbs_lock_class cq_lock_class = { .name = "CQ-uobj" }; 52static struct uverbs_lock_class cq_lock_class = { .name = "CQ-uobj" };
52static struct uverbs_lock_class qp_lock_class = { .name = "QP-uobj" }; 53static struct uverbs_lock_class qp_lock_class = { .name = "QP-uobj" };
53static struct uverbs_lock_class ah_lock_class = { .name = "AH-uobj" }; 54static struct uverbs_lock_class ah_lock_class = { .name = "AH-uobj" };
@@ -1049,6 +1050,126 @@ ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
1049 return in_len; 1050 return in_len;
1050} 1051}
1051 1052
1053ssize_t ib_uverbs_alloc_mw(struct ib_uverbs_file *file,
1054 const char __user *buf, int in_len,
1055 int out_len)
1056{
1057 struct ib_uverbs_alloc_mw cmd;
1058 struct ib_uverbs_alloc_mw_resp resp;
1059 struct ib_uobject *uobj;
1060 struct ib_pd *pd;
1061 struct ib_mw *mw;
1062 int ret;
1063
1064 if (out_len < sizeof(resp))
1065 return -ENOSPC;
1066
1067 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1068 return -EFAULT;
1069
1070 uobj = kmalloc(sizeof(*uobj), GFP_KERNEL);
1071 if (!uobj)
1072 return -ENOMEM;
1073
1074 init_uobj(uobj, 0, file->ucontext, &mw_lock_class);
1075 down_write(&uobj->mutex);
1076
1077 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1078 if (!pd) {
1079 ret = -EINVAL;
1080 goto err_free;
1081 }
1082
1083 mw = pd->device->alloc_mw(pd, cmd.mw_type);
1084 if (IS_ERR(mw)) {
1085 ret = PTR_ERR(mw);
1086 goto err_put;
1087 }
1088
1089 mw->device = pd->device;
1090 mw->pd = pd;
1091 mw->uobject = uobj;
1092 atomic_inc(&pd->usecnt);
1093
1094 uobj->object = mw;
1095 ret = idr_add_uobj(&ib_uverbs_mw_idr, uobj);
1096 if (ret)
1097 goto err_unalloc;
1098
1099 memset(&resp, 0, sizeof(resp));
1100 resp.rkey = mw->rkey;
1101 resp.mw_handle = uobj->id;
1102
1103 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1104 &resp, sizeof(resp))) {
1105 ret = -EFAULT;
1106 goto err_copy;
1107 }
1108
1109 put_pd_read(pd);
1110
1111 mutex_lock(&file->mutex);
1112 list_add_tail(&uobj->list, &file->ucontext->mw_list);
1113 mutex_unlock(&file->mutex);
1114
1115 uobj->live = 1;
1116
1117 up_write(&uobj->mutex);
1118
1119 return in_len;
1120
1121err_copy:
1122 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1123
1124err_unalloc:
1125 ib_dealloc_mw(mw);
1126
1127err_put:
1128 put_pd_read(pd);
1129
1130err_free:
1131 put_uobj_write(uobj);
1132 return ret;
1133}
1134
1135ssize_t ib_uverbs_dealloc_mw(struct ib_uverbs_file *file,
1136 const char __user *buf, int in_len,
1137 int out_len)
1138{
1139 struct ib_uverbs_dealloc_mw cmd;
1140 struct ib_mw *mw;
1141 struct ib_uobject *uobj;
1142 int ret = -EINVAL;
1143
1144 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1145 return -EFAULT;
1146
1147 uobj = idr_write_uobj(&ib_uverbs_mw_idr, cmd.mw_handle, file->ucontext);
1148 if (!uobj)
1149 return -EINVAL;
1150
1151 mw = uobj->object;
1152
1153 ret = ib_dealloc_mw(mw);
1154 if (!ret)
1155 uobj->live = 0;
1156
1157 put_uobj_write(uobj);
1158
1159 if (ret)
1160 return ret;
1161
1162 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
1163
1164 mutex_lock(&file->mutex);
1165 list_del(&uobj->list);
1166 mutex_unlock(&file->mutex);
1167
1168 put_uobj(uobj);
1169
1170 return in_len;
1171}
1172
1052ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file, 1173ssize_t ib_uverbs_create_comp_channel(struct ib_uverbs_file *file,
1053 const char __user *buf, int in_len, 1174 const char __user *buf, int in_len,
1054 int out_len) 1175 int out_len)
diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index 6f2ce6fa98f8..2c6f0f2ecd9d 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -87,6 +87,8 @@ static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
87 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd, 87 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
88 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr, 88 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
89 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr, 89 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
90 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw,
91 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw,
90 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel, 92 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
91 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq, 93 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
92 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq, 94 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq,
@@ -201,6 +203,15 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
201 kfree(uobj); 203 kfree(uobj);
202 } 204 }
203 205
206 /* Remove MWs before QPs, in order to support type 2A MWs. */
207 list_for_each_entry_safe(uobj, tmp, &context->mw_list, list) {
208 struct ib_mw *mw = uobj->object;
209
210 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
211 ib_dealloc_mw(mw);
212 kfree(uobj);
213 }
214
204 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) { 215 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
205 struct ib_qp *qp = uobj->object; 216 struct ib_qp *qp = uobj->object;
206 struct ib_uqp_object *uqp = 217 struct ib_uqp_object *uqp =
@@ -240,8 +251,6 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
240 kfree(uevent); 251 kfree(uevent);
241 } 252 }
242 253
243 /* XXX Free MWs */
244
245 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) { 254 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
246 struct ib_mr *mr = uobj->object; 255 struct ib_mr *mr = uobj->object;
247 256
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 30f199e8579f..a8fdd3381405 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -1099,18 +1099,19 @@ EXPORT_SYMBOL(ib_free_fast_reg_page_list);
1099 1099
1100/* Memory windows */ 1100/* Memory windows */
1101 1101
1102struct ib_mw *ib_alloc_mw(struct ib_pd *pd) 1102struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
1103{ 1103{
1104 struct ib_mw *mw; 1104 struct ib_mw *mw;
1105 1105
1106 if (!pd->device->alloc_mw) 1106 if (!pd->device->alloc_mw)
1107 return ERR_PTR(-ENOSYS); 1107 return ERR_PTR(-ENOSYS);
1108 1108
1109 mw = pd->device->alloc_mw(pd); 1109 mw = pd->device->alloc_mw(pd, type);
1110 if (!IS_ERR(mw)) { 1110 if (!IS_ERR(mw)) {
1111 mw->device = pd->device; 1111 mw->device = pd->device;
1112 mw->pd = pd; 1112 mw->pd = pd;
1113 mw->uobject = NULL; 1113 mw->uobject = NULL;
1114 mw->type = type;
1114 atomic_inc(&pd->usecnt); 1115 atomic_inc(&pd->usecnt);
1115 } 1116 }
1116 1117
diff --git a/drivers/infiniband/hw/amso1100/c2.c b/drivers/infiniband/hw/amso1100/c2.c
index 7275e727e0f5..d53cf519f42a 100644
--- a/drivers/infiniband/hw/amso1100/c2.c
+++ b/drivers/infiniband/hw/amso1100/c2.c
@@ -1238,15 +1238,4 @@ static struct pci_driver c2_pci_driver = {
1238 .remove = c2_remove, 1238 .remove = c2_remove,
1239}; 1239};
1240 1240
1241static int __init c2_init_module(void) 1241module_pci_driver(c2_pci_driver);
1242{
1243 return pci_register_driver(&c2_pci_driver);
1244}
1245
1246static void __exit c2_exit_module(void)
1247{
1248 pci_unregister_driver(&c2_pci_driver);
1249}
1250
1251module_init(c2_init_module);
1252module_exit(c2_exit_module);
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c
index 145d82a64d0a..9c12da0cbd32 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c
@@ -738,7 +738,7 @@ static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
738 return ibmr; 738 return ibmr;
739} 739}
740 740
741static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd) 741static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
742{ 742{
743 struct iwch_dev *rhp; 743 struct iwch_dev *rhp;
744 struct iwch_pd *php; 744 struct iwch_pd *php;
@@ -747,6 +747,9 @@ static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
747 u32 stag = 0; 747 u32 stag = 0;
748 int ret; 748 int ret;
749 749
750 if (type != IB_MW_TYPE_1)
751 return ERR_PTR(-EINVAL);
752
750 php = to_iwch_pd(pd); 753 php = to_iwch_pd(pd);
751 rhp = php->rhp; 754 rhp = php->rhp;
752 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 755 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c
index 6de8463f453b..e5649e8b215d 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_qp.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c
@@ -567,18 +567,19 @@ int iwch_bind_mw(struct ib_qp *qp,
567 if (mw_bind->send_flags & IB_SEND_SIGNALED) 567 if (mw_bind->send_flags & IB_SEND_SIGNALED)
568 t3_wr_flags = T3_COMPLETION_FLAG; 568 t3_wr_flags = T3_COMPLETION_FLAG;
569 569
570 sgl.addr = mw_bind->addr; 570 sgl.addr = mw_bind->bind_info.addr;
571 sgl.lkey = mw_bind->mr->lkey; 571 sgl.lkey = mw_bind->bind_info.mr->lkey;
572 sgl.length = mw_bind->length; 572 sgl.length = mw_bind->bind_info.length;
573 wqe->bind.reserved = 0; 573 wqe->bind.reserved = 0;
574 wqe->bind.type = TPT_VATO; 574 wqe->bind.type = TPT_VATO;
575 575
576 /* TBD: check perms */ 576 /* TBD: check perms */
577 wqe->bind.perms = iwch_ib_to_tpt_bind_access(mw_bind->mw_access_flags); 577 wqe->bind.perms = iwch_ib_to_tpt_bind_access(
578 wqe->bind.mr_stag = cpu_to_be32(mw_bind->mr->lkey); 578 mw_bind->bind_info.mw_access_flags);
579 wqe->bind.mr_stag = cpu_to_be32(mw_bind->bind_info.mr->lkey);
579 wqe->bind.mw_stag = cpu_to_be32(mw->rkey); 580 wqe->bind.mw_stag = cpu_to_be32(mw->rkey);
580 wqe->bind.mw_len = cpu_to_be32(mw_bind->length); 581 wqe->bind.mw_len = cpu_to_be32(mw_bind->bind_info.length);
581 wqe->bind.mw_va = cpu_to_be64(mw_bind->addr); 582 wqe->bind.mw_va = cpu_to_be64(mw_bind->bind_info.addr);
582 err = iwch_sgl2pbl_map(rhp, &sgl, 1, &pbl_addr, &page_size); 583 err = iwch_sgl2pbl_map(rhp, &sgl, 1, &pbl_addr, &page_size);
583 if (err) { 584 if (err) {
584 spin_unlock_irqrestore(&qhp->lock, flag); 585 spin_unlock_irqrestore(&qhp->lock, flag);
diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index c13745cde7fa..565bfb161c1a 100644
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -143,14 +143,28 @@ static void connect_reply_upcall(struct c4iw_ep *ep, int status);
143static LIST_HEAD(timeout_list); 143static LIST_HEAD(timeout_list);
144static spinlock_t timeout_lock; 144static spinlock_t timeout_lock;
145 145
146static void deref_qp(struct c4iw_ep *ep)
147{
148 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
149 clear_bit(QP_REFERENCED, &ep->com.flags);
150}
151
152static void ref_qp(struct c4iw_ep *ep)
153{
154 set_bit(QP_REFERENCED, &ep->com.flags);
155 c4iw_qp_add_ref(&ep->com.qp->ibqp);
156}
157
146static void start_ep_timer(struct c4iw_ep *ep) 158static void start_ep_timer(struct c4iw_ep *ep)
147{ 159{
148 PDBG("%s ep %p\n", __func__, ep); 160 PDBG("%s ep %p\n", __func__, ep);
149 if (timer_pending(&ep->timer)) { 161 if (timer_pending(&ep->timer)) {
150 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep); 162 pr_err("%s timer already started! ep %p\n",
151 del_timer_sync(&ep->timer); 163 __func__, ep);
152 } else 164 return;
153 c4iw_get_ep(&ep->com); 165 }
166 clear_bit(TIMEOUT, &ep->com.flags);
167 c4iw_get_ep(&ep->com);
154 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 168 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
155 ep->timer.data = (unsigned long)ep; 169 ep->timer.data = (unsigned long)ep;
156 ep->timer.function = ep_timeout; 170 ep->timer.function = ep_timeout;
@@ -159,14 +173,10 @@ static void start_ep_timer(struct c4iw_ep *ep)
159 173
160static void stop_ep_timer(struct c4iw_ep *ep) 174static void stop_ep_timer(struct c4iw_ep *ep)
161{ 175{
162 PDBG("%s ep %p\n", __func__, ep); 176 PDBG("%s ep %p stopping\n", __func__, ep);
163 if (!timer_pending(&ep->timer)) {
164 WARN(1, "%s timer stopped when its not running! "
165 "ep %p state %u\n", __func__, ep, ep->com.state);
166 return;
167 }
168 del_timer_sync(&ep->timer); 177 del_timer_sync(&ep->timer);
169 c4iw_put_ep(&ep->com); 178 if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
179 c4iw_put_ep(&ep->com);
170} 180}
171 181
172static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb, 182static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
@@ -271,11 +281,13 @@ void _c4iw_free_ep(struct kref *kref)
271 281
272 ep = container_of(kref, struct c4iw_ep, com.kref); 282 ep = container_of(kref, struct c4iw_ep, com.kref);
273 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]); 283 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
284 if (test_bit(QP_REFERENCED, &ep->com.flags))
285 deref_qp(ep);
274 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) { 286 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
287 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
275 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 288 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
276 dst_release(ep->dst); 289 dst_release(ep->dst);
277 cxgb4_l2t_release(ep->l2t); 290 cxgb4_l2t_release(ep->l2t);
278 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
279 } 291 }
280 kfree(ep); 292 kfree(ep);
281} 293}
@@ -687,7 +699,7 @@ static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
687 memset(mpa, 0, sizeof(*mpa)); 699 memset(mpa, 0, sizeof(*mpa));
688 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 700 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
689 mpa->flags = MPA_REJECT; 701 mpa->flags = MPA_REJECT;
690 mpa->revision = mpa_rev; 702 mpa->revision = ep->mpa_attr.version;
691 mpa->private_data_size = htons(plen); 703 mpa->private_data_size = htons(plen);
692 704
693 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 705 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
@@ -863,7 +875,6 @@ static void close_complete_upcall(struct c4iw_ep *ep)
863 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 875 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
864 ep->com.cm_id->rem_ref(ep->com.cm_id); 876 ep->com.cm_id->rem_ref(ep->com.cm_id);
865 ep->com.cm_id = NULL; 877 ep->com.cm_id = NULL;
866 ep->com.qp = NULL;
867 set_bit(CLOSE_UPCALL, &ep->com.history); 878 set_bit(CLOSE_UPCALL, &ep->com.history);
868 } 879 }
869} 880}
@@ -906,7 +917,6 @@ static void peer_abort_upcall(struct c4iw_ep *ep)
906 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 917 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
907 ep->com.cm_id->rem_ref(ep->com.cm_id); 918 ep->com.cm_id->rem_ref(ep->com.cm_id);
908 ep->com.cm_id = NULL; 919 ep->com.cm_id = NULL;
909 ep->com.qp = NULL;
910 set_bit(ABORT_UPCALL, &ep->com.history); 920 set_bit(ABORT_UPCALL, &ep->com.history);
911 } 921 }
912} 922}
@@ -946,7 +956,6 @@ static void connect_reply_upcall(struct c4iw_ep *ep, int status)
946 if (status < 0) { 956 if (status < 0) {
947 ep->com.cm_id->rem_ref(ep->com.cm_id); 957 ep->com.cm_id->rem_ref(ep->com.cm_id);
948 ep->com.cm_id = NULL; 958 ep->com.cm_id = NULL;
949 ep->com.qp = NULL;
950 } 959 }
951} 960}
952 961
@@ -1291,11 +1300,13 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1291 if (mpa->revision > mpa_rev) { 1300 if (mpa->revision > mpa_rev) {
1292 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d," 1301 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1293 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1302 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1303 stop_ep_timer(ep);
1294 abort_connection(ep, skb, GFP_KERNEL); 1304 abort_connection(ep, skb, GFP_KERNEL);
1295 return; 1305 return;
1296 } 1306 }
1297 1307
1298 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) { 1308 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1309 stop_ep_timer(ep);
1299 abort_connection(ep, skb, GFP_KERNEL); 1310 abort_connection(ep, skb, GFP_KERNEL);
1300 return; 1311 return;
1301 } 1312 }
@@ -1306,6 +1317,7 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1306 * Fail if there's too much private data. 1317 * Fail if there's too much private data.
1307 */ 1318 */
1308 if (plen > MPA_MAX_PRIVATE_DATA) { 1319 if (plen > MPA_MAX_PRIVATE_DATA) {
1320 stop_ep_timer(ep);
1309 abort_connection(ep, skb, GFP_KERNEL); 1321 abort_connection(ep, skb, GFP_KERNEL);
1310 return; 1322 return;
1311 } 1323 }
@@ -1314,6 +1326,7 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1314 * If plen does not account for pkt size 1326 * If plen does not account for pkt size
1315 */ 1327 */
1316 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1328 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1329 stop_ep_timer(ep);
1317 abort_connection(ep, skb, GFP_KERNEL); 1330 abort_connection(ep, skb, GFP_KERNEL);
1318 return; 1331 return;
1319 } 1332 }
@@ -1391,30 +1404,33 @@ static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1391 skb_pull(skb, sizeof(*hdr)); 1404 skb_pull(skb, sizeof(*hdr));
1392 skb_trim(skb, dlen); 1405 skb_trim(skb, dlen);
1393 1406
1394 ep->rcv_seq += dlen;
1395 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1396
1397 /* update RX credits */ 1407 /* update RX credits */
1398 update_rx_credits(ep, dlen); 1408 update_rx_credits(ep, dlen);
1399 1409
1400 switch (state_read(&ep->com)) { 1410 switch (state_read(&ep->com)) {
1401 case MPA_REQ_SENT: 1411 case MPA_REQ_SENT:
1412 ep->rcv_seq += dlen;
1402 process_mpa_reply(ep, skb); 1413 process_mpa_reply(ep, skb);
1403 break; 1414 break;
1404 case MPA_REQ_WAIT: 1415 case MPA_REQ_WAIT:
1416 ep->rcv_seq += dlen;
1405 process_mpa_request(ep, skb); 1417 process_mpa_request(ep, skb);
1406 break; 1418 break;
1407 case MPA_REP_SENT: 1419 case FPDU_MODE: {
1420 struct c4iw_qp_attributes attrs;
1421 BUG_ON(!ep->com.qp);
1422 if (status)
1423 pr_err("%s Unexpected streaming data." \
1424 " qpid %u ep %p state %d tid %u status %d\n",
1425 __func__, ep->com.qp->wq.sq.qid, ep,
1426 state_read(&ep->com), ep->hwtid, status);
1427 attrs.next_state = C4IW_QP_STATE_ERROR;
1428 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1429 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1430 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
1408 break; 1431 break;
1432 }
1409 default: 1433 default:
1410 pr_err("%s Unexpected streaming data." \
1411 " ep %p state %d tid %u status %d\n",
1412 __func__, ep, state_read(&ep->com), ep->hwtid, status);
1413
1414 /*
1415 * The ep will timeout and inform the ULP of the failure.
1416 * See ep_timeout().
1417 */
1418 break; 1434 break;
1419 } 1435 }
1420 return 0; 1436 return 0;
@@ -1437,6 +1453,7 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1437 mutex_lock(&ep->com.mutex); 1453 mutex_lock(&ep->com.mutex);
1438 switch (ep->com.state) { 1454 switch (ep->com.state) {
1439 case ABORTING: 1455 case ABORTING:
1456 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1440 __state_set(&ep->com, DEAD); 1457 __state_set(&ep->com, DEAD);
1441 release = 1; 1458 release = 1;
1442 break; 1459 break;
@@ -1475,11 +1492,11 @@ static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1475 V_FW_OFLD_CONNECTION_WR_ASTID(atid)); 1492 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1476 req->tcb.cplrxdataack_cplpassacceptrpl = 1493 req->tcb.cplrxdataack_cplpassacceptrpl =
1477 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK); 1494 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1478 req->tcb.tx_max = jiffies; 1495 req->tcb.tx_max = (__force __be32) jiffies;
1479 req->tcb.rcv_adv = htons(1); 1496 req->tcb.rcv_adv = htons(1);
1480 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx); 1497 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1481 wscale = compute_wscale(rcv_win); 1498 wscale = compute_wscale(rcv_win);
1482 req->tcb.opt0 = TCAM_BYPASS(1) | 1499 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
1483 (nocong ? NO_CONG(1) : 0) | 1500 (nocong ? NO_CONG(1) : 0) |
1484 KEEP_ALIVE(1) | 1501 KEEP_ALIVE(1) |
1485 DELACK(1) | 1502 DELACK(1) |
@@ -1490,20 +1507,20 @@ static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1490 SMAC_SEL(ep->smac_idx) | 1507 SMAC_SEL(ep->smac_idx) |
1491 DSCP(ep->tos) | 1508 DSCP(ep->tos) |
1492 ULP_MODE(ULP_MODE_TCPDDP) | 1509 ULP_MODE(ULP_MODE_TCPDDP) |
1493 RCV_BUFSIZ(rcv_win >> 10); 1510 RCV_BUFSIZ(rcv_win >> 10));
1494 req->tcb.opt2 = PACE(1) | 1511 req->tcb.opt2 = (__force __be32) (PACE(1) |
1495 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) | 1512 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1496 RX_CHANNEL(0) | 1513 RX_CHANNEL(0) |
1497 CCTRL_ECN(enable_ecn) | 1514 CCTRL_ECN(enable_ecn) |
1498 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid); 1515 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
1499 if (enable_tcp_timestamps) 1516 if (enable_tcp_timestamps)
1500 req->tcb.opt2 |= TSTAMPS_EN(1); 1517 req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
1501 if (enable_tcp_sack) 1518 if (enable_tcp_sack)
1502 req->tcb.opt2 |= SACK_EN(1); 1519 req->tcb.opt2 |= (__force __be32) SACK_EN(1);
1503 if (wscale && enable_tcp_window_scaling) 1520 if (wscale && enable_tcp_window_scaling)
1504 req->tcb.opt2 |= WND_SCALE_EN(1); 1521 req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
1505 req->tcb.opt0 = cpu_to_be64(req->tcb.opt0); 1522 req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
1506 req->tcb.opt2 = cpu_to_be32(req->tcb.opt2); 1523 req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
1507 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 1524 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1508 set_bit(ACT_OFLD_CONN, &ep->com.history); 1525 set_bit(ACT_OFLD_CONN, &ep->com.history);
1509 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1526 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
@@ -1993,6 +2010,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1993 2010
1994 init_timer(&child_ep->timer); 2011 init_timer(&child_ep->timer);
1995 cxgb4_insert_tid(t, child_ep, hwtid); 2012 cxgb4_insert_tid(t, child_ep, hwtid);
2013 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
1996 accept_cr(child_ep, peer_ip, skb, req); 2014 accept_cr(child_ep, peer_ip, skb, req);
1997 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history); 2015 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
1998 goto out; 2016 goto out;
@@ -2018,7 +2036,6 @@ static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2018 ntohs(req->tcp_opt)); 2036 ntohs(req->tcp_opt));
2019 2037
2020 set_emss(ep, ntohs(req->tcp_opt)); 2038 set_emss(ep, ntohs(req->tcp_opt));
2021 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
2022 2039
2023 dst_confirm(ep->dst); 2040 dst_confirm(ep->dst);
2024 state_set(&ep->com, MPA_REQ_WAIT); 2041 state_set(&ep->com, MPA_REQ_WAIT);
@@ -2163,7 +2180,7 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2163 break; 2180 break;
2164 case MPA_REQ_SENT: 2181 case MPA_REQ_SENT:
2165 stop_ep_timer(ep); 2182 stop_ep_timer(ep);
2166 if (mpa_rev == 2 && ep->tried_with_mpa_v1) 2183 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2167 connect_reply_upcall(ep, -ECONNRESET); 2184 connect_reply_upcall(ep, -ECONNRESET);
2168 else { 2185 else {
2169 /* 2186 /*
@@ -2235,9 +2252,8 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2235out: 2252out:
2236 if (release) 2253 if (release)
2237 release_ep_resources(ep); 2254 release_ep_resources(ep);
2238 2255 else if (ep->retry_with_mpa_v1) {
2239 /* retry with mpa-v1 */ 2256 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2240 if (ep && ep->retry_with_mpa_v1) {
2241 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 2257 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2242 dst_release(ep->dst); 2258 dst_release(ep->dst);
2243 cxgb4_l2t_release(ep->l2t); 2259 cxgb4_l2t_release(ep->l2t);
@@ -2430,6 +2446,7 @@ int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2430 cm_id->add_ref(cm_id); 2446 cm_id->add_ref(cm_id);
2431 ep->com.cm_id = cm_id; 2447 ep->com.cm_id = cm_id;
2432 ep->com.qp = qp; 2448 ep->com.qp = qp;
2449 ref_qp(ep);
2433 2450
2434 /* bind QP to EP and move to RTS */ 2451 /* bind QP to EP and move to RTS */
2435 attrs.mpa_attr = ep->mpa_attr; 2452 attrs.mpa_attr = ep->mpa_attr;
@@ -2460,7 +2477,6 @@ int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2460 return 0; 2477 return 0;
2461err1: 2478err1:
2462 ep->com.cm_id = NULL; 2479 ep->com.cm_id = NULL;
2463 ep->com.qp = NULL;
2464 cm_id->rem_ref(cm_id); 2480 cm_id->rem_ref(cm_id);
2465err: 2481err:
2466 c4iw_put_ep(&ep->com); 2482 c4iw_put_ep(&ep->com);
@@ -2501,6 +2517,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2501 ep->com.cm_id = cm_id; 2517 ep->com.cm_id = cm_id;
2502 ep->com.qp = get_qhp(dev, conn_param->qpn); 2518 ep->com.qp = get_qhp(dev, conn_param->qpn);
2503 BUG_ON(!ep->com.qp); 2519 BUG_ON(!ep->com.qp);
2520 ref_qp(ep);
2504 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn, 2521 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2505 ep->com.qp, cm_id); 2522 ep->com.qp, cm_id);
2506 2523
@@ -2756,7 +2773,8 @@ static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2756 struct c4iw_ep *ep; 2773 struct c4iw_ep *ep;
2757 int atid = be32_to_cpu(req->tid); 2774 int atid = be32_to_cpu(req->tid);
2758 2775
2759 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, req->tid); 2776 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
2777 (__force u32) req->tid);
2760 if (!ep) 2778 if (!ep)
2761 return; 2779 return;
2762 2780
@@ -2800,7 +2818,7 @@ static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2800 struct cpl_pass_accept_req *cpl; 2818 struct cpl_pass_accept_req *cpl;
2801 int ret; 2819 int ret;
2802 2820
2803 rpl_skb = (struct sk_buff *)cpu_to_be64(req->cookie); 2821 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
2804 BUG_ON(!rpl_skb); 2822 BUG_ON(!rpl_skb);
2805 if (req->retval) { 2823 if (req->retval) {
2806 PDBG("%s passive open failure %d\n", __func__, req->retval); 2824 PDBG("%s passive open failure %d\n", __func__, req->retval);
@@ -2811,7 +2829,8 @@ static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2811 } else { 2829 } else {
2812 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb); 2830 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
2813 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 2831 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
2814 htonl(req->tid))); 2832 (__force u32) htonl(
2833 (__force u32) req->tid)));
2815 ret = pass_accept_req(dev, rpl_skb); 2834 ret = pass_accept_req(dev, rpl_skb);
2816 if (!ret) 2835 if (!ret)
2817 kfree_skb(rpl_skb); 2836 kfree_skb(rpl_skb);
@@ -2857,10 +2876,10 @@ static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
2857 struct tcp_options_received tmp_opt; 2876 struct tcp_options_received tmp_opt;
2858 2877
2859 /* Store values from cpl_rx_pkt in temporary location. */ 2878 /* Store values from cpl_rx_pkt in temporary location. */
2860 vlantag = cpl->vlan; 2879 vlantag = (__force u16) cpl->vlan;
2861 len = cpl->len; 2880 len = (__force u16) cpl->len;
2862 l2info = cpl->l2info; 2881 l2info = (__force u32) cpl->l2info;
2863 hdr_len = cpl->hdr_len; 2882 hdr_len = (__force u16) cpl->hdr_len;
2864 intf = cpl->iff; 2883 intf = cpl->iff;
2865 2884
2866 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header)); 2885 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
@@ -2871,19 +2890,24 @@ static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
2871 */ 2890 */
2872 memset(&tmp_opt, 0, sizeof(tmp_opt)); 2891 memset(&tmp_opt, 0, sizeof(tmp_opt));
2873 tcp_clear_options(&tmp_opt); 2892 tcp_clear_options(&tmp_opt);
2874 tcp_parse_options(skb, &tmp_opt, 0, 0, NULL); 2893 tcp_parse_options(skb, &tmp_opt, NULL, 0, NULL);
2875 2894
2876 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req)); 2895 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
2877 memset(req, 0, sizeof(*req)); 2896 memset(req, 0, sizeof(*req));
2878 req->l2info = cpu_to_be16(V_SYN_INTF(intf) | 2897 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
2879 V_SYN_MAC_IDX(G_RX_MACIDX(htonl(l2info))) | 2898 V_SYN_MAC_IDX(G_RX_MACIDX(
2899 (__force int) htonl(l2info))) |
2880 F_SYN_XACT_MATCH); 2900 F_SYN_XACT_MATCH);
2881 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(htonl(l2info))) | 2901 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
2882 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(htons(hdr_len))) | 2902 (__force int) htonl(l2info))) |
2883 V_IP_HDR_LEN(G_RX_IPHDR_LEN(htons(hdr_len))) | 2903 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
2884 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(htonl(l2info)))); 2904 (__force int) htons(hdr_len))) |
2885 req->vlan = vlantag; 2905 V_IP_HDR_LEN(G_RX_IPHDR_LEN(
2886 req->len = len; 2906 (__force int) htons(hdr_len))) |
2907 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(
2908 (__force int) htonl(l2info))));
2909 req->vlan = (__force __be16) vlantag;
2910 req->len = (__force __be16) len;
2887 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) | 2911 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
2888 PASS_OPEN_TOS(tos)); 2912 PASS_OPEN_TOS(tos));
2889 req->tcpopt.mss = htons(tmp_opt.mss_clamp); 2913 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
@@ -2912,7 +2936,7 @@ static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
2912 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1)); 2936 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
2913 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16))); 2937 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
2914 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL); 2938 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
2915 req->le.filter = filter; 2939 req->le.filter = (__force __be32) filter;
2916 req->le.lport = lport; 2940 req->le.lport = lport;
2917 req->le.pport = rport; 2941 req->le.pport = rport;
2918 req->le.u.ipv4.lip = laddr; 2942 req->le.u.ipv4.lip = laddr;
@@ -2938,7 +2962,7 @@ static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
2938 * TP will ignore any value > 0 for MSS index. 2962 * TP will ignore any value > 0 for MSS index.
2939 */ 2963 */
2940 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF)); 2964 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
2941 req->cookie = cpu_to_be64((u64)skb); 2965 req->cookie = (unsigned long)skb;
2942 2966
2943 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id); 2967 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
2944 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb); 2968 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
@@ -2988,7 +3012,8 @@ static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
2988 /* 3012 /*
2989 * Calculate the server tid from filter hit index from cpl_rx_pkt. 3013 * Calculate the server tid from filter hit index from cpl_rx_pkt.
2990 */ 3014 */
2991 stid = cpu_to_be32(rss->hash_val) - dev->rdev.lldi.tids->sftid_base 3015 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val)
3016 - dev->rdev.lldi.tids->sftid_base
2992 + dev->rdev.lldi.tids->nstids; 3017 + dev->rdev.lldi.tids->nstids;
2993 3018
2994 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid); 3019 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
@@ -3049,10 +3074,10 @@ static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3049 3074
3050 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan; 3075 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3051 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step]; 3076 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3052 window = htons(tcph->window); 3077 window = (__force u16) htons((__force u16)tcph->window);
3053 3078
3054 /* Calcuate filter portion for LE region. */ 3079 /* Calcuate filter portion for LE region. */
3055 filter = cpu_to_be32(select_ntuple(dev, dst, e)); 3080 filter = (__force unsigned int) cpu_to_be32(select_ntuple(dev, dst, e));
3056 3081
3057 /* 3082 /*
3058 * Synthesize the cpl_pass_accept_req. We have everything except the 3083 * Synthesize the cpl_pass_accept_req. We have everything except the
@@ -3175,11 +3200,16 @@ static DECLARE_WORK(skb_work, process_work);
3175static void ep_timeout(unsigned long arg) 3200static void ep_timeout(unsigned long arg)
3176{ 3201{
3177 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 3202 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3203 int kickit = 0;
3178 3204
3179 spin_lock(&timeout_lock); 3205 spin_lock(&timeout_lock);
3180 list_add_tail(&ep->entry, &timeout_list); 3206 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
3207 list_add_tail(&ep->entry, &timeout_list);
3208 kickit = 1;
3209 }
3181 spin_unlock(&timeout_lock); 3210 spin_unlock(&timeout_lock);
3182 queue_work(workq, &skb_work); 3211 if (kickit)
3212 queue_work(workq, &skb_work);
3183} 3213}
3184 3214
3185/* 3215/*
@@ -3268,8 +3298,14 @@ static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3268 3298
3269 /* 3299 /*
3270 * Wake up any threads in rdma_init() or rdma_fini(). 3300 * Wake up any threads in rdma_init() or rdma_fini().
3301 * However, if we are on MPAv2 and want to retry with MPAv1
3302 * then, don't wake up yet.
3271 */ 3303 */
3272 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 3304 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
3305 if (ep->com.state != MPA_REQ_SENT)
3306 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3307 } else
3308 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3273 sched(dev, skb); 3309 sched(dev, skb);
3274 return 0; 3310 return 0;
3275} 3311}
diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c
index ba11c76c0b5a..80069ad595c1 100644
--- a/drivers/infiniband/hw/cxgb4/device.c
+++ b/drivers/infiniband/hw/cxgb4/device.c
@@ -533,7 +533,7 @@ static int c4iw_rdev_open(struct c4iw_rdev *rdev)
533 PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p qpshift %lu " 533 PDBG("udb len 0x%x udb base %p db_reg %p gts_reg %p qpshift %lu "
534 "qpmask 0x%x cqshift %lu cqmask 0x%x\n", 534 "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
535 (unsigned)pci_resource_len(rdev->lldi.pdev, 2), 535 (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
536 (void *)pci_resource_start(rdev->lldi.pdev, 2), 536 (void *)(unsigned long)pci_resource_start(rdev->lldi.pdev, 2),
537 rdev->lldi.db_reg, 537 rdev->lldi.db_reg,
538 rdev->lldi.gts_reg, 538 rdev->lldi.gts_reg,
539 rdev->qpshift, rdev->qpmask, 539 rdev->qpshift, rdev->qpmask,
@@ -797,7 +797,8 @@ static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
797 "RSS %#llx, FL %#llx, len %u\n", 797 "RSS %#llx, FL %#llx, len %u\n",
798 pci_name(ctx->lldi.pdev), gl->va, 798 pci_name(ctx->lldi.pdev), gl->va,
799 (unsigned long long)be64_to_cpu(*rsp), 799 (unsigned long long)be64_to_cpu(*rsp),
800 (unsigned long long)be64_to_cpu(*(u64 *)gl->va), 800 (unsigned long long)be64_to_cpu(
801 *(__force __be64 *)gl->va),
801 gl->tot_len); 802 gl->tot_len);
802 803
803 return 0; 804 return 0;
diff --git a/drivers/infiniband/hw/cxgb4/ev.c b/drivers/infiniband/hw/cxgb4/ev.c
index cf2f6b47617a..1a840b2211dd 100644
--- a/drivers/infiniband/hw/cxgb4/ev.c
+++ b/drivers/infiniband/hw/cxgb4/ev.c
@@ -46,9 +46,11 @@ static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
46 46
47 if ((qhp->attr.state == C4IW_QP_STATE_ERROR) || 47 if ((qhp->attr.state == C4IW_QP_STATE_ERROR) ||
48 (qhp->attr.state == C4IW_QP_STATE_TERMINATE)) { 48 (qhp->attr.state == C4IW_QP_STATE_TERMINATE)) {
49 PDBG("%s AE received after RTS - " 49 pr_err("%s AE after RTS - qpid 0x%x opcode %d status 0x%x "\
50 "qp state %d qpid 0x%x status 0x%x\n", __func__, 50 "type %d wrid.hi 0x%x wrid.lo 0x%x\n",
51 qhp->attr.state, qhp->wq.sq.qid, CQE_STATUS(err_cqe)); 51 __func__, CQE_QPID(err_cqe), CQE_OPCODE(err_cqe),
52 CQE_STATUS(err_cqe), CQE_TYPE(err_cqe),
53 CQE_WRID_HI(err_cqe), CQE_WRID_LOW(err_cqe));
52 return; 54 return;
53 } 55 }
54 56
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
index 9c1644fb0259..4c07fc069766 100644
--- a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -716,6 +716,8 @@ enum c4iw_ep_flags {
716 ABORT_REQ_IN_PROGRESS = 1, 716 ABORT_REQ_IN_PROGRESS = 1,
717 RELEASE_RESOURCES = 2, 717 RELEASE_RESOURCES = 2,
718 CLOSE_SENT = 3, 718 CLOSE_SENT = 3,
719 TIMEOUT = 4,
720 QP_REFERENCED = 5,
719}; 721};
720 722
721enum c4iw_ep_history { 723enum c4iw_ep_history {
@@ -866,7 +868,7 @@ struct ib_fast_reg_page_list *c4iw_alloc_fastreg_pbl(
866 int page_list_len); 868 int page_list_len);
867struct ib_mr *c4iw_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth); 869struct ib_mr *c4iw_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth);
868int c4iw_dealloc_mw(struct ib_mw *mw); 870int c4iw_dealloc_mw(struct ib_mw *mw);
869struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd); 871struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
870struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, 872struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start,
871 u64 length, u64 virt, int acc, 873 u64 length, u64 virt, int acc,
872 struct ib_udata *udata); 874 struct ib_udata *udata);
diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c
index afd81790ab3c..903a92d6f91d 100644
--- a/drivers/infiniband/hw/cxgb4/mem.c
+++ b/drivers/infiniband/hw/cxgb4/mem.c
@@ -650,7 +650,7 @@ err:
650 return ERR_PTR(err); 650 return ERR_PTR(err);
651} 651}
652 652
653struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd) 653struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
654{ 654{
655 struct c4iw_dev *rhp; 655 struct c4iw_dev *rhp;
656 struct c4iw_pd *php; 656 struct c4iw_pd *php;
@@ -659,6 +659,9 @@ struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd)
659 u32 stag = 0; 659 u32 stag = 0;
660 int ret; 660 int ret;
661 661
662 if (type != IB_MW_TYPE_1)
663 return ERR_PTR(-EINVAL);
664
662 php = to_c4iw_pd(pd); 665 php = to_c4iw_pd(pd);
663 rhp = php->rhp; 666 rhp = php->rhp;
664 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 667 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index 05bfe53bff64..17ba4f8bc12d 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -1383,6 +1383,7 @@ err:
1383 qhp->ep = NULL; 1383 qhp->ep = NULL;
1384 set_state(qhp, C4IW_QP_STATE_ERROR); 1384 set_state(qhp, C4IW_QP_STATE_ERROR);
1385 free = 1; 1385 free = 1;
1386 abort = 1;
1386 wake_up(&qhp->wait); 1387 wake_up(&qhp->wait);
1387 BUG_ON(!ep); 1388 BUG_ON(!ep);
1388 flush_qp(qhp); 1389 flush_qp(qhp);
diff --git a/drivers/infiniband/hw/ehca/ehca_iverbs.h b/drivers/infiniband/hw/ehca/ehca_iverbs.h
index 8f7f282ead65..22f79afa7fc1 100644
--- a/drivers/infiniband/hw/ehca/ehca_iverbs.h
+++ b/drivers/infiniband/hw/ehca/ehca_iverbs.h
@@ -95,7 +95,7 @@ int ehca_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
95 95
96int ehca_dereg_mr(struct ib_mr *mr); 96int ehca_dereg_mr(struct ib_mr *mr);
97 97
98struct ib_mw *ehca_alloc_mw(struct ib_pd *pd); 98struct ib_mw *ehca_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
99 99
100int ehca_bind_mw(struct ib_qp *qp, struct ib_mw *mw, 100int ehca_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
101 struct ib_mw_bind *mw_bind); 101 struct ib_mw_bind *mw_bind);
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c
index 87844869dcc2..bcfb0c183620 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.c
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c
@@ -688,7 +688,7 @@ dereg_mr_exit0:
688 688
689/*----------------------------------------------------------------------*/ 689/*----------------------------------------------------------------------*/
690 690
691struct ib_mw *ehca_alloc_mw(struct ib_pd *pd) 691struct ib_mw *ehca_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
692{ 692{
693 struct ib_mw *ib_mw; 693 struct ib_mw *ib_mw;
694 u64 h_ret; 694 u64 h_ret;
@@ -698,6 +698,9 @@ struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
698 container_of(pd->device, struct ehca_shca, ib_device); 698 container_of(pd->device, struct ehca_shca, ib_device);
699 struct ehca_mw_hipzout_parms hipzout; 699 struct ehca_mw_hipzout_parms hipzout;
700 700
701 if (type != IB_MW_TYPE_1)
702 return ERR_PTR(-EINVAL);
703
701 e_mw = ehca_mw_new(); 704 e_mw = ehca_mw_new();
702 if (!e_mw) { 705 if (!e_mw) {
703 ib_mw = ERR_PTR(-ENOMEM); 706 ib_mw = ERR_PTR(-ENOMEM);
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 0a903c129f0a..934792c477bc 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -1999,16 +1999,17 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
1999 goto demux_err; 1999 goto demux_err;
2000 err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1); 2000 err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1);
2001 if (err) 2001 if (err)
2002 goto demux_err; 2002 goto free_pv;
2003 } 2003 }
2004 mlx4_ib_master_tunnels(dev, 1); 2004 mlx4_ib_master_tunnels(dev, 1);
2005 return 0; 2005 return 0;
2006 2006
2007free_pv:
2008 free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
2007demux_err: 2009demux_err:
2008 while (i > 0) { 2010 while (--i >= 0) {
2009 free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1); 2011 free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
2010 mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); 2012 mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
2011 --i;
2012 } 2013 }
2013 mlx4_ib_device_unregister_sysfs(dev); 2014 mlx4_ib_device_unregister_sysfs(dev);
2014 2015
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index e7d81c0d1ac5..23d734349d8e 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -137,6 +137,14 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
137 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; 137 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
138 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) 138 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
139 props->device_cap_flags |= IB_DEVICE_XRC; 139 props->device_cap_flags |= IB_DEVICE_XRC;
140 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
141 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
142 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
143 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
144 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
145 else
146 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
147 }
140 148
141 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) & 149 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
142 0xffffff; 150 0xffffff;
@@ -1434,6 +1442,17 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
1434 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; 1442 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1435 } 1443 }
1436 1444
1445 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1446 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1447 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1448 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1449 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1450
1451 ibdev->ib_dev.uverbs_cmd_mask |=
1452 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1453 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1454 }
1455
1437 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) { 1456 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1438 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd; 1457 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1439 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd; 1458 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
@@ -1601,8 +1620,7 @@ static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1601 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags); 1620 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1602 } 1621 }
1603out: 1622out:
1604 if (dm) 1623 kfree(dm);
1605 kfree(dm);
1606 return; 1624 return;
1607} 1625}
1608 1626
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index dcd845bc30f0..f61ec26500c4 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -116,6 +116,11 @@ struct mlx4_ib_mr {
116 struct ib_umem *umem; 116 struct ib_umem *umem;
117}; 117};
118 118
119struct mlx4_ib_mw {
120 struct ib_mw ibmw;
121 struct mlx4_mw mmw;
122};
123
119struct mlx4_ib_fast_reg_page_list { 124struct mlx4_ib_fast_reg_page_list {
120 struct ib_fast_reg_page_list ibfrpl; 125 struct ib_fast_reg_page_list ibfrpl;
121 __be64 *mapped_page_list; 126 __be64 *mapped_page_list;
@@ -533,6 +538,11 @@ static inline struct mlx4_ib_mr *to_mmr(struct ib_mr *ibmr)
533 return container_of(ibmr, struct mlx4_ib_mr, ibmr); 538 return container_of(ibmr, struct mlx4_ib_mr, ibmr);
534} 539}
535 540
541static inline struct mlx4_ib_mw *to_mmw(struct ib_mw *ibmw)
542{
543 return container_of(ibmw, struct mlx4_ib_mw, ibmw);
544}
545
536static inline struct mlx4_ib_fast_reg_page_list *to_mfrpl(struct ib_fast_reg_page_list *ibfrpl) 546static inline struct mlx4_ib_fast_reg_page_list *to_mfrpl(struct ib_fast_reg_page_list *ibfrpl)
537{ 547{
538 return container_of(ibfrpl, struct mlx4_ib_fast_reg_page_list, ibfrpl); 548 return container_of(ibfrpl, struct mlx4_ib_fast_reg_page_list, ibfrpl);
@@ -581,6 +591,10 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
581 u64 virt_addr, int access_flags, 591 u64 virt_addr, int access_flags,
582 struct ib_udata *udata); 592 struct ib_udata *udata);
583int mlx4_ib_dereg_mr(struct ib_mr *mr); 593int mlx4_ib_dereg_mr(struct ib_mr *mr);
594struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
595int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
596 struct ib_mw_bind *mw_bind);
597int mlx4_ib_dealloc_mw(struct ib_mw *mw);
584struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd, 598struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
585 int max_page_list_len); 599 int max_page_list_len);
586struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device *ibdev, 600struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
@@ -652,12 +666,12 @@ int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
652int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr, 666int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr,
653 u8 *mac, int *is_mcast, u8 port); 667 u8 *mac, int *is_mcast, u8 port);
654 668
655static inline int mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah) 669static inline bool mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah)
656{ 670{
657 u8 port = be32_to_cpu(ah->av.ib.port_pd) >> 24 & 3; 671 u8 port = be32_to_cpu(ah->av.ib.port_pd) >> 24 & 3;
658 672
659 if (rdma_port_get_link_layer(ah->ibah.device, port) == IB_LINK_LAYER_ETHERNET) 673 if (rdma_port_get_link_layer(ah->ibah.device, port) == IB_LINK_LAYER_ETHERNET)
660 return 1; 674 return true;
661 675
662 return !!(ah->av.ib.g_slid & 0x80); 676 return !!(ah->av.ib.g_slid & 0x80);
663} 677}
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index bbaf6176f207..e471f089ff00 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -41,9 +41,19 @@ static u32 convert_access(int acc)
41 (acc & IB_ACCESS_REMOTE_WRITE ? MLX4_PERM_REMOTE_WRITE : 0) | 41 (acc & IB_ACCESS_REMOTE_WRITE ? MLX4_PERM_REMOTE_WRITE : 0) |
42 (acc & IB_ACCESS_REMOTE_READ ? MLX4_PERM_REMOTE_READ : 0) | 42 (acc & IB_ACCESS_REMOTE_READ ? MLX4_PERM_REMOTE_READ : 0) |
43 (acc & IB_ACCESS_LOCAL_WRITE ? MLX4_PERM_LOCAL_WRITE : 0) | 43 (acc & IB_ACCESS_LOCAL_WRITE ? MLX4_PERM_LOCAL_WRITE : 0) |
44 (acc & IB_ACCESS_MW_BIND ? MLX4_PERM_BIND_MW : 0) |
44 MLX4_PERM_LOCAL_READ; 45 MLX4_PERM_LOCAL_READ;
45} 46}
46 47
48static enum mlx4_mw_type to_mlx4_type(enum ib_mw_type type)
49{
50 switch (type) {
51 case IB_MW_TYPE_1: return MLX4_MW_TYPE_1;
52 case IB_MW_TYPE_2: return MLX4_MW_TYPE_2;
53 default: return -1;
54 }
55}
56
47struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc) 57struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
48{ 58{
49 struct mlx4_ib_mr *mr; 59 struct mlx4_ib_mr *mr;
@@ -68,7 +78,7 @@ struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
68 return &mr->ibmr; 78 return &mr->ibmr;
69 79
70err_mr: 80err_mr:
71 mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr); 81 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
72 82
73err_free: 83err_free:
74 kfree(mr); 84 kfree(mr);
@@ -163,7 +173,7 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
163 return &mr->ibmr; 173 return &mr->ibmr;
164 174
165err_mr: 175err_mr:
166 mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr); 176 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
167 177
168err_umem: 178err_umem:
169 ib_umem_release(mr->umem); 179 ib_umem_release(mr->umem);
@@ -177,8 +187,11 @@ err_free:
177int mlx4_ib_dereg_mr(struct ib_mr *ibmr) 187int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
178{ 188{
179 struct mlx4_ib_mr *mr = to_mmr(ibmr); 189 struct mlx4_ib_mr *mr = to_mmr(ibmr);
190 int ret;
180 191
181 mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr); 192 ret = mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr);
193 if (ret)
194 return ret;
182 if (mr->umem) 195 if (mr->umem)
183 ib_umem_release(mr->umem); 196 ib_umem_release(mr->umem);
184 kfree(mr); 197 kfree(mr);
@@ -186,6 +199,70 @@ int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
186 return 0; 199 return 0;
187} 200}
188 201
202struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
203{
204 struct mlx4_ib_dev *dev = to_mdev(pd->device);
205 struct mlx4_ib_mw *mw;
206 int err;
207
208 mw = kmalloc(sizeof(*mw), GFP_KERNEL);
209 if (!mw)
210 return ERR_PTR(-ENOMEM);
211
212 err = mlx4_mw_alloc(dev->dev, to_mpd(pd)->pdn,
213 to_mlx4_type(type), &mw->mmw);
214 if (err)
215 goto err_free;
216
217 err = mlx4_mw_enable(dev->dev, &mw->mmw);
218 if (err)
219 goto err_mw;
220
221 mw->ibmw.rkey = mw->mmw.key;
222
223 return &mw->ibmw;
224
225err_mw:
226 mlx4_mw_free(dev->dev, &mw->mmw);
227
228err_free:
229 kfree(mw);
230
231 return ERR_PTR(err);
232}
233
234int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
235 struct ib_mw_bind *mw_bind)
236{
237 struct ib_send_wr wr;
238 struct ib_send_wr *bad_wr;
239 int ret;
240
241 memset(&wr, 0, sizeof(wr));
242 wr.opcode = IB_WR_BIND_MW;
243 wr.wr_id = mw_bind->wr_id;
244 wr.send_flags = mw_bind->send_flags;
245 wr.wr.bind_mw.mw = mw;
246 wr.wr.bind_mw.bind_info = mw_bind->bind_info;
247 wr.wr.bind_mw.rkey = ib_inc_rkey(mw->rkey);
248
249 ret = mlx4_ib_post_send(qp, &wr, &bad_wr);
250 if (!ret)
251 mw->rkey = wr.wr.bind_mw.rkey;
252
253 return ret;
254}
255
256int mlx4_ib_dealloc_mw(struct ib_mw *ibmw)
257{
258 struct mlx4_ib_mw *mw = to_mmw(ibmw);
259
260 mlx4_mw_free(to_mdev(ibmw->device)->dev, &mw->mmw);
261 kfree(mw);
262
263 return 0;
264}
265
189struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd, 266struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
190 int max_page_list_len) 267 int max_page_list_len)
191{ 268{
@@ -212,7 +289,7 @@ struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
212 return &mr->ibmr; 289 return &mr->ibmr;
213 290
214err_mr: 291err_mr:
215 mlx4_mr_free(dev->dev, &mr->mmr); 292 (void) mlx4_mr_free(dev->dev, &mr->mmr);
216 293
217err_free: 294err_free:
218 kfree(mr); 295 kfree(mr);
@@ -291,7 +368,7 @@ struct ib_fmr *mlx4_ib_fmr_alloc(struct ib_pd *pd, int acc,
291 return &fmr->ibfmr; 368 return &fmr->ibfmr;
292 369
293err_mr: 370err_mr:
294 mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr); 371 (void) mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr);
295 372
296err_free: 373err_free:
297 kfree(fmr); 374 kfree(fmr);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 19e0637220b9..35cced2a4da8 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -104,6 +104,7 @@ static const __be32 mlx4_ib_opcode[] = {
104 [IB_WR_FAST_REG_MR] = cpu_to_be32(MLX4_OPCODE_FMR), 104 [IB_WR_FAST_REG_MR] = cpu_to_be32(MLX4_OPCODE_FMR),
105 [IB_WR_MASKED_ATOMIC_CMP_AND_SWP] = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_CS), 105 [IB_WR_MASKED_ATOMIC_CMP_AND_SWP] = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_CS),
106 [IB_WR_MASKED_ATOMIC_FETCH_AND_ADD] = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_FA), 106 [IB_WR_MASKED_ATOMIC_FETCH_AND_ADD] = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_FA),
107 [IB_WR_BIND_MW] = cpu_to_be32(MLX4_OPCODE_BIND_MW),
107}; 108};
108 109
109static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp) 110static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
@@ -1746,11 +1747,11 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
1746 int header_size; 1747 int header_size;
1747 int spc; 1748 int spc;
1748 int i; 1749 int i;
1749 int is_eth;
1750 int is_vlan = 0;
1751 int is_grh;
1752 u16 vlan;
1753 int err = 0; 1750 int err = 0;
1751 u16 vlan = 0xffff;
1752 bool is_eth;
1753 bool is_vlan = false;
1754 bool is_grh;
1754 1755
1755 send_size = 0; 1756 send_size = 0;
1756 for (i = 0; i < wr->num_sge; ++i) 1757 for (i = 0; i < wr->num_sge; ++i)
@@ -1953,9 +1954,12 @@ static int mlx4_wq_overflow(struct mlx4_ib_wq *wq, int nreq, struct ib_cq *ib_cq
1953 1954
1954static __be32 convert_access(int acc) 1955static __be32 convert_access(int acc)
1955{ 1956{
1956 return (acc & IB_ACCESS_REMOTE_ATOMIC ? cpu_to_be32(MLX4_WQE_FMR_PERM_ATOMIC) : 0) | 1957 return (acc & IB_ACCESS_REMOTE_ATOMIC ?
1957 (acc & IB_ACCESS_REMOTE_WRITE ? cpu_to_be32(MLX4_WQE_FMR_PERM_REMOTE_WRITE) : 0) | 1958 cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC) : 0) |
1958 (acc & IB_ACCESS_REMOTE_READ ? cpu_to_be32(MLX4_WQE_FMR_PERM_REMOTE_READ) : 0) | 1959 (acc & IB_ACCESS_REMOTE_WRITE ?
1960 cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE) : 0) |
1961 (acc & IB_ACCESS_REMOTE_READ ?
1962 cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ) : 0) |
1959 (acc & IB_ACCESS_LOCAL_WRITE ? cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_WRITE) : 0) | 1963 (acc & IB_ACCESS_LOCAL_WRITE ? cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_WRITE) : 0) |
1960 cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_READ); 1964 cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_READ);
1961} 1965}
@@ -1981,12 +1985,28 @@ static void set_fmr_seg(struct mlx4_wqe_fmr_seg *fseg, struct ib_send_wr *wr)
1981 fseg->reserved[1] = 0; 1985 fseg->reserved[1] = 0;
1982} 1986}
1983 1987
1988static void set_bind_seg(struct mlx4_wqe_bind_seg *bseg, struct ib_send_wr *wr)
1989{
1990 bseg->flags1 =
1991 convert_access(wr->wr.bind_mw.bind_info.mw_access_flags) &
1992 cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ |
1993 MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE |
1994 MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC);
1995 bseg->flags2 = 0;
1996 if (wr->wr.bind_mw.mw->type == IB_MW_TYPE_2)
1997 bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_TYPE_2);
1998 if (wr->wr.bind_mw.bind_info.mw_access_flags & IB_ZERO_BASED)
1999 bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_ZERO_BASED);
2000 bseg->new_rkey = cpu_to_be32(wr->wr.bind_mw.rkey);
2001 bseg->lkey = cpu_to_be32(wr->wr.bind_mw.bind_info.mr->lkey);
2002 bseg->addr = cpu_to_be64(wr->wr.bind_mw.bind_info.addr);
2003 bseg->length = cpu_to_be64(wr->wr.bind_mw.bind_info.length);
2004}
2005
1984static void set_local_inv_seg(struct mlx4_wqe_local_inval_seg *iseg, u32 rkey) 2006static void set_local_inv_seg(struct mlx4_wqe_local_inval_seg *iseg, u32 rkey)
1985{ 2007{
1986 iseg->flags = 0; 2008 memset(iseg, 0, sizeof(*iseg));
1987 iseg->mem_key = cpu_to_be32(rkey); 2009 iseg->mem_key = cpu_to_be32(rkey);
1988 iseg->guest_id = 0;
1989 iseg->pa = 0;
1990} 2010}
1991 2011
1992static __always_inline void set_raddr_seg(struct mlx4_wqe_raddr_seg *rseg, 2012static __always_inline void set_raddr_seg(struct mlx4_wqe_raddr_seg *rseg,
@@ -2291,6 +2311,13 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
2291 size += sizeof (struct mlx4_wqe_fmr_seg) / 16; 2311 size += sizeof (struct mlx4_wqe_fmr_seg) / 16;
2292 break; 2312 break;
2293 2313
2314 case IB_WR_BIND_MW:
2315 ctrl->srcrb_flags |=
2316 cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
2317 set_bind_seg(wqe, wr);
2318 wqe += sizeof(struct mlx4_wqe_bind_seg);
2319 size += sizeof(struct mlx4_wqe_bind_seg) / 16;
2320 break;
2294 default: 2321 default:
2295 /* No extra segments required for sends */ 2322 /* No extra segments required for sends */
2296 break; 2323 break;
diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
index 5b2a01dfb907..97516eb363b7 100644
--- a/drivers/infiniband/hw/mlx4/sysfs.c
+++ b/drivers/infiniband/hw/mlx4/sysfs.c
@@ -732,7 +732,7 @@ int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *dev)
732 dev->ports_parent = 732 dev->ports_parent =
733 kobject_create_and_add("ports", 733 kobject_create_and_add("ports",
734 kobject_get(dev->iov_parent)); 734 kobject_get(dev->iov_parent));
735 if (!dev->iov_parent) { 735 if (!dev->ports_parent) {
736 ret = -ENOMEM; 736 ret = -ENOMEM;
737 goto err_ports; 737 goto err_ports;
738 } 738 }
diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c
index 07e4fbad987a..8f67fe2e91e6 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.c
+++ b/drivers/infiniband/hw/nes/nes_verbs.c
@@ -55,7 +55,8 @@ static void nes_unregister_ofa_device(struct nes_ib_device *nesibdev);
55/** 55/**
56 * nes_alloc_mw 56 * nes_alloc_mw
57 */ 57 */
58static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) { 58static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd, enum ib_mw_type type)
59{
59 struct nes_pd *nespd = to_nespd(ibpd); 60 struct nes_pd *nespd = to_nespd(ibpd);
60 struct nes_vnic *nesvnic = to_nesvnic(ibpd->device); 61 struct nes_vnic *nesvnic = to_nesvnic(ibpd->device);
61 struct nes_device *nesdev = nesvnic->nesdev; 62 struct nes_device *nesdev = nesvnic->nesdev;
@@ -71,6 +72,9 @@ static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) {
71 u32 driver_key = 0; 72 u32 driver_key = 0;
72 u8 stag_key = 0; 73 u8 stag_key = 0;
73 74
75 if (type != IB_MW_TYPE_1)
76 return ERR_PTR(-EINVAL);
77
74 get_random_bytes(&next_stag_index, sizeof(next_stag_index)); 78 get_random_bytes(&next_stag_index, sizeof(next_stag_index));
75 stag_key = (u8)next_stag_index; 79 stag_key = (u8)next_stag_index;
76 80
@@ -244,20 +248,19 @@ static int nes_bind_mw(struct ib_qp *ibqp, struct ib_mw *ibmw,
244 if (ibmw_bind->send_flags & IB_SEND_SIGNALED) 248 if (ibmw_bind->send_flags & IB_SEND_SIGNALED)
245 wqe_misc |= NES_IWARP_SQ_WQE_SIGNALED_COMPL; 249 wqe_misc |= NES_IWARP_SQ_WQE_SIGNALED_COMPL;
246 250
247 if (ibmw_bind->mw_access_flags & IB_ACCESS_REMOTE_WRITE) { 251 if (ibmw_bind->bind_info.mw_access_flags & IB_ACCESS_REMOTE_WRITE)
248 wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_WRITE; 252 wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_WRITE;
249 } 253 if (ibmw_bind->bind_info.mw_access_flags & IB_ACCESS_REMOTE_READ)
250 if (ibmw_bind->mw_access_flags & IB_ACCESS_REMOTE_READ) {
251 wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_READ; 254 wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_READ;
252 }
253 255
254 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_MISC_IDX, wqe_misc); 256 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_MISC_IDX, wqe_misc);
255 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MR_IDX, ibmw_bind->mr->lkey); 257 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MR_IDX,
258 ibmw_bind->bind_info.mr->lkey);
256 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MW_IDX, ibmw->rkey); 259 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MW_IDX, ibmw->rkey);
257 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_LENGTH_LOW_IDX, 260 set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_LENGTH_LOW_IDX,
258 ibmw_bind->length); 261 ibmw_bind->bind_info.length);
259 wqe->wqe_words[NES_IWARP_SQ_BIND_WQE_LENGTH_HIGH_IDX] = 0; 262 wqe->wqe_words[NES_IWARP_SQ_BIND_WQE_LENGTH_HIGH_IDX] = 0;
260 u64temp = (u64)ibmw_bind->addr; 263 u64temp = (u64)ibmw_bind->bind_info.addr;
261 set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_VA_FBO_LOW_IDX, u64temp); 264 set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_VA_FBO_LOW_IDX, u64temp);
262 265
263 head++; 266 head++;
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index 35275099cafd..a6a2cc2ba260 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -268,8 +268,9 @@ static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp)
268 qpp = &q->next) 268 qpp = &q->next)
269 if (q == qp) { 269 if (q == qp) {
270 atomic_dec(&qp->refcount); 270 atomic_dec(&qp->refcount);
271 *qpp = qp->next; 271 rcu_assign_pointer(*qpp,
272 rcu_assign_pointer(qp->next, NULL); 272 rcu_dereference_protected(qp->next,
273 lockdep_is_held(&dev->qpt_lock)));
273 break; 274 break;
274 } 275 }
275 } 276 }
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 07ca6fd5546b..eb71aaa26a9a 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -117,6 +117,8 @@ enum {
117#define IPOIB_OP_CM (0) 117#define IPOIB_OP_CM (0)
118#endif 118#endif
119 119
120#define IPOIB_QPN_MASK ((__force u32) cpu_to_be32(0xFFFFFF))
121
120/* structs */ 122/* structs */
121 123
122struct ipoib_header { 124struct ipoib_header {
@@ -760,4 +762,6 @@ extern int ipoib_debug_level;
760 762
761#define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff) 763#define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
762 764
765extern const char ipoib_driver_version[];
766
763#endif /* _IPOIB_H */ 767#endif /* _IPOIB_H */
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
index ca131335417b..c4b3940845e6 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
@@ -39,7 +39,24 @@
39static void ipoib_get_drvinfo(struct net_device *netdev, 39static void ipoib_get_drvinfo(struct net_device *netdev,
40 struct ethtool_drvinfo *drvinfo) 40 struct ethtool_drvinfo *drvinfo)
41{ 41{
42 strlcpy(drvinfo->driver, "ipoib", sizeof(drvinfo->driver)); 42 struct ipoib_dev_priv *priv = netdev_priv(netdev);
43 struct ib_device_attr *attr;
44
45 attr = kmalloc(sizeof(*attr), GFP_KERNEL);
46 if (attr && !ib_query_device(priv->ca, attr))
47 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
48 "%d.%d.%d", (int)(attr->fw_ver >> 32),
49 (int)(attr->fw_ver >> 16) & 0xffff,
50 (int)attr->fw_ver & 0xffff);
51 kfree(attr);
52
53 strlcpy(drvinfo->bus_info, dev_name(priv->ca->dma_device),
54 sizeof(drvinfo->bus_info));
55
56 strlcpy(drvinfo->version, ipoib_driver_version,
57 sizeof(drvinfo->version));
58
59 strlcpy(drvinfo->driver, "ib_ipoib", sizeof(drvinfo->driver));
43} 60}
44 61
45static int ipoib_get_coalesce(struct net_device *dev, 62static int ipoib_get_coalesce(struct net_device *dev,
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 6fdc9e78da0d..8534afd04e7c 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -49,9 +49,14 @@
49#include <linux/jhash.h> 49#include <linux/jhash.h>
50#include <net/arp.h> 50#include <net/arp.h>
51 51
52#define DRV_VERSION "1.0.0"
53
54const char ipoib_driver_version[] = DRV_VERSION;
55
52MODULE_AUTHOR("Roland Dreier"); 56MODULE_AUTHOR("Roland Dreier");
53MODULE_DESCRIPTION("IP-over-InfiniBand net driver"); 57MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
54MODULE_LICENSE("Dual BSD/GPL"); 58MODULE_LICENSE("Dual BSD/GPL");
59MODULE_VERSION(DRV_VERSION);
55 60
56int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE; 61int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
57int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE; 62int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
@@ -505,6 +510,9 @@ static void path_rec_completion(int status,
505 510
506 spin_unlock_irqrestore(&priv->lock, flags); 511 spin_unlock_irqrestore(&priv->lock, flags);
507 512
513 if (IS_ERR_OR_NULL(ah))
514 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
515
508 if (old_ah) 516 if (old_ah)
509 ipoib_put_ah(old_ah); 517 ipoib_put_ah(old_ah);
510 518
@@ -844,10 +852,10 @@ static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
844 * different subnets. 852 * different subnets.
845 */ 853 */
846 /* qpn octets[1:4) & port GUID octets[12:20) */ 854 /* qpn octets[1:4) & port GUID octets[12:20) */
847 u32 *daddr_32 = (u32 *) daddr; 855 u32 *d32 = (u32 *) daddr;
848 u32 hv; 856 u32 hv;
849 857
850 hv = jhash_3words(daddr_32[3], daddr_32[4], 0xFFFFFF & daddr_32[0], 0); 858 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
851 return hv & htbl->mask; 859 return hv & htbl->mask;
852} 860}
853 861
@@ -1688,6 +1696,8 @@ static void ipoib_remove_one(struct ib_device *device)
1688 return; 1696 return;
1689 1697
1690 dev_list = ib_get_client_data(device, &ipoib_client); 1698 dev_list = ib_get_client_data(device, &ipoib_client);
1699 if (!dev_list)
1700 return;
1691 1701
1692 list_for_each_entry_safe(priv, tmp, dev_list, list) { 1702 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1693 ib_unregister_event_handler(&priv->event_handler); 1703 ib_unregister_event_handler(&priv->event_handler);
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index ef7d3be46c31..5babdb35bda7 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -94,7 +94,7 @@
94 94
95 /* support up to 512KB in one RDMA */ 95 /* support up to 512KB in one RDMA */
96#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K) 96#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
97#define ISER_DEF_CMD_PER_LUN 128 97#define ISER_DEF_CMD_PER_LUN ISCSI_DEF_XMIT_CMDS_MAX
98 98
99/* QP settings */ 99/* QP settings */
100/* Maximal bounds on received asynchronous PDUs */ 100/* Maximal bounds on received asynchronous PDUs */
diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c
index 2033a928d34d..be1edb04b085 100644
--- a/drivers/infiniband/ulp/iser/iser_memory.c
+++ b/drivers/infiniband/ulp/iser/iser_memory.c
@@ -369,10 +369,11 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task,
369 regd_buf = &iser_task->rdma_regd[cmd_dir]; 369 regd_buf = &iser_task->rdma_regd[cmd_dir];
370 370
371 aligned_len = iser_data_buf_aligned_len(mem, ibdev); 371 aligned_len = iser_data_buf_aligned_len(mem, ibdev);
372 if (aligned_len != mem->dma_nents) { 372 if (aligned_len != mem->dma_nents ||
373 (!ib_conn->fmr_pool && mem->dma_nents > 1)) {
373 iscsi_conn->fmr_unalign_cnt++; 374 iscsi_conn->fmr_unalign_cnt++;
374 iser_warn("rdma alignment violation %d/%d aligned\n", 375 iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n",
375 aligned_len, mem->size); 376 aligned_len, mem->size);
376 iser_data_buf_dump(mem, ibdev); 377 iser_data_buf_dump(mem, ibdev);
377 378
378 /* unmap the command data before accessing it */ 379 /* unmap the command data before accessing it */
@@ -404,7 +405,7 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task,
404 } else { /* use FMR for multiple dma entries */ 405 } else { /* use FMR for multiple dma entries */
405 iser_page_vec_build(mem, ib_conn->page_vec, ibdev); 406 iser_page_vec_build(mem, ib_conn->page_vec, ibdev);
406 err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg); 407 err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg);
407 if (err) { 408 if (err && err != -EAGAIN) {
408 iser_data_buf_dump(mem, ibdev); 409 iser_data_buf_dump(mem, ibdev);
409 iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", 410 iser_err("mem->dma_nents = %d (dlength = 0x%x)\n",
410 mem->dma_nents, 411 mem->dma_nents,
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index 95a49affee44..4debadc53106 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -242,10 +242,14 @@ static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
242 IB_ACCESS_REMOTE_READ); 242 IB_ACCESS_REMOTE_READ);
243 243
244 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, &params); 244 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, &params);
245 if (IS_ERR(ib_conn->fmr_pool)) { 245 ret = PTR_ERR(ib_conn->fmr_pool);
246 ret = PTR_ERR(ib_conn->fmr_pool); 246 if (IS_ERR(ib_conn->fmr_pool) && ret != -ENOSYS) {
247 ib_conn->fmr_pool = NULL; 247 ib_conn->fmr_pool = NULL;
248 goto out_err; 248 goto out_err;
249 } else if (ret == -ENOSYS) {
250 ib_conn->fmr_pool = NULL;
251 iser_warn("FMRs are not supported, using unaligned mode\n");
252 ret = 0;
249 } 253 }
250 254
251 memset(&init_attr, 0, sizeof init_attr); 255 memset(&init_attr, 0, sizeof init_attr);
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index d5088ce78290..7ccf3284dda3 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -700,23 +700,24 @@ static int srp_reconnect_target(struct srp_target_port *target)
700 struct Scsi_Host *shost = target->scsi_host; 700 struct Scsi_Host *shost = target->scsi_host;
701 int i, ret; 701 int i, ret;
702 702
703 if (target->state != SRP_TARGET_LIVE)
704 return -EAGAIN;
705
706 scsi_target_block(&shost->shost_gendev); 703 scsi_target_block(&shost->shost_gendev);
707 704
708 srp_disconnect_target(target); 705 srp_disconnect_target(target);
709 /* 706 /*
710 * Now get a new local CM ID so that we avoid confusing the 707 * Now get a new local CM ID so that we avoid confusing the target in
711 * target in case things are really fouled up. 708 * case things are really fouled up. Doing so also ensures that all CM
709 * callbacks will have finished before a new QP is allocated.
712 */ 710 */
713 ret = srp_new_cm_id(target); 711 ret = srp_new_cm_id(target);
714 if (ret) 712 /*
715 goto unblock; 713 * Whether or not creating a new CM ID succeeded, create a new
716 714 * QP. This guarantees that all completion callback function
717 ret = srp_create_target_ib(target); 715 * invocations have finished before request resetting starts.
718 if (ret) 716 */
719 goto unblock; 717 if (ret == 0)
718 ret = srp_create_target_ib(target);
719 else
720 srp_create_target_ib(target);
720 721
721 for (i = 0; i < SRP_CMD_SQ_SIZE; ++i) { 722 for (i = 0; i < SRP_CMD_SQ_SIZE; ++i) {
722 struct srp_request *req = &target->req_ring[i]; 723 struct srp_request *req = &target->req_ring[i];
@@ -728,11 +729,12 @@ static int srp_reconnect_target(struct srp_target_port *target)
728 for (i = 0; i < SRP_SQ_SIZE; ++i) 729 for (i = 0; i < SRP_SQ_SIZE; ++i)
729 list_add(&target->tx_ring[i]->list, &target->free_tx); 730 list_add(&target->tx_ring[i]->list, &target->free_tx);
730 731
731 ret = srp_connect_target(target); 732 if (ret == 0)
733 ret = srp_connect_target(target);
732 734
733unblock:
734 scsi_target_unblock(&shost->shost_gendev, ret == 0 ? SDEV_RUNNING : 735 scsi_target_unblock(&shost->shost_gendev, ret == 0 ? SDEV_RUNNING :
735 SDEV_TRANSPORT_OFFLINE); 736 SDEV_TRANSPORT_OFFLINE);
737 target->transport_offline = !!ret;
736 738
737 if (ret) 739 if (ret)
738 goto err; 740 goto err;
@@ -1352,6 +1354,12 @@ static int srp_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *scmnd)
1352 unsigned long flags; 1354 unsigned long flags;
1353 int len; 1355 int len;
1354 1356
1357 if (unlikely(target->transport_offline)) {
1358 scmnd->result = DID_NO_CONNECT << 16;
1359 scmnd->scsi_done(scmnd);
1360 return 0;
1361 }
1362
1355 spin_lock_irqsave(&target->lock, flags); 1363 spin_lock_irqsave(&target->lock, flags);
1356 iu = __srp_get_tx_iu(target, SRP_IU_CMD); 1364 iu = __srp_get_tx_iu(target, SRP_IU_CMD);
1357 if (!iu) 1365 if (!iu)
@@ -1695,6 +1703,9 @@ static int srp_send_tsk_mgmt(struct srp_target_port *target,
1695 struct srp_iu *iu; 1703 struct srp_iu *iu;
1696 struct srp_tsk_mgmt *tsk_mgmt; 1704 struct srp_tsk_mgmt *tsk_mgmt;
1697 1705
1706 if (!target->connected || target->qp_in_error)
1707 return -1;
1708
1698 init_completion(&target->tsk_mgmt_done); 1709 init_completion(&target->tsk_mgmt_done);
1699 1710
1700 spin_lock_irq(&target->lock); 1711 spin_lock_irq(&target->lock);
@@ -1736,7 +1747,7 @@ static int srp_abort(struct scsi_cmnd *scmnd)
1736 1747
1737 shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n"); 1748 shost_printk(KERN_ERR, target->scsi_host, "SRP abort called\n");
1738 1749
1739 if (!req || target->qp_in_error || !srp_claim_req(target, req, scmnd)) 1750 if (!req || !srp_claim_req(target, req, scmnd))
1740 return FAILED; 1751 return FAILED;
1741 srp_send_tsk_mgmt(target, req->index, scmnd->device->lun, 1752 srp_send_tsk_mgmt(target, req->index, scmnd->device->lun,
1742 SRP_TSK_ABORT_TASK); 1753 SRP_TSK_ABORT_TASK);
@@ -1754,8 +1765,6 @@ static int srp_reset_device(struct scsi_cmnd *scmnd)
1754 1765
1755 shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n"); 1766 shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
1756 1767
1757 if (target->qp_in_error)
1758 return FAILED;
1759 if (srp_send_tsk_mgmt(target, SRP_TAG_NO_REQ, scmnd->device->lun, 1768 if (srp_send_tsk_mgmt(target, SRP_TAG_NO_REQ, scmnd->device->lun,
1760 SRP_TSK_LUN_RESET)) 1769 SRP_TSK_LUN_RESET))
1761 return FAILED; 1770 return FAILED;
@@ -1972,7 +1981,6 @@ static int srp_add_target(struct srp_host *host, struct srp_target_port *target)
1972 spin_unlock(&host->target_lock); 1981 spin_unlock(&host->target_lock);
1973 1982
1974 target->state = SRP_TARGET_LIVE; 1983 target->state = SRP_TARGET_LIVE;
1975 target->connected = false;
1976 1984
1977 scsi_scan_target(&target->scsi_host->shost_gendev, 1985 scsi_scan_target(&target->scsi_host->shost_gendev,
1978 0, target->scsi_id, SCAN_WILD_CARD, 0); 1986 0, target->scsi_id, SCAN_WILD_CARD, 0);
diff --git a/drivers/infiniband/ulp/srp/ib_srp.h b/drivers/infiniband/ulp/srp/ib_srp.h
index de2d0b3c0bfe..66fbedda4571 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.h
+++ b/drivers/infiniband/ulp/srp/ib_srp.h
@@ -140,6 +140,7 @@ struct srp_target_port {
140 unsigned int cmd_sg_cnt; 140 unsigned int cmd_sg_cnt;
141 unsigned int indirect_size; 141 unsigned int indirect_size;
142 bool allow_ext_sg; 142 bool allow_ext_sg;
143 bool transport_offline;
143 144
144 /* Everything above this point is used in the hot path of 145 /* Everything above this point is used in the hot path of
145 * command processing. Try to keep them packed into cachelines. 146 * command processing. Try to keep them packed into cachelines.
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_main.c b/drivers/net/ethernet/mellanox/mlx4/en_main.c
index b2cca58de910..fc27800e9c38 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_main.c
@@ -198,7 +198,7 @@ static void mlx4_en_remove(struct mlx4_dev *dev, void *endev_ptr)
198 198
199 flush_workqueue(mdev->workqueue); 199 flush_workqueue(mdev->workqueue);
200 destroy_workqueue(mdev->workqueue); 200 destroy_workqueue(mdev->workqueue);
201 mlx4_mr_free(dev, &mdev->mr); 201 (void) mlx4_mr_free(dev, &mdev->mr);
202 iounmap(mdev->uar_map); 202 iounmap(mdev->uar_map);
203 mlx4_uar_free(dev, &mdev->priv_uar); 203 mlx4_uar_free(dev, &mdev->priv_uar);
204 mlx4_pd_free(dev, mdev->priv_pdn); 204 mlx4_pd_free(dev, mdev->priv_pdn);
@@ -303,7 +303,7 @@ static void *mlx4_en_add(struct mlx4_dev *dev)
303 return mdev; 303 return mdev;
304 304
305err_mr: 305err_mr:
306 mlx4_mr_free(dev, &mdev->mr); 306 (void) mlx4_mr_free(dev, &mdev->mr);
307err_map: 307err_map:
308 if (!mdev->uar_map) 308 if (!mdev->uar_map)
309 iounmap(mdev->uar_map); 309 iounmap(mdev->uar_map);
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 38b62c78d5da..50917eb3013e 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -762,15 +762,19 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
762 u64 flags; 762 u64 flags;
763 int err = 0; 763 int err = 0;
764 u8 field; 764 u8 field;
765 u32 bmme_flags;
765 766
766 err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP, 767 err = mlx4_cmd_box(dev, 0, outbox->dma, 0, 0, MLX4_CMD_QUERY_DEV_CAP,
767 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); 768 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
768 if (err) 769 if (err)
769 return err; 770 return err;
770 771
771 /* add port mng change event capability unconditionally to slaves */ 772 /* add port mng change event capability and disable mw type 1
773 * unconditionally to slaves
774 */
772 MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET); 775 MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
773 flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV; 776 flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV;
777 flags &= ~MLX4_DEV_CAP_FLAG_MEM_WINDOW;
774 MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET); 778 MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET);
775 779
776 /* For guests, report Blueflame disabled */ 780 /* For guests, report Blueflame disabled */
@@ -778,6 +782,11 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
778 field &= 0x7f; 782 field &= 0x7f;
779 MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_BF_OFFSET); 783 MLX4_PUT(outbox->buf, field, QUERY_DEV_CAP_BF_OFFSET);
780 784
785 /* For guests, disable mw type 2 */
786 MLX4_GET(bmme_flags, outbox, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
787 bmme_flags &= ~MLX4_BMME_FLAG_TYPE_2_WIN;
788 MLX4_PUT(outbox->buf, bmme_flags, QUERY_DEV_CAP_BMME_FLAGS_OFFSET);
789
781 return 0; 790 return 0;
782} 791}
783 792
@@ -1203,6 +1212,7 @@ int mlx4_INIT_HCA(struct mlx4_dev *dev, struct mlx4_init_hca_param *param)
1203#define INIT_HCA_FS_IB_NUM_ADDRS_OFFSET (INIT_HCA_FS_PARAM_OFFSET + 0x26) 1212#define INIT_HCA_FS_IB_NUM_ADDRS_OFFSET (INIT_HCA_FS_PARAM_OFFSET + 0x26)
1204#define INIT_HCA_TPT_OFFSET 0x0f0 1213#define INIT_HCA_TPT_OFFSET 0x0f0
1205#define INIT_HCA_DMPT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x00) 1214#define INIT_HCA_DMPT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x00)
1215#define INIT_HCA_TPT_MW_OFFSET (INIT_HCA_TPT_OFFSET + 0x08)
1206#define INIT_HCA_LOG_MPT_SZ_OFFSET (INIT_HCA_TPT_OFFSET + 0x0b) 1216#define INIT_HCA_LOG_MPT_SZ_OFFSET (INIT_HCA_TPT_OFFSET + 0x0b)
1207#define INIT_HCA_MTT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x10) 1217#define INIT_HCA_MTT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x10)
1208#define INIT_HCA_CMPT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x18) 1218#define INIT_HCA_CMPT_BASE_OFFSET (INIT_HCA_TPT_OFFSET + 0x18)
@@ -1319,6 +1329,7 @@ int mlx4_INIT_HCA(struct mlx4_dev *dev, struct mlx4_init_hca_param *param)
1319 /* TPT attributes */ 1329 /* TPT attributes */
1320 1330
1321 MLX4_PUT(inbox, param->dmpt_base, INIT_HCA_DMPT_BASE_OFFSET); 1331 MLX4_PUT(inbox, param->dmpt_base, INIT_HCA_DMPT_BASE_OFFSET);
1332 MLX4_PUT(inbox, param->mw_enabled, INIT_HCA_TPT_MW_OFFSET);
1322 MLX4_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET); 1333 MLX4_PUT(inbox, param->log_mpt_sz, INIT_HCA_LOG_MPT_SZ_OFFSET);
1323 MLX4_PUT(inbox, param->mtt_base, INIT_HCA_MTT_BASE_OFFSET); 1334 MLX4_PUT(inbox, param->mtt_base, INIT_HCA_MTT_BASE_OFFSET);
1324 MLX4_PUT(inbox, param->cmpt_base, INIT_HCA_CMPT_BASE_OFFSET); 1335 MLX4_PUT(inbox, param->cmpt_base, INIT_HCA_CMPT_BASE_OFFSET);
@@ -1415,6 +1426,7 @@ int mlx4_QUERY_HCA(struct mlx4_dev *dev,
1415 /* TPT attributes */ 1426 /* TPT attributes */
1416 1427
1417 MLX4_GET(param->dmpt_base, outbox, INIT_HCA_DMPT_BASE_OFFSET); 1428 MLX4_GET(param->dmpt_base, outbox, INIT_HCA_DMPT_BASE_OFFSET);
1429 MLX4_GET(param->mw_enabled, outbox, INIT_HCA_TPT_MW_OFFSET);
1418 MLX4_GET(param->log_mpt_sz, outbox, INIT_HCA_LOG_MPT_SZ_OFFSET); 1430 MLX4_GET(param->log_mpt_sz, outbox, INIT_HCA_LOG_MPT_SZ_OFFSET);
1419 MLX4_GET(param->mtt_base, outbox, INIT_HCA_MTT_BASE_OFFSET); 1431 MLX4_GET(param->mtt_base, outbox, INIT_HCA_MTT_BASE_OFFSET);
1420 MLX4_GET(param->cmpt_base, outbox, INIT_HCA_CMPT_BASE_OFFSET); 1432 MLX4_GET(param->cmpt_base, outbox, INIT_HCA_CMPT_BASE_OFFSET);
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h
index 3af33ff669cc..151c2bb380a6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.h
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.h
@@ -170,6 +170,7 @@ struct mlx4_init_hca_param {
170 u8 log_mc_table_sz; 170 u8 log_mc_table_sz;
171 u8 log_mpt_sz; 171 u8 log_mpt_sz;
172 u8 log_uar_sz; 172 u8 log_uar_sz;
173 u8 mw_enabled; /* Enable memory windows */
173 u8 uar_page_sz; /* log pg sz in 4k chunks */ 174 u8 uar_page_sz; /* log pg sz in 4k chunks */
174 u8 steering_mode; /* for QUERY_HCA */ 175 u8 steering_mode; /* for QUERY_HCA */
175 u64 dev_cap_enabled; 176 u64 dev_cap_enabled;
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index b9dde139dac5..d180bc46826a 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -1431,6 +1431,10 @@ static int mlx4_init_hca(struct mlx4_dev *dev)
1431 1431
1432 init_hca.log_uar_sz = ilog2(dev->caps.num_uars); 1432 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1433 init_hca.uar_page_sz = PAGE_SHIFT - 12; 1433 init_hca.uar_page_sz = PAGE_SHIFT - 12;
1434 init_hca.mw_enabled = 0;
1435 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1436 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1437 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1434 1438
1435 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size); 1439 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1436 if (err) 1440 if (err)
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index ed4a6959e828..cf883345af88 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -60,6 +60,8 @@
60#define MLX4_FS_MGM_LOG_ENTRY_SIZE 7 60#define MLX4_FS_MGM_LOG_ENTRY_SIZE 7
61#define MLX4_FS_NUM_MCG (1 << 17) 61#define MLX4_FS_NUM_MCG (1 << 17)
62 62
63#define INIT_HCA_TPT_MW_ENABLE (1 << 7)
64
63#define MLX4_NUM_UP 8 65#define MLX4_NUM_UP 8
64#define MLX4_NUM_TC 8 66#define MLX4_NUM_TC 8
65#define MLX4_RATELIMIT_UNITS 3 /* 100 Mbps */ 67#define MLX4_RATELIMIT_UNITS 3 /* 100 Mbps */
@@ -113,10 +115,10 @@ enum {
113 MLX4_NUM_CMPTS = MLX4_CMPT_NUM_TYPE << MLX4_CMPT_SHIFT 115 MLX4_NUM_CMPTS = MLX4_CMPT_NUM_TYPE << MLX4_CMPT_SHIFT
114}; 116};
115 117
116enum mlx4_mr_state { 118enum mlx4_mpt_state {
117 MLX4_MR_DISABLED = 0, 119 MLX4_MPT_DISABLED = 0,
118 MLX4_MR_EN_HW, 120 MLX4_MPT_EN_HW,
119 MLX4_MR_EN_SW 121 MLX4_MPT_EN_SW
120}; 122};
121 123
122#define MLX4_COMM_TIME 10000 124#define MLX4_COMM_TIME 10000
@@ -263,6 +265,22 @@ struct mlx4_icm_table {
263 struct mlx4_icm **icm; 265 struct mlx4_icm **icm;
264}; 266};
265 267
268#define MLX4_MPT_FLAG_SW_OWNS (0xfUL << 28)
269#define MLX4_MPT_FLAG_FREE (0x3UL << 28)
270#define MLX4_MPT_FLAG_MIO (1 << 17)
271#define MLX4_MPT_FLAG_BIND_ENABLE (1 << 15)
272#define MLX4_MPT_FLAG_PHYSICAL (1 << 9)
273#define MLX4_MPT_FLAG_REGION (1 << 8)
274
275#define MLX4_MPT_PD_FLAG_FAST_REG (1 << 27)
276#define MLX4_MPT_PD_FLAG_RAE (1 << 28)
277#define MLX4_MPT_PD_FLAG_EN_INV (3 << 24)
278
279#define MLX4_MPT_QP_FLAG_BOUND_QP (1 << 7)
280
281#define MLX4_MPT_STATUS_SW 0xF0
282#define MLX4_MPT_STATUS_HW 0x00
283
266/* 284/*
267 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits. 285 * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
268 */ 286 */
@@ -863,10 +881,10 @@ int __mlx4_cq_alloc_icm(struct mlx4_dev *dev, int *cqn);
863void __mlx4_cq_free_icm(struct mlx4_dev *dev, int cqn); 881void __mlx4_cq_free_icm(struct mlx4_dev *dev, int cqn);
864int __mlx4_srq_alloc_icm(struct mlx4_dev *dev, int *srqn); 882int __mlx4_srq_alloc_icm(struct mlx4_dev *dev, int *srqn);
865void __mlx4_srq_free_icm(struct mlx4_dev *dev, int srqn); 883void __mlx4_srq_free_icm(struct mlx4_dev *dev, int srqn);
866int __mlx4_mr_reserve(struct mlx4_dev *dev); 884int __mlx4_mpt_reserve(struct mlx4_dev *dev);
867void __mlx4_mr_release(struct mlx4_dev *dev, u32 index); 885void __mlx4_mpt_release(struct mlx4_dev *dev, u32 index);
868int __mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index); 886int __mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index);
869void __mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index); 887void __mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index);
870u32 __mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order); 888u32 __mlx4_alloc_mtt_range(struct mlx4_dev *dev, int order);
871void __mlx4_free_mtt_range(struct mlx4_dev *dev, u32 first_seg, int order); 889void __mlx4_free_mtt_range(struct mlx4_dev *dev, u32 first_seg, int order);
872 890
diff --git a/drivers/net/ethernet/mellanox/mlx4/mr.c b/drivers/net/ethernet/mellanox/mlx4/mr.c
index c202d3ad2a0e..602ca9bf78e4 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mr.c
+++ b/drivers/net/ethernet/mellanox/mlx4/mr.c
@@ -44,20 +44,6 @@
44#include "mlx4.h" 44#include "mlx4.h"
45#include "icm.h" 45#include "icm.h"
46 46
47#define MLX4_MPT_FLAG_SW_OWNS (0xfUL << 28)
48#define MLX4_MPT_FLAG_FREE (0x3UL << 28)
49#define MLX4_MPT_FLAG_MIO (1 << 17)
50#define MLX4_MPT_FLAG_BIND_ENABLE (1 << 15)
51#define MLX4_MPT_FLAG_PHYSICAL (1 << 9)
52#define MLX4_MPT_FLAG_REGION (1 << 8)
53
54#define MLX4_MPT_PD_FLAG_FAST_REG (1 << 27)
55#define MLX4_MPT_PD_FLAG_RAE (1 << 28)
56#define MLX4_MPT_PD_FLAG_EN_INV (3 << 24)
57
58#define MLX4_MPT_STATUS_SW 0xF0
59#define MLX4_MPT_STATUS_HW 0x00
60
61static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order) 47static u32 mlx4_buddy_alloc(struct mlx4_buddy *buddy, int order)
62{ 48{
63 int o; 49 int o;
@@ -321,7 +307,7 @@ static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd,
321 mr->size = size; 307 mr->size = size;
322 mr->pd = pd; 308 mr->pd = pd;
323 mr->access = access; 309 mr->access = access;
324 mr->enabled = MLX4_MR_DISABLED; 310 mr->enabled = MLX4_MPT_DISABLED;
325 mr->key = hw_index_to_key(mridx); 311 mr->key = hw_index_to_key(mridx);
326 312
327 return mlx4_mtt_init(dev, npages, page_shift, &mr->mtt); 313 return mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
@@ -335,14 +321,14 @@ static int mlx4_WRITE_MTT(struct mlx4_dev *dev,
335 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); 321 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
336} 322}
337 323
338int __mlx4_mr_reserve(struct mlx4_dev *dev) 324int __mlx4_mpt_reserve(struct mlx4_dev *dev)
339{ 325{
340 struct mlx4_priv *priv = mlx4_priv(dev); 326 struct mlx4_priv *priv = mlx4_priv(dev);
341 327
342 return mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap); 328 return mlx4_bitmap_alloc(&priv->mr_table.mpt_bitmap);
343} 329}
344 330
345static int mlx4_mr_reserve(struct mlx4_dev *dev) 331static int mlx4_mpt_reserve(struct mlx4_dev *dev)
346{ 332{
347 u64 out_param; 333 u64 out_param;
348 334
@@ -353,17 +339,17 @@ static int mlx4_mr_reserve(struct mlx4_dev *dev)
353 return -1; 339 return -1;
354 return get_param_l(&out_param); 340 return get_param_l(&out_param);
355 } 341 }
356 return __mlx4_mr_reserve(dev); 342 return __mlx4_mpt_reserve(dev);
357} 343}
358 344
359void __mlx4_mr_release(struct mlx4_dev *dev, u32 index) 345void __mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
360{ 346{
361 struct mlx4_priv *priv = mlx4_priv(dev); 347 struct mlx4_priv *priv = mlx4_priv(dev);
362 348
363 mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index); 349 mlx4_bitmap_free(&priv->mr_table.mpt_bitmap, index);
364} 350}
365 351
366static void mlx4_mr_release(struct mlx4_dev *dev, u32 index) 352static void mlx4_mpt_release(struct mlx4_dev *dev, u32 index)
367{ 353{
368 u64 in_param; 354 u64 in_param;
369 355
@@ -376,17 +362,17 @@ static void mlx4_mr_release(struct mlx4_dev *dev, u32 index)
376 index); 362 index);
377 return; 363 return;
378 } 364 }
379 __mlx4_mr_release(dev, index); 365 __mlx4_mpt_release(dev, index);
380} 366}
381 367
382int __mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index) 368int __mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index)
383{ 369{
384 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; 370 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
385 371
386 return mlx4_table_get(dev, &mr_table->dmpt_table, index); 372 return mlx4_table_get(dev, &mr_table->dmpt_table, index);
387} 373}
388 374
389static int mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index) 375static int mlx4_mpt_alloc_icm(struct mlx4_dev *dev, u32 index)
390{ 376{
391 u64 param; 377 u64 param;
392 378
@@ -397,17 +383,17 @@ static int mlx4_mr_alloc_icm(struct mlx4_dev *dev, u32 index)
397 MLX4_CMD_TIME_CLASS_A, 383 MLX4_CMD_TIME_CLASS_A,
398 MLX4_CMD_WRAPPED); 384 MLX4_CMD_WRAPPED);
399 } 385 }
400 return __mlx4_mr_alloc_icm(dev, index); 386 return __mlx4_mpt_alloc_icm(dev, index);
401} 387}
402 388
403void __mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index) 389void __mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
404{ 390{
405 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table; 391 struct mlx4_mr_table *mr_table = &mlx4_priv(dev)->mr_table;
406 392
407 mlx4_table_put(dev, &mr_table->dmpt_table, index); 393 mlx4_table_put(dev, &mr_table->dmpt_table, index);
408} 394}
409 395
410static void mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index) 396static void mlx4_mpt_free_icm(struct mlx4_dev *dev, u32 index)
411{ 397{
412 u64 in_param; 398 u64 in_param;
413 399
@@ -420,7 +406,7 @@ static void mlx4_mr_free_icm(struct mlx4_dev *dev, u32 index)
420 index); 406 index);
421 return; 407 return;
422 } 408 }
423 return __mlx4_mr_free_icm(dev, index); 409 return __mlx4_mpt_free_icm(dev, index);
424} 410}
425 411
426int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access, 412int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
@@ -429,41 +415,52 @@ int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
429 u32 index; 415 u32 index;
430 int err; 416 int err;
431 417
432 index = mlx4_mr_reserve(dev); 418 index = mlx4_mpt_reserve(dev);
433 if (index == -1) 419 if (index == -1)
434 return -ENOMEM; 420 return -ENOMEM;
435 421
436 err = mlx4_mr_alloc_reserved(dev, index, pd, iova, size, 422 err = mlx4_mr_alloc_reserved(dev, index, pd, iova, size,
437 access, npages, page_shift, mr); 423 access, npages, page_shift, mr);
438 if (err) 424 if (err)
439 mlx4_mr_release(dev, index); 425 mlx4_mpt_release(dev, index);
440 426
441 return err; 427 return err;
442} 428}
443EXPORT_SYMBOL_GPL(mlx4_mr_alloc); 429EXPORT_SYMBOL_GPL(mlx4_mr_alloc);
444 430
445static void mlx4_mr_free_reserved(struct mlx4_dev *dev, struct mlx4_mr *mr) 431static int mlx4_mr_free_reserved(struct mlx4_dev *dev, struct mlx4_mr *mr)
446{ 432{
447 int err; 433 int err;
448 434
449 if (mr->enabled == MLX4_MR_EN_HW) { 435 if (mr->enabled == MLX4_MPT_EN_HW) {
450 err = mlx4_HW2SW_MPT(dev, NULL, 436 err = mlx4_HW2SW_MPT(dev, NULL,
451 key_to_hw_index(mr->key) & 437 key_to_hw_index(mr->key) &
452 (dev->caps.num_mpts - 1)); 438 (dev->caps.num_mpts - 1));
453 if (err) 439 if (err) {
454 mlx4_warn(dev, "xxx HW2SW_MPT failed (%d)\n", err); 440 mlx4_warn(dev, "HW2SW_MPT failed (%d),", err);
441 mlx4_warn(dev, "MR has MWs bound to it.\n");
442 return err;
443 }
455 444
456 mr->enabled = MLX4_MR_EN_SW; 445 mr->enabled = MLX4_MPT_EN_SW;
457 } 446 }
458 mlx4_mtt_cleanup(dev, &mr->mtt); 447 mlx4_mtt_cleanup(dev, &mr->mtt);
448
449 return 0;
459} 450}
460 451
461void mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr) 452int mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
462{ 453{
463 mlx4_mr_free_reserved(dev, mr); 454 int ret;
455
456 ret = mlx4_mr_free_reserved(dev, mr);
457 if (ret)
458 return ret;
464 if (mr->enabled) 459 if (mr->enabled)
465 mlx4_mr_free_icm(dev, key_to_hw_index(mr->key)); 460 mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
466 mlx4_mr_release(dev, key_to_hw_index(mr->key)); 461 mlx4_mpt_release(dev, key_to_hw_index(mr->key));
462
463 return 0;
467} 464}
468EXPORT_SYMBOL_GPL(mlx4_mr_free); 465EXPORT_SYMBOL_GPL(mlx4_mr_free);
469 466
@@ -473,7 +470,7 @@ int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
473 struct mlx4_mpt_entry *mpt_entry; 470 struct mlx4_mpt_entry *mpt_entry;
474 int err; 471 int err;
475 472
476 err = mlx4_mr_alloc_icm(dev, key_to_hw_index(mr->key)); 473 err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mr->key));
477 if (err) 474 if (err)
478 return err; 475 return err;
479 476
@@ -520,7 +517,7 @@ int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
520 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err); 517 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
521 goto err_cmd; 518 goto err_cmd;
522 } 519 }
523 mr->enabled = MLX4_MR_EN_HW; 520 mr->enabled = MLX4_MPT_EN_HW;
524 521
525 mlx4_free_cmd_mailbox(dev, mailbox); 522 mlx4_free_cmd_mailbox(dev, mailbox);
526 523
@@ -530,7 +527,7 @@ err_cmd:
530 mlx4_free_cmd_mailbox(dev, mailbox); 527 mlx4_free_cmd_mailbox(dev, mailbox);
531 528
532err_table: 529err_table:
533 mlx4_mr_free_icm(dev, key_to_hw_index(mr->key)); 530 mlx4_mpt_free_icm(dev, key_to_hw_index(mr->key));
534 return err; 531 return err;
535} 532}
536EXPORT_SYMBOL_GPL(mlx4_mr_enable); 533EXPORT_SYMBOL_GPL(mlx4_mr_enable);
@@ -657,6 +654,101 @@ int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
657} 654}
658EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt); 655EXPORT_SYMBOL_GPL(mlx4_buf_write_mtt);
659 656
657int mlx4_mw_alloc(struct mlx4_dev *dev, u32 pd, enum mlx4_mw_type type,
658 struct mlx4_mw *mw)
659{
660 u32 index;
661
662 if ((type == MLX4_MW_TYPE_1 &&
663 !(dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)) ||
664 (type == MLX4_MW_TYPE_2 &&
665 !(dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)))
666 return -ENOTSUPP;
667
668 index = mlx4_mpt_reserve(dev);
669 if (index == -1)
670 return -ENOMEM;
671
672 mw->key = hw_index_to_key(index);
673 mw->pd = pd;
674 mw->type = type;
675 mw->enabled = MLX4_MPT_DISABLED;
676
677 return 0;
678}
679EXPORT_SYMBOL_GPL(mlx4_mw_alloc);
680
681int mlx4_mw_enable(struct mlx4_dev *dev, struct mlx4_mw *mw)
682{
683 struct mlx4_cmd_mailbox *mailbox;
684 struct mlx4_mpt_entry *mpt_entry;
685 int err;
686
687 err = mlx4_mpt_alloc_icm(dev, key_to_hw_index(mw->key));
688 if (err)
689 return err;
690
691 mailbox = mlx4_alloc_cmd_mailbox(dev);
692 if (IS_ERR(mailbox)) {
693 err = PTR_ERR(mailbox);
694 goto err_table;
695 }
696 mpt_entry = mailbox->buf;
697
698 memset(mpt_entry, 0, sizeof(*mpt_entry));
699
700 /* Note that the MLX4_MPT_FLAG_REGION bit in mpt_entry->flags is turned
701 * off, thus creating a memory window and not a memory region.
702 */
703 mpt_entry->key = cpu_to_be32(key_to_hw_index(mw->key));
704 mpt_entry->pd_flags = cpu_to_be32(mw->pd);
705 if (mw->type == MLX4_MW_TYPE_2) {
706 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_FREE);
707 mpt_entry->qpn = cpu_to_be32(MLX4_MPT_QP_FLAG_BOUND_QP);
708 mpt_entry->pd_flags |= cpu_to_be32(MLX4_MPT_PD_FLAG_EN_INV);
709 }
710
711 err = mlx4_SW2HW_MPT(dev, mailbox,
712 key_to_hw_index(mw->key) &
713 (dev->caps.num_mpts - 1));
714 if (err) {
715 mlx4_warn(dev, "SW2HW_MPT failed (%d)\n", err);
716 goto err_cmd;
717 }
718 mw->enabled = MLX4_MPT_EN_HW;
719
720 mlx4_free_cmd_mailbox(dev, mailbox);
721
722 return 0;
723
724err_cmd:
725 mlx4_free_cmd_mailbox(dev, mailbox);
726
727err_table:
728 mlx4_mpt_free_icm(dev, key_to_hw_index(mw->key));
729 return err;
730}
731EXPORT_SYMBOL_GPL(mlx4_mw_enable);
732
733void mlx4_mw_free(struct mlx4_dev *dev, struct mlx4_mw *mw)
734{
735 int err;
736
737 if (mw->enabled == MLX4_MPT_EN_HW) {
738 err = mlx4_HW2SW_MPT(dev, NULL,
739 key_to_hw_index(mw->key) &
740 (dev->caps.num_mpts - 1));
741 if (err)
742 mlx4_warn(dev, "xxx HW2SW_MPT failed (%d)\n", err);
743
744 mw->enabled = MLX4_MPT_EN_SW;
745 }
746 if (mw->enabled)
747 mlx4_mpt_free_icm(dev, key_to_hw_index(mw->key));
748 mlx4_mpt_release(dev, key_to_hw_index(mw->key));
749}
750EXPORT_SYMBOL_GPL(mlx4_mw_free);
751
660int mlx4_init_mr_table(struct mlx4_dev *dev) 752int mlx4_init_mr_table(struct mlx4_dev *dev)
661{ 753{
662 struct mlx4_priv *priv = mlx4_priv(dev); 754 struct mlx4_priv *priv = mlx4_priv(dev);
@@ -831,7 +923,7 @@ int mlx4_fmr_alloc(struct mlx4_dev *dev, u32 pd, u32 access, int max_pages,
831 return 0; 923 return 0;
832 924
833err_free: 925err_free:
834 mlx4_mr_free(dev, &fmr->mr); 926 (void) mlx4_mr_free(dev, &fmr->mr);
835 return err; 927 return err;
836} 928}
837EXPORT_SYMBOL_GPL(mlx4_fmr_alloc); 929EXPORT_SYMBOL_GPL(mlx4_fmr_alloc);
@@ -882,17 +974,21 @@ void mlx4_fmr_unmap(struct mlx4_dev *dev, struct mlx4_fmr *fmr,
882 err); 974 err);
883 return; 975 return;
884 } 976 }
885 fmr->mr.enabled = MLX4_MR_EN_SW; 977 fmr->mr.enabled = MLX4_MPT_EN_SW;
886} 978}
887EXPORT_SYMBOL_GPL(mlx4_fmr_unmap); 979EXPORT_SYMBOL_GPL(mlx4_fmr_unmap);
888 980
889int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr) 981int mlx4_fmr_free(struct mlx4_dev *dev, struct mlx4_fmr *fmr)
890{ 982{
983 int ret;
984
891 if (fmr->maps) 985 if (fmr->maps)
892 return -EBUSY; 986 return -EBUSY;
893 987
894 mlx4_mr_free(dev, &fmr->mr); 988 ret = mlx4_mr_free(dev, &fmr->mr);
895 fmr->mr.enabled = MLX4_MR_DISABLED; 989 if (ret)
990 return ret;
991 fmr->mr.enabled = MLX4_MPT_DISABLED;
896 992
897 return 0; 993 return 0;
898} 994}
diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
index 5997adc943d0..083fb48dc3d7 100644
--- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
+++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
@@ -1231,14 +1231,14 @@ static int mpt_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1231 1231
1232 switch (op) { 1232 switch (op) {
1233 case RES_OP_RESERVE: 1233 case RES_OP_RESERVE:
1234 index = __mlx4_mr_reserve(dev); 1234 index = __mlx4_mpt_reserve(dev);
1235 if (index == -1) 1235 if (index == -1)
1236 break; 1236 break;
1237 id = index & mpt_mask(dev); 1237 id = index & mpt_mask(dev);
1238 1238
1239 err = add_res_range(dev, slave, id, 1, RES_MPT, index); 1239 err = add_res_range(dev, slave, id, 1, RES_MPT, index);
1240 if (err) { 1240 if (err) {
1241 __mlx4_mr_release(dev, index); 1241 __mlx4_mpt_release(dev, index);
1242 break; 1242 break;
1243 } 1243 }
1244 set_param_l(out_param, index); 1244 set_param_l(out_param, index);
@@ -1251,7 +1251,7 @@ static int mpt_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1251 if (err) 1251 if (err)
1252 return err; 1252 return err;
1253 1253
1254 err = __mlx4_mr_alloc_icm(dev, mpt->key); 1254 err = __mlx4_mpt_alloc_icm(dev, mpt->key);
1255 if (err) { 1255 if (err) {
1256 res_abort_move(dev, slave, RES_MPT, id); 1256 res_abort_move(dev, slave, RES_MPT, id);
1257 return err; 1257 return err;
@@ -1586,7 +1586,7 @@ static int mpt_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1586 err = rem_res_range(dev, slave, id, 1, RES_MPT, 0); 1586 err = rem_res_range(dev, slave, id, 1, RES_MPT, 0);
1587 if (err) 1587 if (err)
1588 break; 1588 break;
1589 __mlx4_mr_release(dev, index); 1589 __mlx4_mpt_release(dev, index);
1590 break; 1590 break;
1591 case RES_OP_MAP_ICM: 1591 case RES_OP_MAP_ICM:
1592 index = get_param_l(&in_param); 1592 index = get_param_l(&in_param);
@@ -1596,7 +1596,7 @@ static int mpt_free_res(struct mlx4_dev *dev, int slave, int op, int cmd,
1596 if (err) 1596 if (err)
1597 return err; 1597 return err;
1598 1598
1599 __mlx4_mr_free_icm(dev, mpt->key); 1599 __mlx4_mpt_free_icm(dev, mpt->key);
1600 res_end_move(dev, slave, RES_MPT, id); 1600 res_end_move(dev, slave, RES_MPT, id);
1601 return err; 1601 return err;
1602 break; 1602 break;
@@ -1796,6 +1796,26 @@ static int mr_get_mtt_size(struct mlx4_mpt_entry *mpt)
1796 return be32_to_cpu(mpt->mtt_sz); 1796 return be32_to_cpu(mpt->mtt_sz);
1797} 1797}
1798 1798
1799static u32 mr_get_pd(struct mlx4_mpt_entry *mpt)
1800{
1801 return be32_to_cpu(mpt->pd_flags) & 0x00ffffff;
1802}
1803
1804static int mr_is_fmr(struct mlx4_mpt_entry *mpt)
1805{
1806 return be32_to_cpu(mpt->pd_flags) & MLX4_MPT_PD_FLAG_FAST_REG;
1807}
1808
1809static int mr_is_bind_enabled(struct mlx4_mpt_entry *mpt)
1810{
1811 return be32_to_cpu(mpt->flags) & MLX4_MPT_FLAG_BIND_ENABLE;
1812}
1813
1814static int mr_is_region(struct mlx4_mpt_entry *mpt)
1815{
1816 return be32_to_cpu(mpt->flags) & MLX4_MPT_FLAG_REGION;
1817}
1818
1799static int qp_get_mtt_addr(struct mlx4_qp_context *qpc) 1819static int qp_get_mtt_addr(struct mlx4_qp_context *qpc)
1800{ 1820{
1801 return be32_to_cpu(qpc->mtt_base_addr_l) & 0xfffffff8; 1821 return be32_to_cpu(qpc->mtt_base_addr_l) & 0xfffffff8;
@@ -1856,12 +1876,41 @@ int mlx4_SW2HW_MPT_wrapper(struct mlx4_dev *dev, int slave,
1856 int mtt_base = mr_get_mtt_addr(inbox->buf) / dev->caps.mtt_entry_sz; 1876 int mtt_base = mr_get_mtt_addr(inbox->buf) / dev->caps.mtt_entry_sz;
1857 int phys; 1877 int phys;
1858 int id; 1878 int id;
1879 u32 pd;
1880 int pd_slave;
1859 1881
1860 id = index & mpt_mask(dev); 1882 id = index & mpt_mask(dev);
1861 err = mr_res_start_move_to(dev, slave, id, RES_MPT_HW, &mpt); 1883 err = mr_res_start_move_to(dev, slave, id, RES_MPT_HW, &mpt);
1862 if (err) 1884 if (err)
1863 return err; 1885 return err;
1864 1886
1887 /* Disable memory windows for VFs. */
1888 if (!mr_is_region(inbox->buf)) {
1889 err = -EPERM;
1890 goto ex_abort;
1891 }
1892
1893 /* Make sure that the PD bits related to the slave id are zeros. */
1894 pd = mr_get_pd(inbox->buf);
1895 pd_slave = (pd >> 17) & 0x7f;
1896 if (pd_slave != 0 && pd_slave != slave) {
1897 err = -EPERM;
1898 goto ex_abort;
1899 }
1900
1901 if (mr_is_fmr(inbox->buf)) {
1902 /* FMR and Bind Enable are forbidden in slave devices. */
1903 if (mr_is_bind_enabled(inbox->buf)) {
1904 err = -EPERM;
1905 goto ex_abort;
1906 }
1907 /* FMR and Memory Windows are also forbidden. */
1908 if (!mr_is_region(inbox->buf)) {
1909 err = -EPERM;
1910 goto ex_abort;
1911 }
1912 }
1913
1865 phys = mr_phys_mpt(inbox->buf); 1914 phys = mr_phys_mpt(inbox->buf);
1866 if (!phys) { 1915 if (!phys) {
1867 err = get_res(dev, slave, mtt_base, RES_MTT, &mtt); 1916 err = get_res(dev, slave, mtt_base, RES_MTT, &mtt);
@@ -3480,7 +3529,7 @@ static void rem_slave_mrs(struct mlx4_dev *dev, int slave)
3480 while (state != 0) { 3529 while (state != 0) {
3481 switch (state) { 3530 switch (state) {
3482 case RES_MPT_RESERVED: 3531 case RES_MPT_RESERVED:
3483 __mlx4_mr_release(dev, mpt->key); 3532 __mlx4_mpt_release(dev, mpt->key);
3484 spin_lock_irq(mlx4_tlock(dev)); 3533 spin_lock_irq(mlx4_tlock(dev));
3485 rb_erase(&mpt->com.node, 3534 rb_erase(&mpt->com.node,
3486 &tracker->res_tree[RES_MPT]); 3535 &tracker->res_tree[RES_MPT]);
@@ -3491,7 +3540,7 @@ static void rem_slave_mrs(struct mlx4_dev *dev, int slave)
3491 break; 3540 break;
3492 3541
3493 case RES_MPT_MAPPED: 3542 case RES_MPT_MAPPED:
3494 __mlx4_mr_free_icm(dev, mpt->key); 3543 __mlx4_mpt_free_icm(dev, mpt->key);
3495 state = RES_MPT_RESERVED; 3544 state = RES_MPT_RESERVED;
3496 break; 3545 break;
3497 3546
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 6d48fce06b4a..811f91cf5e8c 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -171,6 +171,7 @@ enum {
171#define MLX4_ATTR_EXTENDED_PORT_INFO cpu_to_be16(0xff90) 171#define MLX4_ATTR_EXTENDED_PORT_INFO cpu_to_be16(0xff90)
172 172
173enum { 173enum {
174 MLX4_BMME_FLAG_WIN_TYPE_2B = 1 << 1,
174 MLX4_BMME_FLAG_LOCAL_INV = 1 << 6, 175 MLX4_BMME_FLAG_LOCAL_INV = 1 << 6,
175 MLX4_BMME_FLAG_REMOTE_INV = 1 << 7, 176 MLX4_BMME_FLAG_REMOTE_INV = 1 << 7,
176 MLX4_BMME_FLAG_TYPE_2_WIN = 1 << 9, 177 MLX4_BMME_FLAG_TYPE_2_WIN = 1 << 9,
@@ -238,7 +239,8 @@ enum {
238 MLX4_PERM_LOCAL_WRITE = 1 << 11, 239 MLX4_PERM_LOCAL_WRITE = 1 << 11,
239 MLX4_PERM_REMOTE_READ = 1 << 12, 240 MLX4_PERM_REMOTE_READ = 1 << 12,
240 MLX4_PERM_REMOTE_WRITE = 1 << 13, 241 MLX4_PERM_REMOTE_WRITE = 1 << 13,
241 MLX4_PERM_ATOMIC = 1 << 14 242 MLX4_PERM_ATOMIC = 1 << 14,
243 MLX4_PERM_BIND_MW = 1 << 15,
242}; 244};
243 245
244enum { 246enum {
@@ -504,6 +506,18 @@ struct mlx4_mr {
504 int enabled; 506 int enabled;
505}; 507};
506 508
509enum mlx4_mw_type {
510 MLX4_MW_TYPE_1 = 1,
511 MLX4_MW_TYPE_2 = 2,
512};
513
514struct mlx4_mw {
515 u32 key;
516 u32 pd;
517 enum mlx4_mw_type type;
518 int enabled;
519};
520
507struct mlx4_fmr { 521struct mlx4_fmr {
508 struct mlx4_mr mr; 522 struct mlx4_mr mr;
509 struct mlx4_mpt_entry *mpt; 523 struct mlx4_mpt_entry *mpt;
@@ -802,8 +816,12 @@ u64 mlx4_mtt_addr(struct mlx4_dev *dev, struct mlx4_mtt *mtt);
802 816
803int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access, 817int mlx4_mr_alloc(struct mlx4_dev *dev, u32 pd, u64 iova, u64 size, u32 access,
804 int npages, int page_shift, struct mlx4_mr *mr); 818 int npages, int page_shift, struct mlx4_mr *mr);
805void mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr); 819int mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr);
806int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr); 820int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr);
821int mlx4_mw_alloc(struct mlx4_dev *dev, u32 pd, enum mlx4_mw_type type,
822 struct mlx4_mw *mw);
823void mlx4_mw_free(struct mlx4_dev *dev, struct mlx4_mw *mw);
824int mlx4_mw_enable(struct mlx4_dev *dev, struct mlx4_mw *mw);
807int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt, 825int mlx4_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
808 int start_index, int npages, u64 *page_list); 826 int start_index, int npages, u64 *page_list);
809int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt, 827int mlx4_buf_write_mtt(struct mlx4_dev *dev, struct mlx4_mtt *mtt,
diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h
index 4b4ad6ffef92..67f46ad6920a 100644
--- a/include/linux/mlx4/qp.h
+++ b/include/linux/mlx4/qp.h
@@ -265,6 +265,11 @@ struct mlx4_wqe_lso_seg {
265 __be32 header[0]; 265 __be32 header[0];
266}; 266};
267 267
268enum mlx4_wqe_bind_seg_flags2 {
269 MLX4_WQE_BIND_ZERO_BASED = (1 << 30),
270 MLX4_WQE_BIND_TYPE_2 = (1 << 31),
271};
272
268struct mlx4_wqe_bind_seg { 273struct mlx4_wqe_bind_seg {
269 __be32 flags1; 274 __be32 flags1;
270 __be32 flags2; 275 __be32 flags2;
@@ -277,9 +282,9 @@ struct mlx4_wqe_bind_seg {
277enum { 282enum {
278 MLX4_WQE_FMR_PERM_LOCAL_READ = 1 << 27, 283 MLX4_WQE_FMR_PERM_LOCAL_READ = 1 << 27,
279 MLX4_WQE_FMR_PERM_LOCAL_WRITE = 1 << 28, 284 MLX4_WQE_FMR_PERM_LOCAL_WRITE = 1 << 28,
280 MLX4_WQE_FMR_PERM_REMOTE_READ = 1 << 29, 285 MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ = 1 << 29,
281 MLX4_WQE_FMR_PERM_REMOTE_WRITE = 1 << 30, 286 MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE = 1 << 30,
282 MLX4_WQE_FMR_PERM_ATOMIC = 1 << 31 287 MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC = 1 << 31
283}; 288};
284 289
285struct mlx4_wqe_fmr_seg { 290struct mlx4_wqe_fmr_seg {
@@ -304,12 +309,10 @@ struct mlx4_wqe_fmr_ext_seg {
304}; 309};
305 310
306struct mlx4_wqe_local_inval_seg { 311struct mlx4_wqe_local_inval_seg {
307 __be32 flags; 312 u64 reserved1;
308 u32 reserved1;
309 __be32 mem_key; 313 __be32 mem_key;
310 u32 reserved2[2]; 314 u32 reserved2;
311 __be32 guest_id; 315 u64 reserved3[2];
312 __be64 pa;
313}; 316};
314 317
315struct mlx4_wqe_raddr_seg { 318struct mlx4_wqe_raddr_seg {
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 46bc045bbe15..98cc4b29fc5b 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -115,6 +115,8 @@ enum ib_device_cap_flags {
115 IB_DEVICE_XRC = (1<<20), 115 IB_DEVICE_XRC = (1<<20),
116 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21), 116 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22), 117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
118 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
119 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24)
118}; 120};
119 121
120enum ib_atomic_cap { 122enum ib_atomic_cap {
@@ -715,6 +717,11 @@ enum ib_mig_state {
715 IB_MIG_ARMED 717 IB_MIG_ARMED
716}; 718};
717 719
720enum ib_mw_type {
721 IB_MW_TYPE_1 = 1,
722 IB_MW_TYPE_2 = 2
723};
724
718struct ib_qp_attr { 725struct ib_qp_attr {
719 enum ib_qp_state qp_state; 726 enum ib_qp_state qp_state;
720 enum ib_qp_state cur_qp_state; 727 enum ib_qp_state cur_qp_state;
@@ -758,6 +765,7 @@ enum ib_wr_opcode {
758 IB_WR_FAST_REG_MR, 765 IB_WR_FAST_REG_MR,
759 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 766 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
760 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 767 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
768 IB_WR_BIND_MW,
761}; 769};
762 770
763enum ib_send_flags { 771enum ib_send_flags {
@@ -780,6 +788,23 @@ struct ib_fast_reg_page_list {
780 unsigned int max_page_list_len; 788 unsigned int max_page_list_len;
781}; 789};
782 790
791/**
792 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
793 * @mr: A memory region to bind the memory window to.
794 * @addr: The address where the memory window should begin.
795 * @length: The length of the memory window, in bytes.
796 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
797 *
798 * This struct contains the shared parameters for type 1 and type 2
799 * memory window bind operations.
800 */
801struct ib_mw_bind_info {
802 struct ib_mr *mr;
803 u64 addr;
804 u64 length;
805 int mw_access_flags;
806};
807
783struct ib_send_wr { 808struct ib_send_wr {
784 struct ib_send_wr *next; 809 struct ib_send_wr *next;
785 u64 wr_id; 810 u64 wr_id;
@@ -823,6 +848,12 @@ struct ib_send_wr {
823 int access_flags; 848 int access_flags;
824 u32 rkey; 849 u32 rkey;
825 } fast_reg; 850 } fast_reg;
851 struct {
852 struct ib_mw *mw;
853 /* The new rkey for the memory window. */
854 u32 rkey;
855 struct ib_mw_bind_info bind_info;
856 } bind_mw;
826 } wr; 857 } wr;
827 u32 xrc_remote_srq_num; /* XRC TGT QPs only */ 858 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
828}; 859};
@@ -839,7 +870,8 @@ enum ib_access_flags {
839 IB_ACCESS_REMOTE_WRITE = (1<<1), 870 IB_ACCESS_REMOTE_WRITE = (1<<1),
840 IB_ACCESS_REMOTE_READ = (1<<2), 871 IB_ACCESS_REMOTE_READ = (1<<2),
841 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 872 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
842 IB_ACCESS_MW_BIND = (1<<4) 873 IB_ACCESS_MW_BIND = (1<<4),
874 IB_ZERO_BASED = (1<<5)
843}; 875};
844 876
845struct ib_phys_buf { 877struct ib_phys_buf {
@@ -862,13 +894,16 @@ enum ib_mr_rereg_flags {
862 IB_MR_REREG_ACCESS = (1<<2) 894 IB_MR_REREG_ACCESS = (1<<2)
863}; 895};
864 896
897/**
898 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
899 * @wr_id: Work request id.
900 * @send_flags: Flags from ib_send_flags enum.
901 * @bind_info: More parameters of the bind operation.
902 */
865struct ib_mw_bind { 903struct ib_mw_bind {
866 struct ib_mr *mr; 904 u64 wr_id;
867 u64 wr_id; 905 int send_flags;
868 u64 addr; 906 struct ib_mw_bind_info bind_info;
869 u32 length;
870 int send_flags;
871 int mw_access_flags;
872}; 907};
873 908
874struct ib_fmr_attr { 909struct ib_fmr_attr {
@@ -991,6 +1026,7 @@ struct ib_mw {
991 struct ib_pd *pd; 1026 struct ib_pd *pd;
992 struct ib_uobject *uobject; 1027 struct ib_uobject *uobject;
993 u32 rkey; 1028 u32 rkey;
1029 enum ib_mw_type type;
994}; 1030};
995 1031
996struct ib_fmr { 1032struct ib_fmr {
@@ -1202,7 +1238,8 @@ struct ib_device {
1202 int num_phys_buf, 1238 int num_phys_buf,
1203 int mr_access_flags, 1239 int mr_access_flags,
1204 u64 *iova_start); 1240 u64 *iova_start);
1205 struct ib_mw * (*alloc_mw)(struct ib_pd *pd); 1241 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1242 enum ib_mw_type type);
1206 int (*bind_mw)(struct ib_qp *qp, 1243 int (*bind_mw)(struct ib_qp *qp,
1207 struct ib_mw *mw, 1244 struct ib_mw *mw,
1208 struct ib_mw_bind *mw_bind); 1245 struct ib_mw_bind *mw_bind);
@@ -2019,6 +2056,8 @@ int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2019 * ib_dereg_mr - Deregisters a memory region and removes it from the 2056 * ib_dereg_mr - Deregisters a memory region and removes it from the
2020 * HCA translation table. 2057 * HCA translation table.
2021 * @mr: The memory region to deregister. 2058 * @mr: The memory region to deregister.
2059 *
2060 * This function can fail, if the memory region has memory windows bound to it.
2022 */ 2061 */
2023int ib_dereg_mr(struct ib_mr *mr); 2062int ib_dereg_mr(struct ib_mr *mr);
2024 2063
@@ -2071,10 +2110,22 @@ static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2071} 2110}
2072 2111
2073/** 2112/**
2113 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2114 * for calculating a new rkey for type 2 memory windows.
2115 * @rkey - the rkey to increment.
2116 */
2117static inline u32 ib_inc_rkey(u32 rkey)
2118{
2119 const u32 mask = 0x000000ff;
2120 return ((rkey + 1) & mask) | (rkey & ~mask);
2121}
2122
2123/**
2074 * ib_alloc_mw - Allocates a memory window. 2124 * ib_alloc_mw - Allocates a memory window.
2075 * @pd: The protection domain associated with the memory window. 2125 * @pd: The protection domain associated with the memory window.
2126 * @type: The type of the memory window (1 or 2).
2076 */ 2127 */
2077struct ib_mw *ib_alloc_mw(struct ib_pd *pd); 2128struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2078 2129
2079/** 2130/**
2080 * ib_bind_mw - Posts a work request to the send queue of the specified 2131 * ib_bind_mw - Posts a work request to the send queue of the specified
@@ -2084,6 +2135,10 @@ struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2084 * @mw: The memory window to bind. 2135 * @mw: The memory window to bind.
2085 * @mw_bind: Specifies information about the memory window, including 2136 * @mw_bind: Specifies information about the memory window, including
2086 * its address range, remote access rights, and associated memory region. 2137 * its address range, remote access rights, and associated memory region.
2138 *
2139 * If there is no immediate error, the function will update the rkey member
2140 * of the mw parameter to its new value. The bind operation can still fail
2141 * asynchronously.
2087 */ 2142 */
2088static inline int ib_bind_mw(struct ib_qp *qp, 2143static inline int ib_bind_mw(struct ib_qp *qp,
2089 struct ib_mw *mw, 2144 struct ib_mw *mw,
diff --git a/include/uapi/rdma/ib_user_verbs.h b/include/uapi/rdma/ib_user_verbs.h
index 81aba3a73aa3..805711ea2005 100644
--- a/include/uapi/rdma/ib_user_verbs.h
+++ b/include/uapi/rdma/ib_user_verbs.h
@@ -261,6 +261,22 @@ struct ib_uverbs_dereg_mr {
261 __u32 mr_handle; 261 __u32 mr_handle;
262}; 262};
263 263
264struct ib_uverbs_alloc_mw {
265 __u64 response;
266 __u32 pd_handle;
267 __u8 mw_type;
268 __u8 reserved[3];
269};
270
271struct ib_uverbs_alloc_mw_resp {
272 __u32 mw_handle;
273 __u32 rkey;
274};
275
276struct ib_uverbs_dealloc_mw {
277 __u32 mw_handle;
278};
279
264struct ib_uverbs_create_comp_channel { 280struct ib_uverbs_create_comp_channel {
265 __u64 response; 281 __u64 response;
266}; 282};
diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index 745973b729af..93726560eaa8 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -1086,7 +1086,7 @@ rpcrdma_buffer_create(struct rpcrdma_buffer *buf, struct rpcrdma_ep *ep,
1086 case RPCRDMA_MEMWINDOWS: 1086 case RPCRDMA_MEMWINDOWS:
1087 /* Allocate one extra request's worth, for full cycling */ 1087 /* Allocate one extra request's worth, for full cycling */
1088 for (i = (buf->rb_max_requests+1) * RPCRDMA_MAX_SEGS; i; i--) { 1088 for (i = (buf->rb_max_requests+1) * RPCRDMA_MAX_SEGS; i; i--) {
1089 r->r.mw = ib_alloc_mw(ia->ri_pd); 1089 r->r.mw = ib_alloc_mw(ia->ri_pd, IB_MW_TYPE_1);
1090 if (IS_ERR(r->r.mw)) { 1090 if (IS_ERR(r->r.mw)) {
1091 rc = PTR_ERR(r->r.mw); 1091 rc = PTR_ERR(r->r.mw);
1092 dprintk("RPC: %s: ib_alloc_mw" 1092 dprintk("RPC: %s: ib_alloc_mw"
@@ -1673,12 +1673,12 @@ rpcrdma_register_memwin_external(struct rpcrdma_mr_seg *seg,
1673 1673
1674 *nsegs = 1; 1674 *nsegs = 1;
1675 rpcrdma_map_one(ia, seg, writing); 1675 rpcrdma_map_one(ia, seg, writing);
1676 param.mr = ia->ri_bind_mem; 1676 param.bind_info.mr = ia->ri_bind_mem;
1677 param.wr_id = 0ULL; /* no send cookie */ 1677 param.wr_id = 0ULL; /* no send cookie */
1678 param.addr = seg->mr_dma; 1678 param.bind_info.addr = seg->mr_dma;
1679 param.length = seg->mr_len; 1679 param.bind_info.length = seg->mr_len;
1680 param.send_flags = 0; 1680 param.send_flags = 0;
1681 param.mw_access_flags = mem_priv; 1681 param.bind_info.mw_access_flags = mem_priv;
1682 1682
1683 DECR_CQCOUNT(&r_xprt->rx_ep); 1683 DECR_CQCOUNT(&r_xprt->rx_ep);
1684 rc = ib_bind_mw(ia->ri_id->qp, seg->mr_chunk.rl_mw->r.mw, &param); 1684 rc = ib_bind_mw(ia->ri_id->qp, seg->mr_chunk.rl_mw->r.mw, &param);
@@ -1690,7 +1690,7 @@ rpcrdma_register_memwin_external(struct rpcrdma_mr_seg *seg,
1690 rpcrdma_unmap_one(ia, seg); 1690 rpcrdma_unmap_one(ia, seg);
1691 } else { 1691 } else {
1692 seg->mr_rkey = seg->mr_chunk.rl_mw->r.mw->rkey; 1692 seg->mr_rkey = seg->mr_chunk.rl_mw->r.mw->rkey;
1693 seg->mr_base = param.addr; 1693 seg->mr_base = param.bind_info.addr;
1694 seg->mr_nsegs = 1; 1694 seg->mr_nsegs = 1;
1695 } 1695 }
1696 return rc; 1696 return rc;
@@ -1706,10 +1706,10 @@ rpcrdma_deregister_memwin_external(struct rpcrdma_mr_seg *seg,
1706 int rc; 1706 int rc;
1707 1707
1708 BUG_ON(seg->mr_nsegs != 1); 1708 BUG_ON(seg->mr_nsegs != 1);
1709 param.mr = ia->ri_bind_mem; 1709 param.bind_info.mr = ia->ri_bind_mem;
1710 param.addr = 0ULL; /* unbind */ 1710 param.bind_info.addr = 0ULL; /* unbind */
1711 param.length = 0; 1711 param.bind_info.length = 0;
1712 param.mw_access_flags = 0; 1712 param.bind_info.mw_access_flags = 0;
1713 if (*r) { 1713 if (*r) {
1714 param.wr_id = (u64) (unsigned long) *r; 1714 param.wr_id = (u64) (unsigned long) *r;
1715 param.send_flags = IB_SEND_SIGNALED; 1715 param.send_flags = IB_SEND_SIGNALED;
generated by cgit v1.2.2 (git 2.25.0) at 2026-07-22 13:13:43 -0400