aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/core
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/infiniband/core')
-rw-r--r--drivers/infiniband/core/agent.c29
-rw-r--r--drivers/infiniband/core/cma.c313
-rw-r--r--drivers/infiniband/core/iwcm.c4
-rw-r--r--drivers/infiniband/core/mad.c27
-rw-r--r--drivers/infiniband/core/multicast.c23
-rw-r--r--drivers/infiniband/core/sa_query.c30
-rw-r--r--drivers/infiniband/core/sysfs.c15
-rw-r--r--drivers/infiniband/core/ucma.c92
-rw-r--r--drivers/infiniband/core/ud_header.c140
-rw-r--r--drivers/infiniband/core/user_mad.c2
-rw-r--r--drivers/infiniband/core/uverbs_cmd.c2
-rw-r--r--drivers/infiniband/core/verbs.c16
12 files changed, 606 insertions, 87 deletions
diff --git a/drivers/infiniband/core/agent.c b/drivers/infiniband/core/agent.c
index ae7c2880e624..91916a8d5de4 100644
--- a/drivers/infiniband/core/agent.c
+++ b/drivers/infiniband/core/agent.c
@@ -59,8 +59,8 @@ __ib_get_agent_port(struct ib_device *device, int port_num)
59 struct ib_agent_port_private *entry; 59 struct ib_agent_port_private *entry;
60 60
61 list_for_each_entry(entry, &ib_agent_port_list, port_list) { 61 list_for_each_entry(entry, &ib_agent_port_list, port_list) {
62 if (entry->agent[0]->device == device && 62 if (entry->agent[1]->device == device &&
63 entry->agent[0]->port_num == port_num) 63 entry->agent[1]->port_num == port_num)
64 return entry; 64 return entry;
65 } 65 }
66 return NULL; 66 return NULL;
@@ -155,14 +155,16 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
155 goto error1; 155 goto error1;
156 } 156 }
157 157
158 /* Obtain send only MAD agent for SMI QP */ 158 if (rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_INFINIBAND) {
159 port_priv->agent[0] = ib_register_mad_agent(device, port_num, 159 /* Obtain send only MAD agent for SMI QP */
160 IB_QPT_SMI, NULL, 0, 160 port_priv->agent[0] = ib_register_mad_agent(device, port_num,
161 &agent_send_handler, 161 IB_QPT_SMI, NULL, 0,
162 NULL, NULL); 162 &agent_send_handler,
163 if (IS_ERR(port_priv->agent[0])) { 163 NULL, NULL);
164 ret = PTR_ERR(port_priv->agent[0]); 164 if (IS_ERR(port_priv->agent[0])) {
165 goto error2; 165 ret = PTR_ERR(port_priv->agent[0]);
166 goto error2;
167 }
166 } 168 }
167 169
168 /* Obtain send only MAD agent for GSI QP */ 170 /* Obtain send only MAD agent for GSI QP */
@@ -182,7 +184,8 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
182 return 0; 184 return 0;
183 185
184error3: 186error3:
185 ib_unregister_mad_agent(port_priv->agent[0]); 187 if (port_priv->agent[0])
188 ib_unregister_mad_agent(port_priv->agent[0]);
186error2: 189error2:
187 kfree(port_priv); 190 kfree(port_priv);
188error1: 191error1:
@@ -205,7 +208,9 @@ int ib_agent_port_close(struct ib_device *device, int port_num)
205 spin_unlock_irqrestore(&ib_agent_port_list_lock, flags); 208 spin_unlock_irqrestore(&ib_agent_port_list_lock, flags);
206 209
207 ib_unregister_mad_agent(port_priv->agent[1]); 210 ib_unregister_mad_agent(port_priv->agent[1]);
208 ib_unregister_mad_agent(port_priv->agent[0]); 211 if (port_priv->agent[0])
212 ib_unregister_mad_agent(port_priv->agent[0]);
213
209 kfree(port_priv); 214 kfree(port_priv);
210 return 0; 215 return 0;
211} 216}
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index b930b8110a63..6884da24fde1 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -59,6 +59,7 @@ MODULE_LICENSE("Dual BSD/GPL");
59#define CMA_CM_RESPONSE_TIMEOUT 20 59#define CMA_CM_RESPONSE_TIMEOUT 20
60#define CMA_MAX_CM_RETRIES 15 60#define CMA_MAX_CM_RETRIES 15
61#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24) 61#define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
62#define CMA_IBOE_PACKET_LIFETIME 18
62 63
63static void cma_add_one(struct ib_device *device); 64static void cma_add_one(struct ib_device *device);
64static void cma_remove_one(struct ib_device *device); 65static void cma_remove_one(struct ib_device *device);
@@ -157,6 +158,7 @@ struct cma_multicast {
157 struct list_head list; 158 struct list_head list;
158 void *context; 159 void *context;
159 struct sockaddr_storage addr; 160 struct sockaddr_storage addr;
161 struct kref mcref;
160}; 162};
161 163
162struct cma_work { 164struct cma_work {
@@ -173,6 +175,12 @@ struct cma_ndev_work {
173 struct rdma_cm_event event; 175 struct rdma_cm_event event;
174}; 176};
175 177
178struct iboe_mcast_work {
179 struct work_struct work;
180 struct rdma_id_private *id;
181 struct cma_multicast *mc;
182};
183
176union cma_ip_addr { 184union cma_ip_addr {
177 struct in6_addr ip6; 185 struct in6_addr ip6;
178 struct { 186 struct {
@@ -281,6 +289,8 @@ static void cma_attach_to_dev(struct rdma_id_private *id_priv,
281 atomic_inc(&cma_dev->refcount); 289 atomic_inc(&cma_dev->refcount);
282 id_priv->cma_dev = cma_dev; 290 id_priv->cma_dev = cma_dev;
283 id_priv->id.device = cma_dev->device; 291 id_priv->id.device = cma_dev->device;
292 id_priv->id.route.addr.dev_addr.transport =
293 rdma_node_get_transport(cma_dev->device->node_type);
284 list_add_tail(&id_priv->list, &cma_dev->id_list); 294 list_add_tail(&id_priv->list, &cma_dev->id_list);
285} 295}
286 296
@@ -290,6 +300,14 @@ static inline void cma_deref_dev(struct cma_device *cma_dev)
290 complete(&cma_dev->comp); 300 complete(&cma_dev->comp);
291} 301}
292 302
303static inline void release_mc(struct kref *kref)
304{
305 struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
306
307 kfree(mc->multicast.ib);
308 kfree(mc);
309}
310
293static void cma_detach_from_dev(struct rdma_id_private *id_priv) 311static void cma_detach_from_dev(struct rdma_id_private *id_priv)
294{ 312{
295 list_del(&id_priv->list); 313 list_del(&id_priv->list);
@@ -323,22 +341,63 @@ static int cma_set_qkey(struct rdma_id_private *id_priv)
323 return ret; 341 return ret;
324} 342}
325 343
344static int find_gid_port(struct ib_device *device, union ib_gid *gid, u8 port_num)
345{
346 int i;
347 int err;
348 struct ib_port_attr props;
349 union ib_gid tmp;
350
351 err = ib_query_port(device, port_num, &props);
352 if (err)
353 return 1;
354
355 for (i = 0; i < props.gid_tbl_len; ++i) {
356 err = ib_query_gid(device, port_num, i, &tmp);
357 if (err)
358 return 1;
359 if (!memcmp(&tmp, gid, sizeof tmp))
360 return 0;
361 }
362
363 return -EAGAIN;
364}
365
326static int cma_acquire_dev(struct rdma_id_private *id_priv) 366static int cma_acquire_dev(struct rdma_id_private *id_priv)
327{ 367{
328 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; 368 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
329 struct cma_device *cma_dev; 369 struct cma_device *cma_dev;
330 union ib_gid gid; 370 union ib_gid gid, iboe_gid;
331 int ret = -ENODEV; 371 int ret = -ENODEV;
372 u8 port;
373 enum rdma_link_layer dev_ll = dev_addr->dev_type == ARPHRD_INFINIBAND ?
374 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
332 375
333 rdma_addr_get_sgid(dev_addr, &gid); 376 iboe_addr_get_sgid(dev_addr, &iboe_gid);
377 memcpy(&gid, dev_addr->src_dev_addr +
378 rdma_addr_gid_offset(dev_addr), sizeof gid);
334 list_for_each_entry(cma_dev, &dev_list, list) { 379 list_for_each_entry(cma_dev, &dev_list, list) {
335 ret = ib_find_cached_gid(cma_dev->device, &gid, 380 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
336 &id_priv->id.port_num, NULL); 381 if (rdma_port_get_link_layer(cma_dev->device, port) == dev_ll) {
337 if (!ret) { 382 if (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB &&
338 cma_attach_to_dev(id_priv, cma_dev); 383 rdma_port_get_link_layer(cma_dev->device, port) == IB_LINK_LAYER_ETHERNET)
339 break; 384 ret = find_gid_port(cma_dev->device, &iboe_gid, port);
385 else
386 ret = find_gid_port(cma_dev->device, &gid, port);
387
388 if (!ret) {
389 id_priv->id.port_num = port;
390 goto out;
391 } else if (ret == 1)
392 break;
393 }
340 } 394 }
341 } 395 }
396
397out:
398 if (!ret)
399 cma_attach_to_dev(id_priv, cma_dev);
400
342 return ret; 401 return ret;
343} 402}
344 403
@@ -556,10 +615,16 @@ static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
556{ 615{
557 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; 616 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
558 int ret; 617 int ret;
618 u16 pkey;
619
620 if (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num) ==
621 IB_LINK_LAYER_INFINIBAND)
622 pkey = ib_addr_get_pkey(dev_addr);
623 else
624 pkey = 0xffff;
559 625
560 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num, 626 ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
561 ib_addr_get_pkey(dev_addr), 627 pkey, &qp_attr->pkey_index);
562 &qp_attr->pkey_index);
563 if (ret) 628 if (ret)
564 return ret; 629 return ret;
565 630
@@ -737,8 +802,8 @@ static inline int cma_user_data_offset(enum rdma_port_space ps)
737 802
738static void cma_cancel_route(struct rdma_id_private *id_priv) 803static void cma_cancel_route(struct rdma_id_private *id_priv)
739{ 804{
740 switch (rdma_node_get_transport(id_priv->id.device->node_type)) { 805 switch (rdma_port_get_link_layer(id_priv->id.device, id_priv->id.port_num)) {
741 case RDMA_TRANSPORT_IB: 806 case IB_LINK_LAYER_INFINIBAND:
742 if (id_priv->query) 807 if (id_priv->query)
743 ib_sa_cancel_query(id_priv->query_id, id_priv->query); 808 ib_sa_cancel_query(id_priv->query_id, id_priv->query);
744 break; 809 break;
@@ -816,8 +881,17 @@ static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
816 mc = container_of(id_priv->mc_list.next, 881 mc = container_of(id_priv->mc_list.next,
817 struct cma_multicast, list); 882 struct cma_multicast, list);
818 list_del(&mc->list); 883 list_del(&mc->list);
819 ib_sa_free_multicast(mc->multicast.ib); 884 switch (rdma_port_get_link_layer(id_priv->cma_dev->device, id_priv->id.port_num)) {
820 kfree(mc); 885 case IB_LINK_LAYER_INFINIBAND:
886 ib_sa_free_multicast(mc->multicast.ib);
887 kfree(mc);
888 break;
889 case IB_LINK_LAYER_ETHERNET:
890 kref_put(&mc->mcref, release_mc);
891 break;
892 default:
893 break;
894 }
821 } 895 }
822} 896}
823 897
@@ -833,7 +907,7 @@ void rdma_destroy_id(struct rdma_cm_id *id)
833 mutex_lock(&lock); 907 mutex_lock(&lock);
834 if (id_priv->cma_dev) { 908 if (id_priv->cma_dev) {
835 mutex_unlock(&lock); 909 mutex_unlock(&lock);
836 switch (rdma_node_get_transport(id->device->node_type)) { 910 switch (rdma_node_get_transport(id_priv->id.device->node_type)) {
837 case RDMA_TRANSPORT_IB: 911 case RDMA_TRANSPORT_IB:
838 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib)) 912 if (id_priv->cm_id.ib && !IS_ERR(id_priv->cm_id.ib))
839 ib_destroy_cm_id(id_priv->cm_id.ib); 913 ib_destroy_cm_id(id_priv->cm_id.ib);
@@ -1708,6 +1782,81 @@ static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
1708 return 0; 1782 return 0;
1709} 1783}
1710 1784
1785static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
1786{
1787 struct rdma_route *route = &id_priv->id.route;
1788 struct rdma_addr *addr = &route->addr;
1789 struct cma_work *work;
1790 int ret;
1791 struct sockaddr_in *src_addr = (struct sockaddr_in *)&route->addr.src_addr;
1792 struct sockaddr_in *dst_addr = (struct sockaddr_in *)&route->addr.dst_addr;
1793 struct net_device *ndev = NULL;
1794 u16 vid;
1795
1796 if (src_addr->sin_family != dst_addr->sin_family)
1797 return -EINVAL;
1798
1799 work = kzalloc(sizeof *work, GFP_KERNEL);
1800 if (!work)
1801 return -ENOMEM;
1802
1803 work->id = id_priv;
1804 INIT_WORK(&work->work, cma_work_handler);
1805
1806 route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
1807 if (!route->path_rec) {
1808 ret = -ENOMEM;
1809 goto err1;
1810 }
1811
1812 route->num_paths = 1;
1813
1814 if (addr->dev_addr.bound_dev_if)
1815 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
1816 if (!ndev) {
1817 ret = -ENODEV;
1818 goto err2;
1819 }
1820
1821 vid = rdma_vlan_dev_vlan_id(ndev);
1822
1823 iboe_mac_vlan_to_ll(&route->path_rec->sgid, addr->dev_addr.src_dev_addr, vid);
1824 iboe_mac_vlan_to_ll(&route->path_rec->dgid, addr->dev_addr.dst_dev_addr, vid);
1825
1826 route->path_rec->hop_limit = 1;
1827 route->path_rec->reversible = 1;
1828 route->path_rec->pkey = cpu_to_be16(0xffff);
1829 route->path_rec->mtu_selector = IB_SA_EQ;
1830 route->path_rec->sl = id_priv->tos >> 5;
1831
1832 route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
1833 route->path_rec->rate_selector = IB_SA_EQ;
1834 route->path_rec->rate = iboe_get_rate(ndev);
1835 dev_put(ndev);
1836 route->path_rec->packet_life_time_selector = IB_SA_EQ;
1837 route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
1838 if (!route->path_rec->mtu) {
1839 ret = -EINVAL;
1840 goto err2;
1841 }
1842
1843 work->old_state = CMA_ROUTE_QUERY;
1844 work->new_state = CMA_ROUTE_RESOLVED;
1845 work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
1846 work->event.status = 0;
1847
1848 queue_work(cma_wq, &work->work);
1849
1850 return 0;
1851
1852err2:
1853 kfree(route->path_rec);
1854 route->path_rec = NULL;
1855err1:
1856 kfree(work);
1857 return ret;
1858}
1859
1711int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms) 1860int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1712{ 1861{
1713 struct rdma_id_private *id_priv; 1862 struct rdma_id_private *id_priv;
@@ -1720,7 +1869,16 @@ int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
1720 atomic_inc(&id_priv->refcount); 1869 atomic_inc(&id_priv->refcount);
1721 switch (rdma_node_get_transport(id->device->node_type)) { 1870 switch (rdma_node_get_transport(id->device->node_type)) {
1722 case RDMA_TRANSPORT_IB: 1871 case RDMA_TRANSPORT_IB:
1723 ret = cma_resolve_ib_route(id_priv, timeout_ms); 1872 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
1873 case IB_LINK_LAYER_INFINIBAND:
1874 ret = cma_resolve_ib_route(id_priv, timeout_ms);
1875 break;
1876 case IB_LINK_LAYER_ETHERNET:
1877 ret = cma_resolve_iboe_route(id_priv);
1878 break;
1879 default:
1880 ret = -ENOSYS;
1881 }
1724 break; 1882 break;
1725 case RDMA_TRANSPORT_IWARP: 1883 case RDMA_TRANSPORT_IWARP:
1726 ret = cma_resolve_iw_route(id_priv, timeout_ms); 1884 ret = cma_resolve_iw_route(id_priv, timeout_ms);
@@ -1773,7 +1931,7 @@ port_found:
1773 goto out; 1931 goto out;
1774 1932
1775 id_priv->id.route.addr.dev_addr.dev_type = 1933 id_priv->id.route.addr.dev_addr.dev_type =
1776 (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB) ? 1934 (rdma_port_get_link_layer(cma_dev->device, p) == IB_LINK_LAYER_INFINIBAND) ?
1777 ARPHRD_INFINIBAND : ARPHRD_ETHER; 1935 ARPHRD_INFINIBAND : ARPHRD_ETHER;
1778 1936
1779 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid); 1937 rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
@@ -2758,6 +2916,102 @@ static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
2758 return 0; 2916 return 0;
2759} 2917}
2760 2918
2919static void iboe_mcast_work_handler(struct work_struct *work)
2920{
2921 struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
2922 struct cma_multicast *mc = mw->mc;
2923 struct ib_sa_multicast *m = mc->multicast.ib;
2924
2925 mc->multicast.ib->context = mc;
2926 cma_ib_mc_handler(0, m);
2927 kref_put(&mc->mcref, release_mc);
2928 kfree(mw);
2929}
2930
2931static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
2932{
2933 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
2934 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
2935
2936 if (cma_any_addr(addr)) {
2937 memset(mgid, 0, sizeof *mgid);
2938 } else if (addr->sa_family == AF_INET6) {
2939 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
2940 } else {
2941 mgid->raw[0] = 0xff;
2942 mgid->raw[1] = 0x0e;
2943 mgid->raw[2] = 0;
2944 mgid->raw[3] = 0;
2945 mgid->raw[4] = 0;
2946 mgid->raw[5] = 0;
2947 mgid->raw[6] = 0;
2948 mgid->raw[7] = 0;
2949 mgid->raw[8] = 0;
2950 mgid->raw[9] = 0;
2951 mgid->raw[10] = 0xff;
2952 mgid->raw[11] = 0xff;
2953 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
2954 }
2955}
2956
2957static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
2958 struct cma_multicast *mc)
2959{
2960 struct iboe_mcast_work *work;
2961 struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2962 int err;
2963 struct sockaddr *addr = (struct sockaddr *)&mc->addr;
2964 struct net_device *ndev = NULL;
2965
2966 if (cma_zero_addr((struct sockaddr *)&mc->addr))
2967 return -EINVAL;
2968
2969 work = kzalloc(sizeof *work, GFP_KERNEL);
2970 if (!work)
2971 return -ENOMEM;
2972
2973 mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
2974 if (!mc->multicast.ib) {
2975 err = -ENOMEM;
2976 goto out1;
2977 }
2978
2979 cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
2980
2981 mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
2982 if (id_priv->id.ps == RDMA_PS_UDP)
2983 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
2984
2985 if (dev_addr->bound_dev_if)
2986 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
2987 if (!ndev) {
2988 err = -ENODEV;
2989 goto out2;
2990 }
2991 mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
2992 mc->multicast.ib->rec.hop_limit = 1;
2993 mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
2994 dev_put(ndev);
2995 if (!mc->multicast.ib->rec.mtu) {
2996 err = -EINVAL;
2997 goto out2;
2998 }
2999 iboe_addr_get_sgid(dev_addr, &mc->multicast.ib->rec.port_gid);
3000 work->id = id_priv;
3001 work->mc = mc;
3002 INIT_WORK(&work->work, iboe_mcast_work_handler);
3003 kref_get(&mc->mcref);
3004 queue_work(cma_wq, &work->work);
3005
3006 return 0;
3007
3008out2:
3009 kfree(mc->multicast.ib);
3010out1:
3011 kfree(work);
3012 return err;
3013}
3014
2761int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr, 3015int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2762 void *context) 3016 void *context)
2763{ 3017{
@@ -2784,7 +3038,17 @@ int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
2784 3038
2785 switch (rdma_node_get_transport(id->device->node_type)) { 3039 switch (rdma_node_get_transport(id->device->node_type)) {
2786 case RDMA_TRANSPORT_IB: 3040 case RDMA_TRANSPORT_IB:
2787 ret = cma_join_ib_multicast(id_priv, mc); 3041 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3042 case IB_LINK_LAYER_INFINIBAND:
3043 ret = cma_join_ib_multicast(id_priv, mc);
3044 break;
3045 case IB_LINK_LAYER_ETHERNET:
3046 kref_init(&mc->mcref);
3047 ret = cma_iboe_join_multicast(id_priv, mc);
3048 break;
3049 default:
3050 ret = -EINVAL;
3051 }
2788 break; 3052 break;
2789 default: 3053 default:
2790 ret = -ENOSYS; 3054 ret = -ENOSYS;
@@ -2817,8 +3081,19 @@ void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
2817 ib_detach_mcast(id->qp, 3081 ib_detach_mcast(id->qp,
2818 &mc->multicast.ib->rec.mgid, 3082 &mc->multicast.ib->rec.mgid,
2819 mc->multicast.ib->rec.mlid); 3083 mc->multicast.ib->rec.mlid);
2820 ib_sa_free_multicast(mc->multicast.ib); 3084 if (rdma_node_get_transport(id_priv->cma_dev->device->node_type) == RDMA_TRANSPORT_IB) {
2821 kfree(mc); 3085 switch (rdma_port_get_link_layer(id->device, id->port_num)) {
3086 case IB_LINK_LAYER_INFINIBAND:
3087 ib_sa_free_multicast(mc->multicast.ib);
3088 kfree(mc);
3089 break;
3090 case IB_LINK_LAYER_ETHERNET:
3091 kref_put(&mc->mcref, release_mc);
3092 break;
3093 default:
3094 break;
3095 }
3096 }
2822 return; 3097 return;
2823 } 3098 }
2824 } 3099 }
diff --git a/drivers/infiniband/core/iwcm.c b/drivers/infiniband/core/iwcm.c
index bfead5bc25f6..2a1e9ae134b4 100644
--- a/drivers/infiniband/core/iwcm.c
+++ b/drivers/infiniband/core/iwcm.c
@@ -506,6 +506,8 @@ int iw_cm_accept(struct iw_cm_id *cm_id,
506 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn); 506 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn);
507 if (!qp) { 507 if (!qp) {
508 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 508 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
509 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
510 wake_up_all(&cm_id_priv->connect_wait);
509 return -EINVAL; 511 return -EINVAL;
510 } 512 }
511 cm_id->device->iwcm->add_ref(qp); 513 cm_id->device->iwcm->add_ref(qp);
@@ -565,6 +567,8 @@ int iw_cm_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
565 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn); 567 qp = cm_id->device->iwcm->get_qp(cm_id->device, iw_param->qpn);
566 if (!qp) { 568 if (!qp) {
567 spin_unlock_irqrestore(&cm_id_priv->lock, flags); 569 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
570 clear_bit(IWCM_F_CONNECT_WAIT, &cm_id_priv->flags);
571 wake_up_all(&cm_id_priv->connect_wait);
568 return -EINVAL; 572 return -EINVAL;
569 } 573 }
570 cm_id->device->iwcm->add_ref(qp); 574 cm_id->device->iwcm->add_ref(qp);
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index ef1304f151dc..822cfdcd9f78 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -2598,6 +2598,9 @@ static void cleanup_recv_queue(struct ib_mad_qp_info *qp_info)
2598 struct ib_mad_private *recv; 2598 struct ib_mad_private *recv;
2599 struct ib_mad_list_head *mad_list; 2599 struct ib_mad_list_head *mad_list;
2600 2600
2601 if (!qp_info->qp)
2602 return;
2603
2601 while (!list_empty(&qp_info->recv_queue.list)) { 2604 while (!list_empty(&qp_info->recv_queue.list)) {
2602 2605
2603 mad_list = list_entry(qp_info->recv_queue.list.next, 2606 mad_list = list_entry(qp_info->recv_queue.list.next,
@@ -2639,6 +2642,9 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2639 2642
2640 for (i = 0; i < IB_MAD_QPS_CORE; i++) { 2643 for (i = 0; i < IB_MAD_QPS_CORE; i++) {
2641 qp = port_priv->qp_info[i].qp; 2644 qp = port_priv->qp_info[i].qp;
2645 if (!qp)
2646 continue;
2647
2642 /* 2648 /*
2643 * PKey index for QP1 is irrelevant but 2649 * PKey index for QP1 is irrelevant but
2644 * one is needed for the Reset to Init transition 2650 * one is needed for the Reset to Init transition
@@ -2680,6 +2686,9 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2680 } 2686 }
2681 2687
2682 for (i = 0; i < IB_MAD_QPS_CORE; i++) { 2688 for (i = 0; i < IB_MAD_QPS_CORE; i++) {
2689 if (!port_priv->qp_info[i].qp)
2690 continue;
2691
2683 ret = ib_mad_post_receive_mads(&port_priv->qp_info[i], NULL); 2692 ret = ib_mad_post_receive_mads(&port_priv->qp_info[i], NULL);
2684 if (ret) { 2693 if (ret) {
2685 printk(KERN_ERR PFX "Couldn't post receive WRs\n"); 2694 printk(KERN_ERR PFX "Couldn't post receive WRs\n");
@@ -2758,6 +2767,9 @@ error:
2758 2767
2759static void destroy_mad_qp(struct ib_mad_qp_info *qp_info) 2768static void destroy_mad_qp(struct ib_mad_qp_info *qp_info)
2760{ 2769{
2770 if (!qp_info->qp)
2771 return;
2772
2761 ib_destroy_qp(qp_info->qp); 2773 ib_destroy_qp(qp_info->qp);
2762 kfree(qp_info->snoop_table); 2774 kfree(qp_info->snoop_table);
2763} 2775}
@@ -2773,6 +2785,7 @@ static int ib_mad_port_open(struct ib_device *device,
2773 struct ib_mad_port_private *port_priv; 2785 struct ib_mad_port_private *port_priv;
2774 unsigned long flags; 2786 unsigned long flags;
2775 char name[sizeof "ib_mad123"]; 2787 char name[sizeof "ib_mad123"];
2788 int has_smi;
2776 2789
2777 /* Create new device info */ 2790 /* Create new device info */
2778 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL); 2791 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL);
@@ -2788,7 +2801,11 @@ static int ib_mad_port_open(struct ib_device *device,
2788 init_mad_qp(port_priv, &port_priv->qp_info[0]); 2801 init_mad_qp(port_priv, &port_priv->qp_info[0]);
2789 init_mad_qp(port_priv, &port_priv->qp_info[1]); 2802 init_mad_qp(port_priv, &port_priv->qp_info[1]);
2790 2803
2791 cq_size = (mad_sendq_size + mad_recvq_size) * 2; 2804 cq_size = mad_sendq_size + mad_recvq_size;
2805 has_smi = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_INFINIBAND;
2806 if (has_smi)
2807 cq_size *= 2;
2808
2792 port_priv->cq = ib_create_cq(port_priv->device, 2809 port_priv->cq = ib_create_cq(port_priv->device,
2793 ib_mad_thread_completion_handler, 2810 ib_mad_thread_completion_handler,
2794 NULL, port_priv, cq_size, 0); 2811 NULL, port_priv, cq_size, 0);
@@ -2812,9 +2829,11 @@ static int ib_mad_port_open(struct ib_device *device,
2812 goto error5; 2829 goto error5;
2813 } 2830 }
2814 2831
2815 ret = create_mad_qp(&port_priv->qp_info[0], IB_QPT_SMI); 2832 if (has_smi) {
2816 if (ret) 2833 ret = create_mad_qp(&port_priv->qp_info[0], IB_QPT_SMI);
2817 goto error6; 2834 if (ret)
2835 goto error6;
2836 }
2818 ret = create_mad_qp(&port_priv->qp_info[1], IB_QPT_GSI); 2837 ret = create_mad_qp(&port_priv->qp_info[1], IB_QPT_GSI);
2819 if (ret) 2838 if (ret)
2820 goto error7; 2839 goto error7;
diff --git a/drivers/infiniband/core/multicast.c b/drivers/infiniband/core/multicast.c
index a519801dcfb7..68b4162fd9d2 100644
--- a/drivers/infiniband/core/multicast.c
+++ b/drivers/infiniband/core/multicast.c
@@ -774,6 +774,10 @@ static void mcast_event_handler(struct ib_event_handler *handler,
774 int index; 774 int index;
775 775
776 dev = container_of(handler, struct mcast_device, event_handler); 776 dev = container_of(handler, struct mcast_device, event_handler);
777 if (rdma_port_get_link_layer(dev->device, event->element.port_num) !=
778 IB_LINK_LAYER_INFINIBAND)
779 return;
780
777 index = event->element.port_num - dev->start_port; 781 index = event->element.port_num - dev->start_port;
778 782
779 switch (event->event) { 783 switch (event->event) {
@@ -796,6 +800,7 @@ static void mcast_add_one(struct ib_device *device)
796 struct mcast_device *dev; 800 struct mcast_device *dev;
797 struct mcast_port *port; 801 struct mcast_port *port;
798 int i; 802 int i;
803 int count = 0;
799 804
800 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 805 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
801 return; 806 return;
@@ -813,6 +818,9 @@ static void mcast_add_one(struct ib_device *device)
813 } 818 }
814 819
815 for (i = 0; i <= dev->end_port - dev->start_port; i++) { 820 for (i = 0; i <= dev->end_port - dev->start_port; i++) {
821 if (rdma_port_get_link_layer(device, dev->start_port + i) !=
822 IB_LINK_LAYER_INFINIBAND)
823 continue;
816 port = &dev->port[i]; 824 port = &dev->port[i];
817 port->dev = dev; 825 port->dev = dev;
818 port->port_num = dev->start_port + i; 826 port->port_num = dev->start_port + i;
@@ -820,6 +828,12 @@ static void mcast_add_one(struct ib_device *device)
820 port->table = RB_ROOT; 828 port->table = RB_ROOT;
821 init_completion(&port->comp); 829 init_completion(&port->comp);
822 atomic_set(&port->refcount, 1); 830 atomic_set(&port->refcount, 1);
831 ++count;
832 }
833
834 if (!count) {
835 kfree(dev);
836 return;
823 } 837 }
824 838
825 dev->device = device; 839 dev->device = device;
@@ -843,9 +857,12 @@ static void mcast_remove_one(struct ib_device *device)
843 flush_workqueue(mcast_wq); 857 flush_workqueue(mcast_wq);
844 858
845 for (i = 0; i <= dev->end_port - dev->start_port; i++) { 859 for (i = 0; i <= dev->end_port - dev->start_port; i++) {
846 port = &dev->port[i]; 860 if (rdma_port_get_link_layer(device, dev->start_port + i) ==
847 deref_port(port); 861 IB_LINK_LAYER_INFINIBAND) {
848 wait_for_completion(&port->comp); 862 port = &dev->port[i];
863 deref_port(port);
864 wait_for_completion(&port->comp);
865 }
849 } 866 }
850 867
851 kfree(dev); 868 kfree(dev);
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 7e1ffd8ccd5c..91a660310b7c 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -416,6 +416,9 @@ static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event
416 struct ib_sa_port *port = 416 struct ib_sa_port *port =
417 &sa_dev->port[event->element.port_num - sa_dev->start_port]; 417 &sa_dev->port[event->element.port_num - sa_dev->start_port];
418 418
419 if (rdma_port_get_link_layer(handler->device, port->port_num) != IB_LINK_LAYER_INFINIBAND)
420 return;
421
419 spin_lock_irqsave(&port->ah_lock, flags); 422 spin_lock_irqsave(&port->ah_lock, flags);
420 if (port->sm_ah) 423 if (port->sm_ah)
421 kref_put(&port->sm_ah->ref, free_sm_ah); 424 kref_put(&port->sm_ah->ref, free_sm_ah);
@@ -493,6 +496,7 @@ int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
493{ 496{
494 int ret; 497 int ret;
495 u16 gid_index; 498 u16 gid_index;
499 int force_grh;
496 500
497 memset(ah_attr, 0, sizeof *ah_attr); 501 memset(ah_attr, 0, sizeof *ah_attr);
498 ah_attr->dlid = be16_to_cpu(rec->dlid); 502 ah_attr->dlid = be16_to_cpu(rec->dlid);
@@ -502,7 +506,9 @@ int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
502 ah_attr->port_num = port_num; 506 ah_attr->port_num = port_num;
503 ah_attr->static_rate = rec->rate; 507 ah_attr->static_rate = rec->rate;
504 508
505 if (rec->hop_limit > 1) { 509 force_grh = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_ETHERNET;
510
511 if (rec->hop_limit > 1 || force_grh) {
506 ah_attr->ah_flags = IB_AH_GRH; 512 ah_attr->ah_flags = IB_AH_GRH;
507 ah_attr->grh.dgid = rec->dgid; 513 ah_attr->grh.dgid = rec->dgid;
508 514
@@ -1007,7 +1013,7 @@ static void ib_sa_add_one(struct ib_device *device)
1007 e = device->phys_port_cnt; 1013 e = device->phys_port_cnt;
1008 } 1014 }
1009 1015
1010 sa_dev = kmalloc(sizeof *sa_dev + 1016 sa_dev = kzalloc(sizeof *sa_dev +
1011 (e - s + 1) * sizeof (struct ib_sa_port), 1017 (e - s + 1) * sizeof (struct ib_sa_port),
1012 GFP_KERNEL); 1018 GFP_KERNEL);
1013 if (!sa_dev) 1019 if (!sa_dev)
@@ -1017,9 +1023,12 @@ static void ib_sa_add_one(struct ib_device *device)
1017 sa_dev->end_port = e; 1023 sa_dev->end_port = e;
1018 1024
1019 for (i = 0; i <= e - s; ++i) { 1025 for (i = 0; i <= e - s; ++i) {
1026 spin_lock_init(&sa_dev->port[i].ah_lock);
1027 if (rdma_port_get_link_layer(device, i + 1) != IB_LINK_LAYER_INFINIBAND)
1028 continue;
1029
1020 sa_dev->port[i].sm_ah = NULL; 1030 sa_dev->port[i].sm_ah = NULL;
1021 sa_dev->port[i].port_num = i + s; 1031 sa_dev->port[i].port_num = i + s;
1022 spin_lock_init(&sa_dev->port[i].ah_lock);
1023 1032
1024 sa_dev->port[i].agent = 1033 sa_dev->port[i].agent =
1025 ib_register_mad_agent(device, i + s, IB_QPT_GSI, 1034 ib_register_mad_agent(device, i + s, IB_QPT_GSI,
@@ -1045,13 +1054,15 @@ static void ib_sa_add_one(struct ib_device *device)
1045 goto err; 1054 goto err;
1046 1055
1047 for (i = 0; i <= e - s; ++i) 1056 for (i = 0; i <= e - s; ++i)
1048 update_sm_ah(&sa_dev->port[i].update_task); 1057 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1058 update_sm_ah(&sa_dev->port[i].update_task);
1049 1059
1050 return; 1060 return;
1051 1061
1052err: 1062err:
1053 while (--i >= 0) 1063 while (--i >= 0)
1054 ib_unregister_mad_agent(sa_dev->port[i].agent); 1064 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1065 ib_unregister_mad_agent(sa_dev->port[i].agent);
1055 1066
1056 kfree(sa_dev); 1067 kfree(sa_dev);
1057 1068
@@ -1071,9 +1082,12 @@ static void ib_sa_remove_one(struct ib_device *device)
1071 flush_scheduled_work(); 1082 flush_scheduled_work();
1072 1083
1073 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) { 1084 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1074 ib_unregister_mad_agent(sa_dev->port[i].agent); 1085 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) {
1075 if (sa_dev->port[i].sm_ah) 1086 ib_unregister_mad_agent(sa_dev->port[i].agent);
1076 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah); 1087 if (sa_dev->port[i].sm_ah)
1088 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1089 }
1090
1077 } 1091 }
1078 1092
1079 kfree(sa_dev); 1093 kfree(sa_dev);
diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
index 3627300e2a10..9ab5df72df7b 100644
--- a/drivers/infiniband/core/sysfs.c
+++ b/drivers/infiniband/core/sysfs.c
@@ -222,6 +222,19 @@ static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
222 } 222 }
223} 223}
224 224
225static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
226 char *buf)
227{
228 switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
229 case IB_LINK_LAYER_INFINIBAND:
230 return sprintf(buf, "%s\n", "InfiniBand");
231 case IB_LINK_LAYER_ETHERNET:
232 return sprintf(buf, "%s\n", "Ethernet");
233 default:
234 return sprintf(buf, "%s\n", "Unknown");
235 }
236}
237
225static PORT_ATTR_RO(state); 238static PORT_ATTR_RO(state);
226static PORT_ATTR_RO(lid); 239static PORT_ATTR_RO(lid);
227static PORT_ATTR_RO(lid_mask_count); 240static PORT_ATTR_RO(lid_mask_count);
@@ -230,6 +243,7 @@ static PORT_ATTR_RO(sm_sl);
230static PORT_ATTR_RO(cap_mask); 243static PORT_ATTR_RO(cap_mask);
231static PORT_ATTR_RO(rate); 244static PORT_ATTR_RO(rate);
232static PORT_ATTR_RO(phys_state); 245static PORT_ATTR_RO(phys_state);
246static PORT_ATTR_RO(link_layer);
233 247
234static struct attribute *port_default_attrs[] = { 248static struct attribute *port_default_attrs[] = {
235 &port_attr_state.attr, 249 &port_attr_state.attr,
@@ -240,6 +254,7 @@ static struct attribute *port_default_attrs[] = {
240 &port_attr_cap_mask.attr, 254 &port_attr_cap_mask.attr,
241 &port_attr_rate.attr, 255 &port_attr_rate.attr,
242 &port_attr_phys_state.attr, 256 &port_attr_phys_state.attr,
257 &port_attr_link_layer.attr,
243 NULL 258 NULL
244}; 259};
245 260
diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c
index ac7edc24165c..ca12acf38379 100644
--- a/drivers/infiniband/core/ucma.c
+++ b/drivers/infiniband/core/ucma.c
@@ -40,6 +40,7 @@
40#include <linux/in6.h> 40#include <linux/in6.h>
41#include <linux/miscdevice.h> 41#include <linux/miscdevice.h>
42#include <linux/slab.h> 42#include <linux/slab.h>
43#include <linux/sysctl.h>
43 44
44#include <rdma/rdma_user_cm.h> 45#include <rdma/rdma_user_cm.h>
45#include <rdma/ib_marshall.h> 46#include <rdma/ib_marshall.h>
@@ -50,8 +51,24 @@ MODULE_AUTHOR("Sean Hefty");
50MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access"); 51MODULE_DESCRIPTION("RDMA Userspace Connection Manager Access");
51MODULE_LICENSE("Dual BSD/GPL"); 52MODULE_LICENSE("Dual BSD/GPL");
52 53
53enum { 54static unsigned int max_backlog = 1024;
54 UCMA_MAX_BACKLOG = 128 55
56static struct ctl_table_header *ucma_ctl_table_hdr;
57static ctl_table ucma_ctl_table[] = {
58 {
59 .procname = "max_backlog",
60 .data = &max_backlog,
61 .maxlen = sizeof max_backlog,
62 .mode = 0644,
63 .proc_handler = proc_dointvec,
64 },
65 { }
66};
67
68static struct ctl_path ucma_ctl_path[] = {
69 { .procname = "net" },
70 { .procname = "rdma_ucm" },
71 { }
55}; 72};
56 73
57struct ucma_file { 74struct ucma_file {
@@ -583,6 +600,42 @@ static void ucma_copy_ib_route(struct rdma_ucm_query_route_resp *resp,
583 } 600 }
584} 601}
585 602
603static void ucma_copy_iboe_route(struct rdma_ucm_query_route_resp *resp,
604 struct rdma_route *route)
605{
606 struct rdma_dev_addr *dev_addr;
607 struct net_device *dev;
608 u16 vid = 0;
609
610 resp->num_paths = route->num_paths;
611 switch (route->num_paths) {
612 case 0:
613 dev_addr = &route->addr.dev_addr;
614 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
615 if (dev) {
616 vid = rdma_vlan_dev_vlan_id(dev);
617 dev_put(dev);
618 }
619
620 iboe_mac_vlan_to_ll((union ib_gid *) &resp->ib_route[0].dgid,
621 dev_addr->dst_dev_addr, vid);
622 iboe_addr_get_sgid(dev_addr,
623 (union ib_gid *) &resp->ib_route[0].sgid);
624 resp->ib_route[0].pkey = cpu_to_be16(0xffff);
625 break;
626 case 2:
627 ib_copy_path_rec_to_user(&resp->ib_route[1],
628 &route->path_rec[1]);
629 /* fall through */
630 case 1:
631 ib_copy_path_rec_to_user(&resp->ib_route[0],
632 &route->path_rec[0]);
633 break;
634 default:
635 break;
636 }
637}
638
586static ssize_t ucma_query_route(struct ucma_file *file, 639static ssize_t ucma_query_route(struct ucma_file *file,
587 const char __user *inbuf, 640 const char __user *inbuf,
588 int in_len, int out_len) 641 int in_len, int out_len)
@@ -617,12 +670,17 @@ static ssize_t ucma_query_route(struct ucma_file *file,
617 670
618 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid; 671 resp.node_guid = (__force __u64) ctx->cm_id->device->node_guid;
619 resp.port_num = ctx->cm_id->port_num; 672 resp.port_num = ctx->cm_id->port_num;
620 switch (rdma_node_get_transport(ctx->cm_id->device->node_type)) { 673 if (rdma_node_get_transport(ctx->cm_id->device->node_type) == RDMA_TRANSPORT_IB) {
621 case RDMA_TRANSPORT_IB: 674 switch (rdma_port_get_link_layer(ctx->cm_id->device, ctx->cm_id->port_num)) {
622 ucma_copy_ib_route(&resp, &ctx->cm_id->route); 675 case IB_LINK_LAYER_INFINIBAND:
623 break; 676 ucma_copy_ib_route(&resp, &ctx->cm_id->route);
624 default: 677 break;
625 break; 678 case IB_LINK_LAYER_ETHERNET:
679 ucma_copy_iboe_route(&resp, &ctx->cm_id->route);
680 break;
681 default:
682 break;
683 }
626 } 684 }
627 685
628out: 686out:
@@ -686,8 +744,8 @@ static ssize_t ucma_listen(struct ucma_file *file, const char __user *inbuf,
686 if (IS_ERR(ctx)) 744 if (IS_ERR(ctx))
687 return PTR_ERR(ctx); 745 return PTR_ERR(ctx);
688 746
689 ctx->backlog = cmd.backlog > 0 && cmd.backlog < UCMA_MAX_BACKLOG ? 747 ctx->backlog = cmd.backlog > 0 && cmd.backlog < max_backlog ?
690 cmd.backlog : UCMA_MAX_BACKLOG; 748 cmd.backlog : max_backlog;
691 ret = rdma_listen(ctx->cm_id, ctx->backlog); 749 ret = rdma_listen(ctx->cm_id, ctx->backlog);
692 ucma_put_ctx(ctx); 750 ucma_put_ctx(ctx);
693 return ret; 751 return ret;
@@ -1279,16 +1337,26 @@ static int __init ucma_init(void)
1279 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version); 1337 ret = device_create_file(ucma_misc.this_device, &dev_attr_abi_version);
1280 if (ret) { 1338 if (ret) {
1281 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n"); 1339 printk(KERN_ERR "rdma_ucm: couldn't create abi_version attr\n");
1282 goto err; 1340 goto err1;
1341 }
1342
1343 ucma_ctl_table_hdr = register_sysctl_paths(ucma_ctl_path, ucma_ctl_table);
1344 if (!ucma_ctl_table_hdr) {
1345 printk(KERN_ERR "rdma_ucm: couldn't register sysctl paths\n");
1346 ret = -ENOMEM;
1347 goto err2;
1283 } 1348 }
1284 return 0; 1349 return 0;
1285err: 1350err2:
1351 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1352err1:
1286 misc_deregister(&ucma_misc); 1353 misc_deregister(&ucma_misc);
1287 return ret; 1354 return ret;
1288} 1355}
1289 1356
1290static void __exit ucma_cleanup(void) 1357static void __exit ucma_cleanup(void)
1291{ 1358{
1359 unregister_sysctl_table(ucma_ctl_table_hdr);
1292 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version); 1360 device_remove_file(ucma_misc.this_device, &dev_attr_abi_version);
1293 misc_deregister(&ucma_misc); 1361 misc_deregister(&ucma_misc);
1294 idr_destroy(&ctx_idr); 1362 idr_destroy(&ctx_idr);
diff --git a/drivers/infiniband/core/ud_header.c b/drivers/infiniband/core/ud_header.c
index 650b501eb142..bb7e19280821 100644
--- a/drivers/infiniband/core/ud_header.c
+++ b/drivers/infiniband/core/ud_header.c
@@ -33,6 +33,7 @@
33 33
34#include <linux/errno.h> 34#include <linux/errno.h>
35#include <linux/string.h> 35#include <linux/string.h>
36#include <linux/if_ether.h>
36 37
37#include <rdma/ib_pack.h> 38#include <rdma/ib_pack.h>
38 39
@@ -80,6 +81,40 @@ static const struct ib_field lrh_table[] = {
80 .size_bits = 16 } 81 .size_bits = 16 }
81}; 82};
82 83
84static const struct ib_field eth_table[] = {
85 { STRUCT_FIELD(eth, dmac_h),
86 .offset_words = 0,
87 .offset_bits = 0,
88 .size_bits = 32 },
89 { STRUCT_FIELD(eth, dmac_l),
90 .offset_words = 1,
91 .offset_bits = 0,
92 .size_bits = 16 },
93 { STRUCT_FIELD(eth, smac_h),
94 .offset_words = 1,
95 .offset_bits = 16,
96 .size_bits = 16 },
97 { STRUCT_FIELD(eth, smac_l),
98 .offset_words = 2,
99 .offset_bits = 0,
100 .size_bits = 32 },
101 { STRUCT_FIELD(eth, type),
102 .offset_words = 3,
103 .offset_bits = 0,
104 .size_bits = 16 }
105};
106
107static const struct ib_field vlan_table[] = {
108 { STRUCT_FIELD(vlan, tag),
109 .offset_words = 0,
110 .offset_bits = 0,
111 .size_bits = 16 },
112 { STRUCT_FIELD(vlan, type),
113 .offset_words = 0,
114 .offset_bits = 16,
115 .size_bits = 16 }
116};
117
83static const struct ib_field grh_table[] = { 118static const struct ib_field grh_table[] = {
84 { STRUCT_FIELD(grh, ip_version), 119 { STRUCT_FIELD(grh, ip_version),
85 .offset_words = 0, 120 .offset_words = 0,
@@ -180,38 +215,43 @@ static const struct ib_field deth_table[] = {
180/** 215/**
181 * ib_ud_header_init - Initialize UD header structure 216 * ib_ud_header_init - Initialize UD header structure
182 * @payload_bytes:Length of packet payload 217 * @payload_bytes:Length of packet payload
218 * @lrh_present: specify if LRH is present
219 * @eth_present: specify if Eth header is present
220 * @vlan_present: packet is tagged vlan
183 * @grh_present:GRH flag (if non-zero, GRH will be included) 221 * @grh_present:GRH flag (if non-zero, GRH will be included)
184 * @immediate_present: specify if immediate data should be used 222 * @immediate_present: specify if immediate data is present
185 * @header:Structure to initialize 223 * @header:Structure to initialize
186 *
187 * ib_ud_header_init() initializes the lrh.link_version, lrh.link_next_header,
188 * lrh.packet_length, grh.ip_version, grh.payload_length,
189 * grh.next_header, bth.opcode, bth.pad_count and
190 * bth.transport_header_version fields of a &struct ib_ud_header given
191 * the payload length and whether a GRH will be included.
192 */ 224 */
193void ib_ud_header_init(int payload_bytes, 225void ib_ud_header_init(int payload_bytes,
226 int lrh_present,
227 int eth_present,
228 int vlan_present,
194 int grh_present, 229 int grh_present,
195 int immediate_present, 230 int immediate_present,
196 struct ib_ud_header *header) 231 struct ib_ud_header *header)
197{ 232{
198 u16 packet_length;
199
200 memset(header, 0, sizeof *header); 233 memset(header, 0, sizeof *header);
201 234
202 header->lrh.link_version = 0; 235 if (lrh_present) {
203 header->lrh.link_next_header = 236 u16 packet_length;
204 grh_present ? IB_LNH_IBA_GLOBAL : IB_LNH_IBA_LOCAL; 237
205 packet_length = (IB_LRH_BYTES + 238 header->lrh.link_version = 0;
206 IB_BTH_BYTES + 239 header->lrh.link_next_header =
207 IB_DETH_BYTES + 240 grh_present ? IB_LNH_IBA_GLOBAL : IB_LNH_IBA_LOCAL;
208 payload_bytes + 241 packet_length = (IB_LRH_BYTES +
209 4 + /* ICRC */ 242 IB_BTH_BYTES +
210 3) / 4; /* round up */ 243 IB_DETH_BYTES +
211 244 (grh_present ? IB_GRH_BYTES : 0) +
212 header->grh_present = grh_present; 245 payload_bytes +
246 4 + /* ICRC */
247 3) / 4; /* round up */
248 header->lrh.packet_length = cpu_to_be16(packet_length);
249 }
250
251 if (vlan_present)
252 header->eth.type = cpu_to_be16(ETH_P_8021Q);
253
213 if (grh_present) { 254 if (grh_present) {
214 packet_length += IB_GRH_BYTES / 4;
215 header->grh.ip_version = 6; 255 header->grh.ip_version = 6;
216 header->grh.payload_length = 256 header->grh.payload_length =
217 cpu_to_be16((IB_BTH_BYTES + 257 cpu_to_be16((IB_BTH_BYTES +
@@ -222,19 +262,52 @@ void ib_ud_header_init(int payload_bytes,
222 header->grh.next_header = 0x1b; 262 header->grh.next_header = 0x1b;
223 } 263 }
224 264
225 header->lrh.packet_length = cpu_to_be16(packet_length);
226
227 header->immediate_present = immediate_present;
228 if (immediate_present) 265 if (immediate_present)
229 header->bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE; 266 header->bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
230 else 267 else
231 header->bth.opcode = IB_OPCODE_UD_SEND_ONLY; 268 header->bth.opcode = IB_OPCODE_UD_SEND_ONLY;
232 header->bth.pad_count = (4 - payload_bytes) & 3; 269 header->bth.pad_count = (4 - payload_bytes) & 3;
233 header->bth.transport_header_version = 0; 270 header->bth.transport_header_version = 0;
271
272 header->lrh_present = lrh_present;
273 header->eth_present = eth_present;
274 header->vlan_present = vlan_present;
275 header->grh_present = grh_present;
276 header->immediate_present = immediate_present;
234} 277}
235EXPORT_SYMBOL(ib_ud_header_init); 278EXPORT_SYMBOL(ib_ud_header_init);
236 279
237/** 280/**
281 * ib_lrh_header_pack - Pack LRH header struct into wire format
282 * @lrh:unpacked LRH header struct
283 * @buf:Buffer to pack into
284 *
285 * ib_lrh_header_pack() packs the LRH header structure @lrh into
286 * wire format in the buffer @buf.
287 */
288int ib_lrh_header_pack(struct ib_unpacked_lrh *lrh, void *buf)
289{
290 ib_pack(lrh_table, ARRAY_SIZE(lrh_table), lrh, buf);
291 return 0;
292}
293EXPORT_SYMBOL(ib_lrh_header_pack);
294
295/**
296 * ib_lrh_header_unpack - Unpack LRH structure from wire format
297 * @lrh:unpacked LRH header struct
298 * @buf:Buffer to pack into
299 *
300 * ib_lrh_header_unpack() unpacks the LRH header structure from
301 * wire format (in buf) into @lrh.
302 */
303int ib_lrh_header_unpack(void *buf, struct ib_unpacked_lrh *lrh)
304{
305 ib_unpack(lrh_table, ARRAY_SIZE(lrh_table), buf, lrh);
306 return 0;
307}
308EXPORT_SYMBOL(ib_lrh_header_unpack);
309
310/**
238 * ib_ud_header_pack - Pack UD header struct into wire format 311 * ib_ud_header_pack - Pack UD header struct into wire format
239 * @header:UD header struct 312 * @header:UD header struct
240 * @buf:Buffer to pack into 313 * @buf:Buffer to pack into
@@ -247,10 +320,21 @@ int ib_ud_header_pack(struct ib_ud_header *header,
247{ 320{
248 int len = 0; 321 int len = 0;
249 322
250 ib_pack(lrh_table, ARRAY_SIZE(lrh_table), 323 if (header->lrh_present) {
251 &header->lrh, buf); 324 ib_pack(lrh_table, ARRAY_SIZE(lrh_table),
252 len += IB_LRH_BYTES; 325 &header->lrh, buf + len);
253 326 len += IB_LRH_BYTES;
327 }
328 if (header->eth_present) {
329 ib_pack(eth_table, ARRAY_SIZE(eth_table),
330 &header->eth, buf + len);
331 len += IB_ETH_BYTES;
332 }
333 if (header->vlan_present) {
334 ib_pack(vlan_table, ARRAY_SIZE(vlan_table),
335 &header->vlan, buf + len);
336 len += IB_VLAN_BYTES;
337 }
254 if (header->grh_present) { 338 if (header->grh_present) {
255 ib_pack(grh_table, ARRAY_SIZE(grh_table), 339 ib_pack(grh_table, ARRAY_SIZE(grh_table),
256 &header->grh, buf + len); 340 &header->grh, buf + len);
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index 5fa856909511..cd1996d0ad08 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -1022,7 +1022,7 @@ static int ib_umad_init_port(struct ib_device *device, int port_num,
1022 1022
1023 port->ib_dev = device; 1023 port->ib_dev = device;
1024 port->port_num = port_num; 1024 port->port_num = port_num;
1025 init_MUTEX(&port->sm_sem); 1025 sema_init(&port->sm_sem, 1);
1026 mutex_init(&port->file_mutex); 1026 mutex_init(&port->file_mutex);
1027 INIT_LIST_HEAD(&port->file_list); 1027 INIT_LIST_HEAD(&port->file_list);
1028 1028
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 6fcfbeb24a23..b342248aec05 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -460,6 +460,8 @@ ssize_t ib_uverbs_query_port(struct ib_uverbs_file *file,
460 resp.active_width = attr.active_width; 460 resp.active_width = attr.active_width;
461 resp.active_speed = attr.active_speed; 461 resp.active_speed = attr.active_speed;
462 resp.phys_state = attr.phys_state; 462 resp.phys_state = attr.phys_state;
463 resp.link_layer = rdma_port_get_link_layer(file->device->ib_dev,
464 cmd.port_num);
463 465
464 if (copy_to_user((void __user *) (unsigned long) cmd.response, 466 if (copy_to_user((void __user *) (unsigned long) cmd.response,
465 &resp, sizeof resp)) 467 &resp, sizeof resp))
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index e0fa22238715..af7a8b08b2e9 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -94,6 +94,22 @@ rdma_node_get_transport(enum rdma_node_type node_type)
94} 94}
95EXPORT_SYMBOL(rdma_node_get_transport); 95EXPORT_SYMBOL(rdma_node_get_transport);
96 96
97enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
98{
99 if (device->get_link_layer)
100 return device->get_link_layer(device, port_num);
101
102 switch (rdma_node_get_transport(device->node_type)) {
103 case RDMA_TRANSPORT_IB:
104 return IB_LINK_LAYER_INFINIBAND;
105 case RDMA_TRANSPORT_IWARP:
106 return IB_LINK_LAYER_ETHERNET;
107 default:
108 return IB_LINK_LAYER_UNSPECIFIED;
109 }
110}
111EXPORT_SYMBOL(rdma_port_get_link_layer);
112
97/* Protection domains */ 113/* Protection domains */
98 114
99struct ib_pd *ib_alloc_pd(struct ib_device *device) 115struct ib_pd *ib_alloc_pd(struct ib_device *device)
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-25 00:11:51 -0500