aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband
diff options
context:
space:
mode:
authorStefan Richter <stefanr@s5r6.in-berlin.de>2006-09-17 12:17:19 -0400
committerStefan Richter <stefanr@s5r6.in-berlin.de>2006-09-17 12:19:31 -0400
commit9b4f2e9576658c4e52d95dc8d309f51b2e2db096 (patch)
tree7b1902b0f931783fccc6fee45c6f9c16b4fde5ce /drivers/infiniband
parent3c6c65f5ed5a6d307bd607aecd06d658c0934d88 (diff)
parent803db244b9f71102e366fd689000c1417b9a7508 (diff)
ieee1394: merge from Linus
Conflicts: drivers/ieee1394/hosts.c Patch "lockdep: annotate ieee1394 skb-queue-head locking" was meddling with patch "ieee1394: fix kerneldoc of hpsb_alloc_host". Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Diffstat (limited to 'drivers/infiniband')
-rw-r--r--drivers/infiniband/core/addr.c30
-rw-r--r--drivers/infiniband/core/cache.c3
-rw-r--r--drivers/infiniband/core/cm.c25
-rw-r--r--drivers/infiniband/core/cma.c24
-rw-r--r--drivers/infiniband/core/fmr_pool.c8
-rw-r--r--drivers/infiniband/core/mad.c22
-rw-r--r--drivers/infiniband/core/sa_query.c13
-rw-r--r--drivers/infiniband/core/user_mad.c87
-rw-r--r--drivers/infiniband/core/uverbs.h2
-rw-r--r--drivers/infiniband/core/uverbs_cmd.c42
-rw-r--r--drivers/infiniband/core/uverbs_main.c8
-rw-r--r--drivers/infiniband/hw/ipath/ipath_driver.c76
-rw-r--r--drivers/infiniband/hw/ipath/ipath_keys.c15
-rw-r--r--drivers/infiniband/hw/ipath/ipath_verbs.c5
-rw-r--r--drivers/infiniband/hw/mthca/mthca_allocator.c30
-rw-r--r--drivers/infiniband/hw/mthca/mthca_av.c5
-rw-r--r--drivers/infiniband/hw/mthca/mthca_cmd.c5
-rw-r--r--drivers/infiniband/hw/mthca/mthca_main.c6
-rw-r--r--drivers/infiniband/hw/mthca/mthca_provider.c11
-rw-r--r--drivers/infiniband/hw/mthca/mthca_provider.h4
-rw-r--r--drivers/infiniband/hw/mthca/mthca_qp.c68
-rw-r--r--drivers/infiniband/hw/mthca/mthca_srq.c3
-rw-r--r--drivers/infiniband/ulp/ipoib/Kconfig3
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib.h1
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_main.c23
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_multicast.c9
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.c160
-rw-r--r--drivers/infiniband/ulp/iser/iser_verbs.c2
-rw-r--r--drivers/infiniband/ulp/srp/ib_srp.c36
29 files changed, 394 insertions, 332 deletions
diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c
index d294bbc42f09..1205e8027829 100644
--- a/drivers/infiniband/core/addr.c
+++ b/drivers/infiniband/core/addr.c
@@ -35,6 +35,7 @@
35#include <net/arp.h> 35#include <net/arp.h>
36#include <net/neighbour.h> 36#include <net/neighbour.h>
37#include <net/route.h> 37#include <net/route.h>
38#include <net/netevent.h>
38#include <rdma/ib_addr.h> 39#include <rdma/ib_addr.h>
39 40
40MODULE_AUTHOR("Sean Hefty"); 41MODULE_AUTHOR("Sean Hefty");
@@ -326,25 +327,22 @@ void rdma_addr_cancel(struct rdma_dev_addr *addr)
326} 327}
327EXPORT_SYMBOL(rdma_addr_cancel); 328EXPORT_SYMBOL(rdma_addr_cancel);
328 329
329static int addr_arp_recv(struct sk_buff *skb, struct net_device *dev, 330static int netevent_callback(struct notifier_block *self, unsigned long event,
330 struct packet_type *pkt, struct net_device *orig_dev) 331 void *ctx)
331{ 332{
332 struct arphdr *arp_hdr; 333 if (event == NETEVENT_NEIGH_UPDATE) {
334 struct neighbour *neigh = ctx;
333 335
334 arp_hdr = (struct arphdr *) skb->nh.raw; 336 if (neigh->dev->type == ARPHRD_INFINIBAND &&
335 337 (neigh->nud_state & NUD_VALID)) {
336 if (arp_hdr->ar_op == htons(ARPOP_REQUEST) || 338 set_timeout(jiffies);
337 arp_hdr->ar_op == htons(ARPOP_REPLY)) 339 }
338 set_timeout(jiffies); 340 }
339
340 kfree_skb(skb);
341 return 0; 341 return 0;
342} 342}
343 343
344static struct packet_type addr_arp = { 344static struct notifier_block nb = {
345 .type = __constant_htons(ETH_P_ARP), 345 .notifier_call = netevent_callback
346 .func = addr_arp_recv,
347 .af_packet_priv = (void*) 1,
348}; 346};
349 347
350static int addr_init(void) 348static int addr_init(void)
@@ -353,13 +351,13 @@ static int addr_init(void)
353 if (!addr_wq) 351 if (!addr_wq)
354 return -ENOMEM; 352 return -ENOMEM;
355 353
356 dev_add_pack(&addr_arp); 354 register_netevent_notifier(&nb);
357 return 0; 355 return 0;
358} 356}
359 357
360static void addr_cleanup(void) 358static void addr_cleanup(void)
361{ 359{
362 dev_remove_pack(&addr_arp); 360 unregister_netevent_notifier(&nb);
363 destroy_workqueue(addr_wq); 361 destroy_workqueue(addr_wq);
364} 362}
365 363
diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c
index e05ca2cdc73f..75313ade2e0d 100644
--- a/drivers/infiniband/core/cache.c
+++ b/drivers/infiniband/core/cache.c
@@ -301,7 +301,8 @@ static void ib_cache_event(struct ib_event_handler *handler,
301 event->event == IB_EVENT_PORT_ACTIVE || 301 event->event == IB_EVENT_PORT_ACTIVE ||
302 event->event == IB_EVENT_LID_CHANGE || 302 event->event == IB_EVENT_LID_CHANGE ||
303 event->event == IB_EVENT_PKEY_CHANGE || 303 event->event == IB_EVENT_PKEY_CHANGE ||
304 event->event == IB_EVENT_SM_CHANGE) { 304 event->event == IB_EVENT_SM_CHANGE ||
305 event->event == IB_EVENT_CLIENT_REREGISTER) {
305 work = kmalloc(sizeof *work, GFP_ATOMIC); 306 work = kmalloc(sizeof *work, GFP_ATOMIC);
306 if (work) { 307 if (work) {
307 INIT_WORK(&work->work, ib_cache_task, work); 308 INIT_WORK(&work->work, ib_cache_task, work);
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
index 3f6705f3083a..0de335b7bfc2 100644
--- a/drivers/infiniband/core/cm.c
+++ b/drivers/infiniband/core/cm.c
@@ -701,7 +701,7 @@ static void cm_reset_to_idle(struct cm_id_private *cm_id_priv)
701 } 701 }
702} 702}
703 703
704void ib_destroy_cm_id(struct ib_cm_id *cm_id) 704static void cm_destroy_id(struct ib_cm_id *cm_id, int err)
705{ 705{
706 struct cm_id_private *cm_id_priv; 706 struct cm_id_private *cm_id_priv;
707 struct cm_work *work; 707 struct cm_work *work;
@@ -735,12 +735,22 @@ retest:
735 sizeof cm_id_priv->av.port->cm_dev->ca_guid, 735 sizeof cm_id_priv->av.port->cm_dev->ca_guid,
736 NULL, 0); 736 NULL, 0);
737 break; 737 break;
738 case IB_CM_REQ_RCVD:
739 if (err == -ENOMEM) {
740 /* Do not reject to allow future retries. */
741 cm_reset_to_idle(cm_id_priv);
742 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
743 } else {
744 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
745 ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
746 NULL, 0, NULL, 0);
747 }
748 break;
738 case IB_CM_MRA_REQ_RCVD: 749 case IB_CM_MRA_REQ_RCVD:
739 case IB_CM_REP_SENT: 750 case IB_CM_REP_SENT:
740 case IB_CM_MRA_REP_RCVD: 751 case IB_CM_MRA_REP_RCVD:
741 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg); 752 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
742 /* Fall through */ 753 /* Fall through */
743 case IB_CM_REQ_RCVD:
744 case IB_CM_MRA_REQ_SENT: 754 case IB_CM_MRA_REQ_SENT:
745 case IB_CM_REP_RCVD: 755 case IB_CM_REP_RCVD:
746 case IB_CM_MRA_REP_SENT: 756 case IB_CM_MRA_REP_SENT:
@@ -775,6 +785,11 @@ retest:
775 kfree(cm_id_priv->private_data); 785 kfree(cm_id_priv->private_data);
776 kfree(cm_id_priv); 786 kfree(cm_id_priv);
777} 787}
788
789void ib_destroy_cm_id(struct ib_cm_id *cm_id)
790{
791 cm_destroy_id(cm_id, 0);
792}
778EXPORT_SYMBOL(ib_destroy_cm_id); 793EXPORT_SYMBOL(ib_destroy_cm_id);
779 794
780int ib_cm_listen(struct ib_cm_id *cm_id, __be64 service_id, __be64 service_mask, 795int ib_cm_listen(struct ib_cm_id *cm_id, __be64 service_id, __be64 service_mask,
@@ -960,8 +975,10 @@ int ib_send_cm_req(struct ib_cm_id *cm_id,
960 975
961 cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv-> 976 cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
962 id.local_id); 977 id.local_id);
963 if (IS_ERR(cm_id_priv->timewait_info)) 978 if (IS_ERR(cm_id_priv->timewait_info)) {
979 ret = PTR_ERR(cm_id_priv->timewait_info);
964 goto out; 980 goto out;
981 }
965 982
966 ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av); 983 ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av);
967 if (ret) 984 if (ret)
@@ -1163,7 +1180,7 @@ static void cm_process_work(struct cm_id_private *cm_id_priv,
1163 } 1180 }
1164 cm_deref_id(cm_id_priv); 1181 cm_deref_id(cm_id_priv);
1165 if (ret) 1182 if (ret)
1166 ib_destroy_cm_id(&cm_id_priv->id); 1183 cm_destroy_id(&cm_id_priv->id, ret);
1167} 1184}
1168 1185
1169static void cm_format_mra(struct cm_mra_msg *mra_msg, 1186static void cm_format_mra(struct cm_mra_msg *mra_msg,
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 863f64befc7c..5d625a81193f 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -49,7 +49,7 @@ MODULE_DESCRIPTION("Generic RDMA CM Agent");
49MODULE_LICENSE("Dual BSD/GPL"); 49MODULE_LICENSE("Dual BSD/GPL");
50 50
51#define CMA_CM_RESPONSE_TIMEOUT 20 51#define CMA_CM_RESPONSE_TIMEOUT 20
52#define CMA_MAX_CM_RETRIES 3 52#define CMA_MAX_CM_RETRIES 15
53 53
54static void cma_add_one(struct ib_device *device); 54static void cma_add_one(struct ib_device *device);
55static void cma_remove_one(struct ib_device *device); 55static void cma_remove_one(struct ib_device *device);
@@ -262,14 +262,14 @@ static void cma_detach_from_dev(struct rdma_id_private *id_priv)
262static int cma_acquire_ib_dev(struct rdma_id_private *id_priv) 262static int cma_acquire_ib_dev(struct rdma_id_private *id_priv)
263{ 263{
264 struct cma_device *cma_dev; 264 struct cma_device *cma_dev;
265 union ib_gid *gid; 265 union ib_gid gid;
266 int ret = -ENODEV; 266 int ret = -ENODEV;
267 267
268 gid = ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr); 268 ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid),
269 269
270 mutex_lock(&lock); 270 mutex_lock(&lock);
271 list_for_each_entry(cma_dev, &dev_list, list) { 271 list_for_each_entry(cma_dev, &dev_list, list) {
272 ret = ib_find_cached_gid(cma_dev->device, gid, 272 ret = ib_find_cached_gid(cma_dev->device, &gid,
273 &id_priv->id.port_num, NULL); 273 &id_priv->id.port_num, NULL);
274 if (!ret) { 274 if (!ret) {
275 cma_attach_to_dev(id_priv, cma_dev); 275 cma_attach_to_dev(id_priv, cma_dev);
@@ -812,6 +812,7 @@ static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
812 cma_modify_qp_err(&id_priv->id); 812 cma_modify_qp_err(&id_priv->id);
813 status = ib_event->param.rej_rcvd.reason; 813 status = ib_event->param.rej_rcvd.reason;
814 event = RDMA_CM_EVENT_REJECTED; 814 event = RDMA_CM_EVENT_REJECTED;
815 private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
815 break; 816 break;
816 default: 817 default:
817 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d", 818 printk(KERN_ERR "RDMA CMA: unexpected IB CM event: %d",
@@ -1134,8 +1135,8 @@ static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
1134 struct ib_sa_path_rec path_rec; 1135 struct ib_sa_path_rec path_rec;
1135 1136
1136 memset(&path_rec, 0, sizeof path_rec); 1137 memset(&path_rec, 0, sizeof path_rec);
1137 path_rec.sgid = *ib_addr_get_sgid(addr); 1138 ib_addr_get_sgid(addr, &path_rec.sgid);
1138 path_rec.dgid = *ib_addr_get_dgid(addr); 1139 ib_addr_get_dgid(addr, &path_rec.dgid);
1139 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(addr)); 1140 path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(addr));
1140 path_rec.numb_path = 1; 1141 path_rec.numb_path = 1;
1141 1142
@@ -1263,7 +1264,7 @@ static int cma_bind_loopback(struct rdma_id_private *id_priv)
1263{ 1264{
1264 struct cma_device *cma_dev; 1265 struct cma_device *cma_dev;
1265 struct ib_port_attr port_attr; 1266 struct ib_port_attr port_attr;
1266 union ib_gid *gid; 1267 union ib_gid gid;
1267 u16 pkey; 1268 u16 pkey;
1268 int ret; 1269 int ret;
1269 u8 p; 1270 u8 p;
@@ -1284,8 +1285,7 @@ static int cma_bind_loopback(struct rdma_id_private *id_priv)
1284 } 1285 }
1285 1286
1286port_found: 1287port_found:
1287 gid = ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr); 1288 ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid);
1288 ret = ib_get_cached_gid(cma_dev->device, p, 0, gid);
1289 if (ret) 1289 if (ret)
1290 goto out; 1290 goto out;
1291 1291
@@ -1293,6 +1293,7 @@ port_found:
1293 if (ret) 1293 if (ret)
1294 goto out; 1294 goto out;
1295 1295
1296 ib_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1296 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey); 1297 ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
1297 id_priv->id.port_num = p; 1298 id_priv->id.port_num = p;
1298 cma_attach_to_dev(id_priv, cma_dev); 1299 cma_attach_to_dev(id_priv, cma_dev);
@@ -1339,6 +1340,7 @@ static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1339{ 1340{
1340 struct cma_work *work; 1341 struct cma_work *work;
1341 struct sockaddr_in *src_in, *dst_in; 1342 struct sockaddr_in *src_in, *dst_in;
1343 union ib_gid gid;
1342 int ret; 1344 int ret;
1343 1345
1344 work = kzalloc(sizeof *work, GFP_KERNEL); 1346 work = kzalloc(sizeof *work, GFP_KERNEL);
@@ -1351,8 +1353,8 @@ static int cma_resolve_loopback(struct rdma_id_private *id_priv)
1351 goto err; 1353 goto err;
1352 } 1354 }
1353 1355
1354 ib_addr_set_dgid(&id_priv->id.route.addr.dev_addr, 1356 ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
1355 ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr)); 1357 ib_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
1356 1358
1357 if (cma_zero_addr(&id_priv->id.route.addr.src_addr)) { 1359 if (cma_zero_addr(&id_priv->id.route.addr.src_addr)) {
1358 src_in = (struct sockaddr_in *)&id_priv->id.route.addr.src_addr; 1360 src_in = (struct sockaddr_in *)&id_priv->id.route.addr.src_addr;
diff --git a/drivers/infiniband/core/fmr_pool.c b/drivers/infiniband/core/fmr_pool.c
index 615fe9cc6c56..86a3b2d401db 100644
--- a/drivers/infiniband/core/fmr_pool.c
+++ b/drivers/infiniband/core/fmr_pool.c
@@ -426,7 +426,7 @@ EXPORT_SYMBOL(ib_flush_fmr_pool);
426struct ib_pool_fmr *ib_fmr_pool_map_phys(struct ib_fmr_pool *pool_handle, 426struct ib_pool_fmr *ib_fmr_pool_map_phys(struct ib_fmr_pool *pool_handle,
427 u64 *page_list, 427 u64 *page_list,
428 int list_len, 428 int list_len,
429 u64 *io_virtual_address) 429 u64 io_virtual_address)
430{ 430{
431 struct ib_fmr_pool *pool = pool_handle; 431 struct ib_fmr_pool *pool = pool_handle;
432 struct ib_pool_fmr *fmr; 432 struct ib_pool_fmr *fmr;
@@ -440,7 +440,7 @@ struct ib_pool_fmr *ib_fmr_pool_map_phys(struct ib_fmr_pool *pool_handle,
440 fmr = ib_fmr_cache_lookup(pool, 440 fmr = ib_fmr_cache_lookup(pool,
441 page_list, 441 page_list,
442 list_len, 442 list_len,
443 *io_virtual_address); 443 io_virtual_address);
444 if (fmr) { 444 if (fmr) {
445 /* found in cache */ 445 /* found in cache */
446 ++fmr->ref_count; 446 ++fmr->ref_count;
@@ -464,7 +464,7 @@ struct ib_pool_fmr *ib_fmr_pool_map_phys(struct ib_fmr_pool *pool_handle,
464 spin_unlock_irqrestore(&pool->pool_lock, flags); 464 spin_unlock_irqrestore(&pool->pool_lock, flags);
465 465
466 result = ib_map_phys_fmr(fmr->fmr, page_list, list_len, 466 result = ib_map_phys_fmr(fmr->fmr, page_list, list_len,
467 *io_virtual_address); 467 io_virtual_address);
468 468
469 if (result) { 469 if (result) {
470 spin_lock_irqsave(&pool->pool_lock, flags); 470 spin_lock_irqsave(&pool->pool_lock, flags);
@@ -481,7 +481,7 @@ struct ib_pool_fmr *ib_fmr_pool_map_phys(struct ib_fmr_pool *pool_handle,
481 fmr->ref_count = 1; 481 fmr->ref_count = 1;
482 482
483 if (pool->cache_bucket) { 483 if (pool->cache_bucket) {
484 fmr->io_virtual_address = *io_virtual_address; 484 fmr->io_virtual_address = io_virtual_address;
485 fmr->page_list_len = list_len; 485 fmr->page_list_len = list_len;
486 memcpy(fmr->page_list, page_list, list_len * sizeof(*page_list)); 486 memcpy(fmr->page_list, page_list, list_len * sizeof(*page_list));
487 487
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index 5ed4dab52a6f..1c3cfbbe6a97 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -167,6 +167,15 @@ static int is_vendor_method_in_use(
167 return 0; 167 return 0;
168} 168}
169 169
170int ib_response_mad(struct ib_mad *mad)
171{
172 return ((mad->mad_hdr.method & IB_MGMT_METHOD_RESP) ||
173 (mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS) ||
174 ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_BM) &&
175 (mad->mad_hdr.attr_mod & IB_BM_ATTR_MOD_RESP)));
176}
177EXPORT_SYMBOL(ib_response_mad);
178
170/* 179/*
171 * ib_register_mad_agent - Register to send/receive MADs 180 * ib_register_mad_agent - Register to send/receive MADs
172 */ 181 */
@@ -570,13 +579,6 @@ int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent)
570} 579}
571EXPORT_SYMBOL(ib_unregister_mad_agent); 580EXPORT_SYMBOL(ib_unregister_mad_agent);
572 581
573static inline int response_mad(struct ib_mad *mad)
574{
575 /* Trap represses are responses although response bit is reset */
576 return ((mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS) ||
577 (mad->mad_hdr.method & IB_MGMT_METHOD_RESP));
578}
579
580static void dequeue_mad(struct ib_mad_list_head *mad_list) 582static void dequeue_mad(struct ib_mad_list_head *mad_list)
581{ 583{
582 struct ib_mad_queue *mad_queue; 584 struct ib_mad_queue *mad_queue;
@@ -723,7 +725,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
723 switch (ret) 725 switch (ret)
724 { 726 {
725 case IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY: 727 case IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY:
726 if (response_mad(&mad_priv->mad.mad) && 728 if (ib_response_mad(&mad_priv->mad.mad) &&
727 mad_agent_priv->agent.recv_handler) { 729 mad_agent_priv->agent.recv_handler) {
728 local->mad_priv = mad_priv; 730 local->mad_priv = mad_priv;
729 local->recv_mad_agent = mad_agent_priv; 731 local->recv_mad_agent = mad_agent_priv;
@@ -1551,7 +1553,7 @@ find_mad_agent(struct ib_mad_port_private *port_priv,
1551 unsigned long flags; 1553 unsigned long flags;
1552 1554
1553 spin_lock_irqsave(&port_priv->reg_lock, flags); 1555 spin_lock_irqsave(&port_priv->reg_lock, flags);
1554 if (response_mad(mad)) { 1556 if (ib_response_mad(mad)) {
1555 u32 hi_tid; 1557 u32 hi_tid;
1556 struct ib_mad_agent_private *entry; 1558 struct ib_mad_agent_private *entry;
1557 1559
@@ -1799,7 +1801,7 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
1799 } 1801 }
1800 1802
1801 /* Complete corresponding request */ 1803 /* Complete corresponding request */
1802 if (response_mad(mad_recv_wc->recv_buf.mad)) { 1804 if (ib_response_mad(mad_recv_wc->recv_buf.mad)) {
1803 spin_lock_irqsave(&mad_agent_priv->lock, flags); 1805 spin_lock_irqsave(&mad_agent_priv->lock, flags);
1804 mad_send_wr = ib_find_send_mad(mad_agent_priv, mad_recv_wc); 1806 mad_send_wr = ib_find_send_mad(mad_agent_priv, mad_recv_wc);
1805 if (!mad_send_wr) { 1807 if (!mad_send_wr) {
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index e911c99ff843..d6b84226bba7 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -405,7 +405,8 @@ static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event
405 event->event == IB_EVENT_PORT_ACTIVE || 405 event->event == IB_EVENT_PORT_ACTIVE ||
406 event->event == IB_EVENT_LID_CHANGE || 406 event->event == IB_EVENT_LID_CHANGE ||
407 event->event == IB_EVENT_PKEY_CHANGE || 407 event->event == IB_EVENT_PKEY_CHANGE ||
408 event->event == IB_EVENT_SM_CHANGE) { 408 event->event == IB_EVENT_SM_CHANGE ||
409 event->event == IB_EVENT_CLIENT_REREGISTER) {
409 struct ib_sa_device *sa_dev; 410 struct ib_sa_device *sa_dev;
410 sa_dev = container_of(handler, typeof(*sa_dev), event_handler); 411 sa_dev = container_of(handler, typeof(*sa_dev), event_handler);
411 412
@@ -488,13 +489,13 @@ static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
488 spin_unlock_irqrestore(&tid_lock, flags); 489 spin_unlock_irqrestore(&tid_lock, flags);
489} 490}
490 491
491static int send_mad(struct ib_sa_query *query, int timeout_ms) 492static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
492{ 493{
493 unsigned long flags; 494 unsigned long flags;
494 int ret, id; 495 int ret, id;
495 496
496retry: 497retry:
497 if (!idr_pre_get(&query_idr, GFP_ATOMIC)) 498 if (!idr_pre_get(&query_idr, gfp_mask))
498 return -ENOMEM; 499 return -ENOMEM;
499 spin_lock_irqsave(&idr_lock, flags); 500 spin_lock_irqsave(&idr_lock, flags);
500 ret = idr_get_new(&query_idr, query, &id); 501 ret = idr_get_new(&query_idr, query, &id);
@@ -630,7 +631,7 @@ int ib_sa_path_rec_get(struct ib_device *device, u8 port_num,
630 631
631 *sa_query = &query->sa_query; 632 *sa_query = &query->sa_query;
632 633
633 ret = send_mad(&query->sa_query, timeout_ms); 634 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
634 if (ret < 0) 635 if (ret < 0)
635 goto err2; 636 goto err2;
636 637
@@ -752,7 +753,7 @@ int ib_sa_service_rec_query(struct ib_device *device, u8 port_num, u8 method,
752 753
753 *sa_query = &query->sa_query; 754 *sa_query = &query->sa_query;
754 755
755 ret = send_mad(&query->sa_query, timeout_ms); 756 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
756 if (ret < 0) 757 if (ret < 0)
757 goto err2; 758 goto err2;
758 759
@@ -844,7 +845,7 @@ int ib_sa_mcmember_rec_query(struct ib_device *device, u8 port_num,
844 845
845 *sa_query = &query->sa_query; 846 *sa_query = &query->sa_query;
846 847
847 ret = send_mad(&query->sa_query, timeout_ms); 848 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
848 if (ret < 0) 849 if (ret < 0)
849 goto err2; 850 goto err2;
850 851
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index afe70a549c2f..1273f8807e84 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -112,8 +112,10 @@ struct ib_umad_device {
112struct ib_umad_file { 112struct ib_umad_file {
113 struct ib_umad_port *port; 113 struct ib_umad_port *port;
114 struct list_head recv_list; 114 struct list_head recv_list;
115 struct list_head send_list;
115 struct list_head port_list; 116 struct list_head port_list;
116 spinlock_t recv_lock; 117 spinlock_t recv_lock;
118 spinlock_t send_lock;
117 wait_queue_head_t recv_wait; 119 wait_queue_head_t recv_wait;
118 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS]; 120 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
119 int agents_dead; 121 int agents_dead;
@@ -177,12 +179,21 @@ static int queue_packet(struct ib_umad_file *file,
177 return ret; 179 return ret;
178} 180}
179 181
182static void dequeue_send(struct ib_umad_file *file,
183 struct ib_umad_packet *packet)
184 {
185 spin_lock_irq(&file->send_lock);
186 list_del(&packet->list);
187 spin_unlock_irq(&file->send_lock);
188 }
189
180static void send_handler(struct ib_mad_agent *agent, 190static void send_handler(struct ib_mad_agent *agent,
181 struct ib_mad_send_wc *send_wc) 191 struct ib_mad_send_wc *send_wc)
182{ 192{
183 struct ib_umad_file *file = agent->context; 193 struct ib_umad_file *file = agent->context;
184 struct ib_umad_packet *packet = send_wc->send_buf->context[0]; 194 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
185 195
196 dequeue_send(file, packet);
186 ib_destroy_ah(packet->msg->ah); 197 ib_destroy_ah(packet->msg->ah);
187 ib_free_send_mad(packet->msg); 198 ib_free_send_mad(packet->msg);
188 199
@@ -370,6 +381,51 @@ static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
370 return 0; 381 return 0;
371} 382}
372 383
384static int same_destination(struct ib_user_mad_hdr *hdr1,
385 struct ib_user_mad_hdr *hdr2)
386{
387 if (!hdr1->grh_present && !hdr2->grh_present)
388 return (hdr1->lid == hdr2->lid);
389
390 if (hdr1->grh_present && hdr2->grh_present)
391 return !memcmp(hdr1->gid, hdr2->gid, 16);
392
393 return 0;
394}
395
396static int is_duplicate(struct ib_umad_file *file,
397 struct ib_umad_packet *packet)
398{
399 struct ib_umad_packet *sent_packet;
400 struct ib_mad_hdr *sent_hdr, *hdr;
401
402 hdr = (struct ib_mad_hdr *) packet->mad.data;
403 list_for_each_entry(sent_packet, &file->send_list, list) {
404 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
405
406 if ((hdr->tid != sent_hdr->tid) ||
407 (hdr->mgmt_class != sent_hdr->mgmt_class))
408 continue;
409
410 /*
411 * No need to be overly clever here. If two new operations have
412 * the same TID, reject the second as a duplicate. This is more
413 * restrictive than required by the spec.
414 */
415 if (!ib_response_mad((struct ib_mad *) hdr)) {
416 if (!ib_response_mad((struct ib_mad *) sent_hdr))
417 return 1;
418 continue;
419 } else if (!ib_response_mad((struct ib_mad *) sent_hdr))
420 continue;
421
422 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
423 return 1;
424 }
425
426 return 0;
427}
428
373static ssize_t ib_umad_write(struct file *filp, const char __user *buf, 429static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
374 size_t count, loff_t *pos) 430 size_t count, loff_t *pos)
375{ 431{
@@ -379,7 +435,6 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
379 struct ib_ah_attr ah_attr; 435 struct ib_ah_attr ah_attr;
380 struct ib_ah *ah; 436 struct ib_ah *ah;
381 struct ib_rmpp_mad *rmpp_mad; 437 struct ib_rmpp_mad *rmpp_mad;
382 u8 method;
383 __be64 *tid; 438 __be64 *tid;
384 int ret, data_len, hdr_len, copy_offset, rmpp_active; 439 int ret, data_len, hdr_len, copy_offset, rmpp_active;
385 440
@@ -473,28 +528,36 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
473 } 528 }
474 529
475 /* 530 /*
476 * If userspace is generating a request that will generate a 531 * Set the high-order part of the transaction ID to make MADs from
477 * response, we need to make sure the high-order part of the 532 * different agents unique, and allow routing responses back to the
478 * transaction ID matches the agent being used to send the 533 * original requestor.
479 * MAD.
480 */ 534 */
481 method = ((struct ib_mad_hdr *) packet->msg->mad)->method; 535 if (!ib_response_mad(packet->msg->mad)) {
482
483 if (!(method & IB_MGMT_METHOD_RESP) &&
484 method != IB_MGMT_METHOD_TRAP_REPRESS &&
485 method != IB_MGMT_METHOD_SEND) {
486 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid; 536 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
487 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 | 537 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
488 (be64_to_cpup(tid) & 0xffffffff)); 538 (be64_to_cpup(tid) & 0xffffffff));
539 rmpp_mad->mad_hdr.tid = *tid;
540 }
541
542 spin_lock_irq(&file->send_lock);
543 ret = is_duplicate(file, packet);
544 if (!ret)
545 list_add_tail(&packet->list, &file->send_list);
546 spin_unlock_irq(&file->send_lock);
547 if (ret) {
548 ret = -EINVAL;
549 goto err_msg;
489 } 550 }
490 551
491 ret = ib_post_send_mad(packet->msg, NULL); 552 ret = ib_post_send_mad(packet->msg, NULL);
492 if (ret) 553 if (ret)
493 goto err_msg; 554 goto err_send;
494 555
495 up_read(&file->port->mutex); 556 up_read(&file->port->mutex);
496 return count; 557 return count;
497 558
559err_send:
560 dequeue_send(file, packet);
498err_msg: 561err_msg:
499 ib_free_send_mad(packet->msg); 562 ib_free_send_mad(packet->msg);
500err_ah: 563err_ah:
@@ -657,7 +720,9 @@ static int ib_umad_open(struct inode *inode, struct file *filp)
657 } 720 }
658 721
659 spin_lock_init(&file->recv_lock); 722 spin_lock_init(&file->recv_lock);
723 spin_lock_init(&file->send_lock);
660 INIT_LIST_HEAD(&file->recv_list); 724 INIT_LIST_HEAD(&file->recv_list);
725 INIT_LIST_HEAD(&file->send_list);
661 init_waitqueue_head(&file->recv_wait); 726 init_waitqueue_head(&file->recv_wait);
662 727
663 file->port = port; 728 file->port = port;
diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h
index bb9bee56a824..102a59c033ff 100644
--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h
@@ -42,6 +42,7 @@
42#include <linux/kref.h> 42#include <linux/kref.h>
43#include <linux/idr.h> 43#include <linux/idr.h>
44#include <linux/mutex.h> 44#include <linux/mutex.h>
45#include <linux/completion.h>
45 46
46#include <rdma/ib_verbs.h> 47#include <rdma/ib_verbs.h>
47#include <rdma/ib_user_verbs.h> 48#include <rdma/ib_user_verbs.h>
@@ -69,6 +70,7 @@
69 70
70struct ib_uverbs_device { 71struct ib_uverbs_device {
71 struct kref ref; 72 struct kref ref;
73 struct completion comp;
72 int devnum; 74 int devnum;
73 struct cdev *dev; 75 struct cdev *dev;
74 struct class_device *class_dev; 76 struct class_device *class_dev;
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index bdf5d5098190..30923eb68ec7 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -42,6 +42,13 @@
42 42
43#include "uverbs.h" 43#include "uverbs.h"
44 44
45static struct lock_class_key pd_lock_key;
46static struct lock_class_key mr_lock_key;
47static struct lock_class_key cq_lock_key;
48static struct lock_class_key qp_lock_key;
49static struct lock_class_key ah_lock_key;
50static struct lock_class_key srq_lock_key;
51
45#define INIT_UDATA(udata, ibuf, obuf, ilen, olen) \ 52#define INIT_UDATA(udata, ibuf, obuf, ilen, olen) \
46 do { \ 53 do { \
47 (udata)->inbuf = (void __user *) (ibuf); \ 54 (udata)->inbuf = (void __user *) (ibuf); \
@@ -76,12 +83,13 @@
76 */ 83 */
77 84
78static void init_uobj(struct ib_uobject *uobj, u64 user_handle, 85static void init_uobj(struct ib_uobject *uobj, u64 user_handle,
79 struct ib_ucontext *context) 86 struct ib_ucontext *context, struct lock_class_key *key)
80{ 87{
81 uobj->user_handle = user_handle; 88 uobj->user_handle = user_handle;
82 uobj->context = context; 89 uobj->context = context;
83 kref_init(&uobj->ref); 90 kref_init(&uobj->ref);
84 init_rwsem(&uobj->mutex); 91 init_rwsem(&uobj->mutex);
92 lockdep_set_class(&uobj->mutex, key);
85 uobj->live = 0; 93 uobj->live = 0;
86} 94}
87 95
@@ -470,7 +478,7 @@ ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
470 if (!uobj) 478 if (!uobj)
471 return -ENOMEM; 479 return -ENOMEM;
472 480
473 init_uobj(uobj, 0, file->ucontext); 481 init_uobj(uobj, 0, file->ucontext, &pd_lock_key);
474 down_write(&uobj->mutex); 482 down_write(&uobj->mutex);
475 483
476 pd = file->device->ib_dev->alloc_pd(file->device->ib_dev, 484 pd = file->device->ib_dev->alloc_pd(file->device->ib_dev,
@@ -591,7 +599,7 @@ ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
591 if (!obj) 599 if (!obj)
592 return -ENOMEM; 600 return -ENOMEM;
593 601
594 init_uobj(&obj->uobject, 0, file->ucontext); 602 init_uobj(&obj->uobject, 0, file->ucontext, &mr_lock_key);
595 down_write(&obj->uobject.mutex); 603 down_write(&obj->uobject.mutex);
596 604
597 /* 605 /*
@@ -770,7 +778,7 @@ ssize_t ib_uverbs_create_cq(struct ib_uverbs_file *file,
770 if (!obj) 778 if (!obj)
771 return -ENOMEM; 779 return -ENOMEM;
772 780
773 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext); 781 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &cq_lock_key);
774 down_write(&obj->uobject.mutex); 782 down_write(&obj->uobject.mutex);
775 783
776 if (cmd.comp_channel >= 0) { 784 if (cmd.comp_channel >= 0) {
@@ -1051,13 +1059,14 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1051 if (!obj) 1059 if (!obj)
1052 return -ENOMEM; 1060 return -ENOMEM;
1053 1061
1054 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext); 1062 init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
1055 down_write(&obj->uevent.uobject.mutex); 1063 down_write(&obj->uevent.uobject.mutex);
1056 1064
1065 srq = cmd.is_srq ? idr_read_srq(cmd.srq_handle, file->ucontext) : NULL;
1057 pd = idr_read_pd(cmd.pd_handle, file->ucontext); 1066 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1058 scq = idr_read_cq(cmd.send_cq_handle, file->ucontext); 1067 scq = idr_read_cq(cmd.send_cq_handle, file->ucontext);
1059 rcq = idr_read_cq(cmd.recv_cq_handle, file->ucontext); 1068 rcq = cmd.recv_cq_handle == cmd.send_cq_handle ?
1060 srq = cmd.is_srq ? idr_read_srq(cmd.srq_handle, file->ucontext) : NULL; 1069 scq : idr_read_cq(cmd.recv_cq_handle, file->ucontext);
1061 1070
1062 if (!pd || !scq || !rcq || (cmd.is_srq && !srq)) { 1071 if (!pd || !scq || !rcq || (cmd.is_srq && !srq)) {
1063 ret = -EINVAL; 1072 ret = -EINVAL;
@@ -1125,7 +1134,8 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
1125 1134
1126 put_pd_read(pd); 1135 put_pd_read(pd);
1127 put_cq_read(scq); 1136 put_cq_read(scq);
1128 put_cq_read(rcq); 1137 if (rcq != scq)
1138 put_cq_read(rcq);
1129 if (srq) 1139 if (srq)
1130 put_srq_read(srq); 1140 put_srq_read(srq);
1131 1141
@@ -1150,7 +1160,7 @@ err_put:
1150 put_pd_read(pd); 1160 put_pd_read(pd);
1151 if (scq) 1161 if (scq)
1152 put_cq_read(scq); 1162 put_cq_read(scq);
1153 if (rcq) 1163 if (rcq && rcq != scq)
1154 put_cq_read(rcq); 1164 put_cq_read(rcq);
1155 if (srq) 1165 if (srq)
1156 put_srq_read(srq); 1166 put_srq_read(srq);
@@ -1751,7 +1761,7 @@ ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
1751 if (!uobj) 1761 if (!uobj)
1752 return -ENOMEM; 1762 return -ENOMEM;
1753 1763
1754 init_uobj(uobj, cmd.user_handle, file->ucontext); 1764 init_uobj(uobj, cmd.user_handle, file->ucontext, &ah_lock_key);
1755 down_write(&uobj->mutex); 1765 down_write(&uobj->mutex);
1756 1766
1757 pd = idr_read_pd(cmd.pd_handle, file->ucontext); 1767 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
@@ -1775,7 +1785,7 @@ ssize_t ib_uverbs_create_ah(struct ib_uverbs_file *file,
1775 ah = ib_create_ah(pd, &attr); 1785 ah = ib_create_ah(pd, &attr);
1776 if (IS_ERR(ah)) { 1786 if (IS_ERR(ah)) {
1777 ret = PTR_ERR(ah); 1787 ret = PTR_ERR(ah);
1778 goto err; 1788 goto err_put;
1779 } 1789 }
1780 1790
1781 ah->uobject = uobj; 1791 ah->uobject = uobj;
@@ -1811,6 +1821,9 @@ err_copy:
1811err_destroy: 1821err_destroy:
1812 ib_destroy_ah(ah); 1822 ib_destroy_ah(ah);
1813 1823
1824err_put:
1825 put_pd_read(pd);
1826
1814err: 1827err:
1815 put_uobj_write(uobj); 1828 put_uobj_write(uobj);
1816 return ret; 1829 return ret;
@@ -1963,7 +1976,7 @@ ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
1963 if (!obj) 1976 if (!obj)
1964 return -ENOMEM; 1977 return -ENOMEM;
1965 1978
1966 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext); 1979 init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &srq_lock_key);
1967 down_write(&obj->uobject.mutex); 1980 down_write(&obj->uobject.mutex);
1968 1981
1969 pd = idr_read_pd(cmd.pd_handle, file->ucontext); 1982 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
@@ -1984,7 +1997,7 @@ ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
1984 srq = pd->device->create_srq(pd, &attr, &udata); 1997 srq = pd->device->create_srq(pd, &attr, &udata);
1985 if (IS_ERR(srq)) { 1998 if (IS_ERR(srq)) {
1986 ret = PTR_ERR(srq); 1999 ret = PTR_ERR(srq);
1987 goto err; 2000 goto err_put;
1988 } 2001 }
1989 2002
1990 srq->device = pd->device; 2003 srq->device = pd->device;
@@ -2029,6 +2042,9 @@ err_copy:
2029err_destroy: 2042err_destroy:
2030 ib_destroy_srq(srq); 2043 ib_destroy_srq(srq);
2031 2044
2045err_put:
2046 put_pd_read(pd);
2047
2032err: 2048err:
2033 put_uobj_write(&obj->uobject); 2049 put_uobj_write(&obj->uobject);
2034 return ret; 2050 return ret;
diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index e725cccc7cde..4e16314e8e6d 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -122,7 +122,7 @@ static void ib_uverbs_release_dev(struct kref *ref)
122 struct ib_uverbs_device *dev = 122 struct ib_uverbs_device *dev =
123 container_of(ref, struct ib_uverbs_device, ref); 123 container_of(ref, struct ib_uverbs_device, ref);
124 124
125 kfree(dev); 125 complete(&dev->comp);
126} 126}
127 127
128void ib_uverbs_release_ucq(struct ib_uverbs_file *file, 128void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
@@ -740,6 +740,7 @@ static void ib_uverbs_add_one(struct ib_device *device)
740 return; 740 return;
741 741
742 kref_init(&uverbs_dev->ref); 742 kref_init(&uverbs_dev->ref);
743 init_completion(&uverbs_dev->comp);
743 744
744 spin_lock(&map_lock); 745 spin_lock(&map_lock);
745 uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES); 746 uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
@@ -793,6 +794,8 @@ err_cdev:
793 794
794err: 795err:
795 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev); 796 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
797 wait_for_completion(&uverbs_dev->comp);
798 kfree(uverbs_dev);
796 return; 799 return;
797} 800}
798 801
@@ -812,7 +815,10 @@ static void ib_uverbs_remove_one(struct ib_device *device)
812 spin_unlock(&map_lock); 815 spin_unlock(&map_lock);
813 816
814 clear_bit(uverbs_dev->devnum, dev_map); 817 clear_bit(uverbs_dev->devnum, dev_map);
818
815 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev); 819 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
820 wait_for_completion(&uverbs_dev->comp);
821 kfree(uverbs_dev);
816} 822}
817 823
818static int uverbs_event_get_sb(struct file_system_type *fs_type, int flags, 824static int uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index 823131d58b34..f98518d912b5 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -859,6 +859,38 @@ static void ipath_rcv_layer(struct ipath_devdata *dd, u32 etail,
859 __ipath_layer_rcv_lid(dd, hdr); 859 __ipath_layer_rcv_lid(dd, hdr);
860} 860}
861 861
862static void ipath_rcv_hdrerr(struct ipath_devdata *dd,
863 u32 eflags,
864 u32 l,
865 u32 etail,
866 u64 *rc)
867{
868 char emsg[128];
869 struct ipath_message_header *hdr;
870
871 get_rhf_errstring(eflags, emsg, sizeof emsg);
872 hdr = (struct ipath_message_header *)&rc[1];
873 ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u "
874 "tlen=%x opcode=%x egridx=%x: %s\n",
875 eflags, l,
876 ipath_hdrget_rcv_type((__le32 *) rc),
877 ipath_hdrget_length_in_bytes((__le32 *) rc),
878 be32_to_cpu(hdr->bth[0]) >> 24,
879 etail, emsg);
880
881 /* Count local link integrity errors. */
882 if (eflags & (INFINIPATH_RHF_H_ICRCERR | INFINIPATH_RHF_H_VCRCERR)) {
883 u8 n = (dd->ipath_ibcctrl >>
884 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
885 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
886
887 if (++dd->ipath_lli_counter > n) {
888 dd->ipath_lli_counter = 0;
889 dd->ipath_lli_errors++;
890 }
891 }
892}
893
862/* 894/*
863 * ipath_kreceive - receive a packet 895 * ipath_kreceive - receive a packet
864 * @dd: the infinipath device 896 * @dd: the infinipath device
@@ -875,7 +907,6 @@ void ipath_kreceive(struct ipath_devdata *dd)
875 struct ipath_message_header *hdr; 907 struct ipath_message_header *hdr;
876 u32 eflags, i, etype, tlen, pkttot = 0, updegr=0, reloop=0; 908 u32 eflags, i, etype, tlen, pkttot = 0, updegr=0, reloop=0;
877 static u64 totcalls; /* stats, may eventually remove */ 909 static u64 totcalls; /* stats, may eventually remove */
878 char emsg[128];
879 910
880 if (!dd->ipath_hdrqtailptr) { 911 if (!dd->ipath_hdrqtailptr) {
881 ipath_dev_err(dd, 912 ipath_dev_err(dd,
@@ -938,26 +969,9 @@ reloop:
938 "%x\n", etype); 969 "%x\n", etype);
939 } 970 }
940 971
941 if (eflags & ~(INFINIPATH_RHF_H_TIDERR | 972 if (unlikely(eflags))
942 INFINIPATH_RHF_H_IHDRERR)) { 973 ipath_rcv_hdrerr(dd, eflags, l, etail, rc);
943 get_rhf_errstring(eflags, emsg, sizeof emsg); 974 else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
944 ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u "
945 "tlen=%x opcode=%x egridx=%x: %s\n",
946 eflags, l, etype, tlen, bthbytes[0],
947 ipath_hdrget_index((__le32 *) rc), emsg);
948 /* Count local link integrity errors. */
949 if (eflags & (INFINIPATH_RHF_H_ICRCERR |
950 INFINIPATH_RHF_H_VCRCERR)) {
951 u8 n = (dd->ipath_ibcctrl >>
952 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
953 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
954
955 if (++dd->ipath_lli_counter > n) {
956 dd->ipath_lli_counter = 0;
957 dd->ipath_lli_errors++;
958 }
959 }
960 } else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
961 int ret = __ipath_verbs_rcv(dd, rc + 1, 975 int ret = __ipath_verbs_rcv(dd, rc + 1,
962 ebuf, tlen); 976 ebuf, tlen);
963 if (ret == -ENODEV) 977 if (ret == -ENODEV)
@@ -981,25 +995,7 @@ reloop:
981 else if (etype == RCVHQ_RCV_TYPE_EXPECTED) 995 else if (etype == RCVHQ_RCV_TYPE_EXPECTED)
982 ipath_dbg("Bug: Expected TID, opcode %x; ignored\n", 996 ipath_dbg("Bug: Expected TID, opcode %x; ignored\n",
983 be32_to_cpu(hdr->bth[0]) & 0xff); 997 be32_to_cpu(hdr->bth[0]) & 0xff);
984 else if (eflags & (INFINIPATH_RHF_H_TIDERR | 998 else {
985 INFINIPATH_RHF_H_IHDRERR)) {
986 /*
987 * This is a type 3 packet, only the LRH is in the
988 * rcvhdrq, the rest of the header is in the eager
989 * buffer.
990 */
991 u8 opcode;
992 if (ebuf) {
993 bthbytes = (u8 *) ebuf;
994 opcode = *bthbytes;
995 }
996 else
997 opcode = 0;
998 get_rhf_errstring(eflags, emsg, sizeof emsg);
999 ipath_dbg("Err %x (%s), opcode %x, egrbuf %x, "
1000 "len %x\n", eflags, emsg, opcode, etail,
1001 tlen);
1002 } else {
1003 /* 999 /*
1004 * error packet, type of error unknown. 1000 * error packet, type of error unknown.
1005 * Probably type 3, but we don't know, so don't 1001 * Probably type 3, but we don't know, so don't
diff --git a/drivers/infiniband/hw/ipath/ipath_keys.c b/drivers/infiniband/hw/ipath/ipath_keys.c
index 46773c673a1a..a5ca279370aa 100644
--- a/drivers/infiniband/hw/ipath/ipath_keys.c
+++ b/drivers/infiniband/hw/ipath/ipath_keys.c
@@ -197,6 +197,21 @@ int ipath_rkey_ok(struct ipath_ibdev *dev, struct ipath_sge_state *ss,
197 size_t off; 197 size_t off;
198 int ret; 198 int ret;
199 199
200 /*
201 * We use RKEY == zero for physical addresses
202 * (see ipath_get_dma_mr).
203 */
204 if (rkey == 0) {
205 sge->mr = NULL;
206 sge->vaddr = phys_to_virt(vaddr);
207 sge->length = len;
208 sge->sge_length = len;
209 ss->sg_list = NULL;
210 ss->num_sge = 1;
211 ret = 1;
212 goto bail;
213 }
214
200 mr = rkt->table[(rkey >> (32 - ib_ipath_lkey_table_size))]; 215 mr = rkt->table[(rkey >> (32 - ib_ipath_lkey_table_size))];
201 if (unlikely(mr == NULL || mr->lkey != rkey)) { 216 if (unlikely(mr == NULL || mr->lkey != rkey)) {
202 ret = 0; 217 ret = 0;
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index 56ac336dd1ec..d70a9b6b5239 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -191,10 +191,6 @@ void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
191{ 191{
192 struct ipath_sge *sge = &ss->sge; 192 struct ipath_sge *sge = &ss->sge;
193 193
194 while (length > sge->sge_length) {
195 length -= sge->sge_length;
196 ss->sge = *ss->sg_list++;
197 }
198 while (length) { 194 while (length) {
199 u32 len = sge->length; 195 u32 len = sge->length;
200 196
@@ -627,6 +623,7 @@ static int ipath_query_device(struct ib_device *ibdev,
627 props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR | 623 props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
628 IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT | 624 IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
629 IB_DEVICE_SYS_IMAGE_GUID; 625 IB_DEVICE_SYS_IMAGE_GUID;
626 props->page_size_cap = PAGE_SIZE;
630 props->vendor_id = ipath_layer_get_vendorid(dev->dd); 627 props->vendor_id = ipath_layer_get_vendorid(dev->dd);
631 props->vendor_part_id = ipath_layer_get_deviceid(dev->dd); 628 props->vendor_part_id = ipath_layer_get_deviceid(dev->dd);
632 props->hw_ver = ipath_layer_get_pcirev(dev->dd); 629 props->hw_ver = ipath_layer_get_pcirev(dev->dd);
diff --git a/drivers/infiniband/hw/mthca/mthca_allocator.c b/drivers/infiniband/hw/mthca/mthca_allocator.c
index 9ba3211cef7c..f930e55b58fc 100644
--- a/drivers/infiniband/hw/mthca/mthca_allocator.c
+++ b/drivers/infiniband/hw/mthca/mthca_allocator.c
@@ -41,9 +41,11 @@
41/* Trivial bitmap-based allocator */ 41/* Trivial bitmap-based allocator */
42u32 mthca_alloc(struct mthca_alloc *alloc) 42u32 mthca_alloc(struct mthca_alloc *alloc)
43{ 43{
44 unsigned long flags;
44 u32 obj; 45 u32 obj;
45 46
46 spin_lock(&alloc->lock); 47 spin_lock_irqsave(&alloc->lock, flags);
48
47 obj = find_next_zero_bit(alloc->table, alloc->max, alloc->last); 49 obj = find_next_zero_bit(alloc->table, alloc->max, alloc->last);
48 if (obj >= alloc->max) { 50 if (obj >= alloc->max) {
49 alloc->top = (alloc->top + alloc->max) & alloc->mask; 51 alloc->top = (alloc->top + alloc->max) & alloc->mask;
@@ -56,19 +58,24 @@ u32 mthca_alloc(struct mthca_alloc *alloc)
56 } else 58 } else
57 obj = -1; 59 obj = -1;
58 60
59 spin_unlock(&alloc->lock); 61 spin_unlock_irqrestore(&alloc->lock, flags);
60 62
61 return obj; 63 return obj;
62} 64}
63 65
64void mthca_free(struct mthca_alloc *alloc, u32 obj) 66void mthca_free(struct mthca_alloc *alloc, u32 obj)
65{ 67{
68 unsigned long flags;
69
66 obj &= alloc->max - 1; 70 obj &= alloc->max - 1;
67 spin_lock(&alloc->lock); 71
72 spin_lock_irqsave(&alloc->lock, flags);
73
68 clear_bit(obj, alloc->table); 74 clear_bit(obj, alloc->table);
69 alloc->last = min(alloc->last, obj); 75 alloc->last = min(alloc->last, obj);
70 alloc->top = (alloc->top + alloc->max) & alloc->mask; 76 alloc->top = (alloc->top + alloc->max) & alloc->mask;
71 spin_unlock(&alloc->lock); 77
78 spin_unlock_irqrestore(&alloc->lock, flags);
72} 79}
73 80
74int mthca_alloc_init(struct mthca_alloc *alloc, u32 num, u32 mask, 81int mthca_alloc_init(struct mthca_alloc *alloc, u32 num, u32 mask,
@@ -108,14 +115,15 @@ void mthca_alloc_cleanup(struct mthca_alloc *alloc)
108 * serialize access to the array. 115 * serialize access to the array.
109 */ 116 */
110 117
118#define MTHCA_ARRAY_MASK (PAGE_SIZE / sizeof (void *) - 1)
119
111void *mthca_array_get(struct mthca_array *array, int index) 120void *mthca_array_get(struct mthca_array *array, int index)
112{ 121{
113 int p = (index * sizeof (void *)) >> PAGE_SHIFT; 122 int p = (index * sizeof (void *)) >> PAGE_SHIFT;
114 123
115 if (array->page_list[p].page) { 124 if (array->page_list[p].page)
116 int i = index & (PAGE_SIZE / sizeof (void *) - 1); 125 return array->page_list[p].page[index & MTHCA_ARRAY_MASK];
117 return array->page_list[p].page[i]; 126 else
118 } else
119 return NULL; 127 return NULL;
120} 128}
121 129
@@ -130,8 +138,7 @@ int mthca_array_set(struct mthca_array *array, int index, void *value)
130 if (!array->page_list[p].page) 138 if (!array->page_list[p].page)
131 return -ENOMEM; 139 return -ENOMEM;
132 140
133 array->page_list[p].page[index & (PAGE_SIZE / sizeof (void *) - 1)] = 141 array->page_list[p].page[index & MTHCA_ARRAY_MASK] = value;
134 value;
135 ++array->page_list[p].used; 142 ++array->page_list[p].used;
136 143
137 return 0; 144 return 0;
@@ -144,7 +151,8 @@ void mthca_array_clear(struct mthca_array *array, int index)
144 if (--array->page_list[p].used == 0) { 151 if (--array->page_list[p].used == 0) {
145 free_page((unsigned long) array->page_list[p].page); 152 free_page((unsigned long) array->page_list[p].page);
146 array->page_list[p].page = NULL; 153 array->page_list[p].page = NULL;
147 } 154 } else
155 array->page_list[p].page[index & MTHCA_ARRAY_MASK] = NULL;
148 156
149 if (array->page_list[p].used < 0) 157 if (array->page_list[p].used < 0)
150 pr_debug("Array %p index %d page %d with ref count %d < 0\n", 158 pr_debug("Array %p index %d page %d with ref count %d < 0\n",
diff --git a/drivers/infiniband/hw/mthca/mthca_av.c b/drivers/infiniband/hw/mthca/mthca_av.c
index b12aa03be251..e215041b2db9 100644
--- a/drivers/infiniband/hw/mthca/mthca_av.c
+++ b/drivers/infiniband/hw/mthca/mthca_av.c
@@ -303,9 +303,10 @@ int mthca_ah_query(struct ib_ah *ibah, struct ib_ah_attr *attr)
303 memset(attr, 0, sizeof *attr); 303 memset(attr, 0, sizeof *attr);
304 attr->dlid = be16_to_cpu(ah->av->dlid); 304 attr->dlid = be16_to_cpu(ah->av->dlid);
305 attr->sl = be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 28; 305 attr->sl = be32_to_cpu(ah->av->sl_tclass_flowlabel) >> 28;
306 attr->static_rate = ah->av->msg_sr & 0x7;
307 attr->src_path_bits = ah->av->g_slid & 0x7F;
308 attr->port_num = be32_to_cpu(ah->av->port_pd) >> 24; 306 attr->port_num = be32_to_cpu(ah->av->port_pd) >> 24;
307 attr->static_rate = mthca_rate_to_ib(dev, ah->av->msg_sr & 0x7,
308 attr->port_num);
309 attr->src_path_bits = ah->av->g_slid & 0x7F;
309 attr->ah_flags = mthca_ah_grh_present(ah) ? IB_AH_GRH : 0; 310 attr->ah_flags = mthca_ah_grh_present(ah) ? IB_AH_GRH : 0;
310 311
311 if (attr->ah_flags) { 312 if (attr->ah_flags) {
diff --git a/drivers/infiniband/hw/mthca/mthca_cmd.c b/drivers/infiniband/hw/mthca/mthca_cmd.c
index d0f7731802c9..deabc14b4ea4 100644
--- a/drivers/infiniband/hw/mthca/mthca_cmd.c
+++ b/drivers/infiniband/hw/mthca/mthca_cmd.c
@@ -778,11 +778,12 @@ int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status)
778 ((dev->fw_ver & 0xffff0000ull) >> 16) | 778 ((dev->fw_ver & 0xffff0000ull) >> 16) |
779 ((dev->fw_ver & 0x0000ffffull) << 16); 779 ((dev->fw_ver & 0x0000ffffull) << 16);
780 780
781 MTHCA_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
782 dev->cmd.max_cmds = 1 << lg;
783
781 mthca_dbg(dev, "FW version %012llx, max commands %d\n", 784 mthca_dbg(dev, "FW version %012llx, max commands %d\n",
782 (unsigned long long) dev->fw_ver, dev->cmd.max_cmds); 785 (unsigned long long) dev->fw_ver, dev->cmd.max_cmds);
783 786
784 MTHCA_GET(lg, outbox, QUERY_FW_MAX_CMD_OFFSET);
785 dev->cmd.max_cmds = 1 << lg;
786 MTHCA_GET(dev->catas_err.addr, outbox, QUERY_FW_ERR_START_OFFSET); 787 MTHCA_GET(dev->catas_err.addr, outbox, QUERY_FW_ERR_START_OFFSET);
787 MTHCA_GET(dev->catas_err.size, outbox, QUERY_FW_ERR_SIZE_OFFSET); 788 MTHCA_GET(dev->catas_err.size, outbox, QUERY_FW_ERR_SIZE_OFFSET);
788 789
diff --git a/drivers/infiniband/hw/mthca/mthca_main.c b/drivers/infiniband/hw/mthca/mthca_main.c
index 557cde3a4563..7b82c1907f04 100644
--- a/drivers/infiniband/hw/mthca/mthca_main.c
+++ b/drivers/infiniband/hw/mthca/mthca_main.c
@@ -967,12 +967,12 @@ static struct {
967} mthca_hca_table[] = { 967} mthca_hca_table[] = {
968 [TAVOR] = { .latest_fw = MTHCA_FW_VER(3, 4, 0), 968 [TAVOR] = { .latest_fw = MTHCA_FW_VER(3, 4, 0),
969 .flags = 0 }, 969 .flags = 0 },
970 [ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 7, 400), 970 [ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 7, 600),
971 .flags = MTHCA_FLAG_PCIE }, 971 .flags = MTHCA_FLAG_PCIE },
972 [ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 1, 0), 972 [ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 1, 400),
973 .flags = MTHCA_FLAG_MEMFREE | 973 .flags = MTHCA_FLAG_MEMFREE |
974 MTHCA_FLAG_PCIE }, 974 MTHCA_FLAG_PCIE },
975 [SINAI] = { .latest_fw = MTHCA_FW_VER(1, 0, 800), 975 [SINAI] = { .latest_fw = MTHCA_FW_VER(1, 1, 0),
976 .flags = MTHCA_FLAG_MEMFREE | 976 .flags = MTHCA_FLAG_MEMFREE |
977 MTHCA_FLAG_PCIE | 977 MTHCA_FLAG_PCIE |
978 MTHCA_FLAG_SINAI_OPT } 978 MTHCA_FLAG_SINAI_OPT }
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 230ae21db8fd..265b1d1c4a62 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -1287,11 +1287,7 @@ int mthca_register_device(struct mthca_dev *dev)
1287 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | 1287 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1288 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | 1288 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1289 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) | 1289 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1290 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) | 1290 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1291 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1292 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1293 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1294 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1295 dev->ib_dev.node_type = IB_NODE_CA; 1291 dev->ib_dev.node_type = IB_NODE_CA;
1296 dev->ib_dev.phys_port_cnt = dev->limits.num_ports; 1292 dev->ib_dev.phys_port_cnt = dev->limits.num_ports;
1297 dev->ib_dev.dma_device = &dev->pdev->dev; 1293 dev->ib_dev.dma_device = &dev->pdev->dev;
@@ -1316,6 +1312,11 @@ int mthca_register_device(struct mthca_dev *dev)
1316 dev->ib_dev.modify_srq = mthca_modify_srq; 1312 dev->ib_dev.modify_srq = mthca_modify_srq;
1317 dev->ib_dev.query_srq = mthca_query_srq; 1313 dev->ib_dev.query_srq = mthca_query_srq;
1318 dev->ib_dev.destroy_srq = mthca_destroy_srq; 1314 dev->ib_dev.destroy_srq = mthca_destroy_srq;
1315 dev->ib_dev.uverbs_cmd_mask |=
1316 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1317 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1318 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1319 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1319 1320
1320 if (mthca_is_memfree(dev)) 1321 if (mthca_is_memfree(dev))
1321 dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv; 1322 dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.h b/drivers/infiniband/hw/mthca/mthca_provider.h
index 8de2887ba15c..9a5bece3fa5c 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.h
+++ b/drivers/infiniband/hw/mthca/mthca_provider.h
@@ -136,8 +136,8 @@ struct mthca_ah {
136 * We have one global lock that protects dev->cq/qp_table. Each 136 * We have one global lock that protects dev->cq/qp_table. Each
137 * struct mthca_cq/qp also has its own lock. An individual qp lock 137 * struct mthca_cq/qp also has its own lock. An individual qp lock
138 * may be taken inside of an individual cq lock. Both cqs attached to 138 * may be taken inside of an individual cq lock. Both cqs attached to
139 * a qp may be locked, with the send cq locked first. No other 139 * a qp may be locked, with the cq with the lower cqn locked first.
140 * nesting should be done. 140 * No other nesting should be done.
141 * 141 *
142 * Each struct mthca_cq/qp also has an ref count, protected by the 142 * Each struct mthca_cq/qp also has an ref count, protected by the
143 * corresponding table lock. The pointer from the cq/qp_table to the 143 * corresponding table lock. The pointer from the cq/qp_table to the
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index 16c387d8170c..2e8f6f36e0a5 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -99,6 +99,10 @@ enum {
99 MTHCA_QP_BIT_RSC = 1 << 3 99 MTHCA_QP_BIT_RSC = 1 << 3
100}; 100};
101 101
102enum {
103 MTHCA_SEND_DOORBELL_FENCE = 1 << 5
104};
105
102struct mthca_qp_path { 106struct mthca_qp_path {
103 __be32 port_pkey; 107 __be32 port_pkey;
104 u8 rnr_retry; 108 u8 rnr_retry;
@@ -222,9 +226,8 @@ static void *get_send_wqe(struct mthca_qp *qp, int n)
222 (PAGE_SIZE - 1)); 226 (PAGE_SIZE - 1));
223} 227}
224 228
225static void mthca_wq_init(struct mthca_wq *wq) 229static void mthca_wq_reset(struct mthca_wq *wq)
226{ 230{
227 spin_lock_init(&wq->lock);
228 wq->next_ind = 0; 231 wq->next_ind = 0;
229 wq->last_comp = wq->max - 1; 232 wq->last_comp = wq->max - 1;
230 wq->head = 0; 233 wq->head = 0;
@@ -845,10 +848,10 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask)
845 mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq), qp->qpn, 848 mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq), qp->qpn,
846 qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL); 849 qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
847 850
848 mthca_wq_init(&qp->sq); 851 mthca_wq_reset(&qp->sq);
849 qp->sq.last = get_send_wqe(qp, qp->sq.max - 1); 852 qp->sq.last = get_send_wqe(qp, qp->sq.max - 1);
850 853
851 mthca_wq_init(&qp->rq); 854 mthca_wq_reset(&qp->rq);
852 qp->rq.last = get_recv_wqe(qp, qp->rq.max - 1); 855 qp->rq.last = get_recv_wqe(qp, qp->rq.max - 1);
853 856
854 if (mthca_is_memfree(dev)) { 857 if (mthca_is_memfree(dev)) {
@@ -1112,8 +1115,11 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
1112 qp->atomic_rd_en = 0; 1115 qp->atomic_rd_en = 0;
1113 qp->resp_depth = 0; 1116 qp->resp_depth = 0;
1114 qp->sq_policy = send_policy; 1117 qp->sq_policy = send_policy;
1115 mthca_wq_init(&qp->sq); 1118 mthca_wq_reset(&qp->sq);
1116 mthca_wq_init(&qp->rq); 1119 mthca_wq_reset(&qp->rq);
1120
1121 spin_lock_init(&qp->sq.lock);
1122 spin_lock_init(&qp->rq.lock);
1117 1123
1118 ret = mthca_map_memfree(dev, qp); 1124 ret = mthca_map_memfree(dev, qp);
1119 if (ret) 1125 if (ret)
@@ -1257,6 +1263,32 @@ int mthca_alloc_qp(struct mthca_dev *dev,
1257 return 0; 1263 return 0;
1258} 1264}
1259 1265
1266static void mthca_lock_cqs(struct mthca_cq *send_cq, struct mthca_cq *recv_cq)
1267{
1268 if (send_cq == recv_cq)
1269 spin_lock_irq(&send_cq->lock);
1270 else if (send_cq->cqn < recv_cq->cqn) {
1271 spin_lock_irq(&send_cq->lock);
1272 spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
1273 } else {
1274 spin_lock_irq(&recv_cq->lock);
1275 spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
1276 }
1277}
1278
1279static void mthca_unlock_cqs(struct mthca_cq *send_cq, struct mthca_cq *recv_cq)
1280{
1281 if (send_cq == recv_cq)
1282 spin_unlock_irq(&send_cq->lock);
1283 else if (send_cq->cqn < recv_cq->cqn) {
1284 spin_unlock(&recv_cq->lock);
1285 spin_unlock_irq(&send_cq->lock);
1286 } else {
1287 spin_unlock(&send_cq->lock);
1288 spin_unlock_irq(&recv_cq->lock);
1289 }
1290}
1291
1260int mthca_alloc_sqp(struct mthca_dev *dev, 1292int mthca_alloc_sqp(struct mthca_dev *dev,
1261 struct mthca_pd *pd, 1293 struct mthca_pd *pd,
1262 struct mthca_cq *send_cq, 1294 struct mthca_cq *send_cq,
@@ -1309,17 +1341,13 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
1309 * Lock CQs here, so that CQ polling code can do QP lookup 1341 * Lock CQs here, so that CQ polling code can do QP lookup
1310 * without taking a lock. 1342 * without taking a lock.
1311 */ 1343 */
1312 spin_lock_irq(&send_cq->lock); 1344 mthca_lock_cqs(send_cq, recv_cq);
1313 if (send_cq != recv_cq)
1314 spin_lock(&recv_cq->lock);
1315 1345
1316 spin_lock(&dev->qp_table.lock); 1346 spin_lock(&dev->qp_table.lock);
1317 mthca_array_clear(&dev->qp_table.qp, mqpn); 1347 mthca_array_clear(&dev->qp_table.qp, mqpn);
1318 spin_unlock(&dev->qp_table.lock); 1348 spin_unlock(&dev->qp_table.lock);
1319 1349
1320 if (send_cq != recv_cq) 1350 mthca_unlock_cqs(send_cq, recv_cq);
1321 spin_unlock(&recv_cq->lock);
1322 spin_unlock_irq(&send_cq->lock);
1323 1351
1324 err_out: 1352 err_out:
1325 dma_free_coherent(&dev->pdev->dev, sqp->header_buf_size, 1353 dma_free_coherent(&dev->pdev->dev, sqp->header_buf_size,
@@ -1353,9 +1381,7 @@ void mthca_free_qp(struct mthca_dev *dev,
1353 * Lock CQs here, so that CQ polling code can do QP lookup 1381 * Lock CQs here, so that CQ polling code can do QP lookup
1354 * without taking a lock. 1382 * without taking a lock.
1355 */ 1383 */
1356 spin_lock_irq(&send_cq->lock); 1384 mthca_lock_cqs(send_cq, recv_cq);
1357 if (send_cq != recv_cq)
1358 spin_lock(&recv_cq->lock);
1359 1385
1360 spin_lock(&dev->qp_table.lock); 1386 spin_lock(&dev->qp_table.lock);
1361 mthca_array_clear(&dev->qp_table.qp, 1387 mthca_array_clear(&dev->qp_table.qp,
@@ -1363,9 +1389,7 @@ void mthca_free_qp(struct mthca_dev *dev,
1363 --qp->refcount; 1389 --qp->refcount;
1364 spin_unlock(&dev->qp_table.lock); 1390 spin_unlock(&dev->qp_table.lock);
1365 1391
1366 if (send_cq != recv_cq) 1392 mthca_unlock_cqs(send_cq, recv_cq);
1367 spin_unlock(&recv_cq->lock);
1368 spin_unlock_irq(&send_cq->lock);
1369 1393
1370 wait_event(qp->wait, !get_qp_refcount(dev, qp)); 1394 wait_event(qp->wait, !get_qp_refcount(dev, qp));
1371 1395
@@ -1500,7 +1524,7 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1500 int i; 1524 int i;
1501 int size; 1525 int size;
1502 int size0 = 0; 1526 int size0 = 0;
1503 u32 f0 = 0; 1527 u32 f0;
1504 int ind; 1528 int ind;
1505 u8 op0 = 0; 1529 u8 op0 = 0;
1506 1530
@@ -1684,6 +1708,8 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1684 if (!size0) { 1708 if (!size0) {
1685 size0 = size; 1709 size0 = size;
1686 op0 = mthca_opcode[wr->opcode]; 1710 op0 = mthca_opcode[wr->opcode];
1711 f0 = wr->send_flags & IB_SEND_FENCE ?
1712 MTHCA_SEND_DOORBELL_FENCE : 0;
1687 } 1713 }
1688 1714
1689 ++ind; 1715 ++ind;
@@ -1841,7 +1867,7 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
1841 int i; 1867 int i;
1842 int size; 1868 int size;
1843 int size0 = 0; 1869 int size0 = 0;
1844 u32 f0 = 0; 1870 u32 f0;
1845 int ind; 1871 int ind;
1846 u8 op0 = 0; 1872 u8 op0 = 0;
1847 1873
@@ -2049,6 +2075,8 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
2049 if (!size0) { 2075 if (!size0) {
2050 size0 = size; 2076 size0 = size;
2051 op0 = mthca_opcode[wr->opcode]; 2077 op0 = mthca_opcode[wr->opcode];
2078 f0 = wr->send_flags & IB_SEND_FENCE ?
2079 MTHCA_SEND_DOORBELL_FENCE : 0;
2052 } 2080 }
2053 2081
2054 ++ind; 2082 ++ind;
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index fab417c5cf43..b60a9d79ae54 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -370,7 +370,8 @@ int mthca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
370 return -EINVAL; 370 return -EINVAL;
371 371
372 if (attr_mask & IB_SRQ_LIMIT) { 372 if (attr_mask & IB_SRQ_LIMIT) {
373 if (attr->srq_limit > srq->max) 373 u32 max_wr = mthca_is_memfree(dev) ? srq->max - 1 : srq->max;
374 if (attr->srq_limit > max_wr)
374 return -EINVAL; 375 return -EINVAL;
375 376
376 mutex_lock(&srq->mutex); 377 mutex_lock(&srq->mutex);
diff --git a/drivers/infiniband/ulp/ipoib/Kconfig b/drivers/infiniband/ulp/ipoib/Kconfig
index 13d6d01c72c0..d74653d7de1c 100644
--- a/drivers/infiniband/ulp/ipoib/Kconfig
+++ b/drivers/infiniband/ulp/ipoib/Kconfig
@@ -6,8 +6,7 @@ config INFINIBAND_IPOIB
6 transports IP packets over InfiniBand so you can use your IB 6 transports IP packets over InfiniBand so you can use your IB
7 device as a fancy NIC. 7 device as a fancy NIC.
8 8
9 The IPoIB protocol is defined by the IETF ipoib working 9 See Documentation/infiniband/ipoib.txt for more information
10 group: <http://www.ietf.org/html.charters/ipoib-charter.html>.
11 10
12config INFINIBAND_IPOIB_DEBUG 11config INFINIBAND_IPOIB_DEBUG
13 bool "IP-over-InfiniBand debugging" if EMBEDDED 12 bool "IP-over-InfiniBand debugging" if EMBEDDED
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 3f89f5e19036..474aa214ab57 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -212,6 +212,7 @@ struct ipoib_path {
212 212
213struct ipoib_neigh { 213struct ipoib_neigh {
214 struct ipoib_ah *ah; 214 struct ipoib_ah *ah;
215 union ib_gid dgid;
215 struct sk_buff_head queue; 216 struct sk_buff_head queue;
216 217
217 struct neighbour *neighbour; 218 struct neighbour *neighbour;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 1c6ea1c682a5..cf71d2a5515c 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -404,6 +404,8 @@ static void path_rec_completion(int status,
404 list_for_each_entry(neigh, &path->neigh_list, list) { 404 list_for_each_entry(neigh, &path->neigh_list, list) {
405 kref_get(&path->ah->ref); 405 kref_get(&path->ah->ref);
406 neigh->ah = path->ah; 406 neigh->ah = path->ah;
407 memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
408 sizeof(union ib_gid));
407 409
408 while ((skb = __skb_dequeue(&neigh->queue))) 410 while ((skb = __skb_dequeue(&neigh->queue)))
409 __skb_queue_tail(&skqueue, skb); 411 __skb_queue_tail(&skqueue, skb);
@@ -510,6 +512,8 @@ static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
510 if (path->ah) { 512 if (path->ah) {
511 kref_get(&path->ah->ref); 513 kref_get(&path->ah->ref);
512 neigh->ah = path->ah; 514 neigh->ah = path->ah;
515 memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
516 sizeof(union ib_gid));
513 517
514 ipoib_send(dev, skb, path->ah, 518 ipoib_send(dev, skb, path->ah,
515 be32_to_cpup((__be32 *) skb->dst->neighbour->ha)); 519 be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
@@ -633,6 +637,25 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
633 neigh = *to_ipoib_neigh(skb->dst->neighbour); 637 neigh = *to_ipoib_neigh(skb->dst->neighbour);
634 638
635 if (likely(neigh->ah)) { 639 if (likely(neigh->ah)) {
640 if (unlikely(memcmp(&neigh->dgid.raw,
641 skb->dst->neighbour->ha + 4,
642 sizeof(union ib_gid)))) {
643 spin_lock(&priv->lock);
644 /*
645 * It's safe to call ipoib_put_ah() inside
646 * priv->lock here, because we know that
647 * path->ah will always hold one more reference,
648 * so ipoib_put_ah() will never do more than
649 * decrement the ref count.
650 */
651 ipoib_put_ah(neigh->ah);
652 list_del(&neigh->list);
653 ipoib_neigh_free(neigh);
654 spin_unlock(&priv->lock);
655 ipoib_path_lookup(skb, dev);
656 goto out;
657 }
658
636 ipoib_send(dev, skb, neigh->ah, 659 ipoib_send(dev, skb, neigh->ah,
637 be32_to_cpup((__be32 *) skb->dst->neighbour->ha)); 660 be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
638 goto out; 661 goto out;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index ab40488182b3..ec356ce7cdcd 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -264,6 +264,10 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
264 if (!ah) { 264 if (!ah) {
265 ipoib_warn(priv, "ib_address_create failed\n"); 265 ipoib_warn(priv, "ib_address_create failed\n");
266 } else { 266 } else {
267 spin_lock_irq(&priv->lock);
268 mcast->ah = ah;
269 spin_unlock_irq(&priv->lock);
270
267 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT 271 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
268 " AV %p, LID 0x%04x, SL %d\n", 272 " AV %p, LID 0x%04x, SL %d\n",
269 IPOIB_GID_ARG(mcast->mcmember.mgid), 273 IPOIB_GID_ARG(mcast->mcmember.mgid),
@@ -271,10 +275,6 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
271 be16_to_cpu(mcast->mcmember.mlid), 275 be16_to_cpu(mcast->mcmember.mlid),
272 mcast->mcmember.sl); 276 mcast->mcmember.sl);
273 } 277 }
274
275 spin_lock_irq(&priv->lock);
276 mcast->ah = ah;
277 spin_unlock_irq(&priv->lock);
278 } 278 }
279 279
280 /* actually send any queued packets */ 280 /* actually send any queued packets */
@@ -326,6 +326,7 @@ ipoib_mcast_sendonly_join_complete(int status,
326 326
327 /* Clear the busy flag so we try again */ 327 /* Clear the busy flag so we try again */
328 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); 328 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
329 mcast->query = NULL;
329 } 330 }
330 331
331 complete(&mcast->done); 332 complete(&mcast->done);
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c
index b2c033edb03c..1437d7ee3b19 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.c
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.c
@@ -378,21 +378,6 @@ iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
378 return iser_conn_set_full_featured_mode(conn); 378 return iser_conn_set_full_featured_mode(conn);
379} 379}
380 380
381static void
382iscsi_iser_conn_terminate(struct iscsi_conn *conn)
383{
384 struct iscsi_iser_conn *iser_conn = conn->dd_data;
385 struct iser_conn *ib_conn = iser_conn->ib_conn;
386
387 BUG_ON(!ib_conn);
388 /* starts conn teardown process, waits until all previously *
389 * posted buffers get flushed, deallocates all conn resources */
390 iser_conn_terminate(ib_conn);
391 iser_conn->ib_conn = NULL;
392 conn->recv_lock = NULL;
393}
394
395
396static struct iscsi_transport iscsi_iser_transport; 381static struct iscsi_transport iscsi_iser_transport;
397 382
398static struct iscsi_cls_session * 383static struct iscsi_cls_session *
@@ -437,159 +422,50 @@ iscsi_iser_session_create(struct iscsi_transport *iscsit,
437} 422}
438 423
439static int 424static int
440iscsi_iser_conn_set_param(struct iscsi_cls_conn *cls_conn, 425iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
441 enum iscsi_param param, uint32_t value) 426 enum iscsi_param param, char *buf, int buflen)
442{ 427{
443 struct iscsi_conn *conn = cls_conn->dd_data; 428 int value;
444 struct iscsi_session *session = conn->session;
445
446 spin_lock_bh(&session->lock);
447 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
448 conn->stop_stage != STOP_CONN_RECOVER) {
449 printk(KERN_ERR "iscsi_iser: can not change parameter [%d]\n",
450 param);
451 spin_unlock_bh(&session->lock);
452 return 0;
453 }
454 spin_unlock_bh(&session->lock);
455 429
456 switch (param) { 430 switch (param) {
457 case ISCSI_PARAM_MAX_RECV_DLENGTH: 431 case ISCSI_PARAM_MAX_RECV_DLENGTH:
458 /* TBD */ 432 /* TBD */
459 break; 433 break;
460 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
461 conn->max_xmit_dlength = value;
462 break;
463 case ISCSI_PARAM_HDRDGST_EN: 434 case ISCSI_PARAM_HDRDGST_EN:
435 sscanf(buf, "%d", &value);
464 if (value) { 436 if (value) {
465 printk(KERN_ERR "DataDigest wasn't negotiated to None"); 437 printk(KERN_ERR "DataDigest wasn't negotiated to None");
466 return -EPROTO; 438 return -EPROTO;
467 } 439 }
468 break; 440 break;
469 case ISCSI_PARAM_DATADGST_EN: 441 case ISCSI_PARAM_DATADGST_EN:
442 sscanf(buf, "%d", &value);
470 if (value) { 443 if (value) {
471 printk(KERN_ERR "DataDigest wasn't negotiated to None"); 444 printk(KERN_ERR "DataDigest wasn't negotiated to None");
472 return -EPROTO; 445 return -EPROTO;
473 } 446 }
474 break; 447 break;
475 case ISCSI_PARAM_INITIAL_R2T_EN:
476 session->initial_r2t_en = value;
477 break;
478 case ISCSI_PARAM_IMM_DATA_EN:
479 session->imm_data_en = value;
480 break;
481 case ISCSI_PARAM_FIRST_BURST:
482 session->first_burst = value;
483 break;
484 case ISCSI_PARAM_MAX_BURST:
485 session->max_burst = value;
486 break;
487 case ISCSI_PARAM_PDU_INORDER_EN:
488 session->pdu_inorder_en = value;
489 break;
490 case ISCSI_PARAM_DATASEQ_INORDER_EN:
491 session->dataseq_inorder_en = value;
492 break;
493 case ISCSI_PARAM_ERL:
494 session->erl = value;
495 break;
496 case ISCSI_PARAM_IFMARKER_EN: 448 case ISCSI_PARAM_IFMARKER_EN:
449 sscanf(buf, "%d", &value);
497 if (value) { 450 if (value) {
498 printk(KERN_ERR "IFMarker wasn't negotiated to No"); 451 printk(KERN_ERR "IFMarker wasn't negotiated to No");
499 return -EPROTO; 452 return -EPROTO;
500 } 453 }
501 break; 454 break;
502 case ISCSI_PARAM_OFMARKER_EN: 455 case ISCSI_PARAM_OFMARKER_EN:
456 sscanf(buf, "%d", &value);
503 if (value) { 457 if (value) {
504 printk(KERN_ERR "OFMarker wasn't negotiated to No"); 458 printk(KERN_ERR "OFMarker wasn't negotiated to No");
505 return -EPROTO; 459 return -EPROTO;
506 } 460 }
507 break; 461 break;
508 default: 462 default:
509 break; 463 return iscsi_set_param(cls_conn, param, buf, buflen);
510 } 464 }
511 465
512 return 0; 466 return 0;
513} 467}
514 468
515static int
516iscsi_iser_session_get_param(struct iscsi_cls_session *cls_session,
517 enum iscsi_param param, uint32_t *value)
518{
519 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
520 struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
521
522 switch (param) {
523 case ISCSI_PARAM_INITIAL_R2T_EN:
524 *value = session->initial_r2t_en;
525 break;
526 case ISCSI_PARAM_MAX_R2T:
527 *value = session->max_r2t;
528 break;
529 case ISCSI_PARAM_IMM_DATA_EN:
530 *value = session->imm_data_en;
531 break;
532 case ISCSI_PARAM_FIRST_BURST:
533 *value = session->first_burst;
534 break;
535 case ISCSI_PARAM_MAX_BURST:
536 *value = session->max_burst;
537 break;
538 case ISCSI_PARAM_PDU_INORDER_EN:
539 *value = session->pdu_inorder_en;
540 break;
541 case ISCSI_PARAM_DATASEQ_INORDER_EN:
542 *value = session->dataseq_inorder_en;
543 break;
544 case ISCSI_PARAM_ERL:
545 *value = session->erl;
546 break;
547 case ISCSI_PARAM_IFMARKER_EN:
548 *value = 0;
549 break;
550 case ISCSI_PARAM_OFMARKER_EN:
551 *value = 0;
552 break;
553 default:
554 return ISCSI_ERR_PARAM_NOT_FOUND;
555 }
556
557 return 0;
558}
559
560static int
561iscsi_iser_conn_get_param(struct iscsi_cls_conn *cls_conn,
562 enum iscsi_param param, uint32_t *value)
563{
564 struct iscsi_conn *conn = cls_conn->dd_data;
565
566 switch(param) {
567 case ISCSI_PARAM_MAX_RECV_DLENGTH:
568 *value = conn->max_recv_dlength;
569 break;
570 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
571 *value = conn->max_xmit_dlength;
572 break;
573 case ISCSI_PARAM_HDRDGST_EN:
574 *value = 0;
575 break;
576 case ISCSI_PARAM_DATADGST_EN:
577 *value = 0;
578 break;
579 /*case ISCSI_PARAM_TARGET_RECV_DLENGTH:
580 *value = conn->target_recv_dlength;
581 break;
582 case ISCSI_PARAM_INITIATOR_RECV_DLENGTH:
583 *value = conn->initiator_recv_dlength;
584 break;*/
585 default:
586 return ISCSI_ERR_PARAM_NOT_FOUND;
587 }
588
589 return 0;
590}
591
592
593static void 469static void
594iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats) 470iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
595{ 471{
@@ -664,13 +540,13 @@ iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms)
664static void 540static void
665iscsi_iser_ep_disconnect(__u64 ep_handle) 541iscsi_iser_ep_disconnect(__u64 ep_handle)
666{ 542{
667 struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle); 543 struct iser_conn *ib_conn;
668 544
545 ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
669 if (!ib_conn) 546 if (!ib_conn)
670 return; 547 return;
671 548
672 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state); 549 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
673
674 iser_conn_terminate(ib_conn); 550 iser_conn_terminate(ib_conn);
675} 551}
676 552
@@ -701,7 +577,12 @@ static struct iscsi_transport iscsi_iser_transport = {
701 ISCSI_FIRST_BURST | 577 ISCSI_FIRST_BURST |
702 ISCSI_MAX_BURST | 578 ISCSI_MAX_BURST |
703 ISCSI_PDU_INORDER_EN | 579 ISCSI_PDU_INORDER_EN |
704 ISCSI_DATASEQ_INORDER_EN, 580 ISCSI_DATASEQ_INORDER_EN |
581 ISCSI_EXP_STATSN |
582 ISCSI_PERSISTENT_PORT |
583 ISCSI_PERSISTENT_ADDRESS |
584 ISCSI_TARGET_NAME |
585 ISCSI_TPGT,
705 .host_template = &iscsi_iser_sht, 586 .host_template = &iscsi_iser_sht,
706 .conndata_size = sizeof(struct iscsi_conn), 587 .conndata_size = sizeof(struct iscsi_conn),
707 .max_lun = ISCSI_ISER_MAX_LUN, 588 .max_lun = ISCSI_ISER_MAX_LUN,
@@ -713,14 +594,11 @@ static struct iscsi_transport iscsi_iser_transport = {
713 .create_conn = iscsi_iser_conn_create, 594 .create_conn = iscsi_iser_conn_create,
714 .bind_conn = iscsi_iser_conn_bind, 595 .bind_conn = iscsi_iser_conn_bind,
715 .destroy_conn = iscsi_iser_conn_destroy, 596 .destroy_conn = iscsi_iser_conn_destroy,
716 .set_param = iscsi_iser_conn_set_param, 597 .set_param = iscsi_iser_set_param,
717 .get_conn_param = iscsi_iser_conn_get_param, 598 .get_conn_param = iscsi_conn_get_param,
718 .get_session_param = iscsi_iser_session_get_param, 599 .get_session_param = iscsi_session_get_param,
719 .start_conn = iscsi_iser_conn_start, 600 .start_conn = iscsi_iser_conn_start,
720 .stop_conn = iscsi_conn_stop, 601 .stop_conn = iscsi_conn_stop,
721 /* these are called as part of conn recovery */
722 .suspend_conn_recv = NULL, /* FIXME is/how this relvant to iser? */
723 .terminate_conn = iscsi_iser_conn_terminate,
724 /* IO */ 602 /* IO */
725 .send_pdu = iscsi_conn_send_pdu, 603 .send_pdu = iscsi_conn_send_pdu,
726 .get_stats = iscsi_iser_conn_get_stats, 604 .get_stats = iscsi_iser_conn_get_stats,
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index ff117bbf81b4..72febf1f8ff8 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -594,7 +594,7 @@ int iser_reg_page_vec(struct iser_conn *ib_conn,
594 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool, 594 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
595 page_list, 595 page_list,
596 page_vec->length, 596 page_vec->length,
597 &io_addr); 597 io_addr);
598 598
599 if (IS_ERR(mem)) { 599 if (IS_ERR(mem)) {
600 status = (int)PTR_ERR(mem); 600 status = (int)PTR_ERR(mem);
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 4e22afef7206..fd8344cdc0db 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -77,6 +77,14 @@ MODULE_PARM_DESC(topspin_workarounds,
77 77
78static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad }; 78static const u8 topspin_oui[3] = { 0x00, 0x05, 0xad };
79 79
80static int mellanox_workarounds = 1;
81
82module_param(mellanox_workarounds, int, 0444);
83MODULE_PARM_DESC(mellanox_workarounds,
84 "Enable workarounds for Mellanox SRP target bugs if != 0");
85
86static const u8 mellanox_oui[3] = { 0x00, 0x02, 0xc9 };
87
80static void srp_add_one(struct ib_device *device); 88static void srp_add_one(struct ib_device *device);
81static void srp_remove_one(struct ib_device *device); 89static void srp_remove_one(struct ib_device *device);
82static void srp_completion(struct ib_cq *cq, void *target_ptr); 90static void srp_completion(struct ib_cq *cq, void *target_ptr);
@@ -526,8 +534,10 @@ static int srp_reconnect_target(struct srp_target_port *target)
526 while (ib_poll_cq(target->cq, 1, &wc) > 0) 534 while (ib_poll_cq(target->cq, 1, &wc) > 0)
527 ; /* nothing */ 535 ; /* nothing */
528 536
537 spin_lock_irq(target->scsi_host->host_lock);
529 list_for_each_entry_safe(req, tmp, &target->req_queue, list) 538 list_for_each_entry_safe(req, tmp, &target->req_queue, list)
530 srp_reset_req(target, req); 539 srp_reset_req(target, req);
540 spin_unlock_irq(target->scsi_host->host_lock);
531 541
532 target->rx_head = 0; 542 target->rx_head = 0;
533 target->tx_head = 0; 543 target->tx_head = 0;
@@ -567,7 +577,7 @@ err:
567 return ret; 577 return ret;
568} 578}
569 579
570static int srp_map_fmr(struct srp_device *dev, struct scatterlist *scat, 580static int srp_map_fmr(struct srp_target_port *target, struct scatterlist *scat,
571 int sg_cnt, struct srp_request *req, 581 int sg_cnt, struct srp_request *req,
572 struct srp_direct_buf *buf) 582 struct srp_direct_buf *buf)
573{ 583{
@@ -577,10 +587,15 @@ static int srp_map_fmr(struct srp_device *dev, struct scatterlist *scat,
577 int page_cnt; 587 int page_cnt;
578 int i, j; 588 int i, j;
579 int ret; 589 int ret;
590 struct srp_device *dev = target->srp_host->dev;
580 591
581 if (!dev->fmr_pool) 592 if (!dev->fmr_pool)
582 return -ENODEV; 593 return -ENODEV;
583 594
595 if ((sg_dma_address(&scat[0]) & ~dev->fmr_page_mask) &&
596 mellanox_workarounds && !memcmp(&target->ioc_guid, mellanox_oui, 3))
597 return -EINVAL;
598
584 len = page_cnt = 0; 599 len = page_cnt = 0;
585 for (i = 0; i < sg_cnt; ++i) { 600 for (i = 0; i < sg_cnt; ++i) {
586 if (sg_dma_address(&scat[i]) & ~dev->fmr_page_mask) { 601 if (sg_dma_address(&scat[i]) & ~dev->fmr_page_mask) {
@@ -615,9 +630,10 @@ static int srp_map_fmr(struct srp_device *dev, struct scatterlist *scat,
615 (sg_dma_address(&scat[i]) & dev->fmr_page_mask) + j; 630 (sg_dma_address(&scat[i]) & dev->fmr_page_mask) + j;
616 631
617 req->fmr = ib_fmr_pool_map_phys(dev->fmr_pool, 632 req->fmr = ib_fmr_pool_map_phys(dev->fmr_pool,
618 dma_pages, page_cnt, &io_addr); 633 dma_pages, page_cnt, io_addr);
619 if (IS_ERR(req->fmr)) { 634 if (IS_ERR(req->fmr)) {
620 ret = PTR_ERR(req->fmr); 635 ret = PTR_ERR(req->fmr);
636 req->fmr = NULL;
621 goto out; 637 goto out;
622 } 638 }
623 639
@@ -682,7 +698,7 @@ static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
682 buf->va = cpu_to_be64(sg_dma_address(scat)); 698 buf->va = cpu_to_be64(sg_dma_address(scat));
683 buf->key = cpu_to_be32(target->srp_host->dev->mr->rkey); 699 buf->key = cpu_to_be32(target->srp_host->dev->mr->rkey);
684 buf->len = cpu_to_be32(sg_dma_len(scat)); 700 buf->len = cpu_to_be32(sg_dma_len(scat));
685 } else if (srp_map_fmr(target->srp_host->dev, scat, count, req, 701 } else if (srp_map_fmr(target, scat, count, req,
686 (void *) cmd->add_data)) { 702 (void *) cmd->add_data)) {
687 /* 703 /*
688 * FMR mapping failed, and the scatterlist has more 704 * FMR mapping failed, and the scatterlist has more
@@ -783,13 +799,6 @@ static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
783 spin_unlock_irqrestore(target->scsi_host->host_lock, flags); 799 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
784} 800}
785 801
786static void srp_reconnect_work(void *target_ptr)
787{
788 struct srp_target_port *target = target_ptr;
789
790 srp_reconnect_target(target);
791}
792
793static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc) 802static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc)
794{ 803{
795 struct srp_iu *iu; 804 struct srp_iu *iu;
@@ -842,7 +851,6 @@ static void srp_completion(struct ib_cq *cq, void *target_ptr)
842{ 851{
843 struct srp_target_port *target = target_ptr; 852 struct srp_target_port *target = target_ptr;
844 struct ib_wc wc; 853 struct ib_wc wc;
845 unsigned long flags;
846 854
847 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); 855 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
848 while (ib_poll_cq(cq, 1, &wc) > 0) { 856 while (ib_poll_cq(cq, 1, &wc) > 0) {
@@ -850,10 +858,6 @@ static void srp_completion(struct ib_cq *cq, void *target_ptr)
850 printk(KERN_ERR PFX "failed %s status %d\n", 858 printk(KERN_ERR PFX "failed %s status %d\n",
851 wc.wr_id & SRP_OP_RECV ? "receive" : "send", 859 wc.wr_id & SRP_OP_RECV ? "receive" : "send",
852 wc.status); 860 wc.status);
853 spin_lock_irqsave(target->scsi_host->host_lock, flags);
854 if (target->state == SRP_TARGET_LIVE)
855 schedule_work(&target->work);
856 spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
857 break; 861 break;
858 } 862 }
859 863
@@ -1689,8 +1693,6 @@ static ssize_t srp_create_target(struct class_device *class_dev,
1689 target->scsi_host = target_host; 1693 target->scsi_host = target_host;
1690 target->srp_host = host; 1694 target->srp_host = host;
1691 1695
1692 INIT_WORK(&target->work, srp_reconnect_work, target);
1693
1694 INIT_LIST_HEAD(&target->free_reqs); 1696 INIT_LIST_HEAD(&target->free_reqs);
1695 INIT_LIST_HEAD(&target->req_queue); 1697 INIT_LIST_HEAD(&target->req_queue);
1696 for (i = 0; i < SRP_SQ_SIZE; ++i) { 1698 for (i = 0; i < SRP_SQ_SIZE; ++i) {
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-18 23:02:26 -0500