aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2014-08-14 13:09:05 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2014-08-14 13:09:05 -0400
commite3b1fd56f175526db42ae94c457f29c2fa810aca (patch)
tree3e2948ca44fb7fd5348244c2a83eef864b3110b4
parent0680eb1f485ba5aac2ee02c9f0622239c9a4b16c (diff)
parentd087f6ad724dfbcdc3df8e0191b80d9d8d839e71 (diff)
Merge tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband
Pull infiniband/rdma updates from Roland Dreier: "Main set of InfiniBand/RDMA updates for 3.17 merge window: - MR reregistration support - MAD support for RMPP in userspace - iSER and SRP initiator updates - ocrdma hardware driver updates - other fixes..." * tag 'rdma-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (52 commits) IB/srp: Fix return value check in srp_init_module() RDMA/ocrdma: report asic-id in query device RDMA/ocrdma: Update sli data structure for endianness RDMA/ocrdma: Obtain SL from device structure RDMA/uapi: Include socket.h in rdma_user_cm.h IB/srpt: Handle GID change events IB/mlx5: Use ARRAY_SIZE instead of sizeof/sizeof[0] IB/mlx4: Use ARRAY_SIZE instead of sizeof/sizeof[0] RDMA/amso1100: Check for integer overflow in c2_alloc_cq_buf() IPoIB: Remove unnecessary test for NULL before debugfs_remove() IB/mad: Add user space RMPP support IB/mad: add new ioctl to ABI to support new registration options IB/mad: Add dev_notice messages for various umad/mad registration failures IB/mad: Update module to [pr|dev]_* style print messages IB/ipoib: Avoid multicast join attempts with invalid P_key IB/umad: Update module to [pr|dev]_* style print messages IB/ipoib: Avoid flushing the workqueue from worker context IB/ipoib: Use P_Key change event instead of P_Key polling mechanism IB/ipath: Add P_Key change event support mlx4_core: Add support for secure-host and SMP firewall ...
Diffstat
-rw-r--r--Documentation/infiniband/user_mad.txt13
-rw-r--r--drivers/infiniband/core/agent.c16
-rw-r--r--drivers/infiniband/core/cm.c5
-rw-r--r--drivers/infiniband/core/iwcm.c27
-rw-r--r--drivers/infiniband/core/mad.c283
-rw-r--r--drivers/infiniband/core/mad_priv.h3
-rw-r--r--drivers/infiniband/core/sa_query.c2
-rw-r--r--drivers/infiniband/core/user_mad.c188
-rw-r--r--drivers/infiniband/core/uverbs.h1
-rw-r--r--drivers/infiniband/core/uverbs_cmd.c93
-rw-r--r--drivers/infiniband/core/uverbs_main.c1
-rw-r--r--drivers/infiniband/hw/amso1100/c2_cq.c7
-rw-r--r--drivers/infiniband/hw/cxgb4/ev.c1
-rw-r--r--drivers/infiniband/hw/cxgb4/qp.c37
-rw-r--r--drivers/infiniband/hw/cxgb4/t4.h11
-rw-r--r--drivers/infiniband/hw/ipath/ipath_mad.c14
-rw-r--r--drivers/infiniband/hw/mlx4/mad.c2
-rw-r--r--drivers/infiniband/hw/mlx4/main.c8
-rw-r--r--drivers/infiniband/hw/mlx4/mlx4_ib.h4
-rw-r--r--drivers/infiniband/hw/mlx4/mr.c88
-rw-r--r--drivers/infiniband/hw/mlx5/qp.c2
-rw-r--r--drivers/infiniband/hw/mthca/mthca_mad.c2
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma.h26
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_ah.c6
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_hw.c227
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_hw.h2
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_main.c83
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_sli.h295
-rw-r--r--drivers/infiniband/hw/ocrdma/ocrdma_verbs.c36
-rw-r--r--drivers/infiniband/hw/qib/qib_mad.c2
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib.h8
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_fs.c6
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_ib.c133
-rw-r--r--drivers/infiniband/ulp/ipoib/ipoib_main.c9
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.c47
-rw-r--r--drivers/infiniband/ulp/iser/iscsi_iser.h8
-rw-r--r--drivers/infiniband/ulp/iser/iser_verbs.c128
-rw-r--r--drivers/infiniband/ulp/srp/ib_srp.c48
-rw-r--r--drivers/infiniband/ulp/srpt/ib_srpt.c3
-rw-r--r--drivers/net/ethernet/emulex/benet/be.h1
-rw-r--r--drivers/net/ethernet/emulex/benet/be_main.c1
-rw-r--r--drivers/net/ethernet/emulex/benet/be_roce.c18
-rw-r--r--drivers/net/ethernet/emulex/benet/be_roce.h3
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/cmd.c9
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/fw.c91
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/main.c5
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/mlx4.h3
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/mr.c160
-rw-r--r--drivers/net/ethernet/mellanox/mlx4/resource_tracker.c26
-rw-r--r--drivers/scsi/scsi_transport_srp.c3
-rw-r--r--include/linux/mlx4/cmd.h7
-rw-r--r--include/linux/mlx4/device.h17
-rw-r--r--include/rdma/ib_mad.h18
-rw-r--r--include/rdma/ib_verbs.h10
-rw-r--r--include/uapi/rdma/ib_user_mad.h42
-rw-r--r--include/uapi/rdma/ib_user_verbs.h16
-rw-r--r--include/uapi/rdma/rdma_user_cm.h1
57 files changed, 1851 insertions, 455 deletions
diff --git a/Documentation/infiniband/user_mad.txt b/Documentation/infiniband/user_mad.txt
index 8a366959f5cc..7aca13a54a3a 100644
--- a/Documentation/infiniband/user_mad.txt
+++ b/Documentation/infiniband/user_mad.txt
@@ -26,6 +26,11 @@ Creating MAD agents
26 ioctl. Also, all agents registered through a file descriptor will 26 ioctl. Also, all agents registered through a file descriptor will
27 be unregistered when the descriptor is closed. 27 be unregistered when the descriptor is closed.
28 28
29 2014 -- a new registration ioctl is now provided which allows additional
30 fields to be provided during registration.
31 Users of this registration call are implicitly setting the use of
32 pkey_index (see below).
33
29Receiving MADs 34Receiving MADs
30 35
31 MADs are received using read(). The receive side now supports 36 MADs are received using read(). The receive side now supports
@@ -104,10 +109,10 @@ P_Key Index Handling
104 The old ib_umad interface did not allow setting the P_Key index for 109 The old ib_umad interface did not allow setting the P_Key index for
105 MADs that are sent and did not provide a way for obtaining the P_Key 110 MADs that are sent and did not provide a way for obtaining the P_Key
106 index of received MADs. A new layout for struct ib_user_mad_hdr 111 index of received MADs. A new layout for struct ib_user_mad_hdr
107 with a pkey_index member has been defined; however, to preserve 112 with a pkey_index member has been defined; however, to preserve binary
108 binary compatibility with older applications, this new layout will 113 compatibility with older applications, this new layout will not be used
109 not be used unless the IB_USER_MAD_ENABLE_PKEY ioctl is called 114 unless one of IB_USER_MAD_ENABLE_PKEY or IB_USER_MAD_REGISTER_AGENT2 ioctl's
110 before a file descriptor is used for anything else. 115 are called before a file descriptor is used for anything else.
111 116
112 In September 2008, the IB_USER_MAD_ABI_VERSION will be incremented 117 In September 2008, the IB_USER_MAD_ABI_VERSION will be incremented
113 to 6, the new layout of struct ib_user_mad_hdr will be used by 118 to 6, the new layout of struct ib_user_mad_hdr will be used by
diff --git a/drivers/infiniband/core/agent.c b/drivers/infiniband/core/agent.c
index 2bc7f5af64f4..f6d29614cb01 100644
--- a/drivers/infiniband/core/agent.c
+++ b/drivers/infiniband/core/agent.c
@@ -94,14 +94,14 @@ void agent_send_response(struct ib_mad *mad, struct ib_grh *grh,
94 port_priv = ib_get_agent_port(device, port_num); 94 port_priv = ib_get_agent_port(device, port_num);
95 95
96 if (!port_priv) { 96 if (!port_priv) {
97 printk(KERN_ERR SPFX "Unable to find port agent\n"); 97 dev_err(&device->dev, "Unable to find port agent\n");
98 return; 98 return;
99 } 99 }
100 100
101 agent = port_priv->agent[qpn]; 101 agent = port_priv->agent[qpn];
102 ah = ib_create_ah_from_wc(agent->qp->pd, wc, grh, port_num); 102 ah = ib_create_ah_from_wc(agent->qp->pd, wc, grh, port_num);
103 if (IS_ERR(ah)) { 103 if (IS_ERR(ah)) {
104 printk(KERN_ERR SPFX "ib_create_ah_from_wc error %ld\n", 104 dev_err(&device->dev, "ib_create_ah_from_wc error %ld\n",
105 PTR_ERR(ah)); 105 PTR_ERR(ah));
106 return; 106 return;
107 } 107 }
@@ -110,7 +110,7 @@ void agent_send_response(struct ib_mad *mad, struct ib_grh *grh,
110 IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA, 110 IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
111 GFP_KERNEL); 111 GFP_KERNEL);
112 if (IS_ERR(send_buf)) { 112 if (IS_ERR(send_buf)) {
113 printk(KERN_ERR SPFX "ib_create_send_mad error\n"); 113 dev_err(&device->dev, "ib_create_send_mad error\n");
114 goto err1; 114 goto err1;
115 } 115 }
116 116
@@ -125,7 +125,7 @@ void agent_send_response(struct ib_mad *mad, struct ib_grh *grh,
125 } 125 }
126 126
127 if (ib_post_send_mad(send_buf, NULL)) { 127 if (ib_post_send_mad(send_buf, NULL)) {
128 printk(KERN_ERR SPFX "ib_post_send_mad error\n"); 128 dev_err(&device->dev, "ib_post_send_mad error\n");
129 goto err2; 129 goto err2;
130 } 130 }
131 return; 131 return;
@@ -151,7 +151,7 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
151 /* Create new device info */ 151 /* Create new device info */
152 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL); 152 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL);
153 if (!port_priv) { 153 if (!port_priv) {
154 printk(KERN_ERR SPFX "No memory for ib_agent_port_private\n"); 154 dev_err(&device->dev, "No memory for ib_agent_port_private\n");
155 ret = -ENOMEM; 155 ret = -ENOMEM;
156 goto error1; 156 goto error1;
157 } 157 }
@@ -161,7 +161,7 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
161 port_priv->agent[0] = ib_register_mad_agent(device, port_num, 161 port_priv->agent[0] = ib_register_mad_agent(device, port_num,
162 IB_QPT_SMI, NULL, 0, 162 IB_QPT_SMI, NULL, 0,
163 &agent_send_handler, 163 &agent_send_handler,
164 NULL, NULL); 164 NULL, NULL, 0);
165 if (IS_ERR(port_priv->agent[0])) { 165 if (IS_ERR(port_priv->agent[0])) {
166 ret = PTR_ERR(port_priv->agent[0]); 166 ret = PTR_ERR(port_priv->agent[0]);
167 goto error2; 167 goto error2;
@@ -172,7 +172,7 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
172 port_priv->agent[1] = ib_register_mad_agent(device, port_num, 172 port_priv->agent[1] = ib_register_mad_agent(device, port_num,
173 IB_QPT_GSI, NULL, 0, 173 IB_QPT_GSI, NULL, 0,
174 &agent_send_handler, 174 &agent_send_handler,
175 NULL, NULL); 175 NULL, NULL, 0);
176 if (IS_ERR(port_priv->agent[1])) { 176 if (IS_ERR(port_priv->agent[1])) {
177 ret = PTR_ERR(port_priv->agent[1]); 177 ret = PTR_ERR(port_priv->agent[1]);
178 goto error3; 178 goto error3;
@@ -202,7 +202,7 @@ int ib_agent_port_close(struct ib_device *device, int port_num)
202 port_priv = __ib_get_agent_port(device, port_num); 202 port_priv = __ib_get_agent_port(device, port_num);
203 if (port_priv == NULL) { 203 if (port_priv == NULL) {
204 spin_unlock_irqrestore(&ib_agent_port_list_lock, flags); 204 spin_unlock_irqrestore(&ib_agent_port_list_lock, flags);
205 printk(KERN_ERR SPFX "Port %d not found\n", port_num); 205 dev_err(&device->dev, "Port %d not found\n", port_num);
206 return -ENODEV; 206 return -ENODEV;
207 } 207 }
208 list_del(&port_priv->port_list); 208 list_del(&port_priv->port_list);
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
index c3239170d8b7..e28a494e2a3a 100644
--- a/drivers/infiniband/core/cm.c
+++ b/drivers/infiniband/core/cm.c
@@ -3753,7 +3753,7 @@ static void cm_add_one(struct ib_device *ib_device)
3753 struct cm_port *port; 3753 struct cm_port *port;
3754 struct ib_mad_reg_req reg_req = { 3754 struct ib_mad_reg_req reg_req = {
3755 .mgmt_class = IB_MGMT_CLASS_CM, 3755 .mgmt_class = IB_MGMT_CLASS_CM,
3756 .mgmt_class_version = IB_CM_CLASS_VERSION 3756 .mgmt_class_version = IB_CM_CLASS_VERSION,
3757 }; 3757 };
3758 struct ib_port_modify port_modify = { 3758 struct ib_port_modify port_modify = {
3759 .set_port_cap_mask = IB_PORT_CM_SUP 3759 .set_port_cap_mask = IB_PORT_CM_SUP
@@ -3801,7 +3801,8 @@ static void cm_add_one(struct ib_device *ib_device)
3801 0, 3801 0,
3802 cm_send_handler, 3802 cm_send_handler,
3803 cm_recv_handler, 3803 cm_recv_handler,
3804 port); 3804 port,
3805 0);
3805 if (IS_ERR(port->mad_agent)) 3806 if (IS_ERR(port->mad_agent))
3806 goto error2; 3807 goto error2;
3807 3808
diff --git a/drivers/infiniband/core/iwcm.c b/drivers/infiniband/core/iwcm.c
index 3d2e489ab732..ff9163dc1596 100644
--- a/drivers/infiniband/core/iwcm.c
+++ b/drivers/infiniband/core/iwcm.c
@@ -46,6 +46,7 @@
46#include <linux/completion.h> 46#include <linux/completion.h>
47#include <linux/slab.h> 47#include <linux/slab.h>
48#include <linux/module.h> 48#include <linux/module.h>
49#include <linux/sysctl.h>
49 50
50#include <rdma/iw_cm.h> 51#include <rdma/iw_cm.h>
51#include <rdma/ib_addr.h> 52#include <rdma/ib_addr.h>
@@ -65,6 +66,20 @@ struct iwcm_work {
65 struct list_head free_list; 66 struct list_head free_list;
66}; 67};
67 68
69static unsigned int default_backlog = 256;
70
71static struct ctl_table_header *iwcm_ctl_table_hdr;
72static struct ctl_table iwcm_ctl_table[] = {
73 {
74 .procname = "default_backlog",
75 .data = &default_backlog,
76 .maxlen = sizeof(default_backlog),
77 .mode = 0644,
78 .proc_handler = proc_dointvec,
79 },
80 { }
81};
82
68/* 83/*
69 * The following services provide a mechanism for pre-allocating iwcm_work 84 * The following services provide a mechanism for pre-allocating iwcm_work
70 * elements. The design pre-allocates them based on the cm_id type: 85 * elements. The design pre-allocates them based on the cm_id type:
@@ -425,6 +440,9 @@ int iw_cm_listen(struct iw_cm_id *cm_id, int backlog)
425 440
426 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); 441 cm_id_priv = container_of(cm_id, struct iwcm_id_private, id);
427 442
443 if (!backlog)
444 backlog = default_backlog;
445
428 ret = alloc_work_entries(cm_id_priv, backlog); 446 ret = alloc_work_entries(cm_id_priv, backlog);
429 if (ret) 447 if (ret)
430 return ret; 448 return ret;
@@ -1030,11 +1048,20 @@ static int __init iw_cm_init(void)
1030 if (!iwcm_wq) 1048 if (!iwcm_wq)
1031 return -ENOMEM; 1049 return -ENOMEM;
1032 1050
1051 iwcm_ctl_table_hdr = register_net_sysctl(&init_net, "net/iw_cm",
1052 iwcm_ctl_table);
1053 if (!iwcm_ctl_table_hdr) {
1054 pr_err("iw_cm: couldn't register sysctl paths\n");
1055 destroy_workqueue(iwcm_wq);
1056 return -ENOMEM;
1057 }
1058
1033 return 0; 1059 return 0;
1034} 1060}
1035 1061
1036static void __exit iw_cm_cleanup(void) 1062static void __exit iw_cm_cleanup(void)
1037{ 1063{
1064 unregister_net_sysctl_table(iwcm_ctl_table_hdr);
1038 destroy_workqueue(iwcm_wq); 1065 destroy_workqueue(iwcm_wq);
1039} 1066}
1040 1067
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index ab31f136d04b..74c30f4c557e 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -33,6 +33,9 @@
33 * SOFTWARE. 33 * SOFTWARE.
34 * 34 *
35 */ 35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
36#include <linux/dma-mapping.h> 39#include <linux/dma-mapping.h>
37#include <linux/slab.h> 40#include <linux/slab.h>
38#include <linux/module.h> 41#include <linux/module.h>
@@ -195,7 +198,8 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
195 u8 rmpp_version, 198 u8 rmpp_version,
196 ib_mad_send_handler send_handler, 199 ib_mad_send_handler send_handler,
197 ib_mad_recv_handler recv_handler, 200 ib_mad_recv_handler recv_handler,
198 void *context) 201 void *context,
202 u32 registration_flags)
199{ 203{
200 struct ib_mad_port_private *port_priv; 204 struct ib_mad_port_private *port_priv;
201 struct ib_mad_agent *ret = ERR_PTR(-EINVAL); 205 struct ib_mad_agent *ret = ERR_PTR(-EINVAL);
@@ -211,68 +215,109 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
211 215
212 /* Validate parameters */ 216 /* Validate parameters */
213 qpn = get_spl_qp_index(qp_type); 217 qpn = get_spl_qp_index(qp_type);
214 if (qpn == -1) 218 if (qpn == -1) {
219 dev_notice(&device->dev,
220 "ib_register_mad_agent: invalid QP Type %d\n",
221 qp_type);
215 goto error1; 222 goto error1;
223 }
216 224
217 if (rmpp_version && rmpp_version != IB_MGMT_RMPP_VERSION) 225 if (rmpp_version && rmpp_version != IB_MGMT_RMPP_VERSION) {
226 dev_notice(&device->dev,
227 "ib_register_mad_agent: invalid RMPP Version %u\n",
228 rmpp_version);
218 goto error1; 229 goto error1;
230 }
219 231
220 /* Validate MAD registration request if supplied */ 232 /* Validate MAD registration request if supplied */
221 if (mad_reg_req) { 233 if (mad_reg_req) {
222 if (mad_reg_req->mgmt_class_version >= MAX_MGMT_VERSION) 234 if (mad_reg_req->mgmt_class_version >= MAX_MGMT_VERSION) {
235 dev_notice(&device->dev,
236 "ib_register_mad_agent: invalid Class Version %u\n",
237 mad_reg_req->mgmt_class_version);
223 goto error1; 238 goto error1;
224 if (!recv_handler) 239 }
240 if (!recv_handler) {
241 dev_notice(&device->dev,
242 "ib_register_mad_agent: no recv_handler\n");
225 goto error1; 243 goto error1;
244 }
226 if (mad_reg_req->mgmt_class >= MAX_MGMT_CLASS) { 245 if (mad_reg_req->mgmt_class >= MAX_MGMT_CLASS) {
227 /* 246 /*
228 * IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE is the only 247 * IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE is the only
229 * one in this range currently allowed 248 * one in this range currently allowed
230 */ 249 */
231 if (mad_reg_req->mgmt_class != 250 if (mad_reg_req->mgmt_class !=
232 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) 251 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
252 dev_notice(&device->dev,
253 "ib_register_mad_agent: Invalid Mgmt Class 0x%x\n",
254 mad_reg_req->mgmt_class);
233 goto error1; 255 goto error1;
256 }
234 } else if (mad_reg_req->mgmt_class == 0) { 257 } else if (mad_reg_req->mgmt_class == 0) {
235 /* 258 /*
236 * Class 0 is reserved in IBA and is used for 259 * Class 0 is reserved in IBA and is used for
237 * aliasing of IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE 260 * aliasing of IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
238 */ 261 */
262 dev_notice(&device->dev,
263 "ib_register_mad_agent: Invalid Mgmt Class 0\n");
239 goto error1; 264 goto error1;
240 } else if (is_vendor_class(mad_reg_req->mgmt_class)) { 265 } else if (is_vendor_class(mad_reg_req->mgmt_class)) {
241 /* 266 /*
242 * If class is in "new" vendor range, 267 * If class is in "new" vendor range,
243 * ensure supplied OUI is not zero 268 * ensure supplied OUI is not zero
244 */ 269 */
245 if (!is_vendor_oui(mad_reg_req->oui)) 270 if (!is_vendor_oui(mad_reg_req->oui)) {
271 dev_notice(&device->dev,
272 "ib_register_mad_agent: No OUI specified for class 0x%x\n",
273 mad_reg_req->mgmt_class);
246 goto error1; 274 goto error1;
275 }
247 } 276 }
248 /* Make sure class supplied is consistent with RMPP */ 277 /* Make sure class supplied is consistent with RMPP */
249 if (!ib_is_mad_class_rmpp(mad_reg_req->mgmt_class)) { 278 if (!ib_is_mad_class_rmpp(mad_reg_req->mgmt_class)) {
250 if (rmpp_version) 279 if (rmpp_version) {
280 dev_notice(&device->dev,
281 "ib_register_mad_agent: RMPP version for non-RMPP class 0x%x\n",
282 mad_reg_req->mgmt_class);
251 goto error1; 283 goto error1;
284 }
252 } 285 }
286
253 /* Make sure class supplied is consistent with QP type */ 287 /* Make sure class supplied is consistent with QP type */
254 if (qp_type == IB_QPT_SMI) { 288 if (qp_type == IB_QPT_SMI) {
255 if ((mad_reg_req->mgmt_class != 289 if ((mad_reg_req->mgmt_class !=
256 IB_MGMT_CLASS_SUBN_LID_ROUTED) && 290 IB_MGMT_CLASS_SUBN_LID_ROUTED) &&
257 (mad_reg_req->mgmt_class != 291 (mad_reg_req->mgmt_class !=
258 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)) 292 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)) {
293 dev_notice(&device->dev,
294 "ib_register_mad_agent: Invalid SM QP type: class 0x%x\n",
295 mad_reg_req->mgmt_class);
259 goto error1; 296 goto error1;
297 }
260 } else { 298 } else {
261 if ((mad_reg_req->mgmt_class == 299 if ((mad_reg_req->mgmt_class ==
262 IB_MGMT_CLASS_SUBN_LID_ROUTED) || 300 IB_MGMT_CLASS_SUBN_LID_ROUTED) ||
263 (mad_reg_req->mgmt_class == 301 (mad_reg_req->mgmt_class ==
264 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)) 302 IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)) {
303 dev_notice(&device->dev,
304 "ib_register_mad_agent: Invalid GS QP type: class 0x%x\n",
305 mad_reg_req->mgmt_class);
265 goto error1; 306 goto error1;
307 }
266 } 308 }
267 } else { 309 } else {
268 /* No registration request supplied */ 310 /* No registration request supplied */
269 if (!send_handler) 311 if (!send_handler)
270 goto error1; 312 goto error1;
313 if (registration_flags & IB_MAD_USER_RMPP)
314 goto error1;
271 } 315 }
272 316
273 /* Validate device and port */ 317 /* Validate device and port */
274 port_priv = ib_get_mad_port(device, port_num); 318 port_priv = ib_get_mad_port(device, port_num);
275 if (!port_priv) { 319 if (!port_priv) {
320 dev_notice(&device->dev, "ib_register_mad_agent: Invalid port\n");
276 ret = ERR_PTR(-ENODEV); 321 ret = ERR_PTR(-ENODEV);
277 goto error1; 322 goto error1;
278 } 323 }
@@ -280,6 +325,8 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
280 /* Verify the QP requested is supported. For example, Ethernet devices 325 /* Verify the QP requested is supported. For example, Ethernet devices
281 * will not have QP0 */ 326 * will not have QP0 */
282 if (!port_priv->qp_info[qpn].qp) { 327 if (!port_priv->qp_info[qpn].qp) {
328 dev_notice(&device->dev,
329 "ib_register_mad_agent: QP %d not supported\n", qpn);
283 ret = ERR_PTR(-EPROTONOSUPPORT); 330 ret = ERR_PTR(-EPROTONOSUPPORT);
284 goto error1; 331 goto error1;
285 } 332 }
@@ -316,6 +363,7 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
316 mad_agent_priv->agent.context = context; 363 mad_agent_priv->agent.context = context;
317 mad_agent_priv->agent.qp = port_priv->qp_info[qpn].qp; 364 mad_agent_priv->agent.qp = port_priv->qp_info[qpn].qp;
318 mad_agent_priv->agent.port_num = port_num; 365 mad_agent_priv->agent.port_num = port_num;
366 mad_agent_priv->agent.flags = registration_flags;
319 spin_lock_init(&mad_agent_priv->lock); 367 spin_lock_init(&mad_agent_priv->lock);
320 INIT_LIST_HEAD(&mad_agent_priv->send_list); 368 INIT_LIST_HEAD(&mad_agent_priv->send_list);
321 INIT_LIST_HEAD(&mad_agent_priv->wait_list); 369 INIT_LIST_HEAD(&mad_agent_priv->wait_list);
@@ -706,7 +754,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
706 smi_handle_dr_smp_send(smp, device->node_type, port_num) == 754 smi_handle_dr_smp_send(smp, device->node_type, port_num) ==
707 IB_SMI_DISCARD) { 755 IB_SMI_DISCARD) {
708 ret = -EINVAL; 756 ret = -EINVAL;
709 printk(KERN_ERR PFX "Invalid directed route\n"); 757 dev_err(&device->dev, "Invalid directed route\n");
710 goto out; 758 goto out;
711 } 759 }
712 760
@@ -718,7 +766,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
718 local = kmalloc(sizeof *local, GFP_ATOMIC); 766 local = kmalloc(sizeof *local, GFP_ATOMIC);
719 if (!local) { 767 if (!local) {
720 ret = -ENOMEM; 768 ret = -ENOMEM;
721 printk(KERN_ERR PFX "No memory for ib_mad_local_private\n"); 769 dev_err(&device->dev, "No memory for ib_mad_local_private\n");
722 goto out; 770 goto out;
723 } 771 }
724 local->mad_priv = NULL; 772 local->mad_priv = NULL;
@@ -726,7 +774,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
726 mad_priv = kmem_cache_alloc(ib_mad_cache, GFP_ATOMIC); 774 mad_priv = kmem_cache_alloc(ib_mad_cache, GFP_ATOMIC);
727 if (!mad_priv) { 775 if (!mad_priv) {
728 ret = -ENOMEM; 776 ret = -ENOMEM;
729 printk(KERN_ERR PFX "No memory for local response MAD\n"); 777 dev_err(&device->dev, "No memory for local response MAD\n");
730 kfree(local); 778 kfree(local);
731 goto out; 779 goto out;
732 } 780 }
@@ -837,9 +885,9 @@ static int alloc_send_rmpp_list(struct ib_mad_send_wr_private *send_wr,
837 for (left = send_buf->data_len + pad; left > 0; left -= seg_size) { 885 for (left = send_buf->data_len + pad; left > 0; left -= seg_size) {
838 seg = kmalloc(sizeof (*seg) + seg_size, gfp_mask); 886 seg = kmalloc(sizeof (*seg) + seg_size, gfp_mask);
839 if (!seg) { 887 if (!seg) {
840 printk(KERN_ERR "alloc_send_rmpp_segs: RMPP mem " 888 dev_err(&send_buf->mad_agent->device->dev,
841 "alloc failed for len %zd, gfp %#x\n", 889 "alloc_send_rmpp_segs: RMPP mem alloc failed for len %zd, gfp %#x\n",
842 sizeof (*seg) + seg_size, gfp_mask); 890 sizeof (*seg) + seg_size, gfp_mask);
843 free_send_rmpp_list(send_wr); 891 free_send_rmpp_list(send_wr);
844 return -ENOMEM; 892 return -ENOMEM;
845 } 893 }
@@ -862,6 +910,12 @@ static int alloc_send_rmpp_list(struct ib_mad_send_wr_private *send_wr,
862 return 0; 910 return 0;
863} 911}
864 912
913int ib_mad_kernel_rmpp_agent(struct ib_mad_agent *agent)
914{
915 return agent->rmpp_version && !(agent->flags & IB_MAD_USER_RMPP);
916}
917EXPORT_SYMBOL(ib_mad_kernel_rmpp_agent);
918
865struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent, 919struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
866 u32 remote_qpn, u16 pkey_index, 920 u32 remote_qpn, u16 pkey_index,
867 int rmpp_active, 921 int rmpp_active,
@@ -878,10 +932,12 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
878 pad = get_pad_size(hdr_len, data_len); 932 pad = get_pad_size(hdr_len, data_len);
879 message_size = hdr_len + data_len + pad; 933 message_size = hdr_len + data_len + pad;
880 934
881 if ((!mad_agent->rmpp_version && 935 if (ib_mad_kernel_rmpp_agent(mad_agent)) {
882 (rmpp_active || message_size > sizeof(struct ib_mad))) || 936 if (!rmpp_active && message_size > sizeof(struct ib_mad))
883 (!rmpp_active && message_size > sizeof(struct ib_mad))) 937 return ERR_PTR(-EINVAL);
884 return ERR_PTR(-EINVAL); 938 } else
939 if (rmpp_active || message_size > sizeof(struct ib_mad))
940 return ERR_PTR(-EINVAL);
885 941
886 size = rmpp_active ? hdr_len : sizeof(struct ib_mad); 942 size = rmpp_active ? hdr_len : sizeof(struct ib_mad);
887 buf = kzalloc(sizeof *mad_send_wr + size, gfp_mask); 943 buf = kzalloc(sizeof *mad_send_wr + size, gfp_mask);
@@ -1135,7 +1191,7 @@ int ib_post_send_mad(struct ib_mad_send_buf *send_buf,
1135 &mad_agent_priv->send_list); 1191 &mad_agent_priv->send_list);
1136 spin_unlock_irqrestore(&mad_agent_priv->lock, flags); 1192 spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
1137 1193
1138 if (mad_agent_priv->agent.rmpp_version) { 1194 if (ib_mad_kernel_rmpp_agent(&mad_agent_priv->agent)) {
1139 ret = ib_send_rmpp_mad(mad_send_wr); 1195 ret = ib_send_rmpp_mad(mad_send_wr);
1140 if (ret >= 0 && ret != IB_RMPP_RESULT_CONSUMED) 1196 if (ret >= 0 && ret != IB_RMPP_RESULT_CONSUMED)
1141 ret = ib_send_mad(mad_send_wr); 1197 ret = ib_send_mad(mad_send_wr);
@@ -1199,7 +1255,8 @@ EXPORT_SYMBOL(ib_redirect_mad_qp);
1199int ib_process_mad_wc(struct ib_mad_agent *mad_agent, 1255int ib_process_mad_wc(struct ib_mad_agent *mad_agent,
1200 struct ib_wc *wc) 1256 struct ib_wc *wc)
1201{ 1257{
1202 printk(KERN_ERR PFX "ib_process_mad_wc() not implemented yet\n"); 1258 dev_err(&mad_agent->device->dev,
1259 "ib_process_mad_wc() not implemented yet\n");
1203 return 0; 1260 return 0;
1204} 1261}
1205EXPORT_SYMBOL(ib_process_mad_wc); 1262EXPORT_SYMBOL(ib_process_mad_wc);
@@ -1211,7 +1268,7 @@ static int method_in_use(struct ib_mad_mgmt_method_table **method,
1211 1268
1212 for_each_set_bit(i, mad_reg_req->method_mask, IB_MGMT_MAX_METHODS) { 1269 for_each_set_bit(i, mad_reg_req->method_mask, IB_MGMT_MAX_METHODS) {
1213 if ((*method)->agent[i]) { 1270 if ((*method)->agent[i]) {
1214 printk(KERN_ERR PFX "Method %d already in use\n", i); 1271 pr_err("Method %d already in use\n", i);
1215 return -EINVAL; 1272 return -EINVAL;
1216 } 1273 }
1217 } 1274 }
@@ -1223,8 +1280,7 @@ static int allocate_method_table(struct ib_mad_mgmt_method_table **method)
1223 /* Allocate management method table */ 1280 /* Allocate management method table */
1224 *method = kzalloc(sizeof **method, GFP_ATOMIC); 1281 *method = kzalloc(sizeof **method, GFP_ATOMIC);
1225 if (!*method) { 1282 if (!*method) {
1226 printk(KERN_ERR PFX "No memory for " 1283 pr_err("No memory for ib_mad_mgmt_method_table\n");
1227 "ib_mad_mgmt_method_table\n");
1228 return -ENOMEM; 1284 return -ENOMEM;
1229 } 1285 }
1230 1286
@@ -1319,8 +1375,8 @@ static int add_nonoui_reg_req(struct ib_mad_reg_req *mad_reg_req,
1319 /* Allocate management class table for "new" class version */ 1375 /* Allocate management class table for "new" class version */
1320 *class = kzalloc(sizeof **class, GFP_ATOMIC); 1376 *class = kzalloc(sizeof **class, GFP_ATOMIC);
1321 if (!*class) { 1377 if (!*class) {
1322 printk(KERN_ERR PFX "No memory for " 1378 dev_err(&agent_priv->agent.device->dev,
1323 "ib_mad_mgmt_class_table\n"); 1379 "No memory for ib_mad_mgmt_class_table\n");
1324 ret = -ENOMEM; 1380 ret = -ENOMEM;
1325 goto error1; 1381 goto error1;
1326 } 1382 }
@@ -1386,8 +1442,8 @@ static int add_oui_reg_req(struct ib_mad_reg_req *mad_reg_req,
1386 /* Allocate mgmt vendor class table for "new" class version */ 1442 /* Allocate mgmt vendor class table for "new" class version */
1387 vendor = kzalloc(sizeof *vendor, GFP_ATOMIC); 1443 vendor = kzalloc(sizeof *vendor, GFP_ATOMIC);
1388 if (!vendor) { 1444 if (!vendor) {
1389 printk(KERN_ERR PFX "No memory for " 1445 dev_err(&agent_priv->agent.device->dev,
1390 "ib_mad_mgmt_vendor_class_table\n"); 1446 "No memory for ib_mad_mgmt_vendor_class_table\n");
1391 goto error1; 1447 goto error1;
1392 } 1448 }
1393 1449
@@ -1397,8 +1453,8 @@ static int add_oui_reg_req(struct ib_mad_reg_req *mad_reg_req,
1397 /* Allocate table for this management vendor class */ 1453 /* Allocate table for this management vendor class */
1398 vendor_class = kzalloc(sizeof *vendor_class, GFP_ATOMIC); 1454 vendor_class = kzalloc(sizeof *vendor_class, GFP_ATOMIC);
1399 if (!vendor_class) { 1455 if (!vendor_class) {
1400 printk(KERN_ERR PFX "No memory for " 1456 dev_err(&agent_priv->agent.device->dev,
1401 "ib_mad_mgmt_vendor_class\n"); 1457 "No memory for ib_mad_mgmt_vendor_class\n");
1402 goto error2; 1458 goto error2;
1403 } 1459 }
1404 1460
@@ -1429,7 +1485,7 @@ static int add_oui_reg_req(struct ib_mad_reg_req *mad_reg_req,
1429 goto check_in_use; 1485 goto check_in_use;
1430 } 1486 }
1431 } 1487 }
1432 printk(KERN_ERR PFX "All OUI slots in use\n"); 1488 dev_err(&agent_priv->agent.device->dev, "All OUI slots in use\n");
1433 goto error3; 1489 goto error3;
1434 1490
1435check_in_use: 1491check_in_use:
@@ -1640,9 +1696,9 @@ find_mad_agent(struct ib_mad_port_private *port_priv,
1640 if (mad_agent->agent.recv_handler) 1696 if (mad_agent->agent.recv_handler)
1641 atomic_inc(&mad_agent->refcount); 1697 atomic_inc(&mad_agent->refcount);
1642 else { 1698 else {
1643 printk(KERN_NOTICE PFX "No receive handler for client " 1699 dev_notice(&port_priv->device->dev,
1644 "%p on port %d\n", 1700 "No receive handler for client %p on port %d\n",
1645 &mad_agent->agent, port_priv->port_num); 1701 &mad_agent->agent, port_priv->port_num);
1646 mad_agent = NULL; 1702 mad_agent = NULL;
1647 } 1703 }
1648 } 1704 }
@@ -1658,8 +1714,8 @@ static int validate_mad(struct ib_mad *mad, u32 qp_num)
1658 1714
1659 /* Make sure MAD base version is understood */ 1715 /* Make sure MAD base version is understood */
1660 if (mad->mad_hdr.base_version != IB_MGMT_BASE_VERSION) { 1716 if (mad->mad_hdr.base_version != IB_MGMT_BASE_VERSION) {
1661 printk(KERN_ERR PFX "MAD received with unsupported base " 1717 pr_err("MAD received with unsupported base version %d\n",
1662 "version %d\n", mad->mad_hdr.base_version); 1718 mad->mad_hdr.base_version);
1663 goto out; 1719 goto out;
1664 } 1720 }
1665 1721
@@ -1685,6 +1741,7 @@ static int is_data_mad(struct ib_mad_agent_private *mad_agent_priv,
1685 1741
1686 rmpp_mad = (struct ib_rmpp_mad *)mad_hdr; 1742 rmpp_mad = (struct ib_rmpp_mad *)mad_hdr;
1687 return !mad_agent_priv->agent.rmpp_version || 1743 return !mad_agent_priv->agent.rmpp_version ||
1744 !ib_mad_kernel_rmpp_agent(&mad_agent_priv->agent) ||
1688 !(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & 1745 !(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
1689 IB_MGMT_RMPP_FLAG_ACTIVE) || 1746 IB_MGMT_RMPP_FLAG_ACTIVE) ||
1690 (rmpp_mad->rmpp_hdr.rmpp_type == IB_MGMT_RMPP_TYPE_DATA); 1747 (rmpp_mad->rmpp_hdr.rmpp_type == IB_MGMT_RMPP_TYPE_DATA);
@@ -1812,7 +1869,7 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
1812 1869
1813 INIT_LIST_HEAD(&mad_recv_wc->rmpp_list); 1870 INIT_LIST_HEAD(&mad_recv_wc->rmpp_list);
1814 list_add(&mad_recv_wc->recv_buf.list, &mad_recv_wc->rmpp_list); 1871 list_add(&mad_recv_wc->recv_buf.list, &mad_recv_wc->rmpp_list);
1815 if (mad_agent_priv->agent.rmpp_version) { 1872 if (ib_mad_kernel_rmpp_agent(&mad_agent_priv->agent)) {
1816 mad_recv_wc = ib_process_rmpp_recv_wc(mad_agent_priv, 1873 mad_recv_wc = ib_process_rmpp_recv_wc(mad_agent_priv,
1817 mad_recv_wc); 1874 mad_recv_wc);
1818 if (!mad_recv_wc) { 1875 if (!mad_recv_wc) {
@@ -1827,23 +1884,39 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
1827 mad_send_wr = ib_find_send_mad(mad_agent_priv, mad_recv_wc); 1884 mad_send_wr = ib_find_send_mad(mad_agent_priv, mad_recv_wc);
1828 if (!mad_send_wr) { 1885 if (!mad_send_wr) {
1829 spin_unlock_irqrestore(&mad_agent_priv->lock, flags); 1886 spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
1830 ib_free_recv_mad(mad_recv_wc); 1887 if (!ib_mad_kernel_rmpp_agent(&mad_agent_priv->agent)
1831 deref_mad_agent(mad_agent_priv); 1888 && ib_is_mad_class_rmpp(mad_recv_wc->recv_buf.mad->mad_hdr.mgmt_class)
1832 return; 1889 && (ib_get_rmpp_flags(&((struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad)->rmpp_hdr)
1833 } 1890 & IB_MGMT_RMPP_FLAG_ACTIVE)) {
1834 ib_mark_mad_done(mad_send_wr); 1891 /* user rmpp is in effect
1835 spin_unlock_irqrestore(&mad_agent_priv->lock, flags); 1892 * and this is an active RMPP MAD
1893 */
1894 mad_recv_wc->wc->wr_id = 0;
1895 mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent,
1896 mad_recv_wc);
1897 atomic_dec(&mad_agent_priv->refcount);
1898 } else {
1899 /* not user rmpp, revert to normal behavior and
1900 * drop the mad */
1901 ib_free_recv_mad(mad_recv_wc);
1902 deref_mad_agent(mad_agent_priv);
1903 return;
1904 }
1905 } else {
1906 ib_mark_mad_done(mad_send_wr);
1907 spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
1836 1908
1837 /* Defined behavior is to complete response before request */ 1909 /* Defined behavior is to complete response before request */
1838 mad_recv_wc->wc->wr_id = (unsigned long) &mad_send_wr->send_buf; 1910 mad_recv_wc->wc->wr_id = (unsigned long) &mad_send_wr->send_buf;
1839 mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent, 1911 mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent,
1840 mad_recv_wc); 1912 mad_recv_wc);
1841 atomic_dec(&mad_agent_priv->refcount); 1913 atomic_dec(&mad_agent_priv->refcount);
1842 1914
1843 mad_send_wc.status = IB_WC_SUCCESS; 1915 mad_send_wc.status = IB_WC_SUCCESS;
1844 mad_send_wc.vendor_err = 0; 1916 mad_send_wc.vendor_err = 0;
1845 mad_send_wc.send_buf = &mad_send_wr->send_buf; 1917 mad_send_wc.send_buf = &mad_send_wr->send_buf;
1846 ib_mad_complete_send_wr(mad_send_wr, &mad_send_wc); 1918 ib_mad_complete_send_wr(mad_send_wr, &mad_send_wc);
1919 }
1847 } else { 1920 } else {
1848 mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent, 1921 mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent,
1849 mad_recv_wc); 1922 mad_recv_wc);
@@ -1911,8 +1984,8 @@ static void ib_mad_recv_done_handler(struct ib_mad_port_private *port_priv,
1911 1984
1912 response = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL); 1985 response = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL);
1913 if (!response) { 1986 if (!response) {
1914 printk(KERN_ERR PFX "ib_mad_recv_done_handler no memory " 1987 dev_err(&port_priv->device->dev,
1915 "for response buffer\n"); 1988 "ib_mad_recv_done_handler no memory for response buffer\n");
1916 goto out; 1989 goto out;
1917 } 1990 }
1918 1991
@@ -2083,7 +2156,7 @@ void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
2083 2156
2084 mad_agent_priv = mad_send_wr->mad_agent_priv; 2157 mad_agent_priv = mad_send_wr->mad_agent_priv;
2085 spin_lock_irqsave(&mad_agent_priv->lock, flags); 2158 spin_lock_irqsave(&mad_agent_priv->lock, flags);
2086 if (mad_agent_priv->agent.rmpp_version) { 2159 if (ib_mad_kernel_rmpp_agent(&mad_agent_priv->agent)) {
2087 ret = ib_process_rmpp_send_wc(mad_send_wr, mad_send_wc); 2160 ret = ib_process_rmpp_send_wc(mad_send_wr, mad_send_wc);
2088 if (ret == IB_RMPP_RESULT_CONSUMED) 2161 if (ret == IB_RMPP_RESULT_CONSUMED)
2089 goto done; 2162 goto done;
@@ -2176,7 +2249,8 @@ retry:
2176 ret = ib_post_send(qp_info->qp, &queued_send_wr->send_wr, 2249 ret = ib_post_send(qp_info->qp, &queued_send_wr->send_wr,
2177 &bad_send_wr); 2250 &bad_send_wr);
2178 if (ret) { 2251 if (ret) {
2179 printk(KERN_ERR PFX "ib_post_send failed: %d\n", ret); 2252 dev_err(&port_priv->device->dev,
2253 "ib_post_send failed: %d\n", ret);
2180 mad_send_wr = queued_send_wr; 2254 mad_send_wr = queued_send_wr;
2181 wc->status = IB_WC_LOC_QP_OP_ERR; 2255 wc->status = IB_WC_LOC_QP_OP_ERR;
2182 goto retry; 2256 goto retry;
@@ -2248,8 +2322,9 @@ static void mad_error_handler(struct ib_mad_port_private *port_priv,
2248 IB_QP_STATE | IB_QP_CUR_STATE); 2322 IB_QP_STATE | IB_QP_CUR_STATE);
2249 kfree(attr); 2323 kfree(attr);
2250 if (ret) 2324 if (ret)
2251 printk(KERN_ERR PFX "mad_error_handler - " 2325 dev_err(&port_priv->device->dev,
2252 "ib_modify_qp to RTS : %d\n", ret); 2326 "mad_error_handler - ib_modify_qp to RTS : %d\n",
2327 ret);
2253 else 2328 else
2254 mark_sends_for_retry(qp_info); 2329 mark_sends_for_retry(qp_info);
2255 } 2330 }
@@ -2408,7 +2483,8 @@ static void local_completions(struct work_struct *work)
2408 if (local->mad_priv) { 2483 if (local->mad_priv) {
2409 recv_mad_agent = local->recv_mad_agent; 2484 recv_mad_agent = local->recv_mad_agent;
2410 if (!recv_mad_agent) { 2485 if (!recv_mad_agent) {
2411 printk(KERN_ERR PFX "No receive MAD agent for local completion\n"); 2486 dev_err(&mad_agent_priv->agent.device->dev,
2487 "No receive MAD agent for local completion\n");
2412 free_mad = 1; 2488 free_mad = 1;
2413 goto local_send_completion; 2489 goto local_send_completion;
2414 } 2490 }
@@ -2476,7 +2552,7 @@ static int retry_send(struct ib_mad_send_wr_private *mad_send_wr)
2476 2552
2477 mad_send_wr->timeout = msecs_to_jiffies(mad_send_wr->send_buf.timeout_ms); 2553 mad_send_wr->timeout = msecs_to_jiffies(mad_send_wr->send_buf.timeout_ms);
2478 2554
2479 if (mad_send_wr->mad_agent_priv->agent.rmpp_version) { 2555 if (ib_mad_kernel_rmpp_agent(&mad_send_wr->mad_agent_priv->agent)) {
2480 ret = ib_retry_rmpp(mad_send_wr); 2556 ret = ib_retry_rmpp(mad_send_wr);
2481 switch (ret) { 2557 switch (ret) {
2482 case IB_RMPP_RESULT_UNHANDLED: 2558 case IB_RMPP_RESULT_UNHANDLED:
@@ -2589,7 +2665,8 @@ static int ib_mad_post_receive_mads(struct ib_mad_qp_info *qp_info,
2589 } else { 2665 } else {
2590 mad_priv = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL); 2666 mad_priv = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL);
2591 if (!mad_priv) { 2667 if (!mad_priv) {
2592 printk(KERN_ERR PFX "No memory for receive buffer\n"); 2668 dev_err(&qp_info->port_priv->device->dev,
2669 "No memory for receive buffer\n");
2593 ret = -ENOMEM; 2670 ret = -ENOMEM;
2594 break; 2671 break;
2595 } 2672 }
@@ -2625,7 +2702,8 @@ static int ib_mad_post_receive_mads(struct ib_mad_qp_info *qp_info,
2625 sizeof mad_priv->header, 2702 sizeof mad_priv->header,
2626 DMA_FROM_DEVICE); 2703 DMA_FROM_DEVICE);
2627 kmem_cache_free(ib_mad_cache, mad_priv); 2704 kmem_cache_free(ib_mad_cache, mad_priv);
2628 printk(KERN_ERR PFX "ib_post_recv failed: %d\n", ret); 2705 dev_err(&qp_info->port_priv->device->dev,
2706 "ib_post_recv failed: %d\n", ret);
2629 break; 2707 break;
2630 } 2708 }
2631 } while (post); 2709 } while (post);
@@ -2681,7 +2759,8 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2681 2759
2682 attr = kmalloc(sizeof *attr, GFP_KERNEL); 2760 attr = kmalloc(sizeof *attr, GFP_KERNEL);
2683 if (!attr) { 2761 if (!attr) {
2684 printk(KERN_ERR PFX "Couldn't kmalloc ib_qp_attr\n"); 2762 dev_err(&port_priv->device->dev,
2763 "Couldn't kmalloc ib_qp_attr\n");
2685 return -ENOMEM; 2764 return -ENOMEM;
2686 } 2765 }
2687 2766
@@ -2705,16 +2784,18 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2705 ret = ib_modify_qp(qp, attr, IB_QP_STATE | 2784 ret = ib_modify_qp(qp, attr, IB_QP_STATE |
2706 IB_QP_PKEY_INDEX | IB_QP_QKEY); 2785 IB_QP_PKEY_INDEX | IB_QP_QKEY);
2707 if (ret) { 2786 if (ret) {
2708 printk(KERN_ERR PFX "Couldn't change QP%d state to " 2787 dev_err(&port_priv->device->dev,
2709 "INIT: %d\n", i, ret); 2788 "Couldn't change QP%d state to INIT: %d\n",
2789 i, ret);
2710 goto out; 2790 goto out;
2711 } 2791 }
2712 2792
2713 attr->qp_state = IB_QPS_RTR; 2793 attr->qp_state = IB_QPS_RTR;
2714 ret = ib_modify_qp(qp, attr, IB_QP_STATE); 2794 ret = ib_modify_qp(qp, attr, IB_QP_STATE);
2715 if (ret) { 2795 if (ret) {
2716 printk(KERN_ERR PFX "Couldn't change QP%d state to " 2796 dev_err(&port_priv->device->dev,
2717 "RTR: %d\n", i, ret); 2797 "Couldn't change QP%d state to RTR: %d\n",
2798 i, ret);
2718 goto out; 2799 goto out;
2719 } 2800 }
2720 2801
@@ -2722,16 +2803,18 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2722 attr->sq_psn = IB_MAD_SEND_Q_PSN; 2803 attr->sq_psn = IB_MAD_SEND_Q_PSN;
2723 ret = ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_SQ_PSN); 2804 ret = ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_SQ_PSN);
2724 if (ret) { 2805 if (ret) {
2725 printk(KERN_ERR PFX "Couldn't change QP%d state to " 2806 dev_err(&port_priv->device->dev,
2726 "RTS: %d\n", i, ret); 2807 "Couldn't change QP%d state to RTS: %d\n",
2808 i, ret);
2727 goto out; 2809 goto out;
2728 } 2810 }
2729 } 2811 }
2730 2812
2731 ret = ib_req_notify_cq(port_priv->cq, IB_CQ_NEXT_COMP); 2813 ret = ib_req_notify_cq(port_priv->cq, IB_CQ_NEXT_COMP);
2732 if (ret) { 2814 if (ret) {
2733 printk(KERN_ERR PFX "Failed to request completion " 2815 dev_err(&port_priv->device->dev,
2734 "notification: %d\n", ret); 2816 "Failed to request completion notification: %d\n",
2817 ret);
2735 goto out; 2818 goto out;
2736 } 2819 }
2737 2820
@@ -2741,7 +2824,8 @@ static int ib_mad_port_start(struct ib_mad_port_private *port_priv)
2741 2824
2742 ret = ib_mad_post_receive_mads(&port_priv->qp_info[i], NULL); 2825 ret = ib_mad_post_receive_mads(&port_priv->qp_info[i], NULL);
2743 if (ret) { 2826 if (ret) {
2744 printk(KERN_ERR PFX "Couldn't post receive WRs\n"); 2827 dev_err(&port_priv->device->dev,
2828 "Couldn't post receive WRs\n");
2745 goto out; 2829 goto out;
2746 } 2830 }
2747 } 2831 }
@@ -2755,7 +2839,8 @@ static void qp_event_handler(struct ib_event *event, void *qp_context)
2755 struct ib_mad_qp_info *qp_info = qp_context; 2839 struct ib_mad_qp_info *qp_info = qp_context;
2756 2840
2757 /* It's worse than that! He's dead, Jim! */ 2841 /* It's worse than that! He's dead, Jim! */
2758 printk(KERN_ERR PFX "Fatal error (%d) on MAD QP (%d)\n", 2842 dev_err(&qp_info->port_priv->device->dev,
2843 "Fatal error (%d) on MAD QP (%d)\n",
2759 event->event, qp_info->qp->qp_num); 2844 event->event, qp_info->qp->qp_num);
2760} 2845}
2761 2846
@@ -2801,8 +2886,9 @@ static int create_mad_qp(struct ib_mad_qp_info *qp_info,
2801 qp_init_attr.event_handler = qp_event_handler; 2886 qp_init_attr.event_handler = qp_event_handler;
2802 qp_info->qp = ib_create_qp(qp_info->port_priv->pd, &qp_init_attr); 2887 qp_info->qp = ib_create_qp(qp_info->port_priv->pd, &qp_init_attr);
2803 if (IS_ERR(qp_info->qp)) { 2888 if (IS_ERR(qp_info->qp)) {
2804 printk(KERN_ERR PFX "Couldn't create ib_mad QP%d\n", 2889 dev_err(&qp_info->port_priv->device->dev,
2805 get_spl_qp_index(qp_type)); 2890 "Couldn't create ib_mad QP%d\n",
2891 get_spl_qp_index(qp_type));
2806 ret = PTR_ERR(qp_info->qp); 2892 ret = PTR_ERR(qp_info->qp);
2807 goto error; 2893 goto error;
2808 } 2894 }
@@ -2840,7 +2926,7 @@ static int ib_mad_port_open(struct ib_device *device,
2840 /* Create new device info */ 2926 /* Create new device info */
2841 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL); 2927 port_priv = kzalloc(sizeof *port_priv, GFP_KERNEL);
2842 if (!port_priv) { 2928 if (!port_priv) {
2843 printk(KERN_ERR PFX "No memory for ib_mad_port_private\n"); 2929 dev_err(&device->dev, "No memory for ib_mad_port_private\n");
2844 return -ENOMEM; 2930 return -ENOMEM;
2845 } 2931 }
2846 2932
@@ -2860,21 +2946,21 @@ static int ib_mad_port_open(struct ib_device *device,
2860 ib_mad_thread_completion_handler, 2946 ib_mad_thread_completion_handler,
2861 NULL, port_priv, cq_size, 0); 2947 NULL, port_priv, cq_size, 0);
2862 if (IS_ERR(port_priv->cq)) { 2948 if (IS_ERR(port_priv->cq)) {
2863 printk(KERN_ERR PFX "Couldn't create ib_mad CQ\n"); 2949 dev_err(&device->dev, "Couldn't create ib_mad CQ\n");
2864 ret = PTR_ERR(port_priv->cq); 2950 ret = PTR_ERR(port_priv->cq);
2865 goto error3; 2951 goto error3;
2866 } 2952 }
2867 2953
2868 port_priv->pd = ib_alloc_pd(device); 2954 port_priv->pd = ib_alloc_pd(device);
2869 if (IS_ERR(port_priv->pd)) { 2955 if (IS_ERR(port_priv->pd)) {
2870 printk(KERN_ERR PFX "Couldn't create ib_mad PD\n"); 2956 dev_err(&device->dev, "Couldn't create ib_mad PD\n");
2871 ret = PTR_ERR(port_priv->pd); 2957 ret = PTR_ERR(port_priv->pd);
2872 goto error4; 2958 goto error4;
2873 } 2959 }
2874 2960
2875 port_priv->mr = ib_get_dma_mr(port_priv->pd, IB_ACCESS_LOCAL_WRITE); 2961 port_priv->mr = ib_get_dma_mr(port_priv->pd, IB_ACCESS_LOCAL_WRITE);
2876 if (IS_ERR(port_priv->mr)) { 2962 if (IS_ERR(port_priv->mr)) {
2877 printk(KERN_ERR PFX "Couldn't get ib_mad DMA MR\n"); 2963 dev_err(&device->dev, "Couldn't get ib_mad DMA MR\n");
2878 ret = PTR_ERR(port_priv->mr); 2964 ret = PTR_ERR(port_priv->mr);
2879 goto error5; 2965 goto error5;
2880 } 2966 }
@@ -2902,7 +2988,7 @@ static int ib_mad_port_open(struct ib_device *device,
2902 2988
2903 ret = ib_mad_port_start(port_priv); 2989 ret = ib_mad_port_start(port_priv);
2904 if (ret) { 2990 if (ret) {
2905 printk(KERN_ERR PFX "Couldn't start port\n"); 2991 dev_err(&device->dev, "Couldn't start port\n");
2906 goto error9; 2992 goto error9;
2907 } 2993 }
2908 2994
@@ -2946,7 +3032,7 @@ static int ib_mad_port_close(struct ib_device *device, int port_num)
2946 port_priv = __ib_get_mad_port(device, port_num); 3032 port_priv = __ib_get_mad_port(device, port_num);
2947 if (port_priv == NULL) { 3033 if (port_priv == NULL) {
2948 spin_unlock_irqrestore(&ib_mad_port_list_lock, flags); 3034 spin_unlock_irqrestore(&ib_mad_port_list_lock, flags);
2949 printk(KERN_ERR PFX "Port %d not found\n", port_num); 3035 dev_err(&device->dev, "Port %d not found\n", port_num);
2950 return -ENODEV; 3036 return -ENODEV;
2951 } 3037 }
2952 list_del_init(&port_priv->port_list); 3038 list_del_init(&port_priv->port_list);
@@ -2984,14 +3070,12 @@ static void ib_mad_init_device(struct ib_device *device)
2984 3070
2985 for (i = start; i <= end; i++) { 3071 for (i = start; i <= end; i++) {
2986 if (ib_mad_port_open(device, i)) { 3072 if (ib_mad_port_open(device, i)) {
2987 printk(KERN_ERR PFX "Couldn't open %s port %d\n", 3073 dev_err(&device->dev, "Couldn't open port %d\n", i);
2988 device->name, i);
2989 goto error; 3074 goto error;
2990 } 3075 }
2991 if (ib_agent_port_open(device, i)) { 3076 if (ib_agent_port_open(device, i)) {
2992 printk(KERN_ERR PFX "Couldn't open %s port %d " 3077 dev_err(&device->dev,
2993 "for agents\n", 3078 "Couldn't open port %d for agents\n", i);
2994 device->name, i);
2995 goto error_agent; 3079 goto error_agent;
2996 } 3080 }
2997 } 3081 }
@@ -2999,20 +3083,17 @@ static void ib_mad_init_device(struct ib_device *device)
2999 3083
3000error_agent: 3084error_agent:
3001 if (ib_mad_port_close(device, i)) 3085 if (ib_mad_port_close(device, i))
3002 printk(KERN_ERR PFX "Couldn't close %s port %d\n", 3086 dev_err(&device->dev, "Couldn't close port %d\n", i);
3003 device->name, i);
3004 3087
3005error: 3088error:
3006 i--; 3089 i--;
3007 3090
3008 while (i >= start) { 3091 while (i >= start) {
3009 if (ib_agent_port_close(device, i)) 3092 if (ib_agent_port_close(device, i))
3010 printk(KERN_ERR PFX "Couldn't close %s port %d " 3093 dev_err(&device->dev,
3011 "for agents\n", 3094 "Couldn't close port %d for agents\n", i);
3012 device->name, i);
3013 if (ib_mad_port_close(device, i)) 3095 if (ib_mad_port_close(device, i))
3014 printk(KERN_ERR PFX "Couldn't close %s port %d\n", 3096 dev_err(&device->dev, "Couldn't close port %d\n", i);
3015 device->name, i);
3016 i--; 3097 i--;
3017 } 3098 }
3018} 3099}
@@ -3033,12 +3114,12 @@ static void ib_mad_remove_device(struct ib_device *device)
3033 } 3114 }
3034 for (i = 0; i < num_ports; i++, cur_port++) { 3115 for (i = 0; i < num_ports; i++, cur_port++) {
3035 if (ib_agent_port_close(device, cur_port)) 3116 if (ib_agent_port_close(device, cur_port))
3036 printk(KERN_ERR PFX "Couldn't close %s port %d " 3117 dev_err(&device->dev,
3037 "for agents\n", 3118 "Couldn't close port %d for agents\n",
3038 device->name, cur_port); 3119 cur_port);
3039 if (ib_mad_port_close(device, cur_port)) 3120 if (ib_mad_port_close(device, cur_port))
3040 printk(KERN_ERR PFX "Couldn't close %s port %d\n", 3121 dev_err(&device->dev, "Couldn't close port %d\n",
3041 device->name, cur_port); 3122 cur_port);
3042 } 3123 }
3043} 3124}
3044 3125
@@ -3064,7 +3145,7 @@ static int __init ib_mad_init_module(void)
3064 SLAB_HWCACHE_ALIGN, 3145 SLAB_HWCACHE_ALIGN,
3065 NULL); 3146 NULL);
3066 if (!ib_mad_cache) { 3147 if (!ib_mad_cache) {
3067 printk(KERN_ERR PFX "Couldn't create ib_mad cache\n"); 3148 pr_err("Couldn't create ib_mad cache\n");
3068 ret = -ENOMEM; 3149 ret = -ENOMEM;
3069 goto error1; 3150 goto error1;
3070 } 3151 }
@@ -3072,7 +3153,7 @@ static int __init ib_mad_init_module(void)
3072 INIT_LIST_HEAD(&ib_mad_port_list); 3153 INIT_LIST_HEAD(&ib_mad_port_list);
3073 3154
3074 if (ib_register_client(&mad_client)) { 3155 if (ib_register_client(&mad_client)) {
3075 printk(KERN_ERR PFX "Couldn't register ib_mad client\n"); 3156 pr_err("Couldn't register ib_mad client\n");
3076 ret = -EINVAL; 3157 ret = -EINVAL;
3077 goto error2; 3158 goto error2;
3078 } 3159 }
diff --git a/drivers/infiniband/core/mad_priv.h b/drivers/infiniband/core/mad_priv.h
index 9430ab4969c5..d1a0b0ee9444 100644
--- a/drivers/infiniband/core/mad_priv.h
+++ b/drivers/infiniband/core/mad_priv.h
@@ -42,9 +42,6 @@
42#include <rdma/ib_mad.h> 42#include <rdma/ib_mad.h>
43#include <rdma/ib_smi.h> 43#include <rdma/ib_smi.h>
44 44
45
46#define PFX "ib_mad: "
47
48#define IB_MAD_QPS_CORE 2 /* Always QP0 and QP1 as a minimum */ 45#define IB_MAD_QPS_CORE 2 /* Always QP0 and QP1 as a minimum */
49 46
50/* QP and CQ parameters */ 47/* QP and CQ parameters */
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 233eaf541f55..c38f030f0dc9 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -1184,7 +1184,7 @@ static void ib_sa_add_one(struct ib_device *device)
1184 sa_dev->port[i].agent = 1184 sa_dev->port[i].agent =
1185 ib_register_mad_agent(device, i + s, IB_QPT_GSI, 1185 ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1186 NULL, 0, send_handler, 1186 NULL, 0, send_handler,
1187 recv_handler, sa_dev); 1187 recv_handler, sa_dev, 0);
1188 if (IS_ERR(sa_dev->port[i].agent)) 1188 if (IS_ERR(sa_dev->port[i].agent))
1189 goto err; 1189 goto err;
1190 1190
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index 1acb99100556..928cdd20e2d1 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -33,6 +33,8 @@
33 * SOFTWARE. 33 * SOFTWARE.
34 */ 34 */
35 35
36#define pr_fmt(fmt) "user_mad: " fmt
37
36#include <linux/module.h> 38#include <linux/module.h>
37#include <linux/init.h> 39#include <linux/init.h>
38#include <linux/device.h> 40#include <linux/device.h>
@@ -504,13 +506,15 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
504 506
505 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data; 507 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
506 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class); 508 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
507 if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) { 509
508 copy_offset = IB_MGMT_MAD_HDR; 510 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
509 rmpp_active = 0; 511 && ib_mad_kernel_rmpp_agent(agent)) {
510 } else {
511 copy_offset = IB_MGMT_RMPP_HDR; 512 copy_offset = IB_MGMT_RMPP_HDR;
512 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & 513 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
513 IB_MGMT_RMPP_FLAG_ACTIVE; 514 IB_MGMT_RMPP_FLAG_ACTIVE;
515 } else {
516 copy_offset = IB_MGMT_MAD_HDR;
517 rmpp_active = 0;
514 } 518 }
515 519
516 data_len = count - hdr_size(file) - hdr_len; 520 data_len = count - hdr_size(file) - hdr_len;
@@ -556,14 +560,22 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
556 rmpp_mad->mad_hdr.tid = *tid; 560 rmpp_mad->mad_hdr.tid = *tid;
557 } 561 }
558 562
559 spin_lock_irq(&file->send_lock); 563 if (!ib_mad_kernel_rmpp_agent(agent)
560 ret = is_duplicate(file, packet); 564 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
561 if (!ret) 565 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
566 spin_lock_irq(&file->send_lock);
562 list_add_tail(&packet->list, &file->send_list); 567 list_add_tail(&packet->list, &file->send_list);
563 spin_unlock_irq(&file->send_lock); 568 spin_unlock_irq(&file->send_lock);
564 if (ret) { 569 } else {
565 ret = -EINVAL; 570 spin_lock_irq(&file->send_lock);
566 goto err_msg; 571 ret = is_duplicate(file, packet);
572 if (!ret)
573 list_add_tail(&packet->list, &file->send_list);
574 spin_unlock_irq(&file->send_lock);
575 if (ret) {
576 ret = -EINVAL;
577 goto err_msg;
578 }
567 } 579 }
568 580
569 ret = ib_post_send_mad(packet->msg, NULL); 581 ret = ib_post_send_mad(packet->msg, NULL);
@@ -614,6 +626,8 @@ static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
614 mutex_lock(&file->mutex); 626 mutex_lock(&file->mutex);
615 627
616 if (!file->port->ib_dev) { 628 if (!file->port->ib_dev) {
629 dev_notice(file->port->dev,
630 "ib_umad_reg_agent: invalid device\n");
617 ret = -EPIPE; 631 ret = -EPIPE;
618 goto out; 632 goto out;
619 } 633 }
@@ -624,6 +638,9 @@ static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
624 } 638 }
625 639
626 if (ureq.qpn != 0 && ureq.qpn != 1) { 640 if (ureq.qpn != 0 && ureq.qpn != 1) {
641 dev_notice(file->port->dev,
642 "ib_umad_reg_agent: invalid QPN %d specified\n",
643 ureq.qpn);
627 ret = -EINVAL; 644 ret = -EINVAL;
628 goto out; 645 goto out;
629 } 646 }
@@ -632,11 +649,15 @@ static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
632 if (!__get_agent(file, agent_id)) 649 if (!__get_agent(file, agent_id))
633 goto found; 650 goto found;
634 651
652 dev_notice(file->port->dev,
653 "ib_umad_reg_agent: Max Agents (%u) reached\n",
654 IB_UMAD_MAX_AGENTS);
635 ret = -ENOMEM; 655 ret = -ENOMEM;
636 goto out; 656 goto out;
637 657
638found: 658found:
639 if (ureq.mgmt_class) { 659 if (ureq.mgmt_class) {
660 memset(&req, 0, sizeof(req));
640 req.mgmt_class = ureq.mgmt_class; 661 req.mgmt_class = ureq.mgmt_class;
641 req.mgmt_class_version = ureq.mgmt_class_version; 662 req.mgmt_class_version = ureq.mgmt_class_version;
642 memcpy(req.oui, ureq.oui, sizeof req.oui); 663 memcpy(req.oui, ureq.oui, sizeof req.oui);
@@ -657,7 +678,7 @@ found:
657 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI, 678 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
658 ureq.mgmt_class ? &req : NULL, 679 ureq.mgmt_class ? &req : NULL,
659 ureq.rmpp_version, 680 ureq.rmpp_version,
660 send_handler, recv_handler, file); 681 send_handler, recv_handler, file, 0);
661 if (IS_ERR(agent)) { 682 if (IS_ERR(agent)) {
662 ret = PTR_ERR(agent); 683 ret = PTR_ERR(agent);
663 agent = NULL; 684 agent = NULL;
@@ -673,10 +694,11 @@ found:
673 if (!file->already_used) { 694 if (!file->already_used) {
674 file->already_used = 1; 695 file->already_used = 1;
675 if (!file->use_pkey_index) { 696 if (!file->use_pkey_index) {
676 printk(KERN_WARNING "user_mad: process %s did not enable " 697 dev_warn(file->port->dev,
677 "P_Key index support.\n", current->comm); 698 "process %s did not enable P_Key index support.\n",
678 printk(KERN_WARNING "user_mad: Documentation/infiniband/user_mad.txt " 699 current->comm);
679 "has info on the new ABI.\n"); 700 dev_warn(file->port->dev,
701 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
680 } 702 }
681 } 703 }
682 704
@@ -694,6 +716,119 @@ out:
694 return ret; 716 return ret;
695} 717}
696 718
719static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
720{
721 struct ib_user_mad_reg_req2 ureq;
722 struct ib_mad_reg_req req;
723 struct ib_mad_agent *agent = NULL;
724 int agent_id;
725 int ret;
726
727 mutex_lock(&file->port->file_mutex);
728 mutex_lock(&file->mutex);
729
730 if (!file->port->ib_dev) {
731 dev_notice(file->port->dev,
732 "ib_umad_reg_agent2: invalid device\n");
733 ret = -EPIPE;
734 goto out;
735 }
736
737 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
738 ret = -EFAULT;
739 goto out;
740 }
741
742 if (ureq.qpn != 0 && ureq.qpn != 1) {
743 dev_notice(file->port->dev,
744 "ib_umad_reg_agent2: invalid QPN %d specified\n",
745 ureq.qpn);
746 ret = -EINVAL;
747 goto out;
748 }
749
750 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
751 dev_notice(file->port->dev,
752 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
753 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
754 ret = -EINVAL;
755
756 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
757 (u32 __user *) (arg + offsetof(struct
758 ib_user_mad_reg_req2, flags))))
759 ret = -EFAULT;
760
761 goto out;
762 }
763
764 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
765 if (!__get_agent(file, agent_id))
766 goto found;
767
768 dev_notice(file->port->dev,
769 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
770 IB_UMAD_MAX_AGENTS);
771 ret = -ENOMEM;
772 goto out;
773
774found:
775 if (ureq.mgmt_class) {
776 memset(&req, 0, sizeof(req));
777 req.mgmt_class = ureq.mgmt_class;
778 req.mgmt_class_version = ureq.mgmt_class_version;
779 if (ureq.oui & 0xff000000) {
780 dev_notice(file->port->dev,
781 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
782 ureq.oui);
783 ret = -EINVAL;
784 goto out;
785 }
786 req.oui[2] = ureq.oui & 0x0000ff;
787 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
788 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
789 memcpy(req.method_mask, ureq.method_mask,
790 sizeof(req.method_mask));
791 }
792
793 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
794 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
795 ureq.mgmt_class ? &req : NULL,
796 ureq.rmpp_version,
797 send_handler, recv_handler, file,
798 ureq.flags);
799 if (IS_ERR(agent)) {
800 ret = PTR_ERR(agent);
801 agent = NULL;
802 goto out;
803 }
804
805 if (put_user(agent_id,
806 (u32 __user *)(arg +
807 offsetof(struct ib_user_mad_reg_req2, id)))) {
808 ret = -EFAULT;
809 goto out;
810 }
811
812 if (!file->already_used) {
813 file->already_used = 1;
814 file->use_pkey_index = 1;
815 }
816
817 file->agent[agent_id] = agent;
818 ret = 0;
819
820out:
821 mutex_unlock(&file->mutex);
822
823 if (ret && agent)
824 ib_unregister_mad_agent(agent);
825
826 mutex_unlock(&file->port->file_mutex);
827
828 return ret;
829}
830
831
697static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg) 832static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
698{ 833{
699 struct ib_mad_agent *agent = NULL; 834 struct ib_mad_agent *agent = NULL;
@@ -749,6 +884,8 @@ static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
749 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg); 884 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
750 case IB_USER_MAD_ENABLE_PKEY: 885 case IB_USER_MAD_ENABLE_PKEY:
751 return ib_umad_enable_pkey(filp->private_data); 886 return ib_umad_enable_pkey(filp->private_data);
887 case IB_USER_MAD_REGISTER_AGENT2:
888 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
752 default: 889 default:
753 return -ENOIOCTLCMD; 890 return -ENOIOCTLCMD;
754 } 891 }
@@ -765,6 +902,8 @@ static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
765 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg)); 902 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
766 case IB_USER_MAD_ENABLE_PKEY: 903 case IB_USER_MAD_ENABLE_PKEY:
767 return ib_umad_enable_pkey(filp->private_data); 904 return ib_umad_enable_pkey(filp->private_data);
905 case IB_USER_MAD_REGISTER_AGENT2:
906 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
768 default: 907 default:
769 return -ENOIOCTLCMD; 908 return -ENOIOCTLCMD;
770 } 909 }
@@ -983,7 +1122,7 @@ static CLASS_ATTR_STRING(abi_version, S_IRUGO,
983 1122
984static dev_t overflow_maj; 1123static dev_t overflow_maj;
985static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS); 1124static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
986static int find_overflow_devnum(void) 1125static int find_overflow_devnum(struct ib_device *device)
987{ 1126{
988 int ret; 1127 int ret;
989 1128
@@ -991,7 +1130,8 @@ static int find_overflow_devnum(void)
991 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2, 1130 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
992 "infiniband_mad"); 1131 "infiniband_mad");
993 if (ret) { 1132 if (ret) {
994 printk(KERN_ERR "user_mad: couldn't register dynamic device number\n"); 1133 dev_err(&device->dev,
1134 "couldn't register dynamic device number\n");
995 return ret; 1135 return ret;
996 } 1136 }
997 } 1137 }
@@ -1014,7 +1154,7 @@ static int ib_umad_init_port(struct ib_device *device, int port_num,
1014 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS); 1154 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1015 if (devnum >= IB_UMAD_MAX_PORTS) { 1155 if (devnum >= IB_UMAD_MAX_PORTS) {
1016 spin_unlock(&port_lock); 1156 spin_unlock(&port_lock);
1017 devnum = find_overflow_devnum(); 1157 devnum = find_overflow_devnum(device);
1018 if (devnum < 0) 1158 if (devnum < 0)
1019 return -1; 1159 return -1;
1020 1160
@@ -1200,14 +1340,14 @@ static int __init ib_umad_init(void)
1200 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2, 1340 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1201 "infiniband_mad"); 1341 "infiniband_mad");
1202 if (ret) { 1342 if (ret) {
1203 printk(KERN_ERR "user_mad: couldn't register device number\n"); 1343 pr_err("couldn't register device number\n");
1204 goto out; 1344 goto out;
1205 } 1345 }
1206 1346
1207 umad_class = class_create(THIS_MODULE, "infiniband_mad"); 1347 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1208 if (IS_ERR(umad_class)) { 1348 if (IS_ERR(umad_class)) {
1209 ret = PTR_ERR(umad_class); 1349 ret = PTR_ERR(umad_class);
1210 printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n"); 1350 pr_err("couldn't create class infiniband_mad\n");
1211 goto out_chrdev; 1351 goto out_chrdev;
1212 } 1352 }
1213 1353
@@ -1215,13 +1355,13 @@ static int __init ib_umad_init(void)
1215 1355
1216 ret = class_create_file(umad_class, &class_attr_abi_version.attr); 1356 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1217 if (ret) { 1357 if (ret) {
1218 printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n"); 1358 pr_err("couldn't create abi_version attribute\n");
1219 goto out_class; 1359 goto out_class;
1220 } 1360 }
1221 1361
1222 ret = ib_register_client(&umad_client); 1362 ret = ib_register_client(&umad_client);
1223 if (ret) { 1363 if (ret) {
1224 printk(KERN_ERR "user_mad: couldn't register ib_umad client\n"); 1364 pr_err("couldn't register ib_umad client\n");
1225 goto out_class; 1365 goto out_class;
1226 } 1366 }
1227 1367
diff --git a/drivers/infiniband/core/uverbs.h b/drivers/infiniband/core/uverbs.h
index a283274a5a09..643c08a025a5 100644
--- a/drivers/infiniband/core/uverbs.h
+++ b/drivers/infiniband/core/uverbs.h
@@ -221,6 +221,7 @@ IB_UVERBS_DECLARE_CMD(query_port);
221IB_UVERBS_DECLARE_CMD(alloc_pd); 221IB_UVERBS_DECLARE_CMD(alloc_pd);
222IB_UVERBS_DECLARE_CMD(dealloc_pd); 222IB_UVERBS_DECLARE_CMD(dealloc_pd);
223IB_UVERBS_DECLARE_CMD(reg_mr); 223IB_UVERBS_DECLARE_CMD(reg_mr);
224IB_UVERBS_DECLARE_CMD(rereg_mr);
224IB_UVERBS_DECLARE_CMD(dereg_mr); 225IB_UVERBS_DECLARE_CMD(dereg_mr);
225IB_UVERBS_DECLARE_CMD(alloc_mw); 226IB_UVERBS_DECLARE_CMD(alloc_mw);
226IB_UVERBS_DECLARE_CMD(dealloc_mw); 227IB_UVERBS_DECLARE_CMD(dealloc_mw);
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index ea6203ee7bcc..0600c50e6215 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -1002,6 +1002,99 @@ err_free:
1002 return ret; 1002 return ret;
1003} 1003}
1004 1004
1005ssize_t ib_uverbs_rereg_mr(struct ib_uverbs_file *file,
1006 const char __user *buf, int in_len,
1007 int out_len)
1008{
1009 struct ib_uverbs_rereg_mr cmd;
1010 struct ib_uverbs_rereg_mr_resp resp;
1011 struct ib_udata udata;
1012 struct ib_pd *pd = NULL;
1013 struct ib_mr *mr;
1014 struct ib_pd *old_pd;
1015 int ret;
1016 struct ib_uobject *uobj;
1017
1018 if (out_len < sizeof(resp))
1019 return -ENOSPC;
1020
1021 if (copy_from_user(&cmd, buf, sizeof(cmd)))
1022 return -EFAULT;
1023
1024 INIT_UDATA(&udata, buf + sizeof(cmd),
1025 (unsigned long) cmd.response + sizeof(resp),
1026 in_len - sizeof(cmd), out_len - sizeof(resp));
1027
1028 if (cmd.flags & ~IB_MR_REREG_SUPPORTED || !cmd.flags)
1029 return -EINVAL;
1030
1031 if ((cmd.flags & IB_MR_REREG_TRANS) &&
1032 (!cmd.start || !cmd.hca_va || 0 >= cmd.length ||
1033 (cmd.start & ~PAGE_MASK) != (cmd.hca_va & ~PAGE_MASK)))
1034 return -EINVAL;
1035
1036 uobj = idr_write_uobj(&ib_uverbs_mr_idr, cmd.mr_handle,
1037 file->ucontext);
1038
1039 if (!uobj)
1040 return -EINVAL;
1041
1042 mr = uobj->object;
1043
1044 if (cmd.flags & IB_MR_REREG_ACCESS) {
1045 ret = ib_check_mr_access(cmd.access_flags);
1046 if (ret)
1047 goto put_uobjs;
1048 }
1049
1050 if (cmd.flags & IB_MR_REREG_PD) {
1051 pd = idr_read_pd(cmd.pd_handle, file->ucontext);
1052 if (!pd) {
1053 ret = -EINVAL;
1054 goto put_uobjs;
1055 }
1056 }
1057
1058 if (atomic_read(&mr->usecnt)) {
1059 ret = -EBUSY;
1060 goto put_uobj_pd;
1061 }
1062
1063 old_pd = mr->pd;
1064 ret = mr->device->rereg_user_mr(mr, cmd.flags, cmd.start,
1065 cmd.length, cmd.hca_va,
1066 cmd.access_flags, pd, &udata);
1067 if (!ret) {
1068 if (cmd.flags & IB_MR_REREG_PD) {
1069 atomic_inc(&pd->usecnt);
1070 mr->pd = pd;
1071 atomic_dec(&old_pd->usecnt);
1072 }
1073 } else {
1074 goto put_uobj_pd;
1075 }
1076
1077 memset(&resp, 0, sizeof(resp));
1078 resp.lkey = mr->lkey;
1079 resp.rkey = mr->rkey;
1080
1081 if (copy_to_user((void __user *)(unsigned long)cmd.response,
1082 &resp, sizeof(resp)))
1083 ret = -EFAULT;
1084 else
1085 ret = in_len;
1086
1087put_uobj_pd:
1088 if (cmd.flags & IB_MR_REREG_PD)
1089 put_pd_read(pd);
1090
1091put_uobjs:
1092
1093 put_uobj_write(mr->uobject);
1094
1095 return ret;
1096}
1097
1005ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file, 1098ssize_t ib_uverbs_dereg_mr(struct ib_uverbs_file *file,
1006 const char __user *buf, int in_len, 1099 const char __user *buf, int in_len,
1007 int out_len) 1100 int out_len)
diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index 08219fb3338b..c73b22a257fe 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -87,6 +87,7 @@ static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
87 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd, 87 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
88 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd, 88 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
89 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr, 89 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
90 [IB_USER_VERBS_CMD_REREG_MR] = ib_uverbs_rereg_mr,
90 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr, 91 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
91 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw, 92 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw,
92 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw, 93 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw,
diff --git a/drivers/infiniband/hw/amso1100/c2_cq.c b/drivers/infiniband/hw/amso1100/c2_cq.c
index 49e0e8533f74..1b63185b4ad4 100644
--- a/drivers/infiniband/hw/amso1100/c2_cq.c
+++ b/drivers/infiniband/hw/amso1100/c2_cq.c
@@ -260,11 +260,14 @@ static void c2_free_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq)
260 mq->msg_pool.host, dma_unmap_addr(mq, mapping)); 260 mq->msg_pool.host, dma_unmap_addr(mq, mapping));
261} 261}
262 262
263static int c2_alloc_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq, int q_size, 263static int c2_alloc_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq,
264 int msg_size) 264 size_t q_size, size_t msg_size)
265{ 265{
266 u8 *pool_start; 266 u8 *pool_start;
267 267
268 if (q_size > SIZE_MAX / msg_size)
269 return -EINVAL;
270
268 pool_start = dma_alloc_coherent(&c2dev->pcidev->dev, q_size * msg_size, 271 pool_start = dma_alloc_coherent(&c2dev->pcidev->dev, q_size * msg_size,
269 &mq->host_dma, GFP_KERNEL); 272 &mq->host_dma, GFP_KERNEL);
270 if (!pool_start) 273 if (!pool_start)
diff --git a/drivers/infiniband/hw/cxgb4/ev.c b/drivers/infiniband/hw/cxgb4/ev.c
index fbe6051af254..c9df0549f51d 100644
--- a/drivers/infiniband/hw/cxgb4/ev.c
+++ b/drivers/infiniband/hw/cxgb4/ev.c
@@ -227,6 +227,7 @@ int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid)
227 227
228 chp = get_chp(dev, qid); 228 chp = get_chp(dev, qid);
229 if (chp) { 229 if (chp) {
230 t4_clear_cq_armed(&chp->cq);
230 spin_lock_irqsave(&chp->comp_handler_lock, flag); 231 spin_lock_irqsave(&chp->comp_handler_lock, flag);
231 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context); 232 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
232 spin_unlock_irqrestore(&chp->comp_handler_lock, flag); 233 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index c158fcc02bca..41cd6882b648 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -1105,7 +1105,7 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
1105 struct c4iw_cq *schp) 1105 struct c4iw_cq *schp)
1106{ 1106{
1107 int count; 1107 int count;
1108 int flushed; 1108 int rq_flushed, sq_flushed;
1109 unsigned long flag; 1109 unsigned long flag;
1110 1110
1111 PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp); 1111 PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
@@ -1123,27 +1123,40 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
1123 1123
1124 c4iw_flush_hw_cq(rchp); 1124 c4iw_flush_hw_cq(rchp);
1125 c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count); 1125 c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count);
1126 flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count); 1126 rq_flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
1127 spin_unlock(&qhp->lock); 1127 spin_unlock(&qhp->lock);
1128 spin_unlock_irqrestore(&rchp->lock, flag); 1128 spin_unlock_irqrestore(&rchp->lock, flag);
1129 if (flushed) {
1130 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1131 (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
1132 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1133 }
1134 1129
1135 /* locking hierarchy: cq lock first, then qp lock. */ 1130 /* locking hierarchy: cq lock first, then qp lock. */
1136 spin_lock_irqsave(&schp->lock, flag); 1131 spin_lock_irqsave(&schp->lock, flag);
1137 spin_lock(&qhp->lock); 1132 spin_lock(&qhp->lock);
1138 if (schp != rchp) 1133 if (schp != rchp)
1139 c4iw_flush_hw_cq(schp); 1134 c4iw_flush_hw_cq(schp);
1140 flushed = c4iw_flush_sq(qhp); 1135 sq_flushed = c4iw_flush_sq(qhp);
1141 spin_unlock(&qhp->lock); 1136 spin_unlock(&qhp->lock);
1142 spin_unlock_irqrestore(&schp->lock, flag); 1137 spin_unlock_irqrestore(&schp->lock, flag);
1143 if (flushed) { 1138
1144 spin_lock_irqsave(&schp->comp_handler_lock, flag); 1139 if (schp == rchp) {
1145 (*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context); 1140 if (t4_clear_cq_armed(&rchp->cq) &&
1146 spin_unlock_irqrestore(&schp->comp_handler_lock, flag); 1141 (rq_flushed || sq_flushed)) {
1142 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1143 (*rchp->ibcq.comp_handler)(&rchp->ibcq,
1144 rchp->ibcq.cq_context);
1145 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1146 }
1147 } else {
1148 if (t4_clear_cq_armed(&rchp->cq) && rq_flushed) {
1149 spin_lock_irqsave(&rchp->comp_handler_lock, flag);
1150 (*rchp->ibcq.comp_handler)(&rchp->ibcq,
1151 rchp->ibcq.cq_context);
1152 spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
1153 }
1154 if (t4_clear_cq_armed(&schp->cq) && sq_flushed) {
1155 spin_lock_irqsave(&schp->comp_handler_lock, flag);
1156 (*schp->ibcq.comp_handler)(&schp->ibcq,
1157 schp->ibcq.cq_context);
1158 spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
1159 }
1147 } 1160 }
1148} 1161}
1149 1162
diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h
index df5edfa31a8f..c04e5134b30c 100644
--- a/drivers/infiniband/hw/cxgb4/t4.h
+++ b/drivers/infiniband/hw/cxgb4/t4.h
@@ -524,6 +524,10 @@ static inline int t4_wq_db_enabled(struct t4_wq *wq)
524 return !wq->rq.queue[wq->rq.size].status.db_off; 524 return !wq->rq.queue[wq->rq.size].status.db_off;
525} 525}
526 526
527enum t4_cq_flags {
528 CQ_ARMED = 1,
529};
530
527struct t4_cq { 531struct t4_cq {
528 struct t4_cqe *queue; 532 struct t4_cqe *queue;
529 dma_addr_t dma_addr; 533 dma_addr_t dma_addr;
@@ -544,12 +548,19 @@ struct t4_cq {
544 u16 cidx_inc; 548 u16 cidx_inc;
545 u8 gen; 549 u8 gen;
546 u8 error; 550 u8 error;
551 unsigned long flags;
547}; 552};
548 553
554static inline int t4_clear_cq_armed(struct t4_cq *cq)
555{
556 return test_and_clear_bit(CQ_ARMED, &cq->flags);
557}
558
549static inline int t4_arm_cq(struct t4_cq *cq, int se) 559static inline int t4_arm_cq(struct t4_cq *cq, int se)
550{ 560{
551 u32 val; 561 u32 val;
552 562
563 set_bit(CQ_ARMED, &cq->flags);
553 while (cq->cidx_inc > CIDXINC_MASK) { 564 while (cq->cidx_inc > CIDXINC_MASK) {
554 val = SEINTARM(0) | CIDXINC(CIDXINC_MASK) | TIMERREG(7) | 565 val = SEINTARM(0) | CIDXINC(CIDXINC_MASK) | TIMERREG(7) |
555 INGRESSQID(cq->cqid); 566 INGRESSQID(cq->cqid);
diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c
index 43f2d0424d4f..e890e5ba0e01 100644
--- a/drivers/infiniband/hw/ipath/ipath_mad.c
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c
@@ -726,7 +726,7 @@ bail:
726 * @dd: the infinipath device 726 * @dd: the infinipath device
727 * @pkeys: the PKEY table 727 * @pkeys: the PKEY table
728 */ 728 */
729static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys) 729static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys, u8 port)
730{ 730{
731 struct ipath_portdata *pd; 731 struct ipath_portdata *pd;
732 int i; 732 int i;
@@ -759,6 +759,7 @@ static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
759 } 759 }
760 if (changed) { 760 if (changed) {
761 u64 pkey; 761 u64 pkey;
762 struct ib_event event;
762 763
763 pkey = (u64) dd->ipath_pkeys[0] | 764 pkey = (u64) dd->ipath_pkeys[0] |
764 ((u64) dd->ipath_pkeys[1] << 16) | 765 ((u64) dd->ipath_pkeys[1] << 16) |
@@ -768,12 +769,17 @@ static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
768 (unsigned long long) pkey); 769 (unsigned long long) pkey);
769 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey, 770 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
770 pkey); 771 pkey);
772
773 event.event = IB_EVENT_PKEY_CHANGE;
774 event.device = &dd->verbs_dev->ibdev;
775 event.element.port_num = port;
776 ib_dispatch_event(&event);
771 } 777 }
772 return 0; 778 return 0;
773} 779}
774 780
775static int recv_subn_set_pkeytable(struct ib_smp *smp, 781static int recv_subn_set_pkeytable(struct ib_smp *smp,
776 struct ib_device *ibdev) 782 struct ib_device *ibdev, u8 port)
777{ 783{
778 u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff); 784 u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
779 __be16 *p = (__be16 *) smp->data; 785 __be16 *p = (__be16 *) smp->data;
@@ -784,7 +790,7 @@ static int recv_subn_set_pkeytable(struct ib_smp *smp,
784 for (i = 0; i < n; i++) 790 for (i = 0; i < n; i++)
785 q[i] = be16_to_cpu(p[i]); 791 q[i] = be16_to_cpu(p[i]);
786 792
787 if (startpx != 0 || set_pkeys(dev->dd, q) != 0) 793 if (startpx != 0 || set_pkeys(dev->dd, q, port) != 0)
788 smp->status |= IB_SMP_INVALID_FIELD; 794 smp->status |= IB_SMP_INVALID_FIELD;
789 795
790 return recv_subn_get_pkeytable(smp, ibdev); 796 return recv_subn_get_pkeytable(smp, ibdev);
@@ -1342,7 +1348,7 @@ static int process_subn(struct ib_device *ibdev, int mad_flags,
1342 ret = recv_subn_set_portinfo(smp, ibdev, port_num); 1348 ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1343 goto bail; 1349 goto bail;
1344 case IB_SMP_ATTR_PKEY_TABLE: 1350 case IB_SMP_ATTR_PKEY_TABLE:
1345 ret = recv_subn_set_pkeytable(smp, ibdev); 1351 ret = recv_subn_set_pkeytable(smp, ibdev, port_num);
1346 goto bail; 1352 goto bail;
1347 case IB_SMP_ATTR_SM_INFO: 1353 case IB_SMP_ATTR_SM_INFO:
1348 if (dev->port_cap_flags & IB_PORT_SM_DISABLED) { 1354 if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 287ad0564acd..82a7dd87089b 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -891,7 +891,7 @@ int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
891 agent = ib_register_mad_agent(&dev->ib_dev, p + 1, 891 agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
892 q ? IB_QPT_GSI : IB_QPT_SMI, 892 q ? IB_QPT_GSI : IB_QPT_SMI,
893 NULL, 0, send_handler, 893 NULL, 0, send_handler,
894 NULL, NULL); 894 NULL, NULL, 0);
895 if (IS_ERR(agent)) { 895 if (IS_ERR(agent)) {
896 ret = PTR_ERR(agent); 896 ret = PTR_ERR(agent);
897 goto err; 897 goto err;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 0f7027e7db13..e1e558a3d692 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -910,8 +910,7 @@ static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
910 const struct default_rules *pdefault_rules = default_table; 910 const struct default_rules *pdefault_rules = default_table;
911 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port); 911 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
912 912
913 for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++, 913 for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
914 pdefault_rules++) {
915 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS]; 914 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
916 memset(&field_types, 0, sizeof(field_types)); 915 memset(&field_types, 0, sizeof(field_types));
917 916
@@ -965,8 +964,7 @@ static int __mlx4_ib_create_default_rules(
965 int size = 0; 964 int size = 0;
966 int i; 965 int i;
967 966
968 for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/ 967 for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
969 sizeof(pdefault_rules->rules_create_list[0]); i++) {
970 int ret; 968 int ret;
971 union ib_flow_spec ib_spec; 969 union ib_flow_spec ib_spec;
972 switch (pdefault_rules->rules_create_list[i]) { 970 switch (pdefault_rules->rules_create_list[i]) {
@@ -2007,6 +2005,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
2007 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | 2005 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2008 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | 2006 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2009 (1ull << IB_USER_VERBS_CMD_REG_MR) | 2007 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2008 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
2010 (1ull << IB_USER_VERBS_CMD_DEREG_MR) | 2009 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2011 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | 2010 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2012 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | 2011 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
@@ -2059,6 +2058,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
2059 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq; 2058 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2060 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr; 2059 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2061 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr; 2060 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2061 ibdev->ib_dev.rereg_user_mr = mlx4_ib_rereg_user_mr;
2062 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr; 2062 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2063 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr; 2063 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2064 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list; 2064 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 369da3ca5d64..e8cad3926bfc 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -788,5 +788,9 @@ int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn);
788void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count); 788void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count);
789int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, 789int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
790 int is_attach); 790 int is_attach);
791int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
792 u64 start, u64 length, u64 virt_addr,
793 int mr_access_flags, struct ib_pd *pd,
794 struct ib_udata *udata);
791 795
792#endif /* MLX4_IB_H */ 796#endif /* MLX4_IB_H */
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index cb2a8727f3fb..9b0e80e59b08 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -144,8 +144,10 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
144 if (!mr) 144 if (!mr)
145 return ERR_PTR(-ENOMEM); 145 return ERR_PTR(-ENOMEM);
146 146
147 /* Force registering the memory as writable. */
148 /* Used for memory re-registeration. HCA protects the access */
147 mr->umem = ib_umem_get(pd->uobject->context, start, length, 149 mr->umem = ib_umem_get(pd->uobject->context, start, length,
148 access_flags, 0); 150 access_flags | IB_ACCESS_LOCAL_WRITE, 0);
149 if (IS_ERR(mr->umem)) { 151 if (IS_ERR(mr->umem)) {
150 err = PTR_ERR(mr->umem); 152 err = PTR_ERR(mr->umem);
151 goto err_free; 153 goto err_free;
@@ -183,6 +185,90 @@ err_free:
183 return ERR_PTR(err); 185 return ERR_PTR(err);
184} 186}
185 187
188int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
189 u64 start, u64 length, u64 virt_addr,
190 int mr_access_flags, struct ib_pd *pd,
191 struct ib_udata *udata)
192{
193 struct mlx4_ib_dev *dev = to_mdev(mr->device);
194 struct mlx4_ib_mr *mmr = to_mmr(mr);
195 struct mlx4_mpt_entry *mpt_entry;
196 struct mlx4_mpt_entry **pmpt_entry = &mpt_entry;
197 int err;
198
199 /* Since we synchronize this call and mlx4_ib_dereg_mr via uverbs,
200 * we assume that the calls can't run concurrently. Otherwise, a
201 * race exists.
202 */
203 err = mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry);
204
205 if (err)
206 return err;
207
208 if (flags & IB_MR_REREG_PD) {
209 err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry,
210 to_mpd(pd)->pdn);
211
212 if (err)
213 goto release_mpt_entry;
214 }
215
216 if (flags & IB_MR_REREG_ACCESS) {
217 err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry,
218 convert_access(mr_access_flags));
219
220 if (err)
221 goto release_mpt_entry;
222 }
223
224 if (flags & IB_MR_REREG_TRANS) {
225 int shift;
226 int err;
227 int n;
228
229 mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
230 ib_umem_release(mmr->umem);
231 mmr->umem = ib_umem_get(mr->uobject->context, start, length,
232 mr_access_flags |
233 IB_ACCESS_LOCAL_WRITE,
234 0);
235 if (IS_ERR(mmr->umem)) {
236 err = PTR_ERR(mmr->umem);
237 mmr->umem = NULL;
238 goto release_mpt_entry;
239 }
240 n = ib_umem_page_count(mmr->umem);
241 shift = ilog2(mmr->umem->page_size);
242
243 mmr->mmr.iova = virt_addr;
244 mmr->mmr.size = length;
245 err = mlx4_mr_rereg_mem_write(dev->dev, &mmr->mmr,
246 virt_addr, length, n, shift,
247 *pmpt_entry);
248 if (err) {
249 ib_umem_release(mmr->umem);
250 goto release_mpt_entry;
251 }
252
253 err = mlx4_ib_umem_write_mtt(dev, &mmr->mmr.mtt, mmr->umem);
254 if (err) {
255 mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
256 ib_umem_release(mmr->umem);
257 goto release_mpt_entry;
258 }
259 }
260
261 /* If we couldn't transfer the MR to the HCA, just remember to
262 * return a failure. But dereg_mr will free the resources.
263 */
264 err = mlx4_mr_hw_write_mpt(dev->dev, &mmr->mmr, pmpt_entry);
265
266release_mpt_entry:
267 mlx4_mr_hw_put_mpt(dev->dev, pmpt_entry);
268
269 return err;
270}
271
186int mlx4_ib_dereg_mr(struct ib_mr *ibmr) 272int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
187{ 273{
188 struct mlx4_ib_mr *mr = to_mmr(ibmr); 274 struct mlx4_ib_mr *mr = to_mmr(ibmr);
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 7efe6e3f3542..8c574b63d77b 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -2501,7 +2501,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
2501 spin_lock_irqsave(&qp->sq.lock, flags); 2501 spin_lock_irqsave(&qp->sq.lock, flags);
2502 2502
2503 for (nreq = 0; wr; nreq++, wr = wr->next) { 2503 for (nreq = 0; wr; nreq++, wr = wr->next) {
2504 if (unlikely(wr->opcode >= sizeof(mlx5_ib_opcode) / sizeof(mlx5_ib_opcode[0]))) { 2504 if (unlikely(wr->opcode >= ARRAY_SIZE(mlx5_ib_opcode))) {
2505 mlx5_ib_warn(dev, "\n"); 2505 mlx5_ib_warn(dev, "\n");
2506 err = -EINVAL; 2506 err = -EINVAL;
2507 *bad_wr = wr; 2507 *bad_wr = wr;
diff --git a/drivers/infiniband/hw/mthca/mthca_mad.c b/drivers/infiniband/hw/mthca/mthca_mad.c
index b6f7f457fc55..8881fa376e06 100644
--- a/drivers/infiniband/hw/mthca/mthca_mad.c
+++ b/drivers/infiniband/hw/mthca/mthca_mad.c
@@ -294,7 +294,7 @@ int mthca_create_agents(struct mthca_dev *dev)
294 agent = ib_register_mad_agent(&dev->ib_dev, p + 1, 294 agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
295 q ? IB_QPT_GSI : IB_QPT_SMI, 295 q ? IB_QPT_GSI : IB_QPT_SMI,
296 NULL, 0, send_handler, 296 NULL, 0, send_handler,
297 NULL, NULL); 297 NULL, NULL, 0);
298 if (IS_ERR(agent)) { 298 if (IS_ERR(agent)) {
299 ret = PTR_ERR(agent); 299 ret = PTR_ERR(agent);
300 goto err; 300 goto err;
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
index 19011dbb930f..b43456ae124b 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
@@ -40,7 +40,7 @@
40#include <be_roce.h> 40#include <be_roce.h>
41#include "ocrdma_sli.h" 41#include "ocrdma_sli.h"
42 42
43#define OCRDMA_ROCE_DRV_VERSION "10.2.145.0u" 43#define OCRDMA_ROCE_DRV_VERSION "10.2.287.0u"
44 44
45#define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver" 45#define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver"
46#define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA" 46#define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA"
@@ -137,6 +137,7 @@ struct mqe_ctx {
137 u16 cqe_status; 137 u16 cqe_status;
138 u16 ext_status; 138 u16 ext_status;
139 bool cmd_done; 139 bool cmd_done;
140 bool fw_error_state;
140}; 141};
141 142
142struct ocrdma_hw_mr { 143struct ocrdma_hw_mr {
@@ -235,7 +236,10 @@ struct ocrdma_dev {
235 struct list_head entry; 236 struct list_head entry;
236 struct rcu_head rcu; 237 struct rcu_head rcu;
237 int id; 238 int id;
238 u64 stag_arr[OCRDMA_MAX_STAG]; 239 u64 *stag_arr;
240 u8 sl; /* service level */
241 bool pfc_state;
242 atomic_t update_sl;
239 u16 pvid; 243 u16 pvid;
240 u32 asic_id; 244 u32 asic_id;
241 245
@@ -518,4 +522,22 @@ static inline u8 ocrdma_get_asic_type(struct ocrdma_dev *dev)
518 OCRDMA_SLI_ASIC_GEN_NUM_SHIFT; 522 OCRDMA_SLI_ASIC_GEN_NUM_SHIFT;
519} 523}
520 524
525static inline u8 ocrdma_get_pfc_prio(u8 *pfc, u8 prio)
526{
527 return *(pfc + prio);
528}
529
530static inline u8 ocrdma_get_app_prio(u8 *app_prio, u8 prio)
531{
532 return *(app_prio + prio);
533}
534
535static inline u8 ocrdma_is_enabled_and_synced(u32 state)
536{ /* May also be used to interpret TC-state, QCN-state
537 * Appl-state and Logical-link-state in future.
538 */
539 return (state & OCRDMA_STATE_FLAG_ENABLED) &&
540 (state & OCRDMA_STATE_FLAG_SYNC);
541}
542
521#endif 543#endif
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
index d4cc01f10c01..40f8536c10b0 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
@@ -35,6 +35,8 @@
35#include "ocrdma_ah.h" 35#include "ocrdma_ah.h"
36#include "ocrdma_hw.h" 36#include "ocrdma_hw.h"
37 37
38#define OCRDMA_VID_PCP_SHIFT 0xD
39
38static inline int set_av_attr(struct ocrdma_dev *dev, struct ocrdma_ah *ah, 40static inline int set_av_attr(struct ocrdma_dev *dev, struct ocrdma_ah *ah,
39 struct ib_ah_attr *attr, int pdid) 41 struct ib_ah_attr *attr, int pdid)
40{ 42{
@@ -55,7 +57,7 @@ static inline int set_av_attr(struct ocrdma_dev *dev, struct ocrdma_ah *ah,
55 if (vlan_tag && (vlan_tag < 0x1000)) { 57 if (vlan_tag && (vlan_tag < 0x1000)) {
56 eth.eth_type = cpu_to_be16(0x8100); 58 eth.eth_type = cpu_to_be16(0x8100);
57 eth.roce_eth_type = cpu_to_be16(OCRDMA_ROCE_ETH_TYPE); 59 eth.roce_eth_type = cpu_to_be16(OCRDMA_ROCE_ETH_TYPE);
58 vlan_tag |= (attr->sl & 7) << 13; 60 vlan_tag |= (dev->sl & 0x07) << OCRDMA_VID_PCP_SHIFT;
59 eth.vlan_tag = cpu_to_be16(vlan_tag); 61 eth.vlan_tag = cpu_to_be16(vlan_tag);
60 eth_sz = sizeof(struct ocrdma_eth_vlan); 62 eth_sz = sizeof(struct ocrdma_eth_vlan);
61 vlan_enabled = true; 63 vlan_enabled = true;
@@ -100,6 +102,8 @@ struct ib_ah *ocrdma_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
100 if (!(attr->ah_flags & IB_AH_GRH)) 102 if (!(attr->ah_flags & IB_AH_GRH))
101 return ERR_PTR(-EINVAL); 103 return ERR_PTR(-EINVAL);
102 104
105 if (atomic_cmpxchg(&dev->update_sl, 1, 0))
106 ocrdma_init_service_level(dev);
103 ah = kzalloc(sizeof(*ah), GFP_ATOMIC); 107 ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
104 if (!ah) 108 if (!ah)
105 return ERR_PTR(-ENOMEM); 109 return ERR_PTR(-ENOMEM);
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
index 3bbf2010a821..dd35ae558ae1 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -525,7 +525,7 @@ static int ocrdma_mbx_mq_cq_create(struct ocrdma_dev *dev,
525 525
526 cmd->ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS; 526 cmd->ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
527 cmd->eqn = eq->id; 527 cmd->eqn = eq->id;
528 cmd->cqe_count = cq->size / sizeof(struct ocrdma_mcqe); 528 cmd->pdid_cqecnt = cq->size / sizeof(struct ocrdma_mcqe);
529 529
530 ocrdma_build_q_pages(&cmd->pa[0], cq->size / OCRDMA_MIN_Q_PAGE_SIZE, 530 ocrdma_build_q_pages(&cmd->pa[0], cq->size / OCRDMA_MIN_Q_PAGE_SIZE,
531 cq->dma, PAGE_SIZE_4K); 531 cq->dma, PAGE_SIZE_4K);
@@ -661,7 +661,7 @@ static void ocrdma_dispatch_ibevent(struct ocrdma_dev *dev,
661{ 661{
662 struct ocrdma_qp *qp = NULL; 662 struct ocrdma_qp *qp = NULL;
663 struct ocrdma_cq *cq = NULL; 663 struct ocrdma_cq *cq = NULL;
664 struct ib_event ib_evt = { 0 }; 664 struct ib_event ib_evt;
665 int cq_event = 0; 665 int cq_event = 0;
666 int qp_event = 1; 666 int qp_event = 1;
667 int srq_event = 0; 667 int srq_event = 0;
@@ -674,6 +674,8 @@ static void ocrdma_dispatch_ibevent(struct ocrdma_dev *dev,
674 if (cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQVALID) 674 if (cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQVALID)
675 cq = dev->cq_tbl[cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQID_MASK]; 675 cq = dev->cq_tbl[cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQID_MASK];
676 676
677 memset(&ib_evt, 0, sizeof(ib_evt));
678
677 ib_evt.device = &dev->ibdev; 679 ib_evt.device = &dev->ibdev;
678 680
679 switch (type) { 681 switch (type) {
@@ -771,6 +773,10 @@ static void ocrdma_process_grp5_aync(struct ocrdma_dev *dev,
771 OCRDMA_AE_PVID_MCQE_TAG_MASK) >> 773 OCRDMA_AE_PVID_MCQE_TAG_MASK) >>
772 OCRDMA_AE_PVID_MCQE_TAG_SHIFT); 774 OCRDMA_AE_PVID_MCQE_TAG_SHIFT);
773 break; 775 break;
776
777 case OCRDMA_ASYNC_EVENT_COS_VALUE:
778 atomic_set(&dev->update_sl, 1);
779 break;
774 default: 780 default:
775 /* Not interested evts. */ 781 /* Not interested evts. */
776 break; 782 break;
@@ -962,8 +968,12 @@ static int ocrdma_wait_mqe_cmpl(struct ocrdma_dev *dev)
962 msecs_to_jiffies(30000)); 968 msecs_to_jiffies(30000));
963 if (status) 969 if (status)
964 return 0; 970 return 0;
965 else 971 else {
972 dev->mqe_ctx.fw_error_state = true;
973 pr_err("%s(%d) mailbox timeout: fw not responding\n",
974 __func__, dev->id);
966 return -1; 975 return -1;
976 }
967} 977}
968 978
969/* issue a mailbox command on the MQ */ 979/* issue a mailbox command on the MQ */
@@ -975,6 +985,8 @@ static int ocrdma_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe)
975 struct ocrdma_mbx_rsp *rsp = NULL; 985 struct ocrdma_mbx_rsp *rsp = NULL;
976 986
977 mutex_lock(&dev->mqe_ctx.lock); 987 mutex_lock(&dev->mqe_ctx.lock);
988 if (dev->mqe_ctx.fw_error_state)
989 goto mbx_err;
978 ocrdma_post_mqe(dev, mqe); 990 ocrdma_post_mqe(dev, mqe);
979 status = ocrdma_wait_mqe_cmpl(dev); 991 status = ocrdma_wait_mqe_cmpl(dev);
980 if (status) 992 if (status)
@@ -1078,7 +1090,8 @@ static void ocrdma_get_attr(struct ocrdma_dev *dev,
1078 OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT; 1090 OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT;
1079 attr->max_mw = rsp->max_mw; 1091 attr->max_mw = rsp->max_mw;
1080 attr->max_mr = rsp->max_mr; 1092 attr->max_mr = rsp->max_mr;
1081 attr->max_mr_size = ~0ull; 1093 attr->max_mr_size = ((u64)rsp->max_mr_size_hi << 32) |
1094 rsp->max_mr_size_lo;
1082 attr->max_fmr = 0; 1095 attr->max_fmr = 0;
1083 attr->max_pages_per_frmr = rsp->max_pages_per_frmr; 1096 attr->max_pages_per_frmr = rsp->max_pages_per_frmr;
1084 attr->max_num_mr_pbl = rsp->max_num_mr_pbl; 1097 attr->max_num_mr_pbl = rsp->max_num_mr_pbl;
@@ -1252,7 +1265,9 @@ static int ocrdma_mbx_get_ctrl_attribs(struct ocrdma_dev *dev)
1252 ctrl_attr_rsp = (struct ocrdma_get_ctrl_attribs_rsp *)dma.va; 1265 ctrl_attr_rsp = (struct ocrdma_get_ctrl_attribs_rsp *)dma.va;
1253 hba_attribs = &ctrl_attr_rsp->ctrl_attribs.hba_attribs; 1266 hba_attribs = &ctrl_attr_rsp->ctrl_attribs.hba_attribs;
1254 1267
1255 dev->hba_port_num = hba_attribs->phy_port; 1268 dev->hba_port_num = (hba_attribs->ptpnum_maxdoms_hbast_cv &
1269 OCRDMA_HBA_ATTRB_PTNUM_MASK)
1270 >> OCRDMA_HBA_ATTRB_PTNUM_SHIFT;
1256 strncpy(dev->model_number, 1271 strncpy(dev->model_number,
1257 hba_attribs->controller_model_number, 31); 1272 hba_attribs->controller_model_number, 31);
1258 } 1273 }
@@ -1302,7 +1317,8 @@ int ocrdma_mbx_get_link_speed(struct ocrdma_dev *dev, u8 *lnk_speed)
1302 goto mbx_err; 1317 goto mbx_err;
1303 1318
1304 rsp = (struct ocrdma_get_link_speed_rsp *)cmd; 1319 rsp = (struct ocrdma_get_link_speed_rsp *)cmd;
1305 *lnk_speed = rsp->phys_port_speed; 1320 *lnk_speed = (rsp->pflt_pps_ld_pnum & OCRDMA_PHY_PS_MASK)
1321 >> OCRDMA_PHY_PS_SHIFT;
1306 1322
1307mbx_err: 1323mbx_err:
1308 kfree(cmd); 1324 kfree(cmd);
@@ -1328,11 +1344,16 @@ static int ocrdma_mbx_get_phy_info(struct ocrdma_dev *dev)
1328 goto mbx_err; 1344 goto mbx_err;
1329 1345
1330 rsp = (struct ocrdma_get_phy_info_rsp *)cmd; 1346 rsp = (struct ocrdma_get_phy_info_rsp *)cmd;
1331 dev->phy.phy_type = le16_to_cpu(rsp->phy_type); 1347 dev->phy.phy_type =
1348 (rsp->ityp_ptyp & OCRDMA_PHY_TYPE_MASK);
1349 dev->phy.interface_type =
1350 (rsp->ityp_ptyp & OCRDMA_IF_TYPE_MASK)
1351 >> OCRDMA_IF_TYPE_SHIFT;
1332 dev->phy.auto_speeds_supported = 1352 dev->phy.auto_speeds_supported =
1333 le16_to_cpu(rsp->auto_speeds_supported); 1353 (rsp->fspeed_aspeed & OCRDMA_ASPEED_SUPP_MASK);
1334 dev->phy.fixed_speeds_supported = 1354 dev->phy.fixed_speeds_supported =
1335 le16_to_cpu(rsp->fixed_speeds_supported); 1355 (rsp->fspeed_aspeed & OCRDMA_FSPEED_SUPP_MASK)
1356 >> OCRDMA_FSPEED_SUPP_SHIFT;
1336mbx_err: 1357mbx_err:
1337 kfree(cmd); 1358 kfree(cmd);
1338 return status; 1359 return status;
@@ -1457,8 +1478,8 @@ static int ocrdma_mbx_create_ah_tbl(struct ocrdma_dev *dev)
1457 1478
1458 pbes = (struct ocrdma_pbe *)dev->av_tbl.pbl.va; 1479 pbes = (struct ocrdma_pbe *)dev->av_tbl.pbl.va;
1459 for (i = 0; i < dev->av_tbl.size / OCRDMA_MIN_Q_PAGE_SIZE; i++) { 1480 for (i = 0; i < dev->av_tbl.size / OCRDMA_MIN_Q_PAGE_SIZE; i++) {
1460 pbes[i].pa_lo = (u32) (pa & 0xffffffff); 1481 pbes[i].pa_lo = (u32)cpu_to_le32(pa & 0xffffffff);
1461 pbes[i].pa_hi = (u32) upper_32_bits(pa); 1482 pbes[i].pa_hi = (u32)cpu_to_le32(upper_32_bits(pa));
1462 pa += PAGE_SIZE; 1483 pa += PAGE_SIZE;
1463 } 1484 }
1464 cmd->tbl_addr[0].lo = (u32)(dev->av_tbl.pbl.pa & 0xFFFFFFFF); 1485 cmd->tbl_addr[0].lo = (u32)(dev->av_tbl.pbl.pa & 0xFFFFFFFF);
@@ -1501,6 +1522,7 @@ static void ocrdma_mbx_delete_ah_tbl(struct ocrdma_dev *dev)
1501 ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd); 1522 ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1502 dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va, 1523 dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
1503 dev->av_tbl.pa); 1524 dev->av_tbl.pa);
1525 dev->av_tbl.va = NULL;
1504 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va, 1526 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
1505 dev->av_tbl.pbl.pa); 1527 dev->av_tbl.pbl.pa);
1506 kfree(cmd); 1528 kfree(cmd);
@@ -1624,14 +1646,16 @@ int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
1624 cmd->cmd.pgsz_pgcnt |= OCRDMA_CREATE_CQ_DPP << 1646 cmd->cmd.pgsz_pgcnt |= OCRDMA_CREATE_CQ_DPP <<
1625 OCRDMA_CREATE_CQ_TYPE_SHIFT; 1647 OCRDMA_CREATE_CQ_TYPE_SHIFT;
1626 cq->phase_change = false; 1648 cq->phase_change = false;
1627 cmd->cmd.cqe_count = (cq->len / cqe_size); 1649 cmd->cmd.pdid_cqecnt = (cq->len / cqe_size);
1628 } else { 1650 } else {
1629 cmd->cmd.cqe_count = (cq->len / cqe_size) - 1; 1651 cmd->cmd.pdid_cqecnt = (cq->len / cqe_size) - 1;
1630 cmd->cmd.ev_cnt_flags |= OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID; 1652 cmd->cmd.ev_cnt_flags |= OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID;
1631 cq->phase_change = true; 1653 cq->phase_change = true;
1632 } 1654 }
1633 1655
1634 cmd->cmd.pd_id = pd_id; /* valid only for v3 */ 1656 /* pd_id valid only for v3 */
1657 cmd->cmd.pdid_cqecnt |= (pd_id <<
1658 OCRDMA_CREATE_CQ_CMD_PDID_SHIFT);
1635 ocrdma_build_q_pages(&cmd->cmd.pa[0], hw_pages, cq->pa, page_size); 1659 ocrdma_build_q_pages(&cmd->cmd.pa[0], hw_pages, cq->pa, page_size);
1636 status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd); 1660 status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1637 if (status) 1661 if (status)
@@ -2206,7 +2230,8 @@ int ocrdma_mbx_create_qp(struct ocrdma_qp *qp, struct ib_qp_init_attr *attrs,
2206 OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK; 2230 OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK;
2207 qp->rq_cq = cq; 2231 qp->rq_cq = cq;
2208 2232
2209 if (pd->dpp_enabled && pd->num_dpp_qp) { 2233 if (pd->dpp_enabled && attrs->cap.max_inline_data && pd->num_dpp_qp &&
2234 (attrs->cap.max_inline_data <= dev->attr.max_inline_data)) {
2210 ocrdma_set_create_qp_dpp_cmd(cmd, pd, qp, enable_dpp_cq, 2235 ocrdma_set_create_qp_dpp_cmd(cmd, pd, qp, enable_dpp_cq,
2211 dpp_cq_id); 2236 dpp_cq_id);
2212 } 2237 }
@@ -2264,6 +2289,8 @@ static int ocrdma_set_av_params(struct ocrdma_qp *qp,
2264 2289
2265 if ((ah_attr->ah_flags & IB_AH_GRH) == 0) 2290 if ((ah_attr->ah_flags & IB_AH_GRH) == 0)
2266 return -EINVAL; 2291 return -EINVAL;
2292 if (atomic_cmpxchg(&qp->dev->update_sl, 1, 0))
2293 ocrdma_init_service_level(qp->dev);
2267 cmd->params.tclass_sq_psn |= 2294 cmd->params.tclass_sq_psn |=
2268 (ah_attr->grh.traffic_class << OCRDMA_QP_PARAMS_TCLASS_SHIFT); 2295 (ah_attr->grh.traffic_class << OCRDMA_QP_PARAMS_TCLASS_SHIFT);
2269 cmd->params.rnt_rc_sl_fl |= 2296 cmd->params.rnt_rc_sl_fl |=
@@ -2297,6 +2324,8 @@ static int ocrdma_set_av_params(struct ocrdma_qp *qp,
2297 cmd->params.vlan_dmac_b4_to_b5 |= 2324 cmd->params.vlan_dmac_b4_to_b5 |=
2298 vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT; 2325 vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT;
2299 cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID; 2326 cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID;
2327 cmd->params.rnt_rc_sl_fl |=
2328 (qp->dev->sl & 0x07) << OCRDMA_QP_PARAMS_SL_SHIFT;
2300 } 2329 }
2301 return 0; 2330 return 0;
2302} 2331}
@@ -2604,6 +2633,168 @@ int ocrdma_mbx_destroy_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq)
2604 return status; 2633 return status;
2605} 2634}
2606 2635
2636static int ocrdma_mbx_get_dcbx_config(struct ocrdma_dev *dev, u32 ptype,
2637 struct ocrdma_dcbx_cfg *dcbxcfg)
2638{
2639 int status = 0;
2640 dma_addr_t pa;
2641 struct ocrdma_mqe cmd;
2642
2643 struct ocrdma_get_dcbx_cfg_req *req = NULL;
2644 struct ocrdma_get_dcbx_cfg_rsp *rsp = NULL;
2645 struct pci_dev *pdev = dev->nic_info.pdev;
2646 struct ocrdma_mqe_sge *mqe_sge = cmd.u.nonemb_req.sge;
2647
2648 memset(&cmd, 0, sizeof(struct ocrdma_mqe));
2649 cmd.hdr.pyld_len = max_t (u32, sizeof(struct ocrdma_get_dcbx_cfg_rsp),
2650 sizeof(struct ocrdma_get_dcbx_cfg_req));
2651 req = dma_alloc_coherent(&pdev->dev, cmd.hdr.pyld_len, &pa, GFP_KERNEL);
2652 if (!req) {
2653 status = -ENOMEM;
2654 goto mem_err;
2655 }
2656
2657 cmd.hdr.spcl_sge_cnt_emb |= (1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
2658 OCRDMA_MQE_HDR_SGE_CNT_MASK;
2659 mqe_sge->pa_lo = (u32) (pa & 0xFFFFFFFFUL);
2660 mqe_sge->pa_hi = (u32) upper_32_bits(pa);
2661 mqe_sge->len = cmd.hdr.pyld_len;
2662
2663 memset(req, 0, sizeof(struct ocrdma_get_dcbx_cfg_req));
2664 ocrdma_init_mch(&req->hdr, OCRDMA_CMD_GET_DCBX_CONFIG,
2665 OCRDMA_SUBSYS_DCBX, cmd.hdr.pyld_len);
2666 req->param_type = ptype;
2667
2668 status = ocrdma_mbx_cmd(dev, &cmd);
2669 if (status)
2670 goto mbx_err;
2671
2672 rsp = (struct ocrdma_get_dcbx_cfg_rsp *)req;
2673 ocrdma_le32_to_cpu(rsp, sizeof(struct ocrdma_get_dcbx_cfg_rsp));
2674 memcpy(dcbxcfg, &rsp->cfg, sizeof(struct ocrdma_dcbx_cfg));
2675
2676mbx_err:
2677 dma_free_coherent(&pdev->dev, cmd.hdr.pyld_len, req, pa);
2678mem_err:
2679 return status;
2680}
2681
2682#define OCRDMA_MAX_SERVICE_LEVEL_INDEX 0x08
2683#define OCRDMA_DEFAULT_SERVICE_LEVEL 0x05
2684
2685static int ocrdma_parse_dcbxcfg_rsp(struct ocrdma_dev *dev, int ptype,
2686 struct ocrdma_dcbx_cfg *dcbxcfg,
2687 u8 *srvc_lvl)
2688{
2689 int status = -EINVAL, indx, slindx;
2690 int ventry_cnt;
2691 struct ocrdma_app_parameter *app_param;
2692 u8 valid, proto_sel;
2693 u8 app_prio, pfc_prio;
2694 u16 proto;
2695
2696 if (!(dcbxcfg->tcv_aev_opv_st & OCRDMA_DCBX_STATE_MASK)) {
2697 pr_info("%s ocrdma%d DCBX is disabled\n",
2698 dev_name(&dev->nic_info.pdev->dev), dev->id);
2699 goto out;
2700 }
2701
2702 if (!ocrdma_is_enabled_and_synced(dcbxcfg->pfc_state)) {
2703 pr_info("%s ocrdma%d priority flow control(%s) is %s%s\n",
2704 dev_name(&dev->nic_info.pdev->dev), dev->id,
2705 (ptype > 0 ? "operational" : "admin"),
2706 (dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_ENABLED) ?
2707 "enabled" : "disabled",
2708 (dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_SYNC) ?
2709 "" : ", not sync'ed");
2710 goto out;
2711 } else {
2712 pr_info("%s ocrdma%d priority flow control is enabled and sync'ed\n",
2713 dev_name(&dev->nic_info.pdev->dev), dev->id);
2714 }
2715
2716 ventry_cnt = (dcbxcfg->tcv_aev_opv_st >>
2717 OCRDMA_DCBX_APP_ENTRY_SHIFT)
2718 & OCRDMA_DCBX_STATE_MASK;
2719
2720 for (indx = 0; indx < ventry_cnt; indx++) {
2721 app_param = &dcbxcfg->app_param[indx];
2722 valid = (app_param->valid_proto_app >>
2723 OCRDMA_APP_PARAM_VALID_SHIFT)
2724 & OCRDMA_APP_PARAM_VALID_MASK;
2725 proto_sel = (app_param->valid_proto_app
2726 >> OCRDMA_APP_PARAM_PROTO_SEL_SHIFT)
2727 & OCRDMA_APP_PARAM_PROTO_SEL_MASK;
2728 proto = app_param->valid_proto_app &
2729 OCRDMA_APP_PARAM_APP_PROTO_MASK;
2730
2731 if (
2732 valid && proto == OCRDMA_APP_PROTO_ROCE &&
2733 proto_sel == OCRDMA_PROTO_SELECT_L2) {
2734 for (slindx = 0; slindx <
2735 OCRDMA_MAX_SERVICE_LEVEL_INDEX; slindx++) {
2736 app_prio = ocrdma_get_app_prio(
2737 (u8 *)app_param->app_prio,
2738 slindx);
2739 pfc_prio = ocrdma_get_pfc_prio(
2740 (u8 *)dcbxcfg->pfc_prio,
2741 slindx);
2742
2743 if (app_prio && pfc_prio) {
2744 *srvc_lvl = slindx;
2745 status = 0;
2746 goto out;
2747 }
2748 }
2749 if (slindx == OCRDMA_MAX_SERVICE_LEVEL_INDEX) {
2750 pr_info("%s ocrdma%d application priority not set for 0x%x protocol\n",
2751 dev_name(&dev->nic_info.pdev->dev),
2752 dev->id, proto);
2753 }
2754 }
2755 }
2756
2757out:
2758 return status;
2759}
2760
2761void ocrdma_init_service_level(struct ocrdma_dev *dev)
2762{
2763 int status = 0, indx;
2764 struct ocrdma_dcbx_cfg dcbxcfg;
2765 u8 srvc_lvl = OCRDMA_DEFAULT_SERVICE_LEVEL;
2766 int ptype = OCRDMA_PARAMETER_TYPE_OPER;
2767
2768 for (indx = 0; indx < 2; indx++) {
2769 status = ocrdma_mbx_get_dcbx_config(dev, ptype, &dcbxcfg);
2770 if (status) {
2771 pr_err("%s(): status=%d\n", __func__, status);
2772 ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
2773 continue;
2774 }
2775
2776 status = ocrdma_parse_dcbxcfg_rsp(dev, ptype,
2777 &dcbxcfg, &srvc_lvl);
2778 if (status) {
2779 ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
2780 continue;
2781 }
2782
2783 break;
2784 }
2785
2786 if (status)
2787 pr_info("%s ocrdma%d service level default\n",
2788 dev_name(&dev->nic_info.pdev->dev), dev->id);
2789 else
2790 pr_info("%s ocrdma%d service level %d\n",
2791 dev_name(&dev->nic_info.pdev->dev), dev->id,
2792 srvc_lvl);
2793
2794 dev->pfc_state = ocrdma_is_enabled_and_synced(dcbxcfg.pfc_state);
2795 dev->sl = srvc_lvl;
2796}
2797
2607int ocrdma_alloc_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah) 2798int ocrdma_alloc_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
2608{ 2799{
2609 int i; 2800 int i;
@@ -2709,13 +2900,15 @@ int ocrdma_init_hw(struct ocrdma_dev *dev)
2709 goto conf_err; 2900 goto conf_err;
2710 status = ocrdma_mbx_get_phy_info(dev); 2901 status = ocrdma_mbx_get_phy_info(dev);
2711 if (status) 2902 if (status)
2712 goto conf_err; 2903 goto info_attrb_err;
2713 status = ocrdma_mbx_get_ctrl_attribs(dev); 2904 status = ocrdma_mbx_get_ctrl_attribs(dev);
2714 if (status) 2905 if (status)
2715 goto conf_err; 2906 goto info_attrb_err;
2716 2907
2717 return 0; 2908 return 0;
2718 2909
2910info_attrb_err:
2911 ocrdma_mbx_delete_ah_tbl(dev);
2719conf_err: 2912conf_err:
2720 ocrdma_destroy_mq(dev); 2913 ocrdma_destroy_mq(dev);
2721mq_err: 2914mq_err:
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.h b/drivers/infiniband/hw/ocrdma/ocrdma_hw.h
index e513f7293142..6eed8f191322 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_hw.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.h
@@ -135,4 +135,6 @@ int ocrdma_get_irq(struct ocrdma_dev *dev, struct ocrdma_eq *eq);
135 135
136int ocrdma_mbx_rdma_stats(struct ocrdma_dev *, bool reset); 136int ocrdma_mbx_rdma_stats(struct ocrdma_dev *, bool reset);
137char *port_speed_string(struct ocrdma_dev *dev); 137char *port_speed_string(struct ocrdma_dev *dev);
138void ocrdma_init_service_level(struct ocrdma_dev *);
139
138#endif /* __OCRDMA_HW_H__ */ 140#endif /* __OCRDMA_HW_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_main.c b/drivers/infiniband/hw/ocrdma/ocrdma_main.c
index 7c504e079744..256a06bc0b68 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_main.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_main.c
@@ -324,6 +324,11 @@ static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
324 if (!dev->qp_tbl) 324 if (!dev->qp_tbl)
325 goto alloc_err; 325 goto alloc_err;
326 } 326 }
327
328 dev->stag_arr = kzalloc(sizeof(u64) * OCRDMA_MAX_STAG, GFP_KERNEL);
329 if (dev->stag_arr == NULL)
330 goto alloc_err;
331
327 spin_lock_init(&dev->av_tbl.lock); 332 spin_lock_init(&dev->av_tbl.lock);
328 spin_lock_init(&dev->flush_q_lock); 333 spin_lock_init(&dev->flush_q_lock);
329 return 0; 334 return 0;
@@ -334,6 +339,7 @@ alloc_err:
334 339
335static void ocrdma_free_resources(struct ocrdma_dev *dev) 340static void ocrdma_free_resources(struct ocrdma_dev *dev)
336{ 341{
342 kfree(dev->stag_arr);
337 kfree(dev->qp_tbl); 343 kfree(dev->qp_tbl);
338 kfree(dev->cq_tbl); 344 kfree(dev->cq_tbl);
339 kfree(dev->sgid_tbl); 345 kfree(dev->sgid_tbl);
@@ -353,15 +359,25 @@ static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
353{ 359{
354 struct ocrdma_dev *dev = dev_get_drvdata(device); 360 struct ocrdma_dev *dev = dev_get_drvdata(device);
355 361
356 return scnprintf(buf, PAGE_SIZE, "%s", &dev->attr.fw_ver[0]); 362 return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->attr.fw_ver[0]);
363}
364
365static ssize_t show_hca_type(struct device *device,
366 struct device_attribute *attr, char *buf)
367{
368 struct ocrdma_dev *dev = dev_get_drvdata(device);
369
370 return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]);
357} 371}
358 372
359static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); 373static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
360static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); 374static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
375static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
361 376
362static struct device_attribute *ocrdma_attributes[] = { 377static struct device_attribute *ocrdma_attributes[] = {
363 &dev_attr_hw_rev, 378 &dev_attr_hw_rev,
364 &dev_attr_fw_ver 379 &dev_attr_fw_ver,
380 &dev_attr_hca_type
365}; 381};
366 382
367static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev) 383static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
@@ -372,6 +388,58 @@ static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
372 device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]); 388 device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]);
373} 389}
374 390
391static void ocrdma_init_ipv4_gids(struct ocrdma_dev *dev,
392 struct net_device *net)
393{
394 struct in_device *in_dev;
395 union ib_gid gid;
396 in_dev = in_dev_get(net);
397 if (in_dev) {
398 for_ifa(in_dev) {
399 ipv6_addr_set_v4mapped(ifa->ifa_address,
400 (struct in6_addr *)&gid);
401 ocrdma_add_sgid(dev, &gid);
402 }
403 endfor_ifa(in_dev);
404 in_dev_put(in_dev);
405 }
406}
407
408static void ocrdma_init_ipv6_gids(struct ocrdma_dev *dev,
409 struct net_device *net)
410{
411#if IS_ENABLED(CONFIG_IPV6)
412 struct inet6_dev *in6_dev;
413 union ib_gid *pgid;
414 struct inet6_ifaddr *ifp;
415 in6_dev = in6_dev_get(net);
416 if (in6_dev) {
417 read_lock_bh(&in6_dev->lock);
418 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
419 pgid = (union ib_gid *)&ifp->addr;
420 ocrdma_add_sgid(dev, pgid);
421 }
422 read_unlock_bh(&in6_dev->lock);
423 in6_dev_put(in6_dev);
424 }
425#endif
426}
427
428static void ocrdma_init_gid_table(struct ocrdma_dev *dev)
429{
430 struct net_device *net_dev;
431
432 for_each_netdev(&init_net, net_dev) {
433 struct net_device *real_dev = rdma_vlan_dev_real_dev(net_dev) ?
434 rdma_vlan_dev_real_dev(net_dev) : net_dev;
435
436 if (real_dev == dev->nic_info.netdev) {
437 ocrdma_init_ipv4_gids(dev, net_dev);
438 ocrdma_init_ipv6_gids(dev, net_dev);
439 }
440 }
441}
442
375static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info) 443static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
376{ 444{
377 int status = 0, i; 445 int status = 0, i;
@@ -399,6 +467,8 @@ static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
399 if (status) 467 if (status)
400 goto alloc_err; 468 goto alloc_err;
401 469
470 ocrdma_init_service_level(dev);
471 ocrdma_init_gid_table(dev);
402 status = ocrdma_register_device(dev); 472 status = ocrdma_register_device(dev);
403 if (status) 473 if (status)
404 goto alloc_err; 474 goto alloc_err;
@@ -508,6 +578,12 @@ static int ocrdma_close(struct ocrdma_dev *dev)
508 return 0; 578 return 0;
509} 579}
510 580
581static void ocrdma_shutdown(struct ocrdma_dev *dev)
582{
583 ocrdma_close(dev);
584 ocrdma_remove(dev);
585}
586
511/* event handling via NIC driver ensures that all the NIC specific 587/* event handling via NIC driver ensures that all the NIC specific
512 * initialization done before RoCE driver notifies 588 * initialization done before RoCE driver notifies
513 * event to stack. 589 * event to stack.
@@ -521,6 +597,9 @@ static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
521 case BE_DEV_DOWN: 597 case BE_DEV_DOWN:
522 ocrdma_close(dev); 598 ocrdma_close(dev);
523 break; 599 break;
600 case BE_DEV_SHUTDOWN:
601 ocrdma_shutdown(dev);
602 break;
524 } 603 }
525} 604}
526 605
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
index 96c9ee602ba4..904989ec5eaa 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
@@ -44,35 +44,39 @@ enum {
44#define OCRDMA_SUBSYS_ROCE 10 44#define OCRDMA_SUBSYS_ROCE 10
45enum { 45enum {
46 OCRDMA_CMD_QUERY_CONFIG = 1, 46 OCRDMA_CMD_QUERY_CONFIG = 1,
47 OCRDMA_CMD_ALLOC_PD, 47 OCRDMA_CMD_ALLOC_PD = 2,
48 OCRDMA_CMD_DEALLOC_PD, 48 OCRDMA_CMD_DEALLOC_PD = 3,
49 49
50 OCRDMA_CMD_CREATE_AH_TBL, 50 OCRDMA_CMD_CREATE_AH_TBL = 4,
51 OCRDMA_CMD_DELETE_AH_TBL, 51 OCRDMA_CMD_DELETE_AH_TBL = 5,
52 52
53 OCRDMA_CMD_CREATE_QP, 53 OCRDMA_CMD_CREATE_QP = 6,
54 OCRDMA_CMD_QUERY_QP, 54 OCRDMA_CMD_QUERY_QP = 7,
55 OCRDMA_CMD_MODIFY_QP, 55 OCRDMA_CMD_MODIFY_QP = 8 ,
56 OCRDMA_CMD_DELETE_QP, 56 OCRDMA_CMD_DELETE_QP = 9,
57 57
58 OCRDMA_CMD_RSVD1, 58 OCRDMA_CMD_RSVD1 = 10,
59 OCRDMA_CMD_ALLOC_LKEY, 59 OCRDMA_CMD_ALLOC_LKEY = 11,
60 OCRDMA_CMD_DEALLOC_LKEY, 60 OCRDMA_CMD_DEALLOC_LKEY = 12,
61 OCRDMA_CMD_REGISTER_NSMR, 61 OCRDMA_CMD_REGISTER_NSMR = 13,
62 OCRDMA_CMD_REREGISTER_NSMR, 62 OCRDMA_CMD_REREGISTER_NSMR = 14,
63 OCRDMA_CMD_REGISTER_NSMR_CONT, 63 OCRDMA_CMD_REGISTER_NSMR_CONT = 15,
64 OCRDMA_CMD_QUERY_NSMR, 64 OCRDMA_CMD_QUERY_NSMR = 16,
65 OCRDMA_CMD_ALLOC_MW, 65 OCRDMA_CMD_ALLOC_MW = 17,
66 OCRDMA_CMD_QUERY_MW, 66 OCRDMA_CMD_QUERY_MW = 18,
67 67
68 OCRDMA_CMD_CREATE_SRQ, 68 OCRDMA_CMD_CREATE_SRQ = 19,
69 OCRDMA_CMD_QUERY_SRQ, 69 OCRDMA_CMD_QUERY_SRQ = 20,
70 OCRDMA_CMD_MODIFY_SRQ, 70 OCRDMA_CMD_MODIFY_SRQ = 21,
71 OCRDMA_CMD_DELETE_SRQ, 71 OCRDMA_CMD_DELETE_SRQ = 22,
72 72
73 OCRDMA_CMD_ATTACH_MCAST, 73 OCRDMA_CMD_ATTACH_MCAST = 23,
74 OCRDMA_CMD_DETACH_MCAST, 74 OCRDMA_CMD_DETACH_MCAST = 24,
75 OCRDMA_CMD_GET_RDMA_STATS, 75
76 OCRDMA_CMD_CREATE_RBQ = 25,
77 OCRDMA_CMD_DESTROY_RBQ = 26,
78
79 OCRDMA_CMD_GET_RDMA_STATS = 27,
76 80
77 OCRDMA_CMD_MAX 81 OCRDMA_CMD_MAX
78}; 82};
@@ -103,7 +107,7 @@ enum {
103 107
104#define OCRDMA_MAX_QP 2048 108#define OCRDMA_MAX_QP 2048
105#define OCRDMA_MAX_CQ 2048 109#define OCRDMA_MAX_CQ 2048
106#define OCRDMA_MAX_STAG 8192 110#define OCRDMA_MAX_STAG 16384
107 111
108enum { 112enum {
109 OCRDMA_DB_RQ_OFFSET = 0xE0, 113 OCRDMA_DB_RQ_OFFSET = 0xE0,
@@ -422,7 +426,12 @@ struct ocrdma_ae_qp_mcqe {
422 426
423#define OCRDMA_ASYNC_RDMA_EVE_CODE 0x14 427#define OCRDMA_ASYNC_RDMA_EVE_CODE 0x14
424#define OCRDMA_ASYNC_GRP5_EVE_CODE 0x5 428#define OCRDMA_ASYNC_GRP5_EVE_CODE 0x5
425#define OCRDMA_ASYNC_EVENT_PVID_STATE 0x3 429
430enum ocrdma_async_grp5_events {
431 OCRDMA_ASYNC_EVENT_QOS_VALUE = 0x01,
432 OCRDMA_ASYNC_EVENT_COS_VALUE = 0x02,
433 OCRDMA_ASYNC_EVENT_PVID_STATE = 0x03
434};
426 435
427enum OCRDMA_ASYNC_EVENT_TYPE { 436enum OCRDMA_ASYNC_EVENT_TYPE {
428 OCRDMA_CQ_ERROR = 0x00, 437 OCRDMA_CQ_ERROR = 0x00,
@@ -525,8 +534,8 @@ struct ocrdma_mbx_query_config {
525 u32 max_ird_ord_per_qp; 534 u32 max_ird_ord_per_qp;
526 u32 max_shared_ird_ord; 535 u32 max_shared_ird_ord;
527 u32 max_mr; 536 u32 max_mr;
528 u32 max_mr_size_lo;
529 u32 max_mr_size_hi; 537 u32 max_mr_size_hi;
538 u32 max_mr_size_lo;
530 u32 max_num_mr_pbl; 539 u32 max_num_mr_pbl;
531 u32 max_mw; 540 u32 max_mw;
532 u32 max_fmr; 541 u32 max_fmr;
@@ -580,17 +589,26 @@ enum {
580 OCRDMA_FN_MODE_RDMA = 0x4 589 OCRDMA_FN_MODE_RDMA = 0x4
581}; 590};
582 591
592enum {
593 OCRDMA_IF_TYPE_MASK = 0xFFFF0000,
594 OCRDMA_IF_TYPE_SHIFT = 0x10,
595 OCRDMA_PHY_TYPE_MASK = 0x0000FFFF,
596 OCRDMA_FUTURE_DETAILS_MASK = 0xFFFF0000,
597 OCRDMA_FUTURE_DETAILS_SHIFT = 0x10,
598 OCRDMA_EX_PHY_DETAILS_MASK = 0x0000FFFF,
599 OCRDMA_FSPEED_SUPP_MASK = 0xFFFF0000,
600 OCRDMA_FSPEED_SUPP_SHIFT = 0x10,
601 OCRDMA_ASPEED_SUPP_MASK = 0x0000FFFF
602};
603
583struct ocrdma_get_phy_info_rsp { 604struct ocrdma_get_phy_info_rsp {
584 struct ocrdma_mqe_hdr hdr; 605 struct ocrdma_mqe_hdr hdr;
585 struct ocrdma_mbx_rsp rsp; 606 struct ocrdma_mbx_rsp rsp;
586 607
587 u16 phy_type; 608 u32 ityp_ptyp;
588 u16 interface_type;
589 u32 misc_params; 609 u32 misc_params;
590 u16 ext_phy_details; 610 u32 ftrdtl_exphydtl;
591 u16 rsvd; 611 u32 fspeed_aspeed;
592 u16 auto_speeds_supported;
593 u16 fixed_speeds_supported;
594 u32 future_use[2]; 612 u32 future_use[2];
595}; 613};
596 614
@@ -603,19 +621,34 @@ enum {
603 OCRDMA_PHY_SPEED_40GBPS = 0x20 621 OCRDMA_PHY_SPEED_40GBPS = 0x20
604}; 622};
605 623
624enum {
625 OCRDMA_PORT_NUM_MASK = 0x3F,
626 OCRDMA_PT_MASK = 0xC0,
627 OCRDMA_PT_SHIFT = 0x6,
628 OCRDMA_LINK_DUP_MASK = 0x0000FF00,
629 OCRDMA_LINK_DUP_SHIFT = 0x8,
630 OCRDMA_PHY_PS_MASK = 0x00FF0000,
631 OCRDMA_PHY_PS_SHIFT = 0x10,
632 OCRDMA_PHY_PFLT_MASK = 0xFF000000,
633 OCRDMA_PHY_PFLT_SHIFT = 0x18,
634 OCRDMA_QOS_LNKSP_MASK = 0xFFFF0000,
635 OCRDMA_QOS_LNKSP_SHIFT = 0x10,
636 OCRDMA_LLST_MASK = 0xFF,
637 OCRDMA_PLFC_MASK = 0x00000400,
638 OCRDMA_PLFC_SHIFT = 0x8,
639 OCRDMA_PLRFC_MASK = 0x00000200,
640 OCRDMA_PLRFC_SHIFT = 0x8,
641 OCRDMA_PLTFC_MASK = 0x00000100,
642 OCRDMA_PLTFC_SHIFT = 0x8
643};
606 644
607struct ocrdma_get_link_speed_rsp { 645struct ocrdma_get_link_speed_rsp {
608 struct ocrdma_mqe_hdr hdr; 646 struct ocrdma_mqe_hdr hdr;
609 struct ocrdma_mbx_rsp rsp; 647 struct ocrdma_mbx_rsp rsp;
610 648
611 u8 pt_port_num; 649 u32 pflt_pps_ld_pnum;
612 u8 link_duplex; 650 u32 qos_lsp;
613 u8 phys_port_speed; 651 u32 res_lls;
614 u8 phys_port_fault;
615 u16 rsvd1;
616 u16 qos_lnk_speed;
617 u8 logical_lnk_status;
618 u8 rsvd2[3];
619}; 652};
620 653
621enum { 654enum {
@@ -666,8 +699,7 @@ struct ocrdma_create_cq_cmd {
666 u32 pgsz_pgcnt; 699 u32 pgsz_pgcnt;
667 u32 ev_cnt_flags; 700 u32 ev_cnt_flags;
668 u32 eqn; 701 u32 eqn;
669 u16 cqe_count; 702 u32 pdid_cqecnt;
670 u16 pd_id;
671 u32 rsvd6; 703 u32 rsvd6;
672 struct ocrdma_pa pa[OCRDMA_CREATE_CQ_MAX_PAGES]; 704 struct ocrdma_pa pa[OCRDMA_CREATE_CQ_MAX_PAGES];
673}; 705};
@@ -678,6 +710,10 @@ struct ocrdma_create_cq {
678}; 710};
679 711
680enum { 712enum {
713 OCRDMA_CREATE_CQ_CMD_PDID_SHIFT = 0x10
714};
715
716enum {
681 OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK = 0xFFFF 717 OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK = 0xFFFF
682}; 718};
683 719
@@ -1231,7 +1267,6 @@ struct ocrdma_destroy_srq {
1231 1267
1232enum { 1268enum {
1233 OCRDMA_ALLOC_PD_ENABLE_DPP = BIT(16), 1269 OCRDMA_ALLOC_PD_ENABLE_DPP = BIT(16),
1234 OCRDMA_PD_MAX_DPP_ENABLED_QP = 8,
1235 OCRDMA_DPP_PAGE_SIZE = 4096 1270 OCRDMA_DPP_PAGE_SIZE = 4096
1236}; 1271};
1237 1272
@@ -1896,12 +1931,62 @@ struct ocrdma_rdma_stats_resp {
1896 struct ocrdma_rx_dbg_stats rx_dbg_stats; 1931 struct ocrdma_rx_dbg_stats rx_dbg_stats;
1897} __packed; 1932} __packed;
1898 1933
1934enum {
1935 OCRDMA_HBA_ATTRB_EPROM_VER_LO_MASK = 0xFF,
1936 OCRDMA_HBA_ATTRB_EPROM_VER_HI_MASK = 0xFF00,
1937 OCRDMA_HBA_ATTRB_EPROM_VER_HI_SHIFT = 0x08,
1938 OCRDMA_HBA_ATTRB_CDBLEN_MASK = 0xFFFF,
1939 OCRDMA_HBA_ATTRB_ASIC_REV_MASK = 0xFF0000,
1940 OCRDMA_HBA_ATTRB_ASIC_REV_SHIFT = 0x10,
1941 OCRDMA_HBA_ATTRB_GUID0_MASK = 0xFF000000,
1942 OCRDMA_HBA_ATTRB_GUID0_SHIFT = 0x18,
1943 OCRDMA_HBA_ATTRB_GUID13_MASK = 0xFF,
1944 OCRDMA_HBA_ATTRB_GUID14_MASK = 0xFF00,
1945 OCRDMA_HBA_ATTRB_GUID14_SHIFT = 0x08,
1946 OCRDMA_HBA_ATTRB_GUID15_MASK = 0xFF0000,
1947 OCRDMA_HBA_ATTRB_GUID15_SHIFT = 0x10,
1948 OCRDMA_HBA_ATTRB_PCNT_MASK = 0xFF000000,
1949 OCRDMA_HBA_ATTRB_PCNT_SHIFT = 0x18,
1950 OCRDMA_HBA_ATTRB_LDTOUT_MASK = 0xFFFF,
1951 OCRDMA_HBA_ATTRB_ISCSI_VER_MASK = 0xFF0000,
1952 OCRDMA_HBA_ATTRB_ISCSI_VER_SHIFT = 0x10,
1953 OCRDMA_HBA_ATTRB_MFUNC_DEV_MASK = 0xFF000000,
1954 OCRDMA_HBA_ATTRB_MFUNC_DEV_SHIFT = 0x18,
1955 OCRDMA_HBA_ATTRB_CV_MASK = 0xFF,
1956 OCRDMA_HBA_ATTRB_HBA_ST_MASK = 0xFF00,
1957 OCRDMA_HBA_ATTRB_HBA_ST_SHIFT = 0x08,
1958 OCRDMA_HBA_ATTRB_MAX_DOMS_MASK = 0xFF0000,
1959 OCRDMA_HBA_ATTRB_MAX_DOMS_SHIFT = 0x10,
1960 OCRDMA_HBA_ATTRB_PTNUM_MASK = 0x3F000000,
1961 OCRDMA_HBA_ATTRB_PTNUM_SHIFT = 0x18,
1962 OCRDMA_HBA_ATTRB_PT_MASK = 0xC0000000,
1963 OCRDMA_HBA_ATTRB_PT_SHIFT = 0x1E,
1964 OCRDMA_HBA_ATTRB_ISCSI_FET_MASK = 0xFF,
1965 OCRDMA_HBA_ATTRB_ASIC_GEN_MASK = 0xFF00,
1966 OCRDMA_HBA_ATTRB_ASIC_GEN_SHIFT = 0x08,
1967 OCRDMA_HBA_ATTRB_PCI_VID_MASK = 0xFFFF,
1968 OCRDMA_HBA_ATTRB_PCI_DID_MASK = 0xFFFF0000,
1969 OCRDMA_HBA_ATTRB_PCI_DID_SHIFT = 0x10,
1970 OCRDMA_HBA_ATTRB_PCI_SVID_MASK = 0xFFFF,
1971 OCRDMA_HBA_ATTRB_PCI_SSID_MASK = 0xFFFF0000,
1972 OCRDMA_HBA_ATTRB_PCI_SSID_SHIFT = 0x10,
1973 OCRDMA_HBA_ATTRB_PCI_BUSNUM_MASK = 0xFF,
1974 OCRDMA_HBA_ATTRB_PCI_DEVNUM_MASK = 0xFF00,
1975 OCRDMA_HBA_ATTRB_PCI_DEVNUM_SHIFT = 0x08,
1976 OCRDMA_HBA_ATTRB_PCI_FUNCNUM_MASK = 0xFF0000,
1977 OCRDMA_HBA_ATTRB_PCI_FUNCNUM_SHIFT = 0x10,
1978 OCRDMA_HBA_ATTRB_IF_TYPE_MASK = 0xFF000000,
1979 OCRDMA_HBA_ATTRB_IF_TYPE_SHIFT = 0x18,
1980 OCRDMA_HBA_ATTRB_NETFIL_MASK =0xFF
1981};
1899 1982
1900struct mgmt_hba_attribs { 1983struct mgmt_hba_attribs {
1901 u8 flashrom_version_string[32]; 1984 u8 flashrom_version_string[32];
1902 u8 manufacturer_name[32]; 1985 u8 manufacturer_name[32];
1903 u32 supported_modes; 1986 u32 supported_modes;
1904 u32 rsvd0[3]; 1987 u32 rsvd_eprom_verhi_verlo;
1988 u32 mbx_ds_ver;
1989 u32 epfw_ds_ver;
1905 u8 ncsi_ver_string[12]; 1990 u8 ncsi_ver_string[12];
1906 u32 default_extended_timeout; 1991 u32 default_extended_timeout;
1907 u8 controller_model_number[32]; 1992 u8 controller_model_number[32];
@@ -1914,34 +1999,26 @@ struct mgmt_hba_attribs {
1914 u8 driver_version_string[32]; 1999 u8 driver_version_string[32];
1915 u8 fw_on_flash_version_string[32]; 2000 u8 fw_on_flash_version_string[32];
1916 u32 functionalities_supported; 2001 u32 functionalities_supported;
1917 u16 max_cdblength; 2002 u32 guid0_asicrev_cdblen;
1918 u8 asic_revision; 2003 u8 generational_guid[12];
1919 u8 generational_guid[16]; 2004 u32 portcnt_guid15;
1920 u8 hba_port_count; 2005 u32 mfuncdev_iscsi_ldtout;
1921 u16 default_link_down_timeout; 2006 u32 ptpnum_maxdoms_hbast_cv;
1922 u8 iscsi_ver_min_max;
1923 u8 multifunction_device;
1924 u8 cache_valid;
1925 u8 hba_status;
1926 u8 max_domains_supported;
1927 u8 phy_port;
1928 u32 firmware_post_status; 2007 u32 firmware_post_status;
1929 u32 hba_mtu[8]; 2008 u32 hba_mtu[8];
1930 u32 rsvd1[4]; 2009 u32 res_asicgen_iscsi_feaures;
2010 u32 rsvd1[3];
1931}; 2011};
1932 2012
1933struct mgmt_controller_attrib { 2013struct mgmt_controller_attrib {
1934 struct mgmt_hba_attribs hba_attribs; 2014 struct mgmt_hba_attribs hba_attribs;
1935 u16 pci_vendor_id; 2015 u32 pci_did_vid;
1936 u16 pci_device_id; 2016 u32 pci_ssid_svid;
1937 u16 pci_sub_vendor_id; 2017 u32 ityp_fnum_devnum_bnum;
1938 u16 pci_sub_system_id; 2018 u32 uid_hi;
1939 u8 pci_bus_number; 2019 u32 uid_lo;
1940 u8 pci_device_number; 2020 u32 res_nnetfil;
1941 u8 pci_function_number; 2021 u32 rsvd0[4];
1942 u8 interface_type;
1943 u64 unique_identifier;
1944 u32 rsvd0[5];
1945}; 2022};
1946 2023
1947struct ocrdma_get_ctrl_attribs_rsp { 2024struct ocrdma_get_ctrl_attribs_rsp {
@@ -1949,5 +2026,79 @@ struct ocrdma_get_ctrl_attribs_rsp {
1949 struct mgmt_controller_attrib ctrl_attribs; 2026 struct mgmt_controller_attrib ctrl_attribs;
1950}; 2027};
1951 2028
2029#define OCRDMA_SUBSYS_DCBX 0x10
2030
2031enum OCRDMA_DCBX_OPCODE {
2032 OCRDMA_CMD_GET_DCBX_CONFIG = 0x01
2033};
2034
2035enum OCRDMA_DCBX_PARAM_TYPE {
2036 OCRDMA_PARAMETER_TYPE_ADMIN = 0x00,
2037 OCRDMA_PARAMETER_TYPE_OPER = 0x01,
2038 OCRDMA_PARAMETER_TYPE_PEER = 0x02
2039};
2040
2041enum OCRDMA_DCBX_APP_PROTO {
2042 OCRDMA_APP_PROTO_ROCE = 0x8915
2043};
2044
2045enum OCRDMA_DCBX_PROTO {
2046 OCRDMA_PROTO_SELECT_L2 = 0x00,
2047 OCRDMA_PROTO_SELECT_L4 = 0x01
2048};
2049
2050enum OCRDMA_DCBX_APP_PARAM {
2051 OCRDMA_APP_PARAM_APP_PROTO_MASK = 0xFFFF,
2052 OCRDMA_APP_PARAM_PROTO_SEL_MASK = 0xFF,
2053 OCRDMA_APP_PARAM_PROTO_SEL_SHIFT = 0x10,
2054 OCRDMA_APP_PARAM_VALID_MASK = 0xFF,
2055 OCRDMA_APP_PARAM_VALID_SHIFT = 0x18
2056};
2057
2058enum OCRDMA_DCBX_STATE_FLAGS {
2059 OCRDMA_STATE_FLAG_ENABLED = 0x01,
2060 OCRDMA_STATE_FLAG_ADDVERTISED = 0x02,
2061 OCRDMA_STATE_FLAG_WILLING = 0x04,
2062 OCRDMA_STATE_FLAG_SYNC = 0x08,
2063 OCRDMA_STATE_FLAG_UNSUPPORTED = 0x40000000,
2064 OCRDMA_STATE_FLAG_NEG_FAILD = 0x80000000
2065};
2066
2067enum OCRDMA_TCV_AEV_OPV_ST {
2068 OCRDMA_DCBX_TC_SUPPORT_MASK = 0xFF,
2069 OCRDMA_DCBX_TC_SUPPORT_SHIFT = 0x18,
2070 OCRDMA_DCBX_APP_ENTRY_SHIFT = 0x10,
2071 OCRDMA_DCBX_OP_PARAM_SHIFT = 0x08,
2072 OCRDMA_DCBX_STATE_MASK = 0xFF
2073};
2074
2075struct ocrdma_app_parameter {
2076 u32 valid_proto_app;
2077 u32 oui;
2078 u32 app_prio[2];
2079};
2080
2081struct ocrdma_dcbx_cfg {
2082 u32 tcv_aev_opv_st;
2083 u32 tc_state;
2084 u32 pfc_state;
2085 u32 qcn_state;
2086 u32 appl_state;
2087 u32 ll_state;
2088 u32 tc_bw[2];
2089 u32 tc_prio[8];
2090 u32 pfc_prio[2];
2091 struct ocrdma_app_parameter app_param[15];
2092};
2093
2094struct ocrdma_get_dcbx_cfg_req {
2095 struct ocrdma_mbx_hdr hdr;
2096 u32 param_type;
2097} __packed;
2098
2099struct ocrdma_get_dcbx_cfg_rsp {
2100 struct ocrdma_mbx_rsp hdr;
2101 struct ocrdma_dcbx_cfg cfg;
2102} __packed;
1952 2103
1953#endif /* __OCRDMA_SLI_H__ */ 2104#endif /* __OCRDMA_SLI_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index edf6211d84b8..acb434d16903 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -69,11 +69,11 @@ int ocrdma_query_device(struct ib_device *ibdev, struct ib_device_attr *attr)
69 memcpy(&attr->fw_ver, &dev->attr.fw_ver[0], 69 memcpy(&attr->fw_ver, &dev->attr.fw_ver[0],
70 min(sizeof(dev->attr.fw_ver), sizeof(attr->fw_ver))); 70 min(sizeof(dev->attr.fw_ver), sizeof(attr->fw_ver)));
71 ocrdma_get_guid(dev, (u8 *)&attr->sys_image_guid); 71 ocrdma_get_guid(dev, (u8 *)&attr->sys_image_guid);
72 attr->max_mr_size = ~0ull; 72 attr->max_mr_size = dev->attr.max_mr_size;
73 attr->page_size_cap = 0xffff000; 73 attr->page_size_cap = 0xffff000;
74 attr->vendor_id = dev->nic_info.pdev->vendor; 74 attr->vendor_id = dev->nic_info.pdev->vendor;
75 attr->vendor_part_id = dev->nic_info.pdev->device; 75 attr->vendor_part_id = dev->nic_info.pdev->device;
76 attr->hw_ver = 0; 76 attr->hw_ver = dev->asic_id;
77 attr->max_qp = dev->attr.max_qp; 77 attr->max_qp = dev->attr.max_qp;
78 attr->max_ah = OCRDMA_MAX_AH; 78 attr->max_ah = OCRDMA_MAX_AH;
79 attr->max_qp_wr = dev->attr.max_wqe; 79 attr->max_qp_wr = dev->attr.max_wqe;
@@ -268,7 +268,8 @@ static struct ocrdma_pd *_ocrdma_alloc_pd(struct ocrdma_dev *dev,
268 pd->dpp_enabled = 268 pd->dpp_enabled =
269 ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R; 269 ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R;
270 pd->num_dpp_qp = 270 pd->num_dpp_qp =
271 pd->dpp_enabled ? OCRDMA_PD_MAX_DPP_ENABLED_QP : 0; 271 pd->dpp_enabled ? (dev->nic_info.db_page_size /
272 dev->attr.wqe_size) : 0;
272 } 273 }
273 274
274retry: 275retry:
@@ -328,7 +329,10 @@ static int ocrdma_dealloc_ucontext_pd(struct ocrdma_ucontext *uctx)
328 struct ocrdma_pd *pd = uctx->cntxt_pd; 329 struct ocrdma_pd *pd = uctx->cntxt_pd;
329 struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device); 330 struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
330 331
331 BUG_ON(uctx->pd_in_use); 332 if (uctx->pd_in_use) {
333 pr_err("%s(%d) Freeing in use pdid=0x%x.\n",
334 __func__, dev->id, pd->id);
335 }
332 uctx->cntxt_pd = NULL; 336 uctx->cntxt_pd = NULL;
333 status = _ocrdma_dealloc_pd(dev, pd); 337 status = _ocrdma_dealloc_pd(dev, pd);
334 return status; 338 return status;
@@ -843,6 +847,13 @@ int ocrdma_dereg_mr(struct ib_mr *ib_mr)
843 if (mr->umem) 847 if (mr->umem)
844 ib_umem_release(mr->umem); 848 ib_umem_release(mr->umem);
845 kfree(mr); 849 kfree(mr);
850
851 /* Don't stop cleanup, in case FW is unresponsive */
852 if (dev->mqe_ctx.fw_error_state) {
853 status = 0;
854 pr_err("%s(%d) fw not responding.\n",
855 __func__, dev->id);
856 }
846 return status; 857 return status;
847} 858}
848 859
@@ -2054,6 +2065,13 @@ int ocrdma_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
2054 } 2065 }
2055 2066
2056 while (wr) { 2067 while (wr) {
2068 if (qp->qp_type == IB_QPT_UD &&
2069 (wr->opcode != IB_WR_SEND &&
2070 wr->opcode != IB_WR_SEND_WITH_IMM)) {
2071 *bad_wr = wr;
2072 status = -EINVAL;
2073 break;
2074 }
2057 if (ocrdma_hwq_free_cnt(&qp->sq) == 0 || 2075 if (ocrdma_hwq_free_cnt(&qp->sq) == 0 ||
2058 wr->num_sge > qp->sq.max_sges) { 2076 wr->num_sge > qp->sq.max_sges) {
2059 *bad_wr = wr; 2077 *bad_wr = wr;
@@ -2488,6 +2506,11 @@ static bool ocrdma_poll_err_scqe(struct ocrdma_qp *qp,
2488 *stop = true; 2506 *stop = true;
2489 expand = false; 2507 expand = false;
2490 } 2508 }
2509 } else if (is_hw_sq_empty(qp)) {
2510 /* Do nothing */
2511 expand = false;
2512 *polled = false;
2513 *stop = false;
2491 } else { 2514 } else {
2492 *polled = true; 2515 *polled = true;
2493 expand = ocrdma_update_err_scqe(ibwc, cqe, qp, status); 2516 expand = ocrdma_update_err_scqe(ibwc, cqe, qp, status);
@@ -2593,6 +2616,11 @@ static bool ocrdma_poll_err_rcqe(struct ocrdma_qp *qp, struct ocrdma_cqe *cqe,
2593 *stop = true; 2616 *stop = true;
2594 expand = false; 2617 expand = false;
2595 } 2618 }
2619 } else if (is_hw_rq_empty(qp)) {
2620 /* Do nothing */
2621 expand = false;
2622 *polled = false;
2623 *stop = false;
2596 } else { 2624 } else {
2597 *polled = true; 2625 *polled = true;
2598 expand = ocrdma_update_err_rcqe(ibwc, cqe, qp, status); 2626 expand = ocrdma_update_err_rcqe(ibwc, cqe, qp, status);
diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
index 22c720e5740d..636be117b578 100644
--- a/drivers/infiniband/hw/qib/qib_mad.c
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -2476,7 +2476,7 @@ int qib_create_agents(struct qib_ibdev *dev)
2476 ibp = &dd->pport[p].ibport_data; 2476 ibp = &dd->pport[p].ibport_data;
2477 agent = ib_register_mad_agent(&dev->ibdev, p + 1, IB_QPT_SMI, 2477 agent = ib_register_mad_agent(&dev->ibdev, p + 1, IB_QPT_SMI,
2478 NULL, 0, send_handler, 2478 NULL, 0, send_handler,
2479 NULL, NULL); 2479 NULL, NULL, 0);
2480 if (IS_ERR(agent)) { 2480 if (IS_ERR(agent)) {
2481 ret = PTR_ERR(agent); 2481 ret = PTR_ERR(agent);
2482 goto err; 2482 goto err;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index c639f90cfda4..3edce617c31b 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -86,7 +86,6 @@ enum {
86 IPOIB_FLAG_INITIALIZED = 1, 86 IPOIB_FLAG_INITIALIZED = 1,
87 IPOIB_FLAG_ADMIN_UP = 2, 87 IPOIB_FLAG_ADMIN_UP = 2,
88 IPOIB_PKEY_ASSIGNED = 3, 88 IPOIB_PKEY_ASSIGNED = 3,
89 IPOIB_PKEY_STOP = 4,
90 IPOIB_FLAG_SUBINTERFACE = 5, 89 IPOIB_FLAG_SUBINTERFACE = 5,
91 IPOIB_MCAST_RUN = 6, 90 IPOIB_MCAST_RUN = 6,
92 IPOIB_STOP_REAPER = 7, 91 IPOIB_STOP_REAPER = 7,
@@ -312,7 +311,6 @@ struct ipoib_dev_priv {
312 struct list_head multicast_list; 311 struct list_head multicast_list;
313 struct rb_root multicast_tree; 312 struct rb_root multicast_tree;
314 313
315 struct delayed_work pkey_poll_task;
316 struct delayed_work mcast_task; 314 struct delayed_work mcast_task;
317 struct work_struct carrier_on_task; 315 struct work_struct carrier_on_task;
318 struct work_struct flush_light; 316 struct work_struct flush_light;
@@ -473,10 +471,11 @@ void ipoib_ib_dev_flush_heavy(struct work_struct *work);
473void ipoib_pkey_event(struct work_struct *work); 471void ipoib_pkey_event(struct work_struct *work);
474void ipoib_ib_dev_cleanup(struct net_device *dev); 472void ipoib_ib_dev_cleanup(struct net_device *dev);
475 473
476int ipoib_ib_dev_open(struct net_device *dev); 474int ipoib_ib_dev_open(struct net_device *dev, int flush);
477int ipoib_ib_dev_up(struct net_device *dev); 475int ipoib_ib_dev_up(struct net_device *dev);
478int ipoib_ib_dev_down(struct net_device *dev, int flush); 476int ipoib_ib_dev_down(struct net_device *dev, int flush);
479int ipoib_ib_dev_stop(struct net_device *dev, int flush); 477int ipoib_ib_dev_stop(struct net_device *dev, int flush);
478void ipoib_pkey_dev_check_presence(struct net_device *dev);
480 479
481int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port); 480int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
482void ipoib_dev_cleanup(struct net_device *dev); 481void ipoib_dev_cleanup(struct net_device *dev);
@@ -532,8 +531,7 @@ int ipoib_set_mode(struct net_device *dev, const char *buf);
532 531
533void ipoib_setup(struct net_device *dev); 532void ipoib_setup(struct net_device *dev);
534 533
535void ipoib_pkey_poll(struct work_struct *work); 534void ipoib_pkey_open(struct ipoib_dev_priv *priv);
536int ipoib_pkey_dev_delay_open(struct net_device *dev);
537void ipoib_drain_cq(struct net_device *dev); 535void ipoib_drain_cq(struct net_device *dev);
538 536
539void ipoib_set_ethtool_ops(struct net_device *dev); 537void ipoib_set_ethtool_ops(struct net_device *dev);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_fs.c b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
index 50061854616e..6bd5740e2691 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_fs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
@@ -281,10 +281,8 @@ void ipoib_delete_debug_files(struct net_device *dev)
281{ 281{
282 struct ipoib_dev_priv *priv = netdev_priv(dev); 282 struct ipoib_dev_priv *priv = netdev_priv(dev);
283 283
284 if (priv->mcg_dentry) 284 debugfs_remove(priv->mcg_dentry);
285 debugfs_remove(priv->mcg_dentry); 285 debugfs_remove(priv->path_dentry);
286 if (priv->path_dentry)
287 debugfs_remove(priv->path_dentry);
288} 286}
289 287
290int ipoib_register_debugfs(void) 288int ipoib_register_debugfs(void)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 6a7003ddb0be..72626c348174 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -664,17 +664,18 @@ static void ipoib_ib_tx_timer_func(unsigned long ctx)
664 drain_tx_cq((struct net_device *)ctx); 664 drain_tx_cq((struct net_device *)ctx);
665} 665}
666 666
667int ipoib_ib_dev_open(struct net_device *dev) 667int ipoib_ib_dev_open(struct net_device *dev, int flush)
668{ 668{
669 struct ipoib_dev_priv *priv = netdev_priv(dev); 669 struct ipoib_dev_priv *priv = netdev_priv(dev);
670 int ret; 670 int ret;
671 671
672 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) { 672 ipoib_pkey_dev_check_presence(dev);
673 ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey); 673
674 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); 674 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
675 ipoib_warn(priv, "P_Key 0x%04x is %s\n", priv->pkey,
676 (!(priv->pkey & 0x7fff) ? "Invalid" : "not found"));
675 return -1; 677 return -1;
676 } 678 }
677 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
678 679
679 ret = ipoib_init_qp(dev); 680 ret = ipoib_init_qp(dev);
680 if (ret) { 681 if (ret) {
@@ -705,16 +706,17 @@ int ipoib_ib_dev_open(struct net_device *dev)
705dev_stop: 706dev_stop:
706 if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) 707 if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
707 napi_enable(&priv->napi); 708 napi_enable(&priv->napi);
708 ipoib_ib_dev_stop(dev, 1); 709 ipoib_ib_dev_stop(dev, flush);
709 return -1; 710 return -1;
710} 711}
711 712
712static void ipoib_pkey_dev_check_presence(struct net_device *dev) 713void ipoib_pkey_dev_check_presence(struct net_device *dev)
713{ 714{
714 struct ipoib_dev_priv *priv = netdev_priv(dev); 715 struct ipoib_dev_priv *priv = netdev_priv(dev);
715 u16 pkey_index = 0;
716 716
717 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) 717 if (!(priv->pkey & 0x7fff) ||
718 ib_find_pkey(priv->ca, priv->port, priv->pkey,
719 &priv->pkey_index))
718 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); 720 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
719 else 721 else
720 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); 722 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
@@ -745,14 +747,6 @@ int ipoib_ib_dev_down(struct net_device *dev, int flush)
745 clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); 747 clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
746 netif_carrier_off(dev); 748 netif_carrier_off(dev);
747 749
748 /* Shutdown the P_Key thread if still active */
749 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
750 mutex_lock(&pkey_mutex);
751 set_bit(IPOIB_PKEY_STOP, &priv->flags);
752 cancel_delayed_work_sync(&priv->pkey_poll_task);
753 mutex_unlock(&pkey_mutex);
754 }
755
756 ipoib_mcast_stop_thread(dev, flush); 750 ipoib_mcast_stop_thread(dev, flush);
757 ipoib_mcast_dev_flush(dev); 751 ipoib_mcast_dev_flush(dev);
758 752
@@ -924,7 +918,7 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
924 (unsigned long) dev); 918 (unsigned long) dev);
925 919
926 if (dev->flags & IFF_UP) { 920 if (dev->flags & IFF_UP) {
927 if (ipoib_ib_dev_open(dev)) { 921 if (ipoib_ib_dev_open(dev, 1)) {
928 ipoib_transport_dev_cleanup(dev); 922 ipoib_transport_dev_cleanup(dev);
929 return -ENODEV; 923 return -ENODEV;
930 } 924 }
@@ -966,13 +960,27 @@ static inline int update_parent_pkey(struct ipoib_dev_priv *priv)
966 960
967 return 1; 961 return 1;
968} 962}
963/*
964 * returns 0 if pkey value was found in a different slot.
965 */
966static inline int update_child_pkey(struct ipoib_dev_priv *priv)
967{
968 u16 old_index = priv->pkey_index;
969
970 priv->pkey_index = 0;
971 ipoib_pkey_dev_check_presence(priv->dev);
972
973 if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) &&
974 (old_index == priv->pkey_index))
975 return 1;
976 return 0;
977}
969 978
970static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, 979static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv,
971 enum ipoib_flush_level level) 980 enum ipoib_flush_level level)
972{ 981{
973 struct ipoib_dev_priv *cpriv; 982 struct ipoib_dev_priv *cpriv;
974 struct net_device *dev = priv->dev; 983 struct net_device *dev = priv->dev;
975 u16 new_index;
976 int result; 984 int result;
977 985
978 down_read(&priv->vlan_rwsem); 986 down_read(&priv->vlan_rwsem);
@@ -986,16 +994,20 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv,
986 994
987 up_read(&priv->vlan_rwsem); 995 up_read(&priv->vlan_rwsem);
988 996
989 if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) { 997 if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) &&
990 /* for non-child devices must check/update the pkey value here */ 998 level != IPOIB_FLUSH_HEAVY) {
991 if (level == IPOIB_FLUSH_HEAVY &&
992 !test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
993 update_parent_pkey(priv);
994 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); 999 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
995 return; 1000 return;
996 } 1001 }
997 1002
998 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { 1003 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
1004 /* interface is down. update pkey and leave. */
1005 if (level == IPOIB_FLUSH_HEAVY) {
1006 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
1007 update_parent_pkey(priv);
1008 else
1009 update_child_pkey(priv);
1010 }
999 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); 1011 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
1000 return; 1012 return;
1001 } 1013 }
@@ -1005,20 +1017,13 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv,
1005 * (parent) devices should always takes what present in pkey index 0 1017 * (parent) devices should always takes what present in pkey index 0
1006 */ 1018 */
1007 if (test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 1019 if (test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
1008 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) { 1020 result = update_child_pkey(priv);
1009 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); 1021 if (result) {
1010 ipoib_ib_dev_down(dev, 0); 1022 /* restart QP only if P_Key index is changed */
1011 ipoib_ib_dev_stop(dev, 0);
1012 if (ipoib_pkey_dev_delay_open(dev))
1013 return;
1014 }
1015 /* restart QP only if P_Key index is changed */
1016 if (test_and_set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) &&
1017 new_index == priv->pkey_index) {
1018 ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); 1023 ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n");
1019 return; 1024 return;
1020 } 1025 }
1021 priv->pkey_index = new_index; 1026
1022 } else { 1027 } else {
1023 result = update_parent_pkey(priv); 1028 result = update_parent_pkey(priv);
1024 /* restart QP only if P_Key value changed */ 1029 /* restart QP only if P_Key value changed */
@@ -1038,8 +1043,12 @@ static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv,
1038 ipoib_ib_dev_down(dev, 0); 1043 ipoib_ib_dev_down(dev, 0);
1039 1044
1040 if (level == IPOIB_FLUSH_HEAVY) { 1045 if (level == IPOIB_FLUSH_HEAVY) {
1041 ipoib_ib_dev_stop(dev, 0); 1046 if (test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
1042 ipoib_ib_dev_open(dev); 1047 ipoib_ib_dev_stop(dev, 0);
1048 if (ipoib_ib_dev_open(dev, 0) != 0)
1049 return;
1050 if (netif_queue_stopped(dev))
1051 netif_start_queue(dev);
1043 } 1052 }
1044 1053
1045 /* 1054 /*
@@ -1094,54 +1103,4 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
1094 ipoib_transport_dev_cleanup(dev); 1103 ipoib_transport_dev_cleanup(dev);
1095} 1104}
1096 1105
1097/*
1098 * Delayed P_Key Assigment Interim Support
1099 *
1100 * The following is initial implementation of delayed P_Key assigment
1101 * mechanism. It is using the same approach implemented for the multicast
1102 * group join. The single goal of this implementation is to quickly address
1103 * Bug #2507. This implementation will probably be removed when the P_Key
1104 * change async notification is available.
1105 */
1106
1107void ipoib_pkey_poll(struct work_struct *work)
1108{
1109 struct ipoib_dev_priv *priv =
1110 container_of(work, struct ipoib_dev_priv, pkey_poll_task.work);
1111 struct net_device *dev = priv->dev;
1112
1113 ipoib_pkey_dev_check_presence(dev);
1114
1115 if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
1116 ipoib_open(dev);
1117 else {
1118 mutex_lock(&pkey_mutex);
1119 if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
1120 queue_delayed_work(ipoib_workqueue,
1121 &priv->pkey_poll_task,
1122 HZ);
1123 mutex_unlock(&pkey_mutex);
1124 }
1125}
1126
1127int ipoib_pkey_dev_delay_open(struct net_device *dev)
1128{
1129 struct ipoib_dev_priv *priv = netdev_priv(dev);
1130
1131 /* Look for the interface pkey value in the IB Port P_Key table and */
1132 /* set the interface pkey assigment flag */
1133 ipoib_pkey_dev_check_presence(dev);
1134 1106
1135 /* P_Key value not assigned yet - start polling */
1136 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
1137 mutex_lock(&pkey_mutex);
1138 clear_bit(IPOIB_PKEY_STOP, &priv->flags);
1139 queue_delayed_work(ipoib_workqueue,
1140 &priv->pkey_poll_task,
1141 HZ);
1142 mutex_unlock(&pkey_mutex);
1143 return 1;
1144 }
1145
1146 return 0;
1147}
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 4e675f4fecc9..1310acf6bf92 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -108,11 +108,11 @@ int ipoib_open(struct net_device *dev)
108 108
109 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 109 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
110 110
111 if (ipoib_pkey_dev_delay_open(dev)) 111 if (ipoib_ib_dev_open(dev, 1)) {
112 return 0; 112 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
113 113 return 0;
114 if (ipoib_ib_dev_open(dev))
115 goto err_disable; 114 goto err_disable;
115 }
116 116
117 if (ipoib_ib_dev_up(dev)) 117 if (ipoib_ib_dev_up(dev))
118 goto err_stop; 118 goto err_stop;
@@ -1379,7 +1379,6 @@ void ipoib_setup(struct net_device *dev)
1379 INIT_LIST_HEAD(&priv->dead_ahs); 1379 INIT_LIST_HEAD(&priv->dead_ahs);
1380 INIT_LIST_HEAD(&priv->multicast_list); 1380 INIT_LIST_HEAD(&priv->multicast_list);
1381 1381
1382 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
1383 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task); 1382 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
1384 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task); 1383 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
1385 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light); 1384 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c
index eb7973957a6e..61ee91d88380 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.c
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.c
@@ -596,20 +596,28 @@ iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
596 struct iser_conn *ib_conn; 596 struct iser_conn *ib_conn;
597 struct iscsi_endpoint *ep; 597 struct iscsi_endpoint *ep;
598 598
599 ep = iscsi_create_endpoint(sizeof(*ib_conn)); 599 ep = iscsi_create_endpoint(0);
600 if (!ep) 600 if (!ep)
601 return ERR_PTR(-ENOMEM); 601 return ERR_PTR(-ENOMEM);
602 602
603 ib_conn = ep->dd_data; 603 ib_conn = kzalloc(sizeof(*ib_conn), GFP_KERNEL);
604 if (!ib_conn) {
605 err = -ENOMEM;
606 goto failure;
607 }
608
609 ep->dd_data = ib_conn;
604 ib_conn->ep = ep; 610 ib_conn->ep = ep;
605 iser_conn_init(ib_conn); 611 iser_conn_init(ib_conn);
606 612
607 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, 613 err = iser_connect(ib_conn, NULL, dst_addr, non_blocking);
608 non_blocking);
609 if (err) 614 if (err)
610 return ERR_PTR(err); 615 goto failure;
611 616
612 return ep; 617 return ep;
618failure:
619 iscsi_destroy_endpoint(ep);
620 return ERR_PTR(err);
613} 621}
614 622
615static int 623static int
@@ -619,15 +627,16 @@ iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
619 int rc; 627 int rc;
620 628
621 ib_conn = ep->dd_data; 629 ib_conn = ep->dd_data;
622 rc = wait_event_interruptible_timeout(ib_conn->wait, 630 rc = wait_for_completion_interruptible_timeout(&ib_conn->up_completion,
623 ib_conn->state == ISER_CONN_UP, 631 msecs_to_jiffies(timeout_ms));
624 msecs_to_jiffies(timeout_ms));
625
626 /* if conn establishment failed, return error code to iscsi */ 632 /* if conn establishment failed, return error code to iscsi */
627 if (!rc && 633 if (rc == 0) {
628 (ib_conn->state == ISER_CONN_TERMINATING || 634 mutex_lock(&ib_conn->state_mutex);
629 ib_conn->state == ISER_CONN_DOWN)) 635 if (ib_conn->state == ISER_CONN_TERMINATING ||
630 rc = -1; 636 ib_conn->state == ISER_CONN_DOWN)
637 rc = -1;
638 mutex_unlock(&ib_conn->state_mutex);
639 }
631 640
632 iser_info("ib conn %p rc = %d\n", ib_conn, rc); 641 iser_info("ib conn %p rc = %d\n", ib_conn, rc);
633 642
@@ -646,19 +655,25 @@ iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
646 655
647 ib_conn = ep->dd_data; 656 ib_conn = ep->dd_data;
648 iser_info("ep %p ib conn %p state %d\n", ep, ib_conn, ib_conn->state); 657 iser_info("ep %p ib conn %p state %d\n", ep, ib_conn, ib_conn->state);
658 mutex_lock(&ib_conn->state_mutex);
649 iser_conn_terminate(ib_conn); 659 iser_conn_terminate(ib_conn);
650 660
651 /* 661 /*
652 * if iser_conn and iscsi_conn are bound, we must wait iscsi_conn_stop 662 * if iser_conn and iscsi_conn are bound, we must wait for
653 * call and ISER_CONN_DOWN state before freeing the iser resources. 663 * iscsi_conn_stop and flush errors completion before freeing
654 * otherwise we are safe to free resources immediately. 664 * the iser resources. Otherwise we are safe to free resources
665 * immediately.
655 */ 666 */
656 if (ib_conn->iscsi_conn) { 667 if (ib_conn->iscsi_conn) {
657 INIT_WORK(&ib_conn->release_work, iser_release_work); 668 INIT_WORK(&ib_conn->release_work, iser_release_work);
658 queue_work(release_wq, &ib_conn->release_work); 669 queue_work(release_wq, &ib_conn->release_work);
670 mutex_unlock(&ib_conn->state_mutex);
659 } else { 671 } else {
672 ib_conn->state = ISER_CONN_DOWN;
673 mutex_unlock(&ib_conn->state_mutex);
660 iser_conn_release(ib_conn); 674 iser_conn_release(ib_conn);
661 } 675 }
676 iscsi_destroy_endpoint(ep);
662} 677}
663 678
664static umode_t iser_attr_is_visible(int param_type, int param) 679static umode_t iser_attr_is_visible(int param_type, int param)
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index 97cd385bf7f7..c877dad381cb 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -326,7 +326,6 @@ struct iser_conn {
326 struct iser_device *device; /* device context */ 326 struct iser_device *device; /* device context */
327 struct rdma_cm_id *cma_id; /* CMA ID */ 327 struct rdma_cm_id *cma_id; /* CMA ID */
328 struct ib_qp *qp; /* QP */ 328 struct ib_qp *qp; /* QP */
329 wait_queue_head_t wait; /* waitq for conn/disconn */
330 unsigned qp_max_recv_dtos; /* num of rx buffers */ 329 unsigned qp_max_recv_dtos; /* num of rx buffers */
331 unsigned qp_max_recv_dtos_mask; /* above minus 1 */ 330 unsigned qp_max_recv_dtos_mask; /* above minus 1 */
332 unsigned min_posted_rx; /* qp_max_recv_dtos >> 2 */ 331 unsigned min_posted_rx; /* qp_max_recv_dtos >> 2 */
@@ -335,6 +334,9 @@ struct iser_conn {
335 char name[ISER_OBJECT_NAME_SIZE]; 334 char name[ISER_OBJECT_NAME_SIZE];
336 struct work_struct release_work; 335 struct work_struct release_work;
337 struct completion stop_completion; 336 struct completion stop_completion;
337 struct mutex state_mutex;
338 struct completion flush_completion;
339 struct completion up_completion;
338 struct list_head conn_list; /* entry in ig conn list */ 340 struct list_head conn_list; /* entry in ig conn list */
339 341
340 char *login_buf; 342 char *login_buf;
@@ -448,8 +450,8 @@ int iser_reg_rdma_mem_fastreg(struct iscsi_iser_task *task,
448 enum iser_data_dir cmd_dir); 450 enum iser_data_dir cmd_dir);
449 451
450int iser_connect(struct iser_conn *ib_conn, 452int iser_connect(struct iser_conn *ib_conn,
451 struct sockaddr_in *src_addr, 453 struct sockaddr *src_addr,
452 struct sockaddr_in *dst_addr, 454 struct sockaddr *dst_addr,
453 int non_blocking); 455 int non_blocking);
454 456
455int iser_reg_page_vec(struct iser_conn *ib_conn, 457int iser_reg_page_vec(struct iser_conn *ib_conn,
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index ea01075f9f9b..3ef167f97d6f 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -491,10 +491,9 @@ out_err:
491} 491}
492 492
493/** 493/**
494 * releases the QP objects, returns 0 on success, 494 * releases the QP object
495 * -1 on failure
496 */ 495 */
497static int iser_free_ib_conn_res(struct iser_conn *ib_conn) 496static void iser_free_ib_conn_res(struct iser_conn *ib_conn)
498{ 497{
499 int cq_index; 498 int cq_index;
500 BUG_ON(ib_conn == NULL); 499 BUG_ON(ib_conn == NULL);
@@ -513,8 +512,6 @@ static int iser_free_ib_conn_res(struct iser_conn *ib_conn)
513 } 512 }
514 513
515 ib_conn->qp = NULL; 514 ib_conn->qp = NULL;
516
517 return 0;
518} 515}
519 516
520/** 517/**
@@ -568,31 +565,40 @@ static void iser_device_try_release(struct iser_device *device)
568 mutex_unlock(&ig.device_list_mutex); 565 mutex_unlock(&ig.device_list_mutex);
569} 566}
570 567
568/**
569 * Called with state mutex held
570 **/
571static int iser_conn_state_comp_exch(struct iser_conn *ib_conn, 571static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
572 enum iser_ib_conn_state comp, 572 enum iser_ib_conn_state comp,
573 enum iser_ib_conn_state exch) 573 enum iser_ib_conn_state exch)
574{ 574{
575 int ret; 575 int ret;
576 576
577 spin_lock_bh(&ib_conn->lock);
578 if ((ret = (ib_conn->state == comp))) 577 if ((ret = (ib_conn->state == comp)))
579 ib_conn->state = exch; 578 ib_conn->state = exch;
580 spin_unlock_bh(&ib_conn->lock);
581 return ret; 579 return ret;
582} 580}
583 581
584void iser_release_work(struct work_struct *work) 582void iser_release_work(struct work_struct *work)
585{ 583{
586 struct iser_conn *ib_conn; 584 struct iser_conn *ib_conn;
585 int rc;
587 586
588 ib_conn = container_of(work, struct iser_conn, release_work); 587 ib_conn = container_of(work, struct iser_conn, release_work);
589 588
590 /* wait for .conn_stop callback */ 589 /* wait for .conn_stop callback */
591 wait_for_completion(&ib_conn->stop_completion); 590 rc = wait_for_completion_timeout(&ib_conn->stop_completion, 30 * HZ);
591 WARN_ON(rc == 0);
592 592
593 /* wait for the qp`s post send and post receive buffers to empty */ 593 /* wait for the qp`s post send and post receive buffers to empty */
594 wait_event_interruptible(ib_conn->wait, 594 rc = wait_for_completion_timeout(&ib_conn->flush_completion, 30 * HZ);
595 ib_conn->state == ISER_CONN_DOWN); 595 WARN_ON(rc == 0);
596
597 ib_conn->state = ISER_CONN_DOWN;
598
599 mutex_lock(&ib_conn->state_mutex);
600 ib_conn->state = ISER_CONN_DOWN;
601 mutex_unlock(&ib_conn->state_mutex);
596 602
597 iser_conn_release(ib_conn); 603 iser_conn_release(ib_conn);
598} 604}
@@ -604,23 +610,27 @@ void iser_conn_release(struct iser_conn *ib_conn)
604{ 610{
605 struct iser_device *device = ib_conn->device; 611 struct iser_device *device = ib_conn->device;
606 612
607 BUG_ON(ib_conn->state == ISER_CONN_UP);
608
609 mutex_lock(&ig.connlist_mutex); 613 mutex_lock(&ig.connlist_mutex);
610 list_del(&ib_conn->conn_list); 614 list_del(&ib_conn->conn_list);
611 mutex_unlock(&ig.connlist_mutex); 615 mutex_unlock(&ig.connlist_mutex);
616
617 mutex_lock(&ib_conn->state_mutex);
618 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
619
612 iser_free_rx_descriptors(ib_conn); 620 iser_free_rx_descriptors(ib_conn);
613 iser_free_ib_conn_res(ib_conn); 621 iser_free_ib_conn_res(ib_conn);
614 ib_conn->device = NULL; 622 ib_conn->device = NULL;
615 /* on EVENT_ADDR_ERROR there's no device yet for this conn */ 623 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
616 if (device != NULL) 624 if (device != NULL)
617 iser_device_try_release(device); 625 iser_device_try_release(device);
626 mutex_unlock(&ib_conn->state_mutex);
627
618 /* if cma handler context, the caller actually destroy the id */ 628 /* if cma handler context, the caller actually destroy the id */
619 if (ib_conn->cma_id != NULL) { 629 if (ib_conn->cma_id != NULL) {
620 rdma_destroy_id(ib_conn->cma_id); 630 rdma_destroy_id(ib_conn->cma_id);
621 ib_conn->cma_id = NULL; 631 ib_conn->cma_id = NULL;
622 } 632 }
623 iscsi_destroy_endpoint(ib_conn->ep); 633 kfree(ib_conn);
624} 634}
625 635
626/** 636/**
@@ -642,22 +652,31 @@ void iser_conn_terminate(struct iser_conn *ib_conn)
642 ib_conn,err); 652 ib_conn,err);
643} 653}
644 654
655/**
656 * Called with state mutex held
657 **/
645static void iser_connect_error(struct rdma_cm_id *cma_id) 658static void iser_connect_error(struct rdma_cm_id *cma_id)
646{ 659{
647 struct iser_conn *ib_conn; 660 struct iser_conn *ib_conn;
648 661
649 ib_conn = (struct iser_conn *)cma_id->context; 662 ib_conn = (struct iser_conn *)cma_id->context;
650
651 ib_conn->state = ISER_CONN_DOWN; 663 ib_conn->state = ISER_CONN_DOWN;
652 wake_up_interruptible(&ib_conn->wait);
653} 664}
654 665
666/**
667 * Called with state mutex held
668 **/
655static void iser_addr_handler(struct rdma_cm_id *cma_id) 669static void iser_addr_handler(struct rdma_cm_id *cma_id)
656{ 670{
657 struct iser_device *device; 671 struct iser_device *device;
658 struct iser_conn *ib_conn; 672 struct iser_conn *ib_conn;
659 int ret; 673 int ret;
660 674
675 ib_conn = (struct iser_conn *)cma_id->context;
676 if (ib_conn->state != ISER_CONN_PENDING)
677 /* bailout */
678 return;
679
661 device = iser_device_find_by_ib_device(cma_id); 680 device = iser_device_find_by_ib_device(cma_id);
662 if (!device) { 681 if (!device) {
663 iser_err("device lookup/creation failed\n"); 682 iser_err("device lookup/creation failed\n");
@@ -665,7 +684,6 @@ static void iser_addr_handler(struct rdma_cm_id *cma_id)
665 return; 684 return;
666 } 685 }
667 686
668 ib_conn = (struct iser_conn *)cma_id->context;
669 ib_conn->device = device; 687 ib_conn->device = device;
670 688
671 /* connection T10-PI support */ 689 /* connection T10-PI support */
@@ -689,18 +707,27 @@ static void iser_addr_handler(struct rdma_cm_id *cma_id)
689 } 707 }
690} 708}
691 709
710/**
711 * Called with state mutex held
712 **/
692static void iser_route_handler(struct rdma_cm_id *cma_id) 713static void iser_route_handler(struct rdma_cm_id *cma_id)
693{ 714{
694 struct rdma_conn_param conn_param; 715 struct rdma_conn_param conn_param;
695 int ret; 716 int ret;
696 struct iser_cm_hdr req_hdr; 717 struct iser_cm_hdr req_hdr;
718 struct iser_conn *ib_conn = (struct iser_conn *)cma_id->context;
719 struct iser_device *device = ib_conn->device;
720
721 if (ib_conn->state != ISER_CONN_PENDING)
722 /* bailout */
723 return;
697 724
698 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context); 725 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
699 if (ret) 726 if (ret)
700 goto failure; 727 goto failure;
701 728
702 memset(&conn_param, 0, sizeof conn_param); 729 memset(&conn_param, 0, sizeof conn_param);
703 conn_param.responder_resources = 4; 730 conn_param.responder_resources = device->dev_attr.max_qp_rd_atom;
704 conn_param.initiator_depth = 1; 731 conn_param.initiator_depth = 1;
705 conn_param.retry_count = 7; 732 conn_param.retry_count = 7;
706 conn_param.rnr_retry_count = 6; 733 conn_param.rnr_retry_count = 6;
@@ -728,12 +755,16 @@ static void iser_connected_handler(struct rdma_cm_id *cma_id)
728 struct ib_qp_attr attr; 755 struct ib_qp_attr attr;
729 struct ib_qp_init_attr init_attr; 756 struct ib_qp_init_attr init_attr;
730 757
758 ib_conn = (struct iser_conn *)cma_id->context;
759 if (ib_conn->state != ISER_CONN_PENDING)
760 /* bailout */
761 return;
762
731 (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr); 763 (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
732 iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num); 764 iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
733 765
734 ib_conn = (struct iser_conn *)cma_id->context; 766 ib_conn->state = ISER_CONN_UP;
735 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_PENDING, ISER_CONN_UP)) 767 complete(&ib_conn->up_completion);
736 wake_up_interruptible(&ib_conn->wait);
737} 768}
738 769
739static void iser_disconnected_handler(struct rdma_cm_id *cma_id) 770static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
@@ -752,19 +783,25 @@ static void iser_disconnected_handler(struct rdma_cm_id *cma_id)
752 iser_err("iscsi_iser connection isn't bound\n"); 783 iser_err("iscsi_iser connection isn't bound\n");
753 } 784 }
754 785
755 /* Complete the termination process if no posts are pending */ 786 /* Complete the termination process if no posts are pending. This code
787 * block also exists in iser_handle_comp_error(), but it is needed here
788 * for cases of no flushes at all, e.g. discovery over rdma.
789 */
756 if (ib_conn->post_recv_buf_count == 0 && 790 if (ib_conn->post_recv_buf_count == 0 &&
757 (atomic_read(&ib_conn->post_send_buf_count) == 0)) { 791 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
758 ib_conn->state = ISER_CONN_DOWN; 792 complete(&ib_conn->flush_completion);
759 wake_up_interruptible(&ib_conn->wait);
760 } 793 }
761} 794}
762 795
763static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) 796static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
764{ 797{
798 struct iser_conn *ib_conn;
799
800 ib_conn = (struct iser_conn *)cma_id->context;
765 iser_info("event %d status %d conn %p id %p\n", 801 iser_info("event %d status %d conn %p id %p\n",
766 event->event, event->status, cma_id->context, cma_id); 802 event->event, event->status, cma_id->context, cma_id);
767 803
804 mutex_lock(&ib_conn->state_mutex);
768 switch (event->event) { 805 switch (event->event) {
769 case RDMA_CM_EVENT_ADDR_RESOLVED: 806 case RDMA_CM_EVENT_ADDR_RESOLVED:
770 iser_addr_handler(cma_id); 807 iser_addr_handler(cma_id);
@@ -785,24 +822,28 @@ static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *eve
785 case RDMA_CM_EVENT_DISCONNECTED: 822 case RDMA_CM_EVENT_DISCONNECTED:
786 case RDMA_CM_EVENT_DEVICE_REMOVAL: 823 case RDMA_CM_EVENT_DEVICE_REMOVAL:
787 case RDMA_CM_EVENT_ADDR_CHANGE: 824 case RDMA_CM_EVENT_ADDR_CHANGE:
825 case RDMA_CM_EVENT_TIMEWAIT_EXIT:
788 iser_disconnected_handler(cma_id); 826 iser_disconnected_handler(cma_id);
789 break; 827 break;
790 default: 828 default:
791 iser_err("Unexpected RDMA CM event (%d)\n", event->event); 829 iser_err("Unexpected RDMA CM event (%d)\n", event->event);
792 break; 830 break;
793 } 831 }
832 mutex_unlock(&ib_conn->state_mutex);
794 return 0; 833 return 0;
795} 834}
796 835
797void iser_conn_init(struct iser_conn *ib_conn) 836void iser_conn_init(struct iser_conn *ib_conn)
798{ 837{
799 ib_conn->state = ISER_CONN_INIT; 838 ib_conn->state = ISER_CONN_INIT;
800 init_waitqueue_head(&ib_conn->wait);
801 ib_conn->post_recv_buf_count = 0; 839 ib_conn->post_recv_buf_count = 0;
802 atomic_set(&ib_conn->post_send_buf_count, 0); 840 atomic_set(&ib_conn->post_send_buf_count, 0);
803 init_completion(&ib_conn->stop_completion); 841 init_completion(&ib_conn->stop_completion);
842 init_completion(&ib_conn->flush_completion);
843 init_completion(&ib_conn->up_completion);
804 INIT_LIST_HEAD(&ib_conn->conn_list); 844 INIT_LIST_HEAD(&ib_conn->conn_list);
805 spin_lock_init(&ib_conn->lock); 845 spin_lock_init(&ib_conn->lock);
846 mutex_init(&ib_conn->state_mutex);
806} 847}
807 848
808 /** 849 /**
@@ -810,22 +851,21 @@ void iser_conn_init(struct iser_conn *ib_conn)
810 * sleeps until the connection is established or rejected 851 * sleeps until the connection is established or rejected
811 */ 852 */
812int iser_connect(struct iser_conn *ib_conn, 853int iser_connect(struct iser_conn *ib_conn,
813 struct sockaddr_in *src_addr, 854 struct sockaddr *src_addr,
814 struct sockaddr_in *dst_addr, 855 struct sockaddr *dst_addr,
815 int non_blocking) 856 int non_blocking)
816{ 857{
817 struct sockaddr *src, *dst;
818 int err = 0; 858 int err = 0;
819 859
820 sprintf(ib_conn->name, "%pI4:%d", 860 mutex_lock(&ib_conn->state_mutex);
821 &dst_addr->sin_addr.s_addr, dst_addr->sin_port); 861
862 sprintf(ib_conn->name, "%pISp", dst_addr);
863
864 iser_info("connecting to: %s\n", ib_conn->name);
822 865
823 /* the device is known only --after-- address resolution */ 866 /* the device is known only --after-- address resolution */
824 ib_conn->device = NULL; 867 ib_conn->device = NULL;
825 868
826 iser_info("connecting to: %pI4, port 0x%x\n",
827 &dst_addr->sin_addr, dst_addr->sin_port);
828
829 ib_conn->state = ISER_CONN_PENDING; 869 ib_conn->state = ISER_CONN_PENDING;
830 870
831 ib_conn->cma_id = rdma_create_id(iser_cma_handler, 871 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
@@ -837,23 +877,21 @@ int iser_connect(struct iser_conn *ib_conn,
837 goto id_failure; 877 goto id_failure;
838 } 878 }
839 879
840 src = (struct sockaddr *)src_addr; 880 err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
841 dst = (struct sockaddr *)dst_addr;
842 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
843 if (err) { 881 if (err) {
844 iser_err("rdma_resolve_addr failed: %d\n", err); 882 iser_err("rdma_resolve_addr failed: %d\n", err);
845 goto addr_failure; 883 goto addr_failure;
846 } 884 }
847 885
848 if (!non_blocking) { 886 if (!non_blocking) {
849 wait_event_interruptible(ib_conn->wait, 887 wait_for_completion_interruptible(&ib_conn->up_completion);
850 (ib_conn->state != ISER_CONN_PENDING));
851 888
852 if (ib_conn->state != ISER_CONN_UP) { 889 if (ib_conn->state != ISER_CONN_UP) {
853 err = -EIO; 890 err = -EIO;
854 goto connect_failure; 891 goto connect_failure;
855 } 892 }
856 } 893 }
894 mutex_unlock(&ib_conn->state_mutex);
857 895
858 mutex_lock(&ig.connlist_mutex); 896 mutex_lock(&ig.connlist_mutex);
859 list_add(&ib_conn->conn_list, &ig.connlist); 897 list_add(&ib_conn->conn_list, &ig.connlist);
@@ -865,6 +903,7 @@ id_failure:
865addr_failure: 903addr_failure:
866 ib_conn->state = ISER_CONN_DOWN; 904 ib_conn->state = ISER_CONN_DOWN;
867connect_failure: 905connect_failure:
906 mutex_unlock(&ib_conn->state_mutex);
868 iser_conn_release(ib_conn); 907 iser_conn_release(ib_conn);
869 return err; 908 return err;
870} 909}
@@ -1049,18 +1088,19 @@ static void iser_handle_comp_error(struct iser_tx_desc *desc,
1049 1088
1050 if (ib_conn->post_recv_buf_count == 0 && 1089 if (ib_conn->post_recv_buf_count == 0 &&
1051 atomic_read(&ib_conn->post_send_buf_count) == 0) { 1090 atomic_read(&ib_conn->post_send_buf_count) == 0) {
1052 /* getting here when the state is UP means that the conn is * 1091 /**
1053 * being terminated asynchronously from the iSCSI layer's * 1092 * getting here when the state is UP means that the conn is
1054 * perspective. */ 1093 * being terminated asynchronously from the iSCSI layer's
1055 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, 1094 * perspective. It is safe to peek at the connection state
1056 ISER_CONN_TERMINATING)) 1095 * since iscsi_conn_failure is allowed to be called twice.
1096 **/
1097 if (ib_conn->state == ISER_CONN_UP)
1057 iscsi_conn_failure(ib_conn->iscsi_conn, 1098 iscsi_conn_failure(ib_conn->iscsi_conn,
1058 ISCSI_ERR_CONN_FAILED); 1099 ISCSI_ERR_CONN_FAILED);
1059 1100
1060 /* no more non completed posts to the QP, complete the 1101 /* no more non completed posts to the QP, complete the
1061 * termination process w.o worrying on disconnect event */ 1102 * termination process w.o worrying on disconnect event */
1062 ib_conn->state = ISER_CONN_DOWN; 1103 complete(&ib_conn->flush_completion);
1063 wake_up_interruptible(&ib_conn->wait);
1064 } 1104 }
1065} 1105}
1066 1106
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index e3c2c5b4297f..62d2a18e1b41 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -130,6 +130,7 @@ static void srp_send_completion(struct ib_cq *cq, void *target_ptr);
130static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event); 130static int srp_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event);
131 131
132static struct scsi_transport_template *ib_srp_transport_template; 132static struct scsi_transport_template *ib_srp_transport_template;
133static struct workqueue_struct *srp_remove_wq;
133 134
134static struct ib_client srp_client = { 135static struct ib_client srp_client = {
135 .name = "srp", 136 .name = "srp",
@@ -731,7 +732,7 @@ static bool srp_queue_remove_work(struct srp_target_port *target)
731 spin_unlock_irq(&target->lock); 732 spin_unlock_irq(&target->lock);
732 733
733 if (changed) 734 if (changed)
734 queue_work(system_long_wq, &target->remove_work); 735 queue_work(srp_remove_wq, &target->remove_work);
735 736
736 return changed; 737 return changed;
737} 738}
@@ -1643,10 +1644,14 @@ static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
1643 SCSI_SENSE_BUFFERSIZE)); 1644 SCSI_SENSE_BUFFERSIZE));
1644 } 1645 }
1645 1646
1646 if (rsp->flags & (SRP_RSP_FLAG_DOOVER | SRP_RSP_FLAG_DOUNDER)) 1647 if (unlikely(rsp->flags & SRP_RSP_FLAG_DIUNDER))
1647 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_out_res_cnt));
1648 else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
1649 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt)); 1648 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));
1649 else if (unlikely(rsp->flags & SRP_RSP_FLAG_DIOVER))
1650 scsi_set_resid(scmnd, -be32_to_cpu(rsp->data_in_res_cnt));
1651 else if (unlikely(rsp->flags & SRP_RSP_FLAG_DOUNDER))
1652 scsi_set_resid(scmnd, be32_to_cpu(rsp->data_out_res_cnt));
1653 else if (unlikely(rsp->flags & SRP_RSP_FLAG_DOOVER))
1654 scsi_set_resid(scmnd, -be32_to_cpu(rsp->data_out_res_cnt));
1650 1655
1651 srp_free_req(target, req, scmnd, 1656 srp_free_req(target, req, scmnd,
1652 be32_to_cpu(rsp->req_lim_delta)); 1657 be32_to_cpu(rsp->req_lim_delta));
@@ -3261,9 +3266,10 @@ static void srp_remove_one(struct ib_device *device)
3261 spin_unlock(&host->target_lock); 3266 spin_unlock(&host->target_lock);
3262 3267
3263 /* 3268 /*
3264 * Wait for target port removal tasks. 3269 * Wait for tl_err and target port removal tasks.
3265 */ 3270 */
3266 flush_workqueue(system_long_wq); 3271 flush_workqueue(system_long_wq);
3272 flush_workqueue(srp_remove_wq);
3267 3273
3268 kfree(host); 3274 kfree(host);
3269 } 3275 }
@@ -3313,16 +3319,22 @@ static int __init srp_init_module(void)
3313 indirect_sg_entries = cmd_sg_entries; 3319 indirect_sg_entries = cmd_sg_entries;
3314 } 3320 }
3315 3321
3322 srp_remove_wq = create_workqueue("srp_remove");
3323 if (!srp_remove_wq) {
3324 ret = -ENOMEM;
3325 goto out;
3326 }
3327
3328 ret = -ENOMEM;
3316 ib_srp_transport_template = 3329 ib_srp_transport_template =
3317 srp_attach_transport(&ib_srp_transport_functions); 3330 srp_attach_transport(&ib_srp_transport_functions);
3318 if (!ib_srp_transport_template) 3331 if (!ib_srp_transport_template)
3319 return -ENOMEM; 3332 goto destroy_wq;
3320 3333
3321 ret = class_register(&srp_class); 3334 ret = class_register(&srp_class);
3322 if (ret) { 3335 if (ret) {
3323 pr_err("couldn't register class infiniband_srp\n"); 3336 pr_err("couldn't register class infiniband_srp\n");
3324 srp_release_transport(ib_srp_transport_template); 3337 goto release_tr;
3325 return ret;
3326 } 3338 }
3327 3339
3328 ib_sa_register_client(&srp_sa_client); 3340 ib_sa_register_client(&srp_sa_client);
@@ -3330,13 +3342,22 @@ static int __init srp_init_module(void)
3330 ret = ib_register_client(&srp_client); 3342 ret = ib_register_client(&srp_client);
3331 if (ret) { 3343 if (ret) {
3332 pr_err("couldn't register IB client\n"); 3344 pr_err("couldn't register IB client\n");
3333 srp_release_transport(ib_srp_transport_template); 3345 goto unreg_sa;
3334 ib_sa_unregister_client(&srp_sa_client);
3335 class_unregister(&srp_class);
3336 return ret;
3337 } 3346 }
3338 3347
3339 return 0; 3348out:
3349 return ret;
3350
3351unreg_sa:
3352 ib_sa_unregister_client(&srp_sa_client);
3353 class_unregister(&srp_class);
3354
3355release_tr:
3356 srp_release_transport(ib_srp_transport_template);
3357
3358destroy_wq:
3359 destroy_workqueue(srp_remove_wq);
3360 goto out;
3340} 3361}
3341 3362
3342static void __exit srp_cleanup_module(void) 3363static void __exit srp_cleanup_module(void)
@@ -3345,6 +3366,7 @@ static void __exit srp_cleanup_module(void)
3345 ib_sa_unregister_client(&srp_sa_client); 3366 ib_sa_unregister_client(&srp_sa_client);
3346 class_unregister(&srp_class); 3367 class_unregister(&srp_class);
3347 srp_release_transport(ib_srp_transport_template); 3368 srp_release_transport(ib_srp_transport_template);
3369 destroy_workqueue(srp_remove_wq);
3348} 3370}
3349 3371
3350module_init(srp_init_module); 3372module_init(srp_init_module);
diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
index fe09f2788b15..d28a8c284da9 100644
--- a/drivers/infiniband/ulp/srpt/ib_srpt.c
+++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
@@ -198,6 +198,7 @@ static void srpt_event_handler(struct ib_event_handler *handler,
198 case IB_EVENT_PKEY_CHANGE: 198 case IB_EVENT_PKEY_CHANGE:
199 case IB_EVENT_SM_CHANGE: 199 case IB_EVENT_SM_CHANGE:
200 case IB_EVENT_CLIENT_REREGISTER: 200 case IB_EVENT_CLIENT_REREGISTER:
201 case IB_EVENT_GID_CHANGE:
201 /* Refresh port data asynchronously. */ 202 /* Refresh port data asynchronously. */
202 if (event->element.port_num <= sdev->device->phys_port_cnt) { 203 if (event->element.port_num <= sdev->device->phys_port_cnt) {
203 sport = &sdev->port[event->element.port_num - 1]; 204 sport = &sdev->port[event->element.port_num - 1];
@@ -563,7 +564,7 @@ static int srpt_refresh_port(struct srpt_port *sport)
563 &reg_req, 0, 564 &reg_req, 0,
564 srpt_mad_send_handler, 565 srpt_mad_send_handler,
565 srpt_mad_recv_handler, 566 srpt_mad_recv_handler,
566 sport); 567 sport, 0);
567 if (IS_ERR(sport->mad_agent)) { 568 if (IS_ERR(sport->mad_agent)) {
568 ret = PTR_ERR(sport->mad_agent); 569 ret = PTR_ERR(sport->mad_agent);
569 sport->mad_agent = NULL; 570 sport->mad_agent = NULL;
diff --git a/drivers/net/ethernet/emulex/benet/be.h b/drivers/net/ethernet/emulex/benet/be.h
index 811f1351db7a..43e08d0bc3d3 100644
--- a/drivers/net/ethernet/emulex/benet/be.h
+++ b/drivers/net/ethernet/emulex/benet/be.h
@@ -897,5 +897,6 @@ void be_roce_dev_remove(struct be_adapter *);
897 */ 897 */
898void be_roce_dev_open(struct be_adapter *); 898void be_roce_dev_open(struct be_adapter *);
899void be_roce_dev_close(struct be_adapter *); 899void be_roce_dev_close(struct be_adapter *);
900void be_roce_dev_shutdown(struct be_adapter *);
900 901
901#endif /* BE_H */ 902#endif /* BE_H */
diff --git a/drivers/net/ethernet/emulex/benet/be_main.c b/drivers/net/ethernet/emulex/benet/be_main.c
index db4ff14ff18f..9cdeda54674a 100644
--- a/drivers/net/ethernet/emulex/benet/be_main.c
+++ b/drivers/net/ethernet/emulex/benet/be_main.c
@@ -5014,6 +5014,7 @@ static void be_shutdown(struct pci_dev *pdev)
5014 if (!adapter) 5014 if (!adapter)
5015 return; 5015 return;
5016 5016
5017 be_roce_dev_shutdown(adapter);
5017 cancel_delayed_work_sync(&adapter->work); 5018 cancel_delayed_work_sync(&adapter->work);
5018 cancel_delayed_work_sync(&adapter->func_recovery_work); 5019 cancel_delayed_work_sync(&adapter->func_recovery_work);
5019 5020
diff --git a/drivers/net/ethernet/emulex/benet/be_roce.c b/drivers/net/ethernet/emulex/benet/be_roce.c
index 5bf16603a3e9..ef4672dc7357 100644
--- a/drivers/net/ethernet/emulex/benet/be_roce.c
+++ b/drivers/net/ethernet/emulex/benet/be_roce.c
@@ -120,7 +120,8 @@ static void _be_roce_dev_open(struct be_adapter *adapter)
120{ 120{
121 if (ocrdma_drv && adapter->ocrdma_dev && 121 if (ocrdma_drv && adapter->ocrdma_dev &&
122 ocrdma_drv->state_change_handler) 122 ocrdma_drv->state_change_handler)
123 ocrdma_drv->state_change_handler(adapter->ocrdma_dev, 0); 123 ocrdma_drv->state_change_handler(adapter->ocrdma_dev,
124 BE_DEV_UP);
124} 125}
125 126
126void be_roce_dev_open(struct be_adapter *adapter) 127void be_roce_dev_open(struct be_adapter *adapter)
@@ -136,7 +137,8 @@ static void _be_roce_dev_close(struct be_adapter *adapter)
136{ 137{
137 if (ocrdma_drv && adapter->ocrdma_dev && 138 if (ocrdma_drv && adapter->ocrdma_dev &&
138 ocrdma_drv->state_change_handler) 139 ocrdma_drv->state_change_handler)
139 ocrdma_drv->state_change_handler(adapter->ocrdma_dev, 1); 140 ocrdma_drv->state_change_handler(adapter->ocrdma_dev,
141 BE_DEV_DOWN);
140} 142}
141 143
142void be_roce_dev_close(struct be_adapter *adapter) 144void be_roce_dev_close(struct be_adapter *adapter)
@@ -148,6 +150,18 @@ void be_roce_dev_close(struct be_adapter *adapter)
148 } 150 }
149} 151}
150 152
153void be_roce_dev_shutdown(struct be_adapter *adapter)
154{
155 if (be_roce_supported(adapter)) {
156 mutex_lock(&be_adapter_list_lock);
157 if (ocrdma_drv && adapter->ocrdma_dev &&
158 ocrdma_drv->state_change_handler)
159 ocrdma_drv->state_change_handler(adapter->ocrdma_dev,
160 BE_DEV_SHUTDOWN);
161 mutex_unlock(&be_adapter_list_lock);
162 }
163}
164
151int be_roce_register_driver(struct ocrdma_driver *drv) 165int be_roce_register_driver(struct ocrdma_driver *drv)
152{ 166{
153 struct be_adapter *dev; 167 struct be_adapter *dev;
diff --git a/drivers/net/ethernet/emulex/benet/be_roce.h b/drivers/net/ethernet/emulex/benet/be_roce.h
index a3d9e96c18eb..e6f7eb1a7d87 100644
--- a/drivers/net/ethernet/emulex/benet/be_roce.h
+++ b/drivers/net/ethernet/emulex/benet/be_roce.h
@@ -62,7 +62,8 @@ struct ocrdma_driver {
62 62
63enum { 63enum {
64 BE_DEV_UP = 0, 64 BE_DEV_UP = 0,
65 BE_DEV_DOWN = 1 65 BE_DEV_DOWN = 1,
66 BE_DEV_SHUTDOWN = 2
66}; 67};
67 68
68/* APIs for RoCE driver to register callback handlers, 69/* APIs for RoCE driver to register callback handlers,
diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c
index 5d940a26055c..65a4a0f88ea0 100644
--- a/drivers/net/ethernet/mellanox/mlx4/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c
@@ -1311,6 +1311,15 @@ static struct mlx4_cmd_info cmd_info[] = {
1311 .wrapper = mlx4_MAD_IFC_wrapper 1311 .wrapper = mlx4_MAD_IFC_wrapper
1312 }, 1312 },
1313 { 1313 {
1314 .opcode = MLX4_CMD_MAD_DEMUX,
1315 .has_inbox = false,
1316 .has_outbox = false,
1317 .out_is_imm = false,
1318 .encode_slave_id = false,
1319 .verify = NULL,
1320 .wrapper = mlx4_CMD_EPERM_wrapper
1321 },
1322 {
1314 .opcode = MLX4_CMD_QUERY_IF_STAT, 1323 .opcode = MLX4_CMD_QUERY_IF_STAT,
1315 .has_inbox = false, 1324 .has_inbox = false,
1316 .has_outbox = true, 1325 .has_outbox = true,
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 688e1eabab29..494753e44ae3 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -136,7 +136,8 @@ static void dump_dev_cap_flags2(struct mlx4_dev *dev, u64 flags)
136 [7] = "FSM (MAC anti-spoofing) support", 136 [7] = "FSM (MAC anti-spoofing) support",
137 [8] = "Dynamic QP updates support", 137 [8] = "Dynamic QP updates support",
138 [9] = "Device managed flow steering IPoIB support", 138 [9] = "Device managed flow steering IPoIB support",
139 [10] = "TCP/IP offloads/flow-steering for VXLAN support" 139 [10] = "TCP/IP offloads/flow-steering for VXLAN support",
140 [11] = "MAD DEMUX (Secure-Host) support"
140 }; 141 };
141 int i; 142 int i;
142 143
@@ -571,6 +572,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
571#define QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET 0xa0 572#define QUERY_DEV_CAP_MAX_ICM_SZ_OFFSET 0xa0
572#define QUERY_DEV_CAP_FW_REASSIGN_MAC 0x9d 573#define QUERY_DEV_CAP_FW_REASSIGN_MAC 0x9d
573#define QUERY_DEV_CAP_VXLAN 0x9e 574#define QUERY_DEV_CAP_VXLAN 0x9e
575#define QUERY_DEV_CAP_MAD_DEMUX_OFFSET 0xb0
574 576
575 dev_cap->flags2 = 0; 577 dev_cap->flags2 = 0;
576 mailbox = mlx4_alloc_cmd_mailbox(dev); 578 mailbox = mlx4_alloc_cmd_mailbox(dev);
@@ -748,6 +750,11 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
748 MLX4_GET(dev_cap->max_counters, outbox, 750 MLX4_GET(dev_cap->max_counters, outbox,
749 QUERY_DEV_CAP_MAX_COUNTERS_OFFSET); 751 QUERY_DEV_CAP_MAX_COUNTERS_OFFSET);
750 752
753 MLX4_GET(field32, outbox,
754 QUERY_DEV_CAP_MAD_DEMUX_OFFSET);
755 if (field32 & (1 << 0))
756 dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_MAD_DEMUX;
757
751 MLX4_GET(field32, outbox, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET); 758 MLX4_GET(field32, outbox, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
752 if (field32 & (1 << 16)) 759 if (field32 & (1 << 16))
753 dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP; 760 dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_UPDATE_QP;
@@ -2016,3 +2023,85 @@ void mlx4_opreq_action(struct work_struct *work)
2016out: 2023out:
2017 mlx4_free_cmd_mailbox(dev, mailbox); 2024 mlx4_free_cmd_mailbox(dev, mailbox);
2018} 2025}
2026
2027static int mlx4_check_smp_firewall_active(struct mlx4_dev *dev,
2028 struct mlx4_cmd_mailbox *mailbox)
2029{
2030#define MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET 0x10
2031#define MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET 0x20
2032#define MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET 0x40
2033#define MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET 0x70
2034
2035 u32 set_attr_mask, getresp_attr_mask;
2036 u32 trap_attr_mask, traprepress_attr_mask;
2037
2038 MLX4_GET(set_attr_mask, mailbox->buf,
2039 MLX4_CMD_MAD_DEMUX_SET_ATTR_OFFSET);
2040 mlx4_dbg(dev, "SMP firewall set_attribute_mask = 0x%x\n",
2041 set_attr_mask);
2042
2043 MLX4_GET(getresp_attr_mask, mailbox->buf,
2044 MLX4_CMD_MAD_DEMUX_GETRESP_ATTR_OFFSET);
2045 mlx4_dbg(dev, "SMP firewall getresp_attribute_mask = 0x%x\n",
2046 getresp_attr_mask);
2047
2048 MLX4_GET(trap_attr_mask, mailbox->buf,
2049 MLX4_CMD_MAD_DEMUX_TRAP_ATTR_OFFSET);
2050 mlx4_dbg(dev, "SMP firewall trap_attribute_mask = 0x%x\n",
2051 trap_attr_mask);
2052
2053 MLX4_GET(traprepress_attr_mask, mailbox->buf,
2054 MLX4_CMD_MAD_DEMUX_TRAP_REPRESS_ATTR_OFFSET);
2055 mlx4_dbg(dev, "SMP firewall traprepress_attribute_mask = 0x%x\n",
2056 traprepress_attr_mask);
2057
2058 if (set_attr_mask && getresp_attr_mask && trap_attr_mask &&
2059 traprepress_attr_mask)
2060 return 1;
2061
2062 return 0;
2063}
2064
2065int mlx4_config_mad_demux(struct mlx4_dev *dev)
2066{
2067 struct mlx4_cmd_mailbox *mailbox;
2068 int secure_host_active;
2069 int err;
2070
2071 /* Check if mad_demux is supported */
2072 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_MAD_DEMUX))
2073 return 0;
2074
2075 mailbox = mlx4_alloc_cmd_mailbox(dev);
2076 if (IS_ERR(mailbox)) {
2077 mlx4_warn(dev, "Failed to allocate mailbox for cmd MAD_DEMUX");
2078 return -ENOMEM;
2079 }
2080
2081 /* Query mad_demux to find out which MADs are handled by internal sma */
2082 err = mlx4_cmd_box(dev, 0, mailbox->dma, 0x01 /* subn mgmt class */,
2083 MLX4_CMD_MAD_DEMUX_QUERY_RESTR, MLX4_CMD_MAD_DEMUX,
2084 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2085 if (err) {
2086 mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: query restrictions failed (%d)\n",
2087 err);
2088 goto out;
2089 }
2090
2091 secure_host_active = mlx4_check_smp_firewall_active(dev, mailbox);
2092
2093 /* Config mad_demux to handle all MADs returned by the query above */
2094 err = mlx4_cmd(dev, mailbox->dma, 0x01 /* subn mgmt class */,
2095 MLX4_CMD_MAD_DEMUX_CONFIG, MLX4_CMD_MAD_DEMUX,
2096 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
2097 if (err) {
2098 mlx4_warn(dev, "MLX4_CMD_MAD_DEMUX: configure failed (%d)\n", err);
2099 goto out;
2100 }
2101
2102 if (secure_host_active)
2103 mlx4_warn(dev, "HCA operating in secure-host mode. SMP firewall activated.\n");
2104out:
2105 mlx4_free_cmd_mailbox(dev, mailbox);
2106 return err;
2107}
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index 80b8c5f30e4e..0158689906fd 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -1853,6 +1853,11 @@ static int mlx4_setup_hca(struct mlx4_dev *dev)
1853 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n"); 1853 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
1854 goto err_mr_table_free; 1854 goto err_mr_table_free;
1855 } 1855 }
1856 err = mlx4_config_mad_demux(dev);
1857 if (err) {
1858 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
1859 goto err_mcg_table_free;
1860 }
1856 } 1861 }
1857 1862
1858 err = mlx4_init_eq_table(dev); 1863 err = mlx4_init_eq_table(dev);
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
index 13fbcd03c3e4..b508c7887ef8 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h
@@ -274,6 +274,8 @@ struct mlx4_icm_table {
274#define MLX4_MPT_FLAG_PHYSICAL (1 << 9) 274#define MLX4_MPT_FLAG_PHYSICAL (1 << 9)
275#define MLX4_MPT_FLAG_REGION (1 << 8) 275#define MLX4_MPT_FLAG_REGION (1 << 8)
276 276
277#define MLX4_MPT_PD_MASK (0x1FFFFUL)
278#define MLX4_MPT_PD_VF_MASK (0xFE0000UL)
277#define MLX4_MPT_PD_FLAG_FAST_REG (1 << 27) 279#define MLX4_MPT_PD_FLAG_FAST_REG (1 << 27)
278#define MLX4_MPT_PD_FLAG_RAE (1 << 28) 280#define MLX4_MPT_PD_FLAG_RAE (1 << 28)
279#define MLX4_MPT_PD_FLAG_EN_INV (3 << 24) 281#define MLX4_MPT_PD_FLAG_EN_INV (3 << 24)
@@ -1306,5 +1308,6 @@ void mlx4_init_quotas(struct mlx4_dev *dev);
1306int mlx4_get_slave_num_gids(struct mlx4_dev *dev, int slave, int port); 1308int mlx4_get_slave_num_gids(struct mlx4_dev *dev, int slave, int port);
1307/* Returns the VF index of slave */ 1309/* Returns the VF index of slave */
1308int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave); 1310int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave);
1311int mlx4_config_mad_demux(struct mlx4_dev *dev);
1309 1312
1310#endif /* MLX4_H */ 1313#endif /* MLX4_H */
diff --git a/drivers/net/ethernet/mellanox/mlx4/mr.c b/drivers/net/ethernet/mellanox/mlx4/mr.c
index 2839abb878a6..7d717eccb7b0 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mr.c
+++ b/drivers/net/ethernet/mellanox/mlx4/mr.c
@@ -298,6 +298,131 @@ static int mlx4_HW2SW_MPT(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox
298 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED); 298 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
299} 299}
300 300
301int mlx4_mr_hw_get_mpt(struct mlx4_dev *dev, struct mlx4_mr *mmr,
302 struct mlx4_mpt_entry ***mpt_entry)
303{
304 int err;
305 int key = key_to_hw_index(mmr->key) & (dev->caps.num_mpts - 1);
306 struct mlx4_cmd_mailbox *mailbox = NULL;
307
308 /* Make sure that at this point we have single-threaded access only */
309
310 if (mmr->enabled != MLX4_MPT_EN_HW)
311 return -EINVAL;
312
313 err = mlx4_HW2SW_MPT(dev, NULL, key);
314
315 if (err) {
316 mlx4_warn(dev, "HW2SW_MPT failed (%d).", err);
317 mlx4_warn(dev, "Most likely the MR has MWs bound to it.\n");
318 return err;
319 }
320
321 mmr->enabled = MLX4_MPT_EN_SW;
322
323 if (!mlx4_is_mfunc(dev)) {
324 **mpt_entry = mlx4_table_find(
325 &mlx4_priv(dev)->mr_table.dmpt_table,
326 key, NULL);
327 } else {
328 mailbox = mlx4_alloc_cmd_mailbox(dev);
329 if (IS_ERR_OR_NULL(mailbox))
330 return PTR_ERR(mailbox);
331
332 err = mlx4_cmd_box(dev, 0, mailbox->dma, key,
333 0, MLX4_CMD_QUERY_MPT,
334 MLX4_CMD_TIME_CLASS_B,
335 MLX4_CMD_WRAPPED);
336
337 if (err)
338 goto free_mailbox;
339
340 *mpt_entry = (struct mlx4_mpt_entry **)&mailbox->buf;
341 }
342
343 if (!(*mpt_entry) || !(**mpt_entry)) {
344 err = -ENOMEM;
345 goto free_mailbox;
346 }
347
348 return 0;
349
350free_mailbox:
351 mlx4_free_cmd_mailbox(dev, mailbox);
352 return err;
353}
354EXPORT_SYMBOL_GPL(mlx4_mr_hw_get_mpt);
355
356int mlx4_mr_hw_write_mpt(struct mlx4_dev *dev, struct mlx4_mr *mmr,
357 struct mlx4_mpt_entry **mpt_entry)
358{
359 int err;
360
361 if (!mlx4_is_mfunc(dev)) {
362 /* Make sure any changes to this entry are flushed */
363 wmb();
364
365 *(u8 *)(*mpt_entry) = MLX4_MPT_STATUS_HW;
366
367 /* Make sure the new status is written */
368 wmb();
369
370 err = mlx4_SYNC_TPT(dev);
371 } else {
372 int key = key_to_hw_index(mmr->key) & (dev->caps.num_mpts - 1);
373
374 struct mlx4_cmd_mailbox *mailbox =
375 container_of((void *)mpt_entry, struct mlx4_cmd_mailbox,
376 buf);
377
378 err = mlx4_SW2HW_MPT(dev, mailbox, key);
379 }
380
381 mmr->pd = be32_to_cpu((*mpt_entry)->pd_flags) & MLX4_MPT_PD_MASK;
382 if (!err)
383 mmr->enabled = MLX4_MPT_EN_HW;
384 return err;
385}
386EXPORT_SYMBOL_GPL(mlx4_mr_hw_write_mpt);
387
388void mlx4_mr_hw_put_mpt(struct mlx4_dev *dev,
389 struct mlx4_mpt_entry **mpt_entry)
390{
391 if (mlx4_is_mfunc(dev)) {
392 struct mlx4_cmd_mailbox *mailbox =
393 container_of((void *)mpt_entry, struct mlx4_cmd_mailbox,
394 buf);
395 mlx4_free_cmd_mailbox(dev, mailbox);
396 }
397}
398EXPORT_SYMBOL_GPL(mlx4_mr_hw_put_mpt);
399
400int mlx4_mr_hw_change_pd(struct mlx4_dev *dev, struct mlx4_mpt_entry *mpt_entry,
401 u32 pdn)
402{
403 u32 pd_flags = be32_to_cpu(mpt_entry->pd_flags);
404 /* The wrapper function will put the slave's id here */
405 if (mlx4_is_mfunc(dev))
406 pd_flags &= ~MLX4_MPT_PD_VF_MASK;
407 mpt_entry->pd_flags = cpu_to_be32((pd_flags & ~MLX4_MPT_PD_MASK) |
408 (pdn & MLX4_MPT_PD_MASK)
409 | MLX4_MPT_PD_FLAG_EN_INV);
410 return 0;
411}
412EXPORT_SYMBOL_GPL(mlx4_mr_hw_change_pd);
413
414int mlx4_mr_hw_change_access(struct mlx4_dev *dev,
415 struct mlx4_mpt_entry *mpt_entry,
416 u32 access)
417{
418 u32 flags = (be32_to_cpu(mpt_entry->flags) & ~MLX4_PERM_MASK) |
419 (access & MLX4_PERM_MASK);
420
421 mpt_entry->flags = cpu_to_be32(flags);
422 return 0;
423}
424EXPORT_SYMBOL_GPL(mlx4_mr_hw_change_access);
425
301static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd, 426static int mlx4_mr_alloc_reserved(struct mlx4_dev *dev, u32 mridx, u32 pd,
302 u64 iova, u64 size, u32 access, int npages, 427 u64 iova, u64 size, u32 access, int npages,
303 int page_shift, struct mlx4_mr *mr) 428 int page_shift, struct mlx4_mr *mr)
@@ -463,6 +588,41 @@ int mlx4_mr_free(struct mlx4_dev *dev, struct mlx4_mr *mr)
463} 588}
464EXPORT_SYMBOL_GPL(mlx4_mr_free); 589EXPORT_SYMBOL_GPL(mlx4_mr_free);
465 590
591void mlx4_mr_rereg_mem_cleanup(struct mlx4_dev *dev, struct mlx4_mr *mr)
592{
593 mlx4_mtt_cleanup(dev, &mr->mtt);
594}
595EXPORT_SYMBOL_GPL(mlx4_mr_rereg_mem_cleanup);
596
597int mlx4_mr_rereg_mem_write(struct mlx4_dev *dev, struct mlx4_mr *mr,
598 u64 iova, u64 size, int npages,
599 int page_shift, struct mlx4_mpt_entry *mpt_entry)
600{
601 int err;
602
603 mpt_entry->start = cpu_to_be64(mr->iova);
604 mpt_entry->length = cpu_to_be64(mr->size);
605 mpt_entry->entity_size = cpu_to_be32(mr->mtt.page_shift);
606
607 err = mlx4_mtt_init(dev, npages, page_shift, &mr->mtt);
608 if (err)
609 return err;
610
611 if (mr->mtt.order < 0) {
612 mpt_entry->flags |= cpu_to_be32(MLX4_MPT_FLAG_PHYSICAL);
613 mpt_entry->mtt_addr = 0;
614 } else {
615 mpt_entry->mtt_addr = cpu_to_be64(mlx4_mtt_addr(dev,
616 &mr->mtt));
617 if (mr->mtt.page_shift == 0)
618 mpt_entry->mtt_sz = cpu_to_be32(1 << mr->mtt.order);
619 }
620 mr->enabled = MLX4_MPT_EN_SW;
621
622 return 0;
623}
624EXPORT_SYMBOL_GPL(mlx4_mr_rereg_mem_write);
625
466int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr) 626int mlx4_mr_enable(struct mlx4_dev *dev, struct mlx4_mr *mr)
467{ 627{
468 struct mlx4_cmd_mailbox *mailbox; 628 struct mlx4_cmd_mailbox *mailbox;
diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
index 0efc1368e5a8..1089367fed22 100644
--- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
+++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c
@@ -2613,12 +2613,34 @@ int mlx4_QUERY_MPT_wrapper(struct mlx4_dev *dev, int slave,
2613 if (err) 2613 if (err)
2614 return err; 2614 return err;
2615 2615
2616 if (mpt->com.from_state != RES_MPT_HW) { 2616 if (mpt->com.from_state == RES_MPT_MAPPED) {
2617 /* In order to allow rereg in SRIOV, we need to alter the MPT entry. To do
2618 * that, the VF must read the MPT. But since the MPT entry memory is not
2619 * in the VF's virtual memory space, it must use QUERY_MPT to obtain the
2620 * entry contents. To guarantee that the MPT cannot be changed, the driver
2621 * must perform HW2SW_MPT before this query and return the MPT entry to HW
2622 * ownership fofollowing the change. The change here allows the VF to
2623 * perform QUERY_MPT also when the entry is in SW ownership.
2624 */
2625 struct mlx4_mpt_entry *mpt_entry = mlx4_table_find(
2626 &mlx4_priv(dev)->mr_table.dmpt_table,
2627 mpt->key, NULL);
2628
2629 if (NULL == mpt_entry || NULL == outbox->buf) {
2630 err = -EINVAL;
2631 goto out;
2632 }
2633
2634 memcpy(outbox->buf, mpt_entry, sizeof(*mpt_entry));
2635
2636 err = 0;
2637 } else if (mpt->com.from_state == RES_MPT_HW) {
2638 err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2639 } else {
2617 err = -EBUSY; 2640 err = -EBUSY;
2618 goto out; 2641 goto out;
2619 } 2642 }
2620 2643
2621 err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd);
2622 2644
2623out: 2645out:
2624 put_res(dev, slave, id, RES_MPT); 2646 put_res(dev, slave, id, RES_MPT);
diff --git a/drivers/scsi/scsi_transport_srp.c b/drivers/scsi/scsi_transport_srp.c
index 43fea2219f83..ae45bd99baed 100644
--- a/drivers/scsi/scsi_transport_srp.c
+++ b/drivers/scsi/scsi_transport_srp.c
@@ -472,7 +472,8 @@ static void __srp_start_tl_fail_timers(struct srp_rport *rport)
472 if (delay > 0) 472 if (delay > 0)
473 queue_delayed_work(system_long_wq, &rport->reconnect_work, 473 queue_delayed_work(system_long_wq, &rport->reconnect_work,
474 1UL * delay * HZ); 474 1UL * delay * HZ);
475 if (srp_rport_set_state(rport, SRP_RPORT_BLOCKED) == 0) { 475 if ((fast_io_fail_tmo >= 0 || dev_loss_tmo >= 0) &&
476 srp_rport_set_state(rport, SRP_RPORT_BLOCKED) == 0) {
476 pr_debug("%s new state: %d\n", dev_name(&shost->shost_gendev), 477 pr_debug("%s new state: %d\n", dev_name(&shost->shost_gendev),
477 rport->state); 478 rport->state);
478 scsi_target_block(&shost->shost_gendev); 479 scsi_target_block(&shost->shost_gendev);
diff --git a/include/linux/mlx4/cmd.h b/include/linux/mlx4/cmd.h
index c8450366c130..379c02648ab3 100644
--- a/include/linux/mlx4/cmd.h
+++ b/include/linux/mlx4/cmd.h
@@ -116,6 +116,7 @@ enum {
116 /* special QP and management commands */ 116 /* special QP and management commands */
117 MLX4_CMD_CONF_SPECIAL_QP = 0x23, 117 MLX4_CMD_CONF_SPECIAL_QP = 0x23,
118 MLX4_CMD_MAD_IFC = 0x24, 118 MLX4_CMD_MAD_IFC = 0x24,
119 MLX4_CMD_MAD_DEMUX = 0x203,
119 120
120 /* multicast commands */ 121 /* multicast commands */
121 MLX4_CMD_READ_MCG = 0x25, 122 MLX4_CMD_READ_MCG = 0x25,
@@ -186,6 +187,12 @@ enum {
186}; 187};
187 188
188enum { 189enum {
190 MLX4_CMD_MAD_DEMUX_CONFIG = 0,
191 MLX4_CMD_MAD_DEMUX_QUERY_STATE = 1,
192 MLX4_CMD_MAD_DEMUX_QUERY_RESTR = 2, /* Query mad demux restrictions */
193};
194
195enum {
189 MLX4_CMD_WRAPPED, 196 MLX4_CMD_WRAPPED,
190 MLX4_CMD_NATIVE 197 MLX4_CMD_NATIVE
191}; 198};
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index e15b1544ea83..071f6b234604 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -183,6 +183,7 @@ enum {
183 MLX4_DEV_CAP_FLAG2_UPDATE_QP = 1LL << 8, 183 MLX4_DEV_CAP_FLAG2_UPDATE_QP = 1LL << 8,
184 MLX4_DEV_CAP_FLAG2_DMFS_IPOIB = 1LL << 9, 184 MLX4_DEV_CAP_FLAG2_DMFS_IPOIB = 1LL << 9,
185 MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS = 1LL << 10, 185 MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS = 1LL << 10,
186 MLX4_DEV_CAP_FLAG2_MAD_DEMUX = 1LL << 11,
186}; 187};
187 188
188enum { 189enum {
@@ -273,6 +274,7 @@ enum {
273 MLX4_PERM_REMOTE_WRITE = 1 << 13, 274 MLX4_PERM_REMOTE_WRITE = 1 << 13,
274 MLX4_PERM_ATOMIC = 1 << 14, 275 MLX4_PERM_ATOMIC = 1 << 14,
275 MLX4_PERM_BIND_MW = 1 << 15, 276 MLX4_PERM_BIND_MW = 1 << 15,
277 MLX4_PERM_MASK = 0xFC00
276}; 278};
277 279
278enum { 280enum {
@@ -1254,6 +1256,21 @@ int mlx4_vf_smi_enabled(struct mlx4_dev *dev, int slave, int port);
1254int mlx4_vf_get_enable_smi_admin(struct mlx4_dev *dev, int slave, int port); 1256int mlx4_vf_get_enable_smi_admin(struct mlx4_dev *dev, int slave, int port);
1255int mlx4_vf_set_enable_smi_admin(struct mlx4_dev *dev, int slave, int port, 1257int mlx4_vf_set_enable_smi_admin(struct mlx4_dev *dev, int slave, int port,
1256 int enable); 1258 int enable);
1259int mlx4_mr_hw_get_mpt(struct mlx4_dev *dev, struct mlx4_mr *mmr,
1260 struct mlx4_mpt_entry ***mpt_entry);
1261int mlx4_mr_hw_write_mpt(struct mlx4_dev *dev, struct mlx4_mr *mmr,
1262 struct mlx4_mpt_entry **mpt_entry);
1263int mlx4_mr_hw_change_pd(struct mlx4_dev *dev, struct mlx4_mpt_entry *mpt_entry,
1264 u32 pdn);
1265int mlx4_mr_hw_change_access(struct mlx4_dev *dev,
1266 struct mlx4_mpt_entry *mpt_entry,
1267 u32 access);
1268void mlx4_mr_hw_put_mpt(struct mlx4_dev *dev,
1269 struct mlx4_mpt_entry **mpt_entry);
1270void mlx4_mr_rereg_mem_cleanup(struct mlx4_dev *dev, struct mlx4_mr *mr);
1271int mlx4_mr_rereg_mem_write(struct mlx4_dev *dev, struct mlx4_mr *mr,
1272 u64 iova, u64 size, int npages,
1273 int page_shift, struct mlx4_mpt_entry *mpt_entry);
1257 1274
1258/* Returns true if running in low memory profile (kdump kernel) */ 1275/* Returns true if running in low memory profile (kdump kernel) */
1259static inline bool mlx4_low_memory_profile(void) 1276static inline bool mlx4_low_memory_profile(void)
diff --git a/include/rdma/ib_mad.h b/include/rdma/ib_mad.h
index 3d81b90cc315..9bb99e983f58 100644
--- a/include/rdma/ib_mad.h
+++ b/include/rdma/ib_mad.h
@@ -40,6 +40,7 @@
40#include <linux/list.h> 40#include <linux/list.h>
41 41
42#include <rdma/ib_verbs.h> 42#include <rdma/ib_verbs.h>
43#include <uapi/rdma/ib_user_mad.h>
43 44
44/* Management base version */ 45/* Management base version */
45#define IB_MGMT_BASE_VERSION 1 46#define IB_MGMT_BASE_VERSION 1
@@ -355,9 +356,13 @@ typedef void (*ib_mad_recv_handler)(struct ib_mad_agent *mad_agent,
355 * @hi_tid: Access layer assigned transaction ID for this client. 356 * @hi_tid: Access layer assigned transaction ID for this client.
356 * Unsolicited MADs sent by this client will have the upper 32-bits 357 * Unsolicited MADs sent by this client will have the upper 32-bits
357 * of their TID set to this value. 358 * of their TID set to this value.
359 * @flags: registration flags
358 * @port_num: Port number on which QP is registered 360 * @port_num: Port number on which QP is registered
359 * @rmpp_version: If set, indicates the RMPP version used by this agent. 361 * @rmpp_version: If set, indicates the RMPP version used by this agent.
360 */ 362 */
363enum {
364 IB_MAD_USER_RMPP = IB_USER_MAD_USER_RMPP,
365};
361struct ib_mad_agent { 366struct ib_mad_agent {
362 struct ib_device *device; 367 struct ib_device *device;
363 struct ib_qp *qp; 368 struct ib_qp *qp;
@@ -367,6 +372,7 @@ struct ib_mad_agent {
367 ib_mad_snoop_handler snoop_handler; 372 ib_mad_snoop_handler snoop_handler;
368 void *context; 373 void *context;
369 u32 hi_tid; 374 u32 hi_tid;
375 u32 flags;
370 u8 port_num; 376 u8 port_num;
371 u8 rmpp_version; 377 u8 rmpp_version;
372}; 378};
@@ -426,6 +432,7 @@ struct ib_mad_recv_wc {
426 * in the range from 0x30 to 0x4f. Otherwise not used. 432 * in the range from 0x30 to 0x4f. Otherwise not used.
427 * @method_mask: The caller will receive unsolicited MADs for any method 433 * @method_mask: The caller will receive unsolicited MADs for any method
428 * where @method_mask = 1. 434 * where @method_mask = 1.
435 *
429 */ 436 */
430struct ib_mad_reg_req { 437struct ib_mad_reg_req {
431 u8 mgmt_class; 438 u8 mgmt_class;
@@ -451,6 +458,7 @@ struct ib_mad_reg_req {
451 * @recv_handler: The completion callback routine invoked for a received 458 * @recv_handler: The completion callback routine invoked for a received
452 * MAD. 459 * MAD.
453 * @context: User specified context associated with the registration. 460 * @context: User specified context associated with the registration.
461 * @registration_flags: Registration flags to set for this agent
454 */ 462 */
455struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device, 463struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
456 u8 port_num, 464 u8 port_num,
@@ -459,7 +467,8 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
459 u8 rmpp_version, 467 u8 rmpp_version,
460 ib_mad_send_handler send_handler, 468 ib_mad_send_handler send_handler,
461 ib_mad_recv_handler recv_handler, 469 ib_mad_recv_handler recv_handler,
462 void *context); 470 void *context,
471 u32 registration_flags);
463 472
464enum ib_mad_snoop_flags { 473enum ib_mad_snoop_flags {
465 /*IB_MAD_SNOOP_POSTED_SENDS = 1,*/ 474 /*IB_MAD_SNOOP_POSTED_SENDS = 1,*/
@@ -661,4 +670,11 @@ void *ib_get_rmpp_segment(struct ib_mad_send_buf *send_buf, int seg_num);
661 */ 670 */
662void ib_free_send_mad(struct ib_mad_send_buf *send_buf); 671void ib_free_send_mad(struct ib_mad_send_buf *send_buf);
663 672
673/**
674 * ib_mad_kernel_rmpp_agent - Returns if the agent is performing RMPP.
675 * @agent: the agent in question
676 * @return: true if agent is performing rmpp, false otherwise.
677 */
678int ib_mad_kernel_rmpp_agent(struct ib_mad_agent *agent);
679
664#endif /* IB_MAD_H */ 680#endif /* IB_MAD_H */
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 7ccef342f724..ed44cc07a7b3 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -1097,7 +1097,8 @@ struct ib_mr_attr {
1097enum ib_mr_rereg_flags { 1097enum ib_mr_rereg_flags {
1098 IB_MR_REREG_TRANS = 1, 1098 IB_MR_REREG_TRANS = 1,
1099 IB_MR_REREG_PD = (1<<1), 1099 IB_MR_REREG_PD = (1<<1),
1100 IB_MR_REREG_ACCESS = (1<<2) 1100 IB_MR_REREG_ACCESS = (1<<2),
1101 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1101}; 1102};
1102 1103
1103/** 1104/**
@@ -1547,6 +1548,13 @@ struct ib_device {
1547 u64 virt_addr, 1548 u64 virt_addr,
1548 int mr_access_flags, 1549 int mr_access_flags,
1549 struct ib_udata *udata); 1550 struct ib_udata *udata);
1551 int (*rereg_user_mr)(struct ib_mr *mr,
1552 int flags,
1553 u64 start, u64 length,
1554 u64 virt_addr,
1555 int mr_access_flags,
1556 struct ib_pd *pd,
1557 struct ib_udata *udata);
1550 int (*query_mr)(struct ib_mr *mr, 1558 int (*query_mr)(struct ib_mr *mr,
1551 struct ib_mr_attr *mr_attr); 1559 struct ib_mr_attr *mr_attr);
1552 int (*dereg_mr)(struct ib_mr *mr); 1560 int (*dereg_mr)(struct ib_mr *mr);
diff --git a/include/uapi/rdma/ib_user_mad.h b/include/uapi/rdma/ib_user_mad.h
index d6fce1cbdb90..09f809f323ea 100644
--- a/include/uapi/rdma/ib_user_mad.h
+++ b/include/uapi/rdma/ib_user_mad.h
@@ -191,6 +191,45 @@ struct ib_user_mad_reg_req {
191 __u8 rmpp_version; 191 __u8 rmpp_version;
192}; 192};
193 193
194/**
195 * ib_user_mad_reg_req2 - MAD registration request
196 *
197 * @id - Set by the _kernel_; used by userspace to identify the
198 * registered agent in future requests.
199 * @qpn - Queue pair number; must be 0 or 1.
200 * @mgmt_class - Indicates which management class of MADs should be
201 * receive by the caller. This field is only required if
202 * the user wishes to receive unsolicited MADs, otherwise
203 * it should be 0.
204 * @mgmt_class_version - Indicates which version of MADs for the given
205 * management class to receive.
206 * @res - Ignored.
207 * @flags - additional registration flags; Must be in the set of
208 * flags defined in IB_USER_MAD_REG_FLAGS_CAP
209 * @method_mask - The caller wishes to receive unsolicited MADs for the
210 * methods whose bit(s) is(are) set.
211 * @oui - Indicates IEEE OUI to use when mgmt_class is a vendor
212 * class in the range from 0x30 to 0x4f. Otherwise not
213 * used.
214 * @rmpp_version - If set, indicates the RMPP version to use.
215 */
216enum {
217 IB_USER_MAD_USER_RMPP = (1 << 0),
218};
219#define IB_USER_MAD_REG_FLAGS_CAP (IB_USER_MAD_USER_RMPP)
220struct ib_user_mad_reg_req2 {
221 __u32 id;
222 __u32 qpn;
223 __u8 mgmt_class;
224 __u8 mgmt_class_version;
225 __u16 res;
226 __u32 flags;
227 __u64 method_mask[2];
228 __u32 oui;
229 __u8 rmpp_version;
230 __u8 reserved[3];
231};
232
194#define IB_IOCTL_MAGIC 0x1b 233#define IB_IOCTL_MAGIC 0x1b
195 234
196#define IB_USER_MAD_REGISTER_AGENT _IOWR(IB_IOCTL_MAGIC, 1, \ 235#define IB_USER_MAD_REGISTER_AGENT _IOWR(IB_IOCTL_MAGIC, 1, \
@@ -200,4 +239,7 @@ struct ib_user_mad_reg_req {
200 239
201#define IB_USER_MAD_ENABLE_PKEY _IO(IB_IOCTL_MAGIC, 3) 240#define IB_USER_MAD_ENABLE_PKEY _IO(IB_IOCTL_MAGIC, 3)
202 241
242#define IB_USER_MAD_REGISTER_AGENT2 _IOWR(IB_IOCTL_MAGIC, 4, \
243 struct ib_user_mad_reg_req2)
244
203#endif /* IB_USER_MAD_H */ 245#endif /* IB_USER_MAD_H */
diff --git a/include/uapi/rdma/ib_user_verbs.h b/include/uapi/rdma/ib_user_verbs.h
index cbfdd4ca9510..26daf55ff76e 100644
--- a/include/uapi/rdma/ib_user_verbs.h
+++ b/include/uapi/rdma/ib_user_verbs.h
@@ -276,6 +276,22 @@ struct ib_uverbs_reg_mr_resp {
276 __u32 rkey; 276 __u32 rkey;
277}; 277};
278 278
279struct ib_uverbs_rereg_mr {
280 __u64 response;
281 __u32 mr_handle;
282 __u32 flags;
283 __u64 start;
284 __u64 length;
285 __u64 hca_va;
286 __u32 pd_handle;
287 __u32 access_flags;
288};
289
290struct ib_uverbs_rereg_mr_resp {
291 __u32 lkey;
292 __u32 rkey;
293};
294
279struct ib_uverbs_dereg_mr { 295struct ib_uverbs_dereg_mr {
280 __u32 mr_handle; 296 __u32 mr_handle;
281}; 297};
diff --git a/include/uapi/rdma/rdma_user_cm.h b/include/uapi/rdma/rdma_user_cm.h
index 99b80abf360a..3066718eb120 100644
--- a/include/uapi/rdma/rdma_user_cm.h
+++ b/include/uapi/rdma/rdma_user_cm.h
@@ -34,6 +34,7 @@
34#define RDMA_USER_CM_H 34#define RDMA_USER_CM_H
35 35
36#include <linux/types.h> 36#include <linux/types.h>
37#include <linux/socket.h>
37#include <linux/in6.h> 38#include <linux/in6.h>
38#include <rdma/ib_user_verbs.h> 39#include <rdma/ib_user_verbs.h>
39#include <rdma/ib_user_sa.h> 40#include <rdma/ib_user_sa.h>
generated by cgit v1.2.2 (git 2.25.0) at 2025-12-31 01:42:26 -0500