drivers/infiniband/hw/hfi1/mad.c

   1 /*
   2  * Copyright(c) 2015-2018 Intel Corporation.
   3  *
   4  * This file is provided under a dual BSD/GPLv2 license.  When using or
   5  * redistributing this file, you may do so under either license.
   6  *
   7  * GPL LICENSE SUMMARY
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of version 2 of the GNU General Public License as
  11  * published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * General Public License for more details.
  17  *
  18  * BSD LICENSE
  19  *
  20  * Redistribution and use in source and binary forms, with or without
  21  * modification, are permitted provided that the following conditions
  22  * are met:
  23  *
  24  *  - Redistributions of source code must retain the above copyright
  25  *    notice, this list of conditions and the following disclaimer.
  26  *  - Redistributions in binary form must reproduce the above copyright
  27  *    notice, this list of conditions and the following disclaimer in
  28  *    the documentation and/or other materials provided with the
  29  *    distribution.
  30  *  - Neither the name of Intel Corporation nor the names of its
  31  *    contributors may be used to endorse or promote products derived
  32  *    from this software without specific prior written permission.
  33  *
  34  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  35  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  36  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  37  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  38  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  39  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  40  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  41  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  42  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  43  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  44  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  45  *
  46  */
  47
  48 #include <linux/net.h>
  49 #include <rdma/opa_addr.h>
  50 #define OPA_NUM_PKEY_BLOCKS_PER_SMP (OPA_SMP_DR_DATA_SIZE \
  51                         / (OPA_PARTITION_TABLE_BLK_SIZE * sizeof(u16)))
  52
  53 #include "hfi.h"
  54 #include "mad.h"
  55 #include "trace.h"
  56 #include "qp.h"
  57 #include "vnic.h"
  58
  59 /* the reset value from the FM is supposed to be 0xffff, handle both */
  60 #define OPA_LINK_WIDTH_RESET_OLD 0x0fff
  61 #define OPA_LINK_WIDTH_RESET 0xffff
  62
  63 struct trap_node {
  64         struct list_head list;
  65         struct opa_mad_notice_attr data;
  66         __be64 tid;
  67         int len;
  68         u32 retry;
  69         u8 in_use;
  70         u8 repress;
  71 };
  72
  73 static int smp_length_check(u32 data_size, u32 request_len)
  74 {
  75         if (unlikely(request_len < data_size))
  76                 return -EINVAL;
  77
  78         return 0;
  79 }
  80
  81 static int reply(struct ib_mad_hdr *smp)
  82 {
  83         /*
  84          * The verbs framework will handle the directed/LID route
  85          * packet changes.
  86          */
  87         smp->method = IB_MGMT_METHOD_GET_RESP;
  88         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
  89                 smp->status |= IB_SMP_DIRECTION;
  90         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
  91 }
  92
  93 static inline void clear_opa_smp_data(struct opa_smp *smp)
  94 {
  95         void *data = opa_get_smp_data(smp);
  96         size_t size = opa_get_smp_data_size(smp);
  97
  98         memset(data, 0, size);
  99 }
 100
 101 static u16 hfi1_lookup_pkey_value(struct hfi1_ibport *ibp, int pkey_idx)
 102 {
 103         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
 104
 105         if (pkey_idx < ARRAY_SIZE(ppd->pkeys))
 106                 return ppd->pkeys[pkey_idx];
 107
 108         return 0;
 109 }
 110
 111 void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port)
 112 {
 113         struct ib_event event;
 114
 115         event.event = IB_EVENT_PKEY_CHANGE;
 116         event.device = &dd->verbs_dev.rdi.ibdev;
 117         event.element.port_num = port;
 118         ib_dispatch_event(&event);
 119 }
 120
 121 /*
 122  * If the port is down, clean up all pending traps.  We need to be careful
 123  * with the given trap, because it may be queued.
 124  */
 125 static void cleanup_traps(struct hfi1_ibport *ibp, struct trap_node *trap)
 126 {
 127         struct trap_node *node, *q;
 128         unsigned long flags;
 129         struct list_head trap_list;
 130         int i;
 131
 132         for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) {
 133                 spin_lock_irqsave(&ibp->rvp.lock, flags);
 134                 list_replace_init(&ibp->rvp.trap_lists[i].list, &trap_list);
 135                 ibp->rvp.trap_lists[i].list_len = 0;
 136                 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 137
 138                 /*
 139                  * Remove all items from the list, freeing all the non-given
 140                  * traps.
 141                  */
 142                 list_for_each_entry_safe(node, q, &trap_list, list) {
 143                         list_del(&node->list);
 144                         if (node != trap)
 145                                 kfree(node);
 146                 }
 147         }
 148
 149         /*
 150          * If this wasn't on one of the lists it would not be freed.  If it
 151          * was on the list, it is now safe to free.
 152          */
 153         kfree(trap);
 154 }
 155
 156 static struct trap_node *check_and_add_trap(struct hfi1_ibport *ibp,
 157                                             struct trap_node *trap)
 158 {
 159         struct trap_node *node;
 160         struct trap_list *trap_list;
 161         unsigned long flags;
 162         unsigned long timeout;
 163         int found = 0;
 164         unsigned int queue_id;
 165         static int trap_count;
 166
 167         queue_id = trap->data.generic_type & 0x0F;
 168         if (queue_id >= RVT_MAX_TRAP_LISTS) {
 169                 trap_count++;
 170                 pr_err_ratelimited("hfi1: Invalid trap 0x%0x dropped. Total dropped: %d\n",
 171                                    trap->data.generic_type, trap_count);
 172                 kfree(trap);
 173                 return NULL;
 174         }
 175
 176         /*
 177          * Since the retry (handle timeout) does not remove a trap request
 178          * from the list, all we have to do is compare the node.
 179          */
 180         spin_lock_irqsave(&ibp->rvp.lock, flags);
 181         trap_list = &ibp->rvp.trap_lists[queue_id];
 182
 183         list_for_each_entry(node, &trap_list->list, list) {
 184                 if (node == trap) {
 185                         node->retry++;
 186                         found = 1;
 187                         break;
 188                 }
 189         }
 190
 191         /* If it is not on the list, add it, limited to RVT-MAX_TRAP_LEN. */
 192         if (!found) {
 193                 if (trap_list->list_len < RVT_MAX_TRAP_LEN) {
 194                         trap_list->list_len++;
 195                         list_add_tail(&trap->list, &trap_list->list);
 196                 } else {
 197                         pr_warn_ratelimited("hfi1: Maximum trap limit reached for 0x%0x traps\n",
 198                                             trap->data.generic_type);
 199                         kfree(trap);
 200                 }
 201         }
 202
 203         /*
 204          * Next check to see if there is a timer pending.  If not, set it up
 205          * and get the first trap from the list.
 206          */
 207         node = NULL;
 208         if (!timer_pending(&ibp->rvp.trap_timer)) {
 209                 /*
 210                  * o14-2
 211                  * If the time out is set we have to wait until it expires
 212                  * before the trap can be sent.
 213                  * This should be > RVT_TRAP_TIMEOUT
 214                  */
 215                 timeout = (RVT_TRAP_TIMEOUT *
 216                            (1UL << ibp->rvp.subnet_timeout)) / 1000;
 217                 mod_timer(&ibp->rvp.trap_timer,
 218                           jiffies + usecs_to_jiffies(timeout));
 219                 node = list_first_entry(&trap_list->list, struct trap_node,
 220                                         list);
 221                 node->in_use = 1;
 222         }
 223         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 224
 225         return node;
 226 }
 227
 228 static void subn_handle_opa_trap_repress(struct hfi1_ibport *ibp,
 229                                          struct opa_smp *smp)
 230 {
 231         struct trap_list *trap_list;
 232         struct trap_node *trap;
 233         unsigned long flags;
 234         int i;
 235
 236         if (smp->attr_id != IB_SMP_ATTR_NOTICE)
 237                 return;
 238
 239         spin_lock_irqsave(&ibp->rvp.lock, flags);
 240         for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) {
 241                 trap_list = &ibp->rvp.trap_lists[i];
 242                 trap = list_first_entry_or_null(&trap_list->list,
 243                                                 struct trap_node, list);
 244                 if (trap && trap->tid == smp->tid) {
 245                         if (trap->in_use) {
 246                                 trap->repress = 1;
 247                         } else {
 248                                 trap_list->list_len--;
 249                                 list_del(&trap->list);
 250                                 kfree(trap);
 251                         }
 252                         break;
 253                 }
 254         }
 255         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 256 }
 257
 258 static void hfi1_update_sm_ah_attr(struct hfi1_ibport *ibp,
 259                                    struct rdma_ah_attr *attr, u32 dlid)
 260 {
 261         rdma_ah_set_dlid(attr, dlid);
 262         rdma_ah_set_port_num(attr, ppd_from_ibp(ibp)->port);
 263         if (dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
 264                 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
 265
 266                 rdma_ah_set_ah_flags(attr, IB_AH_GRH);
 267                 grh->sgid_index = 0;
 268                 grh->hop_limit = 1;
 269                 grh->dgid.global.subnet_prefix =
 270                         ibp->rvp.gid_prefix;
 271                 grh->dgid.global.interface_id = OPA_MAKE_ID(dlid);
 272         }
 273 }
 274
 275 static int hfi1_modify_qp0_ah(struct hfi1_ibport *ibp,
 276                               struct rvt_ah *ah, u32 dlid)
 277 {
 278         struct rdma_ah_attr attr;
 279         struct rvt_qp *qp0;
 280         int ret = -EINVAL;
 281
 282         memset(&attr, 0, sizeof(attr));
 283         attr.type = ah->ibah.type;
 284         hfi1_update_sm_ah_attr(ibp, &attr, dlid);
 285         rcu_read_lock();
 286         qp0 = rcu_dereference(ibp->rvp.qp[0]);
 287         if (qp0)
 288                 ret = rdma_modify_ah(&ah->ibah, &attr);
 289         rcu_read_unlock();
 290         return ret;
 291 }
 292
 293 static struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u32 dlid)
 294 {
 295         struct rdma_ah_attr attr;
 296         struct ib_ah *ah = ERR_PTR(-EINVAL);
 297         struct rvt_qp *qp0;
 298         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
 299         struct hfi1_devdata *dd = dd_from_ppd(ppd);
 300         u8 port_num = ppd->port;
 301
 302         memset(&attr, 0, sizeof(attr));
 303         attr.type = rdma_ah_find_type(&dd->verbs_dev.rdi.ibdev, port_num);
 304         hfi1_update_sm_ah_attr(ibp, &attr, dlid);
 305         rcu_read_lock();
 306         qp0 = rcu_dereference(ibp->rvp.qp[0]);
 307         if (qp0)
 308                 ah = rdma_create_ah(qp0->ibqp.pd, &attr, 0);
 309         rcu_read_unlock();
 310         return ah;
 311 }
 312
 313 static void send_trap(struct hfi1_ibport *ibp, struct trap_node *trap)
 314 {
 315         struct ib_mad_send_buf *send_buf;
 316         struct ib_mad_agent *agent;
 317         struct opa_smp *smp;
 318         unsigned long flags;
 319         int pkey_idx;
 320         u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp;
 321
 322         agent = ibp->rvp.send_agent;
 323         if (!agent) {
 324                 cleanup_traps(ibp, trap);
 325                 return;
 326         }
 327
 328         /* o14-3.2.1 */
 329         if (driver_lstate(ppd_from_ibp(ibp)) != IB_PORT_ACTIVE) {
 330                 cleanup_traps(ibp, trap);
 331                 return;
 332         }
 333
 334         /* Add the trap to the list if necessary and see if we can send it */
 335         trap = check_and_add_trap(ibp, trap);
 336         if (!trap)
 337                 return;
 338
 339         pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
 340         if (pkey_idx < 0) {
 341                 pr_warn("%s: failed to find limited mgmt pkey, defaulting 0x%x\n",
 342                         __func__, hfi1_get_pkey(ibp, 1));
 343                 pkey_idx = 1;
 344         }
 345
 346         send_buf = ib_create_send_mad(agent, qpn, pkey_idx, 0,
 347                                       IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
 348                                       GFP_ATOMIC, IB_MGMT_BASE_VERSION);
 349         if (IS_ERR(send_buf))
 350                 return;
 351
 352         smp = send_buf->mad;
 353         smp->base_version = OPA_MGMT_BASE_VERSION;
 354         smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
 355         smp->class_version = OPA_SM_CLASS_VERSION;
 356         smp->method = IB_MGMT_METHOD_TRAP;
 357
 358         /* Only update the transaction ID for new traps (o13-5). */
 359         if (trap->tid == 0) {
 360                 ibp->rvp.tid++;
 361                 /* make sure that tid != 0 */
 362                 if (ibp->rvp.tid == 0)
 363                         ibp->rvp.tid++;
 364                 trap->tid = cpu_to_be64(ibp->rvp.tid);
 365         }
 366         smp->tid = trap->tid;
 367
 368         smp->attr_id = IB_SMP_ATTR_NOTICE;
 369         /* o14-1: smp->mkey = 0; */
 370
 371         memcpy(smp->route.lid.data, &trap->data, trap->len);
 372
 373         spin_lock_irqsave(&ibp->rvp.lock, flags);
 374         if (!ibp->rvp.sm_ah) {
 375                 if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
 376                         struct ib_ah *ah;
 377
 378                         ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid);
 379                         if (IS_ERR(ah)) {
 380                                 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 381                                 return;
 382                         }
 383                         send_buf->ah = ah;
 384                         ibp->rvp.sm_ah = ibah_to_rvtah(ah);
 385                 } else {
 386                         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 387                         return;
 388                 }
 389         } else {
 390                 send_buf->ah = &ibp->rvp.sm_ah->ibah;
 391         }
 392
 393         /*
 394          * If the trap was repressed while things were getting set up, don't
 395          * bother sending it. This could happen for a retry.
 396          */
 397         if (trap->repress) {
 398                 list_del(&trap->list);
 399                 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 400                 kfree(trap);
 401                 ib_free_send_mad(send_buf);
 402                 return;
 403         }
 404
 405         trap->in_use = 0;
 406         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 407
 408         if (ib_post_send_mad(send_buf, NULL))
 409                 ib_free_send_mad(send_buf);
 410 }
 411
 412 void hfi1_handle_trap_timer(struct timer_list *t)
 413 {
 414         struct hfi1_ibport *ibp = from_timer(ibp, t, rvp.trap_timer);
 415         struct trap_node *trap = NULL;
 416         unsigned long flags;
 417         int i;
 418
 419         /* Find the trap with the highest priority */
 420         spin_lock_irqsave(&ibp->rvp.lock, flags);
 421         for (i = 0; !trap && i < RVT_MAX_TRAP_LISTS; i++) {
 422                 trap = list_first_entry_or_null(&ibp->rvp.trap_lists[i].list,
 423                                                 struct trap_node, list);
 424         }
 425         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 426
 427         if (trap)
 428                 send_trap(ibp, trap);
 429 }
 430
 431 static struct trap_node *create_trap_node(u8 type, __be16 trap_num, u32 lid)
 432 {
 433         struct trap_node *trap;
 434
 435         trap = kzalloc(sizeof(*trap), GFP_ATOMIC);
 436         if (!trap)
 437                 return NULL;
 438
 439         INIT_LIST_HEAD(&trap->list);
 440         trap->data.generic_type = type;
 441         trap->data.prod_type_lsb = IB_NOTICE_PROD_CA;
 442         trap->data.trap_num = trap_num;
 443         trap->data.issuer_lid = cpu_to_be32(lid);
 444
 445         return trap;
 446 }
 447
 448 /*
 449  * Send a bad P_Key trap (ch. 14.3.8).
 450  */
 451 void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl,
 452                    u32 qp1, u32 qp2, u32 lid1, u32 lid2)
 453 {
 454         struct trap_node *trap;
 455         u32 lid = ppd_from_ibp(ibp)->lid;
 456
 457         ibp->rvp.n_pkt_drops++;
 458         ibp->rvp.pkey_violations++;
 459
 460         trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_P_KEY,
 461                                 lid);
 462         if (!trap)
 463                 return;
 464
 465         /* Send violation trap */
 466         trap->data.ntc_257_258.lid1 = cpu_to_be32(lid1);
 467         trap->data.ntc_257_258.lid2 = cpu_to_be32(lid2);
 468         trap->data.ntc_257_258.key = cpu_to_be32(key);
 469         trap->data.ntc_257_258.sl = sl << 3;
 470         trap->data.ntc_257_258.qp1 = cpu_to_be32(qp1);
 471         trap->data.ntc_257_258.qp2 = cpu_to_be32(qp2);
 472
 473         trap->len = sizeof(trap->data);
 474         send_trap(ibp, trap);
 475 }
 476
 477 /*
 478  * Send a bad M_Key trap (ch. 14.3.9).
 479  */
 480 static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
 481                      __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt)
 482 {
 483         struct trap_node *trap;
 484         u32 lid = ppd_from_ibp(ibp)->lid;
 485
 486         trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_M_KEY,
 487                                 lid);
 488         if (!trap)
 489                 return;
 490
 491         /* Send violation trap */
 492         trap->data.ntc_256.lid = trap->data.issuer_lid;
 493         trap->data.ntc_256.method = mad->method;
 494         trap->data.ntc_256.attr_id = mad->attr_id;
 495         trap->data.ntc_256.attr_mod = mad->attr_mod;
 496         trap->data.ntc_256.mkey = mkey;
 497         if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
 498                 trap->data.ntc_256.dr_slid = dr_slid;
 499                 trap->data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
 500                 if (hop_cnt > ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path)) {
 501                         trap->data.ntc_256.dr_trunc_hop |=
 502                                 IB_NOTICE_TRAP_DR_TRUNC;
 503                         hop_cnt = ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path);
 504                 }
 505                 trap->data.ntc_256.dr_trunc_hop |= hop_cnt;
 506                 memcpy(trap->data.ntc_256.dr_rtn_path, return_path,
 507                        hop_cnt);
 508         }
 509
 510         trap->len = sizeof(trap->data);
 511
 512         send_trap(ibp, trap);
 513 }
 514
 515 /*
 516  * Send a Port Capability Mask Changed trap (ch. 14.3.11).
 517  */
 518 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num)
 519 {
 520         struct trap_node *trap;
 521         struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi);
 522         struct hfi1_devdata *dd = dd_from_dev(verbs_dev);
 523         struct hfi1_ibport *ibp = &dd->pport[port_num - 1].ibport_data;
 524         u32 lid = ppd_from_ibp(ibp)->lid;
 525
 526         trap = create_trap_node(IB_NOTICE_TYPE_INFO,
 527                                 OPA_TRAP_CHANGE_CAPABILITY,
 528                                 lid);
 529         if (!trap)
 530                 return;
 531
 532         trap->data.ntc_144.lid = trap->data.issuer_lid;
 533         trap->data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
 534         trap->data.ntc_144.cap_mask3 = cpu_to_be16(ibp->rvp.port_cap3_flags);
 535
 536         trap->len = sizeof(trap->data);
 537         send_trap(ibp, trap);
 538 }
 539
 540 /*
 541  * Send a System Image GUID Changed trap (ch. 14.3.12).
 542  */
 543 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp)
 544 {
 545         struct trap_node *trap;
 546         u32 lid = ppd_from_ibp(ibp)->lid;
 547
 548         trap = create_trap_node(IB_NOTICE_TYPE_INFO, OPA_TRAP_CHANGE_SYSGUID,
 549                                 lid);
 550         if (!trap)
 551                 return;
 552
 553         trap->data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid;
 554         trap->data.ntc_145.lid = trap->data.issuer_lid;
 555
 556         trap->len = sizeof(trap->data);
 557         send_trap(ibp, trap);
 558 }
 559
 560 /*
 561  * Send a Node Description Changed trap (ch. 14.3.13).
 562  */
 563 void hfi1_node_desc_chg(struct hfi1_ibport *ibp)
 564 {
 565         struct trap_node *trap;
 566         u32 lid = ppd_from_ibp(ibp)->lid;
 567
 568         trap = create_trap_node(IB_NOTICE_TYPE_INFO,
 569                                 OPA_TRAP_CHANGE_CAPABILITY,
 570                                 lid);
 571         if (!trap)
 572                 return;
 573
 574         trap->data.ntc_144.lid = trap->data.issuer_lid;
 575         trap->data.ntc_144.change_flags =
 576                 cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG);
 577
 578         trap->len = sizeof(trap->data);
 579         send_trap(ibp, trap);
 580 }
 581
 582 static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am,
 583                                    u8 *data, struct ib_device *ibdev,
 584                                    u8 port, u32 *resp_len, u32 max_len)
 585 {
 586         struct opa_node_description *nd;
 587
 588         if (am || smp_length_check(sizeof(*nd), max_len)) {
 589                 smp->status |= IB_SMP_INVALID_FIELD;
 590                 return reply((struct ib_mad_hdr *)smp);
 591         }
 592
 593         nd = (struct opa_node_description *)data;
 594
 595         memcpy(nd->data, ibdev->node_desc, sizeof(nd->data));
 596
 597         if (resp_len)
 598                 *resp_len += sizeof(*nd);
 599
 600         return reply((struct ib_mad_hdr *)smp);
 601 }
 602
 603 static int __subn_get_opa_nodeinfo(struct opa_smp *smp, u32 am, u8 *data,
 604                                    struct ib_device *ibdev, u8 port,
 605                                    u32 *resp_len, u32 max_len)
 606 {
 607         struct opa_node_info *ni;
 608         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 609         unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
 610
 611         ni = (struct opa_node_info *)data;
 612
 613         /* GUID 0 is illegal */
 614         if (am || pidx >= dd->num_pports || ibdev->node_guid == 0 ||
 615             smp_length_check(sizeof(*ni), max_len) ||
 616             get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX) == 0) {
 617                 smp->status |= IB_SMP_INVALID_FIELD;
 618                 return reply((struct ib_mad_hdr *)smp);
 619         }
 620
 621         ni->port_guid = get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX);
 622         ni->base_version = OPA_MGMT_BASE_VERSION;
 623         ni->class_version = OPA_SM_CLASS_VERSION;
 624         ni->node_type = 1;     /* channel adapter */
 625         ni->num_ports = ibdev->phys_port_cnt;
 626         /* This is already in network order */
 627         ni->system_image_guid = ib_hfi1_sys_image_guid;
 628         ni->node_guid = ibdev->node_guid;
 629         ni->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
 630         ni->device_id = cpu_to_be16(dd->pcidev->device);
 631         ni->revision = cpu_to_be32(dd->minrev);
 632         ni->local_port_num = port;
 633         ni->vendor_id[0] = dd->oui1;
 634         ni->vendor_id[1] = dd->oui2;
 635         ni->vendor_id[2] = dd->oui3;
 636
 637         if (resp_len)
 638                 *resp_len += sizeof(*ni);
 639
 640         return reply((struct ib_mad_hdr *)smp);
 641 }
 642
 643 static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
 644                              u8 port)
 645 {
 646         struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
 647         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 648         unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
 649
 650         /* GUID 0 is illegal */
 651         if (smp->attr_mod || pidx >= dd->num_pports ||
 652             ibdev->node_guid == 0 ||
 653             get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX) == 0) {
 654                 smp->status |= IB_SMP_INVALID_FIELD;
 655                 return reply((struct ib_mad_hdr *)smp);
 656         }
 657
 658         nip->port_guid = get_sguid(to_iport(ibdev, port), HFI1_PORT_GUID_INDEX);
 659         nip->base_version = OPA_MGMT_BASE_VERSION;
 660         nip->class_version = OPA_SM_CLASS_VERSION;
 661         nip->node_type = 1;     /* channel adapter */
 662         nip->num_ports = ibdev->phys_port_cnt;
 663         /* This is already in network order */
 664         nip->sys_guid = ib_hfi1_sys_image_guid;
 665         nip->node_guid = ibdev->node_guid;
 666         nip->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
 667         nip->device_id = cpu_to_be16(dd->pcidev->device);
 668         nip->revision = cpu_to_be32(dd->minrev);
 669         nip->local_port_num = port;
 670         nip->vendor_id[0] = dd->oui1;
 671         nip->vendor_id[1] = dd->oui2;
 672         nip->vendor_id[2] = dd->oui3;
 673
 674         return reply((struct ib_mad_hdr *)smp);
 675 }
 676
 677 static void set_link_width_enabled(struct hfi1_pportdata *ppd, u32 w)
 678 {
 679         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_ENB, w);
 680 }
 681
 682 static void set_link_width_downgrade_enabled(struct hfi1_pportdata *ppd, u32 w)
 683 {
 684         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_DG_ENB, w);
 685 }
 686
 687 static void set_link_speed_enabled(struct hfi1_pportdata *ppd, u32 s)
 688 {
 689         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_SPD_ENB, s);
 690 }
 691
 692 static int check_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
 693                       int mad_flags, __be64 mkey, __be32 dr_slid,
 694                       u8 return_path[], u8 hop_cnt)
 695 {
 696         int valid_mkey = 0;
 697         int ret = 0;
 698
 699         /* Is the mkey in the process of expiring? */
 700         if (ibp->rvp.mkey_lease_timeout &&
 701             time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
 702                 /* Clear timeout and mkey protection field. */
 703                 ibp->rvp.mkey_lease_timeout = 0;
 704                 ibp->rvp.mkeyprot = 0;
 705         }
 706
 707         if ((mad_flags & IB_MAD_IGNORE_MKEY) ||  ibp->rvp.mkey == 0 ||
 708             ibp->rvp.mkey == mkey)
 709                 valid_mkey = 1;
 710
 711         /* Unset lease timeout on any valid Get/Set/TrapRepress */
 712         if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
 713             (mad->method == IB_MGMT_METHOD_GET ||
 714              mad->method == IB_MGMT_METHOD_SET ||
 715              mad->method == IB_MGMT_METHOD_TRAP_REPRESS))
 716                 ibp->rvp.mkey_lease_timeout = 0;
 717
 718         if (!valid_mkey) {
 719                 switch (mad->method) {
 720                 case IB_MGMT_METHOD_GET:
 721                         /* Bad mkey not a violation below level 2 */
 722                         if (ibp->rvp.mkeyprot < 2)
 723                                 break;
 724                         /* fall through */
 725                 case IB_MGMT_METHOD_SET:
 726                 case IB_MGMT_METHOD_TRAP_REPRESS:
 727                         if (ibp->rvp.mkey_violations != 0xFFFF)
 728                                 ++ibp->rvp.mkey_violations;
 729                         if (!ibp->rvp.mkey_lease_timeout &&
 730                             ibp->rvp.mkey_lease_period)
 731                                 ibp->rvp.mkey_lease_timeout = jiffies +
 732                                         ibp->rvp.mkey_lease_period * HZ;
 733                         /* Generate a trap notice. */
 734                         bad_mkey(ibp, mad, mkey, dr_slid, return_path,
 735                                  hop_cnt);
 736                         ret = 1;
 737                 }
 738         }
 739
 740         return ret;
 741 }
 742
 743 /*
 744  * The SMA caches reads from LCB registers in case the LCB is unavailable.
 745  * (The LCB is unavailable in certain link states, for example.)
 746  */
 747 struct lcb_datum {
 748         u32 off;
 749         u64 val;
 750 };
 751
 752 static struct lcb_datum lcb_cache[] = {
 753         { DC_LCB_STS_ROUND_TRIP_LTP_CNT, 0 },
 754 };
 755
 756 static int write_lcb_cache(u32 off, u64 val)
 757 {
 758         int i;
 759
 760         for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
 761                 if (lcb_cache[i].off == off) {
 762                         lcb_cache[i].val = val;
 763                         return 0;
 764                 }
 765         }
 766
 767         pr_warn("%s bad offset 0x%x\n", __func__, off);
 768         return -1;
 769 }
 770
 771 static int read_lcb_cache(u32 off, u64 *val)
 772 {
 773         int i;
 774
 775         for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
 776                 if (lcb_cache[i].off == off) {
 777                         *val = lcb_cache[i].val;
 778                         return 0;
 779                 }
 780         }
 781
 782         pr_warn("%s bad offset 0x%x\n", __func__, off);
 783         return -1;
 784 }
 785
 786 void read_ltp_rtt(struct hfi1_devdata *dd)
 787 {
 788         u64 reg;
 789
 790         if (read_lcb_csr(dd, DC_LCB_STS_ROUND_TRIP_LTP_CNT, &reg))
 791                 dd_dev_err(dd, "%s: unable to read LTP RTT\n", __func__);
 792         else
 793                 write_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, reg);
 794 }
 795
 796 static int __subn_get_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
 797                                    struct ib_device *ibdev, u8 port,
 798                                    u32 *resp_len, u32 max_len)
 799 {
 800         int i;
 801         struct hfi1_devdata *dd;
 802         struct hfi1_pportdata *ppd;
 803         struct hfi1_ibport *ibp;
 804         struct opa_port_info *pi = (struct opa_port_info *)data;
 805         u8 mtu;
 806         u8 credit_rate;
 807         u8 is_beaconing_active;
 808         u32 state;
 809         u32 num_ports = OPA_AM_NPORT(am);
 810         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
 811         u32 buffer_units;
 812         u64 tmp = 0;
 813
 814         if (num_ports != 1 || smp_length_check(sizeof(*pi), max_len)) {
 815                 smp->status |= IB_SMP_INVALID_FIELD;
 816                 return reply((struct ib_mad_hdr *)smp);
 817         }
 818
 819         dd = dd_from_ibdev(ibdev);
 820         /* IB numbers ports from 1, hw from 0 */
 821         ppd = dd->pport + (port - 1);
 822         ibp = &ppd->ibport_data;
 823
 824         if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
 825             ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
 826                 smp->status |= IB_SMP_INVALID_FIELD;
 827                 return reply((struct ib_mad_hdr *)smp);
 828         }
 829
 830         pi->lid = cpu_to_be32(ppd->lid);
 831
 832         /* Only return the mkey if the protection field allows it. */
 833         if (!(smp->method == IB_MGMT_METHOD_GET &&
 834               ibp->rvp.mkey != smp->mkey &&
 835               ibp->rvp.mkeyprot == 1))
 836                 pi->mkey = ibp->rvp.mkey;
 837
 838         pi->subnet_prefix = ibp->rvp.gid_prefix;
 839         pi->sm_lid = cpu_to_be32(ibp->rvp.sm_lid);
 840         pi->ib_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
 841         pi->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
 842         pi->sm_trap_qp = cpu_to_be32(ppd->sm_trap_qp);
 843         pi->sa_qp = cpu_to_be32(ppd->sa_qp);
 844
 845         pi->link_width.enabled = cpu_to_be16(ppd->link_width_enabled);
 846         pi->link_width.supported = cpu_to_be16(ppd->link_width_supported);
 847         pi->link_width.active = cpu_to_be16(ppd->link_width_active);
 848
 849         pi->link_width_downgrade.supported =
 850                         cpu_to_be16(ppd->link_width_downgrade_supported);
 851         pi->link_width_downgrade.enabled =
 852                         cpu_to_be16(ppd->link_width_downgrade_enabled);
 853         pi->link_width_downgrade.tx_active =
 854                         cpu_to_be16(ppd->link_width_downgrade_tx_active);
 855         pi->link_width_downgrade.rx_active =
 856                         cpu_to_be16(ppd->link_width_downgrade_rx_active);
 857
 858         pi->link_speed.supported = cpu_to_be16(ppd->link_speed_supported);
 859         pi->link_speed.active = cpu_to_be16(ppd->link_speed_active);
 860         pi->link_speed.enabled = cpu_to_be16(ppd->link_speed_enabled);
 861
 862         state = driver_lstate(ppd);
 863
 864         if (start_of_sm_config && (state == IB_PORT_INIT))
 865                 ppd->is_sm_config_started = 1;
 866
 867         pi->port_phys_conf = (ppd->port_type & 0xf);
 868
 869         pi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
 870         pi->port_states.ledenable_offlinereason |=
 871                 ppd->is_sm_config_started << 5;
 872         /*
 873          * This pairs with the memory barrier in hfi1_start_led_override to
 874          * ensure that we read the correct state of LED beaconing represented
 875          * by led_override_timer_active
 876          */
 877         smp_rmb();
 878         is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
 879         pi->port_states.ledenable_offlinereason |= is_beaconing_active << 6;
 880         pi->port_states.ledenable_offlinereason |=
 881                 ppd->offline_disabled_reason;
 882
 883         pi->port_states.portphysstate_portstate =
 884                 (driver_pstate(ppd) << 4) | state;
 885
 886         pi->mkeyprotect_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
 887
 888         memset(pi->neigh_mtu.pvlx_to_mtu, 0, sizeof(pi->neigh_mtu.pvlx_to_mtu));
 889         for (i = 0; i < ppd->vls_supported; i++) {
 890                 mtu = mtu_to_enum(dd->vld[i].mtu, HFI1_DEFAULT_ACTIVE_MTU);
 891                 if ((i % 2) == 0)
 892                         pi->neigh_mtu.pvlx_to_mtu[i / 2] |= (mtu << 4);
 893                 else
 894                         pi->neigh_mtu.pvlx_to_mtu[i / 2] |= mtu;
 895         }
 896         /* don't forget VL 15 */
 897         mtu = mtu_to_enum(dd->vld[15].mtu, 2048);
 898         pi->neigh_mtu.pvlx_to_mtu[15 / 2] |= mtu;
 899         pi->smsl = ibp->rvp.sm_sl & OPA_PI_MASK_SMSL;
 900         pi->operational_vls = hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS);
 901         pi->partenforce_filterraw |=
 902                 (ppd->linkinit_reason & OPA_PI_MASK_LINKINIT_REASON);
 903         if (ppd->part_enforce & HFI1_PART_ENFORCE_IN)
 904                 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_IN;
 905         if (ppd->part_enforce & HFI1_PART_ENFORCE_OUT)
 906                 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_OUT;
 907         pi->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
 908         /* P_KeyViolations are counted by hardware. */
 909         pi->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
 910         pi->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
 911
 912         pi->vl.cap = ppd->vls_supported;
 913         pi->vl.high_limit = cpu_to_be16(ibp->rvp.vl_high_limit);
 914         pi->vl.arb_high_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_CAP);
 915         pi->vl.arb_low_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_LOW_CAP);
 916
 917         pi->clientrereg_subnettimeout = ibp->rvp.subnet_timeout;
 918
 919         pi->port_link_mode  = cpu_to_be16(OPA_PORT_LINK_MODE_OPA << 10 |
 920                                           OPA_PORT_LINK_MODE_OPA << 5 |
 921                                           OPA_PORT_LINK_MODE_OPA);
 922
 923         pi->port_ltp_crc_mode = cpu_to_be16(ppd->port_ltp_crc_mode);
 924
 925         pi->port_mode = cpu_to_be16(
 926                                 ppd->is_active_optimize_enabled ?
 927                                         OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE : 0);
 928
 929         pi->port_packet_format.supported =
 930                 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B |
 931                             OPA_PORT_PACKET_FORMAT_16B);
 932         pi->port_packet_format.enabled =
 933                 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B |
 934                             OPA_PORT_PACKET_FORMAT_16B);
 935
 936         /* flit_control.interleave is (OPA V1, version .76):
 937          * bits         use
 938          * ----         ---
 939          * 2            res
 940          * 2            DistanceSupported
 941          * 2            DistanceEnabled
 942          * 5            MaxNextLevelTxEnabled
 943          * 5            MaxNestLevelRxSupported
 944          *
 945          * HFI supports only "distance mode 1" (see OPA V1, version .76,
 946          * section 9.6.2), so set DistanceSupported, DistanceEnabled
 947          * to 0x1.
 948          */
 949         pi->flit_control.interleave = cpu_to_be16(0x1400);
 950
 951         pi->link_down_reason = ppd->local_link_down_reason.sma;
 952         pi->neigh_link_down_reason = ppd->neigh_link_down_reason.sma;
 953         pi->port_error_action = cpu_to_be32(ppd->port_error_action);
 954         pi->mtucap = mtu_to_enum(hfi1_max_mtu, IB_MTU_4096);
 955
 956         /* 32.768 usec. response time (guessing) */
 957         pi->resptimevalue = 3;
 958
 959         pi->local_port_num = port;
 960
 961         /* buffer info for FM */
 962         pi->overall_buffer_space = cpu_to_be16(dd->link_credits);
 963
 964         pi->neigh_node_guid = cpu_to_be64(ppd->neighbor_guid);
 965         pi->neigh_port_num = ppd->neighbor_port_number;
 966         pi->port_neigh_mode =
 967                 (ppd->neighbor_type & OPA_PI_MASK_NEIGH_NODE_TYPE) |
 968                 (ppd->mgmt_allowed ? OPA_PI_MASK_NEIGH_MGMT_ALLOWED : 0) |
 969                 (ppd->neighbor_fm_security ?
 970                         OPA_PI_MASK_NEIGH_FW_AUTH_BYPASS : 0);
 971
 972         /* HFIs shall always return VL15 credits to their
 973          * neighbor in a timely manner, without any credit return pacing.
 974          */
 975         credit_rate = 0;
 976         buffer_units  = (dd->vau) & OPA_PI_MASK_BUF_UNIT_BUF_ALLOC;
 977         buffer_units |= (dd->vcu << 3) & OPA_PI_MASK_BUF_UNIT_CREDIT_ACK;
 978         buffer_units |= (credit_rate << 6) &
 979                                 OPA_PI_MASK_BUF_UNIT_VL15_CREDIT_RATE;
 980         buffer_units |= (dd->vl15_init << 11) & OPA_PI_MASK_BUF_UNIT_VL15_INIT;
 981         pi->buffer_units = cpu_to_be32(buffer_units);
 982
 983         pi->opa_cap_mask = cpu_to_be16(ibp->rvp.port_cap3_flags);
 984         pi->collectivemask_multicastmask = ((OPA_COLLECTIVE_NR & 0x7)
 985                                             << 3 | (OPA_MCAST_NR & 0x7));
 986
 987         /* HFI supports a replay buffer 128 LTPs in size */
 988         pi->replay_depth.buffer = 0x80;
 989         /* read the cached value of DC_LCB_STS_ROUND_TRIP_LTP_CNT */
 990         read_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, &tmp);
 991
 992         /*
 993          * this counter is 16 bits wide, but the replay_depth.wire
 994          * variable is only 8 bits
 995          */
 996         if (tmp > 0xff)
 997                 tmp = 0xff;
 998         pi->replay_depth.wire = tmp;
 999
1000         if (resp_len)
1001                 *resp_len += sizeof(struct opa_port_info);
1002
1003         return reply((struct ib_mad_hdr *)smp);
1004 }
1005
1006 /**
1007  * get_pkeys - return the PKEY table
1008  * @dd: the hfi1_ib device
1009  * @port: the IB port number
1010  * @pkeys: the pkey table is placed here
1011  */
1012 static int get_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1013 {
1014         struct hfi1_pportdata *ppd = dd->pport + port - 1;
1015
1016         memcpy(pkeys, ppd->pkeys, sizeof(ppd->pkeys));
1017
1018         return 0;
1019 }
1020
1021 static int __subn_get_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1022                                     struct ib_device *ibdev, u8 port,
1023                                     u32 *resp_len, u32 max_len)
1024 {
1025         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1026         u32 n_blocks_req = OPA_AM_NBLK(am);
1027         u32 start_block = am & 0x7ff;
1028         __be16 *p;
1029         u16 *q;
1030         int i;
1031         u16 n_blocks_avail;
1032         unsigned npkeys = hfi1_get_npkeys(dd);
1033         size_t size;
1034
1035         if (n_blocks_req == 0) {
1036                 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1037                         port, start_block, n_blocks_req);
1038                 smp->status |= IB_SMP_INVALID_FIELD;
1039                 return reply((struct ib_mad_hdr *)smp);
1040         }
1041
1042         n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
1043
1044         size = (n_blocks_req * OPA_PARTITION_TABLE_BLK_SIZE) * sizeof(u16);
1045
1046         if (smp_length_check(size, max_len)) {
1047                 smp->status |= IB_SMP_INVALID_FIELD;
1048                 return reply((struct ib_mad_hdr *)smp);
1049         }
1050
1051         if (start_block + n_blocks_req > n_blocks_avail ||
1052             n_blocks_req > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1053                 pr_warn("OPA Get PKey AM Invalid : s 0x%x; req 0x%x; "
1054                         "avail 0x%x; blk/smp 0x%lx\n",
1055                         start_block, n_blocks_req, n_blocks_avail,
1056                         OPA_NUM_PKEY_BLOCKS_PER_SMP);
1057                 smp->status |= IB_SMP_INVALID_FIELD;
1058                 return reply((struct ib_mad_hdr *)smp);
1059         }
1060
1061         p = (__be16 *)data;
1062         q = (u16 *)data;
1063         /* get the real pkeys if we are requesting the first block */
1064         if (start_block == 0) {
1065                 get_pkeys(dd, port, q);
1066                 for (i = 0; i < npkeys; i++)
1067                         p[i] = cpu_to_be16(q[i]);
1068                 if (resp_len)
1069                         *resp_len += size;
1070         } else {
1071                 smp->status |= IB_SMP_INVALID_FIELD;
1072         }
1073         return reply((struct ib_mad_hdr *)smp);
1074 }
1075
1076 enum {
1077         HFI_TRANSITION_DISALLOWED,
1078         HFI_TRANSITION_IGNORED,
1079         HFI_TRANSITION_ALLOWED,
1080         HFI_TRANSITION_UNDEFINED,
1081 };
1082
1083 /*
1084  * Use shortened names to improve readability of
1085  * {logical,physical}_state_transitions
1086  */
1087 enum {
1088         __D = HFI_TRANSITION_DISALLOWED,
1089         __I = HFI_TRANSITION_IGNORED,
1090         __A = HFI_TRANSITION_ALLOWED,
1091         __U = HFI_TRANSITION_UNDEFINED,
1092 };
1093
1094 /*
1095  * IB_PORTPHYSSTATE_POLLING (2) through OPA_PORTPHYSSTATE_MAX (11) are
1096  * represented in physical_state_transitions.
1097  */
1098 #define __N_PHYSTATES (OPA_PORTPHYSSTATE_MAX - IB_PORTPHYSSTATE_POLLING + 1)
1099
1100 /*
1101  * Within physical_state_transitions, rows represent "old" states,
1102  * columns "new" states, and physical_state_transitions.allowed[old][new]
1103  * indicates if the transition from old state to new state is legal (see
1104  * OPAg1v1, Table 6-4).
1105  */
1106 static const struct {
1107         u8 allowed[__N_PHYSTATES][__N_PHYSTATES];
1108 } physical_state_transitions = {
1109         {
1110                 /* 2    3    4    5    6    7    8    9   10   11 */
1111         /* 2 */ { __A, __A, __D, __D, __D, __D, __D, __D, __D, __D },
1112         /* 3 */ { __A, __I, __D, __D, __D, __D, __D, __D, __D, __A },
1113         /* 4 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1114         /* 5 */ { __A, __A, __D, __I, __D, __D, __D, __D, __D, __D },
1115         /* 6 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1116         /* 7 */ { __D, __A, __D, __D, __D, __I, __D, __D, __D, __D },
1117         /* 8 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1118         /* 9 */ { __I, __A, __D, __D, __D, __D, __D, __I, __D, __D },
1119         /*10 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
1120         /*11 */ { __D, __A, __D, __D, __D, __D, __D, __D, __D, __I },
1121         }
1122 };
1123
1124 /*
1125  * IB_PORT_DOWN (1) through IB_PORT_ACTIVE_DEFER (5) are represented
1126  * logical_state_transitions
1127  */
1128
1129 #define __N_LOGICAL_STATES (IB_PORT_ACTIVE_DEFER - IB_PORT_DOWN + 1)
1130
1131 /*
1132  * Within logical_state_transitions rows represent "old" states,
1133  * columns "new" states, and logical_state_transitions.allowed[old][new]
1134  * indicates if the transition from old state to new state is legal (see
1135  * OPAg1v1, Table 9-12).
1136  */
1137 static const struct {
1138         u8 allowed[__N_LOGICAL_STATES][__N_LOGICAL_STATES];
1139 } logical_state_transitions = {
1140         {
1141                 /* 1    2    3    4    5 */
1142         /* 1 */ { __I, __D, __D, __D, __U},
1143         /* 2 */ { __D, __I, __A, __D, __U},
1144         /* 3 */ { __D, __D, __I, __A, __U},
1145         /* 4 */ { __D, __D, __I, __I, __U},
1146         /* 5 */ { __U, __U, __U, __U, __U},
1147         }
1148 };
1149
1150 static int logical_transition_allowed(int old, int new)
1151 {
1152         if (old < IB_PORT_NOP || old > IB_PORT_ACTIVE_DEFER ||
1153             new < IB_PORT_NOP || new > IB_PORT_ACTIVE_DEFER) {
1154                 pr_warn("invalid logical state(s) (old %d new %d)\n",
1155                         old, new);
1156                 return HFI_TRANSITION_UNDEFINED;
1157         }
1158
1159         if (new == IB_PORT_NOP)
1160                 return HFI_TRANSITION_ALLOWED; /* always allowed */
1161
1162         /* adjust states for indexing into logical_state_transitions */
1163         old -= IB_PORT_DOWN;
1164         new -= IB_PORT_DOWN;
1165
1166         if (old < 0 || new < 0)
1167                 return HFI_TRANSITION_UNDEFINED;
1168         return logical_state_transitions.allowed[old][new];
1169 }
1170
1171 static int physical_transition_allowed(int old, int new)
1172 {
1173         if (old < IB_PORTPHYSSTATE_NOP || old > OPA_PORTPHYSSTATE_MAX ||
1174             new < IB_PORTPHYSSTATE_NOP || new > OPA_PORTPHYSSTATE_MAX) {
1175                 pr_warn("invalid physical state(s) (old %d new %d)\n",
1176                         old, new);
1177                 return HFI_TRANSITION_UNDEFINED;
1178         }
1179
1180         if (new == IB_PORTPHYSSTATE_NOP)
1181                 return HFI_TRANSITION_ALLOWED; /* always allowed */
1182
1183         /* adjust states for indexing into physical_state_transitions */
1184         old -= IB_PORTPHYSSTATE_POLLING;
1185         new -= IB_PORTPHYSSTATE_POLLING;
1186
1187         if (old < 0 || new < 0)
1188                 return HFI_TRANSITION_UNDEFINED;
1189         return physical_state_transitions.allowed[old][new];
1190 }
1191
1192 static int port_states_transition_allowed(struct hfi1_pportdata *ppd,
1193                                           u32 logical_new, u32 physical_new)
1194 {
1195         u32 physical_old = driver_pstate(ppd);
1196         u32 logical_old = driver_lstate(ppd);
1197         int ret, logical_allowed, physical_allowed;
1198
1199         ret = logical_transition_allowed(logical_old, logical_new);
1200         logical_allowed = ret;
1201
1202         if (ret == HFI_TRANSITION_DISALLOWED ||
1203             ret == HFI_TRANSITION_UNDEFINED) {
1204                 pr_warn("invalid logical state transition %s -> %s\n",
1205                         opa_lstate_name(logical_old),
1206                         opa_lstate_name(logical_new));
1207                 return ret;
1208         }
1209
1210         ret = physical_transition_allowed(physical_old, physical_new);
1211         physical_allowed = ret;
1212
1213         if (ret == HFI_TRANSITION_DISALLOWED ||
1214             ret == HFI_TRANSITION_UNDEFINED) {
1215                 pr_warn("invalid physical state transition %s -> %s\n",
1216                         opa_pstate_name(physical_old),
1217                         opa_pstate_name(physical_new));
1218                 return ret;
1219         }
1220
1221         if (logical_allowed == HFI_TRANSITION_IGNORED &&
1222             physical_allowed == HFI_TRANSITION_IGNORED)
1223                 return HFI_TRANSITION_IGNORED;
1224
1225         /*
1226          * A change request of Physical Port State from
1227          * 'Offline' to 'Polling' should be ignored.
1228          */
1229         if ((physical_old == OPA_PORTPHYSSTATE_OFFLINE) &&
1230             (physical_new == IB_PORTPHYSSTATE_POLLING))
1231                 return HFI_TRANSITION_IGNORED;
1232
1233         /*
1234          * Either physical_allowed or logical_allowed is
1235          * HFI_TRANSITION_ALLOWED.
1236          */
1237         return HFI_TRANSITION_ALLOWED;
1238 }
1239
1240 static int set_port_states(struct hfi1_pportdata *ppd, struct opa_smp *smp,
1241                            u32 logical_state, u32 phys_state, int local_mad)
1242 {
1243         struct hfi1_devdata *dd = ppd->dd;
1244         u32 link_state;
1245         int ret;
1246
1247         ret = port_states_transition_allowed(ppd, logical_state, phys_state);
1248         if (ret == HFI_TRANSITION_DISALLOWED ||
1249             ret == HFI_TRANSITION_UNDEFINED) {
1250                 /* error message emitted above */
1251                 smp->status |= IB_SMP_INVALID_FIELD;
1252                 return 0;
1253         }
1254
1255         if (ret == HFI_TRANSITION_IGNORED)
1256                 return 0;
1257
1258         if ((phys_state != IB_PORTPHYSSTATE_NOP) &&
1259             !(logical_state == IB_PORT_DOWN ||
1260               logical_state == IB_PORT_NOP)){
1261                 pr_warn("SubnSet(OPA_PortInfo) port state invalid: logical_state 0x%x physical_state 0x%x\n",
1262                         logical_state, phys_state);
1263                 smp->status |= IB_SMP_INVALID_FIELD;
1264         }
1265
1266         /*
1267          * Logical state changes are summarized in OPAv1g1 spec.,
1268          * Table 9-12; physical state changes are summarized in
1269          * OPAv1g1 spec., Table 6.4.
1270          */
1271         switch (logical_state) {
1272         case IB_PORT_NOP:
1273                 if (phys_state == IB_PORTPHYSSTATE_NOP)
1274                         break;
1275                 /* FALLTHROUGH */
1276         case IB_PORT_DOWN:
1277                 if (phys_state == IB_PORTPHYSSTATE_NOP) {
1278                         link_state = HLS_DN_DOWNDEF;
1279                 } else if (phys_state == IB_PORTPHYSSTATE_POLLING) {
1280                         link_state = HLS_DN_POLL;
1281                         set_link_down_reason(ppd, OPA_LINKDOWN_REASON_FM_BOUNCE,
1282                                              0, OPA_LINKDOWN_REASON_FM_BOUNCE);
1283                 } else if (phys_state == IB_PORTPHYSSTATE_DISABLED) {
1284                         link_state = HLS_DN_DISABLE;
1285                 } else {
1286                         pr_warn("SubnSet(OPA_PortInfo) invalid physical state 0x%x\n",
1287                                 phys_state);
1288                         smp->status |= IB_SMP_INVALID_FIELD;
1289                         break;
1290                 }
1291
1292                 if ((link_state == HLS_DN_POLL ||
1293                      link_state == HLS_DN_DOWNDEF)) {
1294                         /*
1295                          * Going to poll.  No matter what the current state,
1296                          * always move offline first, then tune and start the
1297                          * link.  This correctly handles a FM link bounce and
1298                          * a link enable.  Going offline is a no-op if already
1299                          * offline.
1300                          */
1301                         set_link_state(ppd, HLS_DN_OFFLINE);
1302                         start_link(ppd);
1303                 } else {
1304                         set_link_state(ppd, link_state);
1305                 }
1306                 if (link_state == HLS_DN_DISABLE &&
1307                     (ppd->offline_disabled_reason >
1308                      HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED) ||
1309                      ppd->offline_disabled_reason ==
1310                      HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
1311                         ppd->offline_disabled_reason =
1312                         HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
1313                 /*
1314                  * Don't send a reply if the response would be sent
1315                  * through the disabled port.
1316                  */
1317                 if (link_state == HLS_DN_DISABLE && !local_mad)
1318                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1319                 break;
1320         case IB_PORT_ARMED:
1321                 ret = set_link_state(ppd, HLS_UP_ARMED);
1322                 if (!ret)
1323                         send_idle_sma(dd, SMA_IDLE_ARM);
1324                 break;
1325         case IB_PORT_ACTIVE:
1326                 if (ppd->neighbor_normal) {
1327                         ret = set_link_state(ppd, HLS_UP_ACTIVE);
1328                         if (ret == 0)
1329                                 send_idle_sma(dd, SMA_IDLE_ACTIVE);
1330                 } else {
1331                         pr_warn("SubnSet(OPA_PortInfo) Cannot move to Active with NeighborNormal 0\n");
1332                         smp->status |= IB_SMP_INVALID_FIELD;
1333                 }
1334                 break;
1335         default:
1336                 pr_warn("SubnSet(OPA_PortInfo) invalid logical state 0x%x\n",
1337                         logical_state);
1338                 smp->status |= IB_SMP_INVALID_FIELD;
1339         }
1340
1341         return 0;
1342 }
1343
1344 /**
1345  * subn_set_opa_portinfo - set port information
1346  * @smp: the incoming SM packet
1347  * @ibdev: the infiniband device
1348  * @port: the port on the device
1349  *
1350  */
1351 static int __subn_set_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
1352                                    struct ib_device *ibdev, u8 port,
1353                                    u32 *resp_len, u32 max_len, int local_mad)
1354 {
1355         struct opa_port_info *pi = (struct opa_port_info *)data;
1356         struct ib_event event;
1357         struct hfi1_devdata *dd;
1358         struct hfi1_pportdata *ppd;
1359         struct hfi1_ibport *ibp;
1360         u8 clientrereg;
1361         unsigned long flags;
1362         u32 smlid;
1363         u32 lid;
1364         u8 ls_old, ls_new, ps_new;
1365         u8 vls;
1366         u8 msl;
1367         u8 crc_enabled;
1368         u16 lse, lwe, mtu;
1369         u32 num_ports = OPA_AM_NPORT(am);
1370         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1371         int ret, i, invalid = 0, call_set_mtu = 0;
1372         int call_link_downgrade_policy = 0;
1373
1374         if (num_ports != 1 ||
1375             smp_length_check(sizeof(*pi), max_len)) {
1376                 smp->status |= IB_SMP_INVALID_FIELD;
1377                 return reply((struct ib_mad_hdr *)smp);
1378         }
1379
1380         lid = be32_to_cpu(pi->lid);
1381         if (lid & 0xFF000000) {
1382                 pr_warn("OPA_PortInfo lid out of range: %X\n", lid);
1383                 smp->status |= IB_SMP_INVALID_FIELD;
1384                 goto get_only;
1385         }
1386
1387
1388         smlid = be32_to_cpu(pi->sm_lid);
1389         if (smlid & 0xFF000000) {
1390                 pr_warn("OPA_PortInfo SM lid out of range: %X\n", smlid);
1391                 smp->status |= IB_SMP_INVALID_FIELD;
1392                 goto get_only;
1393         }
1394
1395         clientrereg = (pi->clientrereg_subnettimeout &
1396                         OPA_PI_MASK_CLIENT_REREGISTER);
1397
1398         dd = dd_from_ibdev(ibdev);
1399         /* IB numbers ports from 1, hw from 0 */
1400         ppd = dd->pport + (port - 1);
1401         ibp = &ppd->ibport_data;
1402         event.device = ibdev;
1403         event.element.port_num = port;
1404
1405         ls_old = driver_lstate(ppd);
1406
1407         ibp->rvp.mkey = pi->mkey;
1408         if (ibp->rvp.gid_prefix != pi->subnet_prefix) {
1409                 ibp->rvp.gid_prefix = pi->subnet_prefix;
1410                 event.event = IB_EVENT_GID_CHANGE;
1411                 ib_dispatch_event(&event);
1412         }
1413         ibp->rvp.mkey_lease_period = be16_to_cpu(pi->mkey_lease_period);
1414
1415         /* Must be a valid unicast LID address. */
1416         if ((lid == 0 && ls_old > IB_PORT_INIT) ||
1417              (hfi1_is_16B_mcast(lid))) {
1418                 smp->status |= IB_SMP_INVALID_FIELD;
1419                 pr_warn("SubnSet(OPA_PortInfo) lid invalid 0x%x\n",
1420                         lid);
1421         } else if (ppd->lid != lid ||
1422                  ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC)) {
1423                 if (ppd->lid != lid)
1424                         hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LID_CHANGE_BIT);
1425                 if (ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC))
1426                         hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LMC_CHANGE_BIT);
1427                 hfi1_set_lid(ppd, lid, pi->mkeyprotect_lmc & OPA_PI_MASK_LMC);
1428                 event.event = IB_EVENT_LID_CHANGE;
1429                 ib_dispatch_event(&event);
1430
1431                 if (HFI1_PORT_GUID_INDEX + 1 < HFI1_GUIDS_PER_PORT) {
1432                         /* Manufacture GID from LID to support extended
1433                          * addresses
1434                          */
1435                         ppd->guids[HFI1_PORT_GUID_INDEX + 1] =
1436                                 be64_to_cpu(OPA_MAKE_ID(lid));
1437                         event.event = IB_EVENT_GID_CHANGE;
1438                         ib_dispatch_event(&event);
1439                 }
1440         }
1441
1442         msl = pi->smsl & OPA_PI_MASK_SMSL;
1443         if (pi->partenforce_filterraw & OPA_PI_MASK_LINKINIT_REASON)
1444                 ppd->linkinit_reason =
1445                         (pi->partenforce_filterraw &
1446                          OPA_PI_MASK_LINKINIT_REASON);
1447
1448         /* Must be a valid unicast LID address. */
1449         if ((smlid == 0 && ls_old > IB_PORT_INIT) ||
1450              (hfi1_is_16B_mcast(smlid))) {
1451                 smp->status |= IB_SMP_INVALID_FIELD;
1452                 pr_warn("SubnSet(OPA_PortInfo) smlid invalid 0x%x\n", smlid);
1453         } else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
1454                 pr_warn("SubnSet(OPA_PortInfo) smlid 0x%x\n", smlid);
1455                 spin_lock_irqsave(&ibp->rvp.lock, flags);
1456                 if (ibp->rvp.sm_ah) {
1457                         if (smlid != ibp->rvp.sm_lid)
1458                                 hfi1_modify_qp0_ah(ibp, ibp->rvp.sm_ah, smlid);
1459                         if (msl != ibp->rvp.sm_sl)
1460                                 rdma_ah_set_sl(&ibp->rvp.sm_ah->attr, msl);
1461                 }
1462                 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1463                 if (smlid != ibp->rvp.sm_lid)
1464                         ibp->rvp.sm_lid = smlid;
1465                 if (msl != ibp->rvp.sm_sl)
1466                         ibp->rvp.sm_sl = msl;
1467                 event.event = IB_EVENT_SM_CHANGE;
1468                 ib_dispatch_event(&event);
1469         }
1470
1471         if (pi->link_down_reason == 0) {
1472                 ppd->local_link_down_reason.sma = 0;
1473                 ppd->local_link_down_reason.latest = 0;
1474         }
1475
1476         if (pi->neigh_link_down_reason == 0) {
1477                 ppd->neigh_link_down_reason.sma = 0;
1478                 ppd->neigh_link_down_reason.latest = 0;
1479         }
1480
1481         ppd->sm_trap_qp = be32_to_cpu(pi->sm_trap_qp);
1482         ppd->sa_qp = be32_to_cpu(pi->sa_qp);
1483
1484         ppd->port_error_action = be32_to_cpu(pi->port_error_action);
1485         lwe = be16_to_cpu(pi->link_width.enabled);
1486         if (lwe) {
1487                 if (lwe == OPA_LINK_WIDTH_RESET ||
1488                     lwe == OPA_LINK_WIDTH_RESET_OLD)
1489                         set_link_width_enabled(ppd, ppd->link_width_supported);
1490                 else if ((lwe & ~ppd->link_width_supported) == 0)
1491                         set_link_width_enabled(ppd, lwe);
1492                 else
1493                         smp->status |= IB_SMP_INVALID_FIELD;
1494         }
1495         lwe = be16_to_cpu(pi->link_width_downgrade.enabled);
1496         /* LWD.E is always applied - 0 means "disabled" */
1497         if (lwe == OPA_LINK_WIDTH_RESET ||
1498             lwe == OPA_LINK_WIDTH_RESET_OLD) {
1499                 set_link_width_downgrade_enabled(ppd,
1500                                                  ppd->
1501                                                  link_width_downgrade_supported
1502                                                  );
1503         } else if ((lwe & ~ppd->link_width_downgrade_supported) == 0) {
1504                 /* only set and apply if something changed */
1505                 if (lwe != ppd->link_width_downgrade_enabled) {
1506                         set_link_width_downgrade_enabled(ppd, lwe);
1507                         call_link_downgrade_policy = 1;
1508                 }
1509         } else {
1510                 smp->status |= IB_SMP_INVALID_FIELD;
1511         }
1512         lse = be16_to_cpu(pi->link_speed.enabled);
1513         if (lse) {
1514                 if (lse & be16_to_cpu(pi->link_speed.supported))
1515                         set_link_speed_enabled(ppd, lse);
1516                 else
1517                         smp->status |= IB_SMP_INVALID_FIELD;
1518         }
1519
1520         ibp->rvp.mkeyprot =
1521                 (pi->mkeyprotect_lmc & OPA_PI_MASK_MKEY_PROT_BIT) >> 6;
1522         ibp->rvp.vl_high_limit = be16_to_cpu(pi->vl.high_limit) & 0xFF;
1523         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_LIMIT,
1524                                     ibp->rvp.vl_high_limit);
1525
1526         if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
1527             ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
1528                 smp->status |= IB_SMP_INVALID_FIELD;
1529                 return reply((struct ib_mad_hdr *)smp);
1530         }
1531         for (i = 0; i < ppd->vls_supported; i++) {
1532                 if ((i % 2) == 0)
1533                         mtu = enum_to_mtu((pi->neigh_mtu.pvlx_to_mtu[i / 2] >>
1534                                            4) & 0xF);
1535                 else
1536                         mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[i / 2] &
1537                                           0xF);
1538                 if (mtu == 0xffff) {
1539                         pr_warn("SubnSet(OPA_PortInfo) mtu invalid %d (0x%x)\n",
1540                                 mtu,
1541                                 (pi->neigh_mtu.pvlx_to_mtu[0] >> 4) & 0xF);
1542                         smp->status |= IB_SMP_INVALID_FIELD;
1543                         mtu = hfi1_max_mtu; /* use a valid MTU */
1544                 }
1545                 if (dd->vld[i].mtu != mtu) {
1546                         dd_dev_info(dd,
1547                                     "MTU change on vl %d from %d to %d\n",
1548                                     i, dd->vld[i].mtu, mtu);
1549                         dd->vld[i].mtu = mtu;
1550                         call_set_mtu++;
1551                 }
1552         }
1553         /* As per OPAV1 spec: VL15 must support and be configured
1554          * for operation with a 2048 or larger MTU.
1555          */
1556         mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[15 / 2] & 0xF);
1557         if (mtu < 2048 || mtu == 0xffff)
1558                 mtu = 2048;
1559         if (dd->vld[15].mtu != mtu) {
1560                 dd_dev_info(dd,
1561                             "MTU change on vl 15 from %d to %d\n",
1562                             dd->vld[15].mtu, mtu);
1563                 dd->vld[15].mtu = mtu;
1564                 call_set_mtu++;
1565         }
1566         if (call_set_mtu)
1567                 set_mtu(ppd);
1568
1569         /* Set operational VLs */
1570         vls = pi->operational_vls & OPA_PI_MASK_OPERATIONAL_VL;
1571         if (vls) {
1572                 if (vls > ppd->vls_supported) {
1573                         pr_warn("SubnSet(OPA_PortInfo) VL's supported invalid %d\n",
1574                                 pi->operational_vls);
1575                         smp->status |= IB_SMP_INVALID_FIELD;
1576                 } else {
1577                         if (hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS,
1578                                             vls) == -EINVAL)
1579                                 smp->status |= IB_SMP_INVALID_FIELD;
1580                 }
1581         }
1582
1583         if (pi->mkey_violations == 0)
1584                 ibp->rvp.mkey_violations = 0;
1585
1586         if (pi->pkey_violations == 0)
1587                 ibp->rvp.pkey_violations = 0;
1588
1589         if (pi->qkey_violations == 0)
1590                 ibp->rvp.qkey_violations = 0;
1591
1592         ibp->rvp.subnet_timeout =
1593                 pi->clientrereg_subnettimeout & OPA_PI_MASK_SUBNET_TIMEOUT;
1594
1595         crc_enabled = be16_to_cpu(pi->port_ltp_crc_mode);
1596         crc_enabled >>= 4;
1597         crc_enabled &= 0xf;
1598
1599         if (crc_enabled != 0)
1600                 ppd->port_crc_mode_enabled = port_ltp_to_cap(crc_enabled);
1601
1602         ppd->is_active_optimize_enabled =
1603                         !!(be16_to_cpu(pi->port_mode)
1604                                         & OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE);
1605
1606         ls_new = pi->port_states.portphysstate_portstate &
1607                         OPA_PI_MASK_PORT_STATE;
1608         ps_new = (pi->port_states.portphysstate_portstate &
1609                         OPA_PI_MASK_PORT_PHYSICAL_STATE) >> 4;
1610
1611         if (ls_old == IB_PORT_INIT) {
1612                 if (start_of_sm_config) {
1613                         if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1614                                 ppd->is_sm_config_started = 1;
1615                 } else if (ls_new == IB_PORT_ARMED) {
1616                         if (ppd->is_sm_config_started == 0) {
1617                                 invalid = 1;
1618                                 smp->status |= IB_SMP_INVALID_FIELD;
1619                         }
1620                 }
1621         }
1622
1623         /* Handle CLIENT_REREGISTER event b/c SM asked us for it */
1624         if (clientrereg) {
1625                 event.event = IB_EVENT_CLIENT_REREGISTER;
1626                 ib_dispatch_event(&event);
1627         }
1628
1629         /*
1630          * Do the port state change now that the other link parameters
1631          * have been set.
1632          * Changing the port physical state only makes sense if the link
1633          * is down or is being set to down.
1634          */
1635
1636         if (!invalid) {
1637                 ret = set_port_states(ppd, smp, ls_new, ps_new, local_mad);
1638                 if (ret)
1639                         return ret;
1640         }
1641
1642         ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len,
1643                                       max_len);
1644
1645         /* restore re-reg bit per o14-12.2.1 */
1646         pi->clientrereg_subnettimeout |= clientrereg;
1647
1648         /*
1649          * Apply the new link downgrade policy.  This may result in a link
1650          * bounce.  Do this after everything else so things are settled.
1651          * Possible problem: if setting the port state above fails, then
1652          * the policy change is not applied.
1653          */
1654         if (call_link_downgrade_policy)
1655                 apply_link_downgrade_policy(ppd, 0);
1656
1657         return ret;
1658
1659 get_only:
1660         return __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len,
1661                                        max_len);
1662 }
1663
1664 /**
1665  * set_pkeys - set the PKEY table for ctxt 0
1666  * @dd: the hfi1_ib device
1667  * @port: the IB port number
1668  * @pkeys: the PKEY table
1669  */
1670 static int set_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1671 {
1672         struct hfi1_pportdata *ppd;
1673         int i;
1674         int changed = 0;
1675         int update_includes_mgmt_partition = 0;
1676
1677         /*
1678          * IB port one/two always maps to context zero/one,
1679          * always a kernel context, no locking needed
1680          * If we get here with ppd setup, no need to check
1681          * that rcd is valid.
1682          */
1683         ppd = dd->pport + (port - 1);
1684         /*
1685          * If the update does not include the management pkey, don't do it.
1686          */
1687         for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1688                 if (pkeys[i] == LIM_MGMT_P_KEY) {
1689                         update_includes_mgmt_partition = 1;
1690                         break;
1691                 }
1692         }
1693
1694         if (!update_includes_mgmt_partition)
1695                 return 1;
1696
1697         for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1698                 u16 key = pkeys[i];
1699                 u16 okey = ppd->pkeys[i];
1700
1701                 if (key == okey)
1702                         continue;
1703                 /*
1704                  * The SM gives us the complete PKey table. We have
1705                  * to ensure that we put the PKeys in the matching
1706                  * slots.
1707                  */
1708                 ppd->pkeys[i] = key;
1709                 changed = 1;
1710         }
1711
1712         if (changed) {
1713                 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
1714                 hfi1_event_pkey_change(dd, port);
1715         }
1716
1717         return 0;
1718 }
1719
1720 static int __subn_set_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1721                                     struct ib_device *ibdev, u8 port,
1722                                     u32 *resp_len, u32 max_len)
1723 {
1724         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1725         u32 n_blocks_sent = OPA_AM_NBLK(am);
1726         u32 start_block = am & 0x7ff;
1727         u16 *p = (u16 *)data;
1728         __be16 *q = (__be16 *)data;
1729         int i;
1730         u16 n_blocks_avail;
1731         unsigned npkeys = hfi1_get_npkeys(dd);
1732         u32 size = 0;
1733
1734         if (n_blocks_sent == 0) {
1735                 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1736                         port, start_block, n_blocks_sent);
1737                 smp->status |= IB_SMP_INVALID_FIELD;
1738                 return reply((struct ib_mad_hdr *)smp);
1739         }
1740
1741         n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
1742
1743         size = sizeof(u16) * (n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE);
1744
1745         if (smp_length_check(size, max_len)) {
1746                 smp->status |= IB_SMP_INVALID_FIELD;
1747                 return reply((struct ib_mad_hdr *)smp);
1748         }
1749
1750         if (start_block + n_blocks_sent > n_blocks_avail ||
1751             n_blocks_sent > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1752                 pr_warn("OPA Set PKey AM Invalid : s 0x%x; req 0x%x; avail 0x%x; blk/smp 0x%lx\n",
1753                         start_block, n_blocks_sent, n_blocks_avail,
1754                         OPA_NUM_PKEY_BLOCKS_PER_SMP);
1755                 smp->status |= IB_SMP_INVALID_FIELD;
1756                 return reply((struct ib_mad_hdr *)smp);
1757         }
1758
1759         for (i = 0; i < n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE; i++)
1760                 p[i] = be16_to_cpu(q[i]);
1761
1762         if (start_block == 0 && set_pkeys(dd, port, p) != 0) {
1763                 smp->status |= IB_SMP_INVALID_FIELD;
1764                 return reply((struct ib_mad_hdr *)smp);
1765         }
1766
1767         return __subn_get_opa_pkeytable(smp, am, data, ibdev, port, resp_len,
1768                                         max_len);
1769 }
1770
1771 #define ILLEGAL_VL 12
1772 /*
1773  * filter_sc2vlt changes mappings to VL15 to ILLEGAL_VL (except
1774  * for SC15, which must map to VL15). If we don't remap things this
1775  * way it is possible for VL15 counters to increment when we try to
1776  * send on a SC which is mapped to an invalid VL.
1777  * When getting the table convert ILLEGAL_VL back to VL15.
1778  */
1779 static void filter_sc2vlt(void *data, bool set)
1780 {
1781         int i;
1782         u8 *pd = data;
1783
1784         for (i = 0; i < OPA_MAX_SCS; i++) {
1785                 if (i == 15)
1786                         continue;
1787
1788                 if (set) {
1789                         if ((pd[i] & 0x1f) == 0xf)
1790                                 pd[i] = ILLEGAL_VL;
1791                 } else {
1792                         if ((pd[i] & 0x1f) == ILLEGAL_VL)
1793                                 pd[i] = 0xf;
1794                 }
1795         }
1796 }
1797
1798 static int set_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1799 {
1800         u64 *val = data;
1801
1802         filter_sc2vlt(data, true);
1803
1804         write_csr(dd, SEND_SC2VLT0, *val++);
1805         write_csr(dd, SEND_SC2VLT1, *val++);
1806         write_csr(dd, SEND_SC2VLT2, *val++);
1807         write_csr(dd, SEND_SC2VLT3, *val++);
1808         write_seqlock_irq(&dd->sc2vl_lock);
1809         memcpy(dd->sc2vl, data, sizeof(dd->sc2vl));
1810         write_sequnlock_irq(&dd->sc2vl_lock);
1811         return 0;
1812 }
1813
1814 static int get_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1815 {
1816         u64 *val = (u64 *)data;
1817
1818         *val++ = read_csr(dd, SEND_SC2VLT0);
1819         *val++ = read_csr(dd, SEND_SC2VLT1);
1820         *val++ = read_csr(dd, SEND_SC2VLT2);
1821         *val++ = read_csr(dd, SEND_SC2VLT3);
1822
1823         filter_sc2vlt((u64 *)data, false);
1824         return 0;
1825 }
1826
1827 static int __subn_get_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1828                                    struct ib_device *ibdev, u8 port,
1829                                    u32 *resp_len, u32 max_len)
1830 {
1831         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1832         u8 *p = data;
1833         size_t size = ARRAY_SIZE(ibp->sl_to_sc); /* == 32 */
1834         unsigned i;
1835
1836         if (am || smp_length_check(size, max_len)) {
1837                 smp->status |= IB_SMP_INVALID_FIELD;
1838                 return reply((struct ib_mad_hdr *)smp);
1839         }
1840
1841         for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1842                 *p++ = ibp->sl_to_sc[i];
1843
1844         if (resp_len)
1845                 *resp_len += size;
1846
1847         return reply((struct ib_mad_hdr *)smp);
1848 }
1849
1850 static int __subn_set_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1851                                    struct ib_device *ibdev, u8 port,
1852                                    u32 *resp_len, u32 max_len)
1853 {
1854         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1855         u8 *p = data;
1856         size_t size = ARRAY_SIZE(ibp->sl_to_sc);
1857         int i;
1858         u8 sc;
1859
1860         if (am || smp_length_check(size, max_len)) {
1861                 smp->status |= IB_SMP_INVALID_FIELD;
1862                 return reply((struct ib_mad_hdr *)smp);
1863         }
1864
1865         for (i = 0; i <  ARRAY_SIZE(ibp->sl_to_sc); i++) {
1866                 sc = *p++;
1867                 if (ibp->sl_to_sc[i] != sc) {
1868                         ibp->sl_to_sc[i] = sc;
1869
1870                         /* Put all stale qps into error state */
1871                         hfi1_error_port_qps(ibp, i);
1872                 }
1873         }
1874
1875         return __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port, resp_len,
1876                                        max_len);
1877 }
1878
1879 static int __subn_get_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1880                                    struct ib_device *ibdev, u8 port,
1881                                    u32 *resp_len, u32 max_len)
1882 {
1883         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1884         u8 *p = data;
1885         size_t size = ARRAY_SIZE(ibp->sc_to_sl); /* == 32 */
1886         unsigned i;
1887
1888         if (am || smp_length_check(size, max_len)) {
1889                 smp->status |= IB_SMP_INVALID_FIELD;
1890                 return reply((struct ib_mad_hdr *)smp);
1891         }
1892
1893         for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1894                 *p++ = ibp->sc_to_sl[i];
1895
1896         if (resp_len)
1897                 *resp_len += size;
1898
1899         return reply((struct ib_mad_hdr *)smp);
1900 }
1901
1902 static int __subn_set_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1903                                    struct ib_device *ibdev, u8 port,
1904                                    u32 *resp_len, u32 max_len)
1905 {
1906         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1907         size_t size = ARRAY_SIZE(ibp->sc_to_sl);
1908         u8 *p = data;
1909         int i;
1910
1911         if (am || smp_length_check(size, max_len)) {
1912                 smp->status |= IB_SMP_INVALID_FIELD;
1913                 return reply((struct ib_mad_hdr *)smp);
1914         }
1915
1916         for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1917                 ibp->sc_to_sl[i] = *p++;
1918
1919         return __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port, resp_len,
1920                                        max_len);
1921 }
1922
1923 static int __subn_get_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1924                                     struct ib_device *ibdev, u8 port,
1925                                     u32 *resp_len, u32 max_len)
1926 {
1927         u32 n_blocks = OPA_AM_NBLK(am);
1928         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1929         void *vp = (void *)data;
1930         size_t size = 4 * sizeof(u64);
1931
1932         if (n_blocks != 1 || smp_length_check(size, max_len)) {
1933                 smp->status |= IB_SMP_INVALID_FIELD;
1934                 return reply((struct ib_mad_hdr *)smp);
1935         }
1936
1937         get_sc2vlt_tables(dd, vp);
1938
1939         if (resp_len)
1940                 *resp_len += size;
1941
1942         return reply((struct ib_mad_hdr *)smp);
1943 }
1944
1945 static int __subn_set_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1946                                     struct ib_device *ibdev, u8 port,
1947                                     u32 *resp_len, u32 max_len)
1948 {
1949         u32 n_blocks = OPA_AM_NBLK(am);
1950         int async_update = OPA_AM_ASYNC(am);
1951         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1952         void *vp = (void *)data;
1953         struct hfi1_pportdata *ppd;
1954         int lstate;
1955         /*
1956          * set_sc2vlt_tables writes the information contained in *data
1957          * to four 64-bit registers SendSC2VLt[0-3]. We need to make
1958          * sure *max_len is not greater than the total size of the four
1959          * SendSC2VLt[0-3] registers.
1960          */
1961         size_t size = 4 * sizeof(u64);
1962
1963         if (n_blocks != 1 || async_update || smp_length_check(size, max_len)) {
1964                 smp->status |= IB_SMP_INVALID_FIELD;
1965                 return reply((struct ib_mad_hdr *)smp);
1966         }
1967
1968         /* IB numbers ports from 1, hw from 0 */
1969         ppd = dd->pport + (port - 1);
1970         lstate = driver_lstate(ppd);
1971         /*
1972          * it's known that async_update is 0 by this point, but include
1973          * the explicit check for clarity
1974          */
1975         if (!async_update &&
1976             (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE)) {
1977                 smp->status |= IB_SMP_INVALID_FIELD;
1978                 return reply((struct ib_mad_hdr *)smp);
1979         }
1980
1981         set_sc2vlt_tables(dd, vp);
1982
1983         return __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port, resp_len,
1984                                         max_len);
1985 }
1986
1987 static int __subn_get_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1988                                      struct ib_device *ibdev, u8 port,
1989                                      u32 *resp_len, u32 max_len)
1990 {
1991         u32 n_blocks = OPA_AM_NPORT(am);
1992         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1993         struct hfi1_pportdata *ppd;
1994         void *vp = (void *)data;
1995         int size = sizeof(struct sc2vlnt);
1996
1997         if (n_blocks != 1 || smp_length_check(size, max_len)) {
1998                 smp->status |= IB_SMP_INVALID_FIELD;
1999                 return reply((struct ib_mad_hdr *)smp);
2000         }
2001
2002         ppd = dd->pport + (port - 1);
2003
2004         fm_get_table(ppd, FM_TBL_SC2VLNT, vp);
2005
2006         if (resp_len)
2007                 *resp_len += size;
2008
2009         return reply((struct ib_mad_hdr *)smp);
2010 }
2011
2012 static int __subn_set_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
2013                                      struct ib_device *ibdev, u8 port,
2014                                      u32 *resp_len, u32 max_len)
2015 {
2016         u32 n_blocks = OPA_AM_NPORT(am);
2017         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2018         struct hfi1_pportdata *ppd;
2019         void *vp = (void *)data;
2020         int lstate;
2021         int size = sizeof(struct sc2vlnt);
2022
2023         if (n_blocks != 1 || smp_length_check(size, max_len)) {
2024                 smp->status |= IB_SMP_INVALID_FIELD;
2025                 return reply((struct ib_mad_hdr *)smp);
2026         }
2027
2028         /* IB numbers ports from 1, hw from 0 */
2029         ppd = dd->pport + (port - 1);
2030         lstate = driver_lstate(ppd);
2031         if (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE) {
2032                 smp->status |= IB_SMP_INVALID_FIELD;
2033                 return reply((struct ib_mad_hdr *)smp);
2034         }
2035
2036         ppd = dd->pport + (port - 1);
2037
2038         fm_set_table(ppd, FM_TBL_SC2VLNT, vp);
2039
2040         return __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
2041                                          resp_len, max_len);
2042 }
2043
2044 static int __subn_get_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
2045                               struct ib_device *ibdev, u8 port,
2046                               u32 *resp_len, u32 max_len)
2047 {
2048         u32 nports = OPA_AM_NPORT(am);
2049         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
2050         u32 lstate;
2051         struct hfi1_ibport *ibp;
2052         struct hfi1_pportdata *ppd;
2053         struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
2054
2055         if (nports != 1 || smp_length_check(sizeof(*psi), max_len)) {
2056                 smp->status |= IB_SMP_INVALID_FIELD;
2057                 return reply((struct ib_mad_hdr *)smp);
2058         }
2059
2060         ibp = to_iport(ibdev, port);
2061         ppd = ppd_from_ibp(ibp);
2062
2063         lstate = driver_lstate(ppd);
2064
2065         if (start_of_sm_config && (lstate == IB_PORT_INIT))
2066                 ppd->is_sm_config_started = 1;
2067
2068         psi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
2069         psi->port_states.ledenable_offlinereason |=
2070                 ppd->is_sm_config_started << 5;
2071         psi->port_states.ledenable_offlinereason |=
2072                 ppd->offline_disabled_reason;
2073
2074         psi->port_states.portphysstate_portstate =
2075                 (driver_pstate(ppd) << 4) | (lstate & 0xf);
2076         psi->link_width_downgrade_tx_active =
2077                 cpu_to_be16(ppd->link_width_downgrade_tx_active);
2078         psi->link_width_downgrade_rx_active =
2079                 cpu_to_be16(ppd->link_width_downgrade_rx_active);
2080         if (resp_len)
2081                 *resp_len += sizeof(struct opa_port_state_info);
2082
2083         return reply((struct ib_mad_hdr *)smp);
2084 }
2085
2086 static int __subn_set_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
2087                               struct ib_device *ibdev, u8 port,
2088                               u32 *resp_len, u32 max_len, int local_mad)
2089 {
2090         u32 nports = OPA_AM_NPORT(am);
2091         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
2092         u32 ls_old;
2093         u8 ls_new, ps_new;
2094         struct hfi1_ibport *ibp;
2095         struct hfi1_pportdata *ppd;
2096         struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
2097         int ret, invalid = 0;
2098
2099         if (nports != 1 || smp_length_check(sizeof(*psi), max_len)) {
2100                 smp->status |= IB_SMP_INVALID_FIELD;
2101                 return reply((struct ib_mad_hdr *)smp);
2102         }
2103
2104         ibp = to_iport(ibdev, port);
2105         ppd = ppd_from_ibp(ibp);
2106
2107         ls_old = driver_lstate(ppd);
2108
2109         ls_new = port_states_to_logical_state(&psi->port_states);
2110         ps_new = port_states_to_phys_state(&psi->port_states);
2111
2112         if (ls_old == IB_PORT_INIT) {
2113                 if (start_of_sm_config) {
2114                         if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
2115                                 ppd->is_sm_config_started = 1;
2116                 } else if (ls_new == IB_PORT_ARMED) {
2117                         if (ppd->is_sm_config_started == 0) {
2118                                 invalid = 1;
2119                                 smp->status |= IB_SMP_INVALID_FIELD;
2120                         }
2121                 }
2122         }
2123
2124         if (!invalid) {
2125                 ret = set_port_states(ppd, smp, ls_new, ps_new, local_mad);
2126                 if (ret)
2127                         return ret;
2128         }
2129
2130         return __subn_get_opa_psi(smp, am, data, ibdev, port, resp_len,
2131                                   max_len);
2132 }
2133
2134 static int __subn_get_opa_cable_info(struct opa_smp *smp, u32 am, u8 *data,
2135                                      struct ib_device *ibdev, u8 port,
2136                                      u32 *resp_len, u32 max_len)
2137 {
2138         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2139         u32 addr = OPA_AM_CI_ADDR(am);
2140         u32 len = OPA_AM_CI_LEN(am) + 1;
2141         int ret;
2142
2143         if (dd->pport->port_type != PORT_TYPE_QSFP ||
2144             smp_length_check(len, max_len)) {
2145                 smp->status |= IB_SMP_INVALID_FIELD;
2146                 return reply((struct ib_mad_hdr *)smp);
2147         }
2148
2149 #define __CI_PAGE_SIZE BIT(7) /* 128 bytes */
2150 #define __CI_PAGE_MASK ~(__CI_PAGE_SIZE - 1)
2151 #define __CI_PAGE_NUM(a) ((a) & __CI_PAGE_MASK)
2152
2153         /*
2154          * check that addr is within spec, and
2155          * addr and (addr + len - 1) are on the same "page"
2156          */
2157         if (addr >= 4096 ||
2158             (__CI_PAGE_NUM(addr) != __CI_PAGE_NUM(addr + len - 1))) {
2159                 smp->status |= IB_SMP_INVALID_FIELD;
2160                 return reply((struct ib_mad_hdr *)smp);
2161         }
2162
2163         ret = get_cable_info(dd, port, addr, len, data);
2164
2165         if (ret == -ENODEV) {
2166                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
2167                 return reply((struct ib_mad_hdr *)smp);
2168         }
2169
2170         /* The address range for the CableInfo SMA query is wider than the
2171          * memory available on the QSFP cable. We want to return a valid
2172          * response, albeit zeroed out, for address ranges beyond available
2173          * memory but that are within the CableInfo query spec
2174          */
2175         if (ret < 0 && ret != -ERANGE) {
2176                 smp->status |= IB_SMP_INVALID_FIELD;
2177                 return reply((struct ib_mad_hdr *)smp);
2178         }
2179
2180         if (resp_len)
2181                 *resp_len += len;
2182
2183         return reply((struct ib_mad_hdr *)smp);
2184 }
2185
2186 static int __subn_get_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
2187                               struct ib_device *ibdev, u8 port, u32 *resp_len,
2188                               u32 max_len)
2189 {
2190         u32 num_ports = OPA_AM_NPORT(am);
2191         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2192         struct hfi1_pportdata *ppd;
2193         struct buffer_control *p = (struct buffer_control *)data;
2194         int size = sizeof(struct buffer_control);
2195
2196         if (num_ports != 1 || smp_length_check(size, max_len)) {
2197                 smp->status |= IB_SMP_INVALID_FIELD;
2198                 return reply((struct ib_mad_hdr *)smp);
2199         }
2200
2201         ppd = dd->pport + (port - 1);
2202         fm_get_table(ppd, FM_TBL_BUFFER_CONTROL, p);
2203         trace_bct_get(dd, p);
2204         if (resp_len)
2205                 *resp_len += size;
2206
2207         return reply((struct ib_mad_hdr *)smp);
2208 }
2209
2210 static int __subn_set_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
2211                               struct ib_device *ibdev, u8 port, u32 *resp_len,
2212                               u32 max_len)
2213 {
2214         u32 num_ports = OPA_AM_NPORT(am);
2215         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2216         struct hfi1_pportdata *ppd;
2217         struct buffer_control *p = (struct buffer_control *)data;
2218
2219         if (num_ports != 1 || smp_length_check(sizeof(*p), max_len)) {
2220                 smp->status |= IB_SMP_INVALID_FIELD;
2221                 return reply((struct ib_mad_hdr *)smp);
2222         }
2223         ppd = dd->pport + (port - 1);
2224         trace_bct_set(dd, p);
2225         if (fm_set_table(ppd, FM_TBL_BUFFER_CONTROL, p) < 0) {
2226                 smp->status |= IB_SMP_INVALID_FIELD;
2227                 return reply((struct ib_mad_hdr *)smp);
2228         }
2229
2230         return __subn_get_opa_bct(smp, am, data, ibdev, port, resp_len,
2231                                   max_len);
2232 }
2233
2234 static int __subn_get_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
2235                                  struct ib_device *ibdev, u8 port,
2236                                  u32 *resp_len, u32 max_len)
2237 {
2238         struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
2239         u32 num_ports = OPA_AM_NPORT(am);
2240         u8 section = (am & 0x00ff0000) >> 16;
2241         u8 *p = data;
2242         int size = 256;
2243
2244         if (num_ports != 1 || smp_length_check(size, max_len)) {
2245                 smp->status |= IB_SMP_INVALID_FIELD;
2246                 return reply((struct ib_mad_hdr *)smp);
2247         }
2248
2249         switch (section) {
2250         case OPA_VLARB_LOW_ELEMENTS:
2251                 fm_get_table(ppd, FM_TBL_VL_LOW_ARB, p);
2252                 break;
2253         case OPA_VLARB_HIGH_ELEMENTS:
2254                 fm_get_table(ppd, FM_TBL_VL_HIGH_ARB, p);
2255                 break;
2256         case OPA_VLARB_PREEMPT_ELEMENTS:
2257                 fm_get_table(ppd, FM_TBL_VL_PREEMPT_ELEMS, p);
2258                 break;
2259         case OPA_VLARB_PREEMPT_MATRIX:
2260                 fm_get_table(ppd, FM_TBL_VL_PREEMPT_MATRIX, p);
2261                 break;
2262         default:
2263                 pr_warn("OPA SubnGet(VL Arb) AM Invalid : 0x%x\n",
2264                         be32_to_cpu(smp->attr_mod));
2265                 smp->status |= IB_SMP_INVALID_FIELD;
2266                 size = 0;
2267                 break;
2268         }
2269
2270         if (size > 0 && resp_len)
2271                 *resp_len += size;
2272
2273         return reply((struct ib_mad_hdr *)smp);
2274 }
2275
2276 static int __subn_set_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
2277                                  struct ib_device *ibdev, u8 port,
2278                                  u32 *resp_len, u32 max_len)
2279 {
2280         struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
2281         u32 num_ports = OPA_AM_NPORT(am);
2282         u8 section = (am & 0x00ff0000) >> 16;
2283         u8 *p = data;
2284         int size = 256;
2285
2286         if (num_ports != 1 || smp_length_check(size, max_len)) {
2287                 smp->status |= IB_SMP_INVALID_FIELD;
2288                 return reply((struct ib_mad_hdr *)smp);
2289         }
2290
2291         switch (section) {
2292         case OPA_VLARB_LOW_ELEMENTS:
2293                 (void)fm_set_table(ppd, FM_TBL_VL_LOW_ARB, p);
2294                 break;
2295         case OPA_VLARB_HIGH_ELEMENTS:
2296                 (void)fm_set_table(ppd, FM_TBL_VL_HIGH_ARB, p);
2297                 break;
2298         /*
2299          * neither OPA_VLARB_PREEMPT_ELEMENTS, or OPA_VLARB_PREEMPT_MATRIX
2300          * can be changed from the default values
2301          */
2302         case OPA_VLARB_PREEMPT_ELEMENTS:
2303                 /* FALLTHROUGH */
2304         case OPA_VLARB_PREEMPT_MATRIX:
2305                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
2306                 break;
2307         default:
2308                 pr_warn("OPA SubnSet(VL Arb) AM Invalid : 0x%x\n",
2309                         be32_to_cpu(smp->attr_mod));
2310                 smp->status |= IB_SMP_INVALID_FIELD;
2311                 break;
2312         }
2313
2314         return __subn_get_opa_vl_arb(smp, am, data, ibdev, port, resp_len,
2315                                      max_len);
2316 }
2317
2318 struct opa_pma_mad {
2319         struct ib_mad_hdr mad_hdr;
2320         u8 data[2024];
2321 } __packed;
2322
2323 struct opa_port_status_req {
2324         __u8 port_num;
2325         __u8 reserved[3];
2326         __be32 vl_select_mask;
2327 };
2328
2329 #define VL_MASK_ALL             0x000080ff
2330
2331 struct opa_port_status_rsp {
2332         __u8 port_num;
2333         __u8 reserved[3];
2334         __be32  vl_select_mask;
2335
2336         /* Data counters */
2337         __be64 port_xmit_data;
2338         __be64 port_rcv_data;
2339         __be64 port_xmit_pkts;
2340         __be64 port_rcv_pkts;
2341         __be64 port_multicast_xmit_pkts;
2342         __be64 port_multicast_rcv_pkts;
2343         __be64 port_xmit_wait;
2344         __be64 sw_port_congestion;
2345         __be64 port_rcv_fecn;
2346         __be64 port_rcv_becn;
2347         __be64 port_xmit_time_cong;
2348         __be64 port_xmit_wasted_bw;
2349         __be64 port_xmit_wait_data;
2350         __be64 port_rcv_bubble;
2351         __be64 port_mark_fecn;
2352         /* Error counters */
2353         __be64 port_rcv_constraint_errors;
2354         __be64 port_rcv_switch_relay_errors;
2355         __be64 port_xmit_discards;
2356         __be64 port_xmit_constraint_errors;
2357         __be64 port_rcv_remote_physical_errors;
2358         __be64 local_link_integrity_errors;
2359         __be64 port_rcv_errors;
2360         __be64 excessive_buffer_overruns;
2361         __be64 fm_config_errors;
2362         __be32 link_error_recovery;
2363         __be32 link_downed;
2364         u8 uncorrectable_errors;
2365
2366         u8 link_quality_indicator; /* 5res, 3bit */
2367         u8 res2[6];
2368         struct _vls_pctrs {
2369                 /* per-VL Data counters */
2370                 __be64 port_vl_xmit_data;
2371                 __be64 port_vl_rcv_data;
2372                 __be64 port_vl_xmit_pkts;
2373                 __be64 port_vl_rcv_pkts;
2374                 __be64 port_vl_xmit_wait;
2375                 __be64 sw_port_vl_congestion;
2376                 __be64 port_vl_rcv_fecn;
2377                 __be64 port_vl_rcv_becn;
2378                 __be64 port_xmit_time_cong;
2379                 __be64 port_vl_xmit_wasted_bw;
2380                 __be64 port_vl_xmit_wait_data;
2381                 __be64 port_vl_rcv_bubble;
2382                 __be64 port_vl_mark_fecn;
2383                 __be64 port_vl_xmit_discards;
2384         } vls[0]; /* real array size defined by # bits set in vl_select_mask */
2385 };
2386
2387 enum counter_selects {
2388         CS_PORT_XMIT_DATA                       = (1 << 31),
2389         CS_PORT_RCV_DATA                        = (1 << 30),
2390         CS_PORT_XMIT_PKTS                       = (1 << 29),
2391         CS_PORT_RCV_PKTS                        = (1 << 28),
2392         CS_PORT_MCAST_XMIT_PKTS                 = (1 << 27),
2393         CS_PORT_MCAST_RCV_PKTS                  = (1 << 26),
2394         CS_PORT_XMIT_WAIT                       = (1 << 25),
2395         CS_SW_PORT_CONGESTION                   = (1 << 24),
2396         CS_PORT_RCV_FECN                        = (1 << 23),
2397         CS_PORT_RCV_BECN                        = (1 << 22),
2398         CS_PORT_XMIT_TIME_CONG                  = (1 << 21),
2399         CS_PORT_XMIT_WASTED_BW                  = (1 << 20),
2400         CS_PORT_XMIT_WAIT_DATA                  = (1 << 19),
2401         CS_PORT_RCV_BUBBLE                      = (1 << 18),
2402         CS_PORT_MARK_FECN                       = (1 << 17),
2403         CS_PORT_RCV_CONSTRAINT_ERRORS           = (1 << 16),
2404         CS_PORT_RCV_SWITCH_RELAY_ERRORS         = (1 << 15),
2405         CS_PORT_XMIT_DISCARDS                   = (1 << 14),
2406         CS_PORT_XMIT_CONSTRAINT_ERRORS          = (1 << 13),
2407         CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS      = (1 << 12),
2408         CS_LOCAL_LINK_INTEGRITY_ERRORS          = (1 << 11),
2409         CS_PORT_RCV_ERRORS                      = (1 << 10),
2410         CS_EXCESSIVE_BUFFER_OVERRUNS            = (1 << 9),
2411         CS_FM_CONFIG_ERRORS                     = (1 << 8),
2412         CS_LINK_ERROR_RECOVERY                  = (1 << 7),
2413         CS_LINK_DOWNED                          = (1 << 6),
2414         CS_UNCORRECTABLE_ERRORS                 = (1 << 5),
2415 };
2416
2417 struct opa_clear_port_status {
2418         __be64 port_select_mask[4];
2419         __be32 counter_select_mask;
2420 };
2421
2422 struct opa_aggregate {
2423         __be16 attr_id;
2424         __be16 err_reqlength;   /* 1 bit, 8 res, 7 bit */
2425         __be32 attr_mod;
2426         u8 data[0];
2427 };
2428
2429 #define MSK_LLI 0x000000f0
2430 #define MSK_LLI_SFT 4
2431 #define MSK_LER 0x0000000f
2432 #define MSK_LER_SFT 0
2433 #define ADD_LLI 8
2434 #define ADD_LER 2
2435
2436 /* Request contains first three fields, response contains those plus the rest */
2437 struct opa_port_data_counters_msg {
2438         __be64 port_select_mask[4];
2439         __be32 vl_select_mask;
2440         __be32 resolution;
2441
2442         /* Response fields follow */
2443         struct _port_dctrs {
2444                 u8 port_number;
2445                 u8 reserved2[3];
2446                 __be32 link_quality_indicator; /* 29res, 3bit */
2447
2448                 /* Data counters */
2449                 __be64 port_xmit_data;
2450                 __be64 port_rcv_data;
2451                 __be64 port_xmit_pkts;
2452                 __be64 port_rcv_pkts;
2453                 __be64 port_multicast_xmit_pkts;
2454                 __be64 port_multicast_rcv_pkts;
2455                 __be64 port_xmit_wait;
2456                 __be64 sw_port_congestion;
2457                 __be64 port_rcv_fecn;
2458                 __be64 port_rcv_becn;
2459                 __be64 port_xmit_time_cong;
2460                 __be64 port_xmit_wasted_bw;
2461                 __be64 port_xmit_wait_data;
2462                 __be64 port_rcv_bubble;
2463                 __be64 port_mark_fecn;
2464
2465                 __be64 port_error_counter_summary;
2466                 /* Sum of error counts/port */
2467
2468                 struct _vls_dctrs {
2469                         /* per-VL Data counters */
2470                         __be64 port_vl_xmit_data;
2471                         __be64 port_vl_rcv_data;
2472                         __be64 port_vl_xmit_pkts;
2473                         __be64 port_vl_rcv_pkts;
2474                         __be64 port_vl_xmit_wait;
2475                         __be64 sw_port_vl_congestion;
2476                         __be64 port_vl_rcv_fecn;
2477                         __be64 port_vl_rcv_becn;
2478                         __be64 port_xmit_time_cong;
2479                         __be64 port_vl_xmit_wasted_bw;
2480                         __be64 port_vl_xmit_wait_data;
2481                         __be64 port_vl_rcv_bubble;
2482                         __be64 port_vl_mark_fecn;
2483                 } vls[0];
2484                 /* array size defined by #bits set in vl_select_mask*/
2485         } port[1]; /* array size defined by  #ports in attribute modifier */
2486 };
2487
2488 struct opa_port_error_counters64_msg {
2489         /*
2490          * Request contains first two fields, response contains the
2491          * whole magilla
2492          */
2493         __be64 port_select_mask[4];
2494         __be32 vl_select_mask;
2495
2496         /* Response-only fields follow */
2497         __be32 reserved1;
2498         struct _port_ectrs {
2499                 u8 port_number;
2500                 u8 reserved2[7];
2501                 __be64 port_rcv_constraint_errors;
2502                 __be64 port_rcv_switch_relay_errors;
2503                 __be64 port_xmit_discards;
2504                 __be64 port_xmit_constraint_errors;
2505                 __be64 port_rcv_remote_physical_errors;
2506                 __be64 local_link_integrity_errors;
2507                 __be64 port_rcv_errors;
2508                 __be64 excessive_buffer_overruns;
2509                 __be64 fm_config_errors;
2510                 __be32 link_error_recovery;
2511                 __be32 link_downed;
2512                 u8 uncorrectable_errors;
2513                 u8 reserved3[7];
2514                 struct _vls_ectrs {
2515                         __be64 port_vl_xmit_discards;
2516                 } vls[0];
2517                 /* array size defined by #bits set in vl_select_mask */
2518         } port[1]; /* array size defined by #ports in attribute modifier */
2519 };
2520
2521 struct opa_port_error_info_msg {
2522         __be64 port_select_mask[4];
2523         __be32 error_info_select_mask;
2524         __be32 reserved1;
2525         struct _port_ei {
2526                 u8 port_number;
2527                 u8 reserved2[7];
2528
2529                 /* PortRcvErrorInfo */
2530                 struct {
2531                         u8 status_and_code;
2532                         union {
2533                                 u8 raw[17];
2534                                 struct {
2535                                         /* EI1to12 format */
2536                                         u8 packet_flit1[8];
2537                                         u8 packet_flit2[8];
2538                                         u8 remaining_flit_bits12;
2539                                 } ei1to12;
2540                                 struct {
2541                                         u8 packet_bytes[8];
2542                                         u8 remaining_flit_bits;
2543                                 } ei13;
2544                         } ei;
2545                         u8 reserved3[6];
2546                 } __packed port_rcv_ei;
2547
2548                 /* ExcessiveBufferOverrunInfo */
2549                 struct {
2550                         u8 status_and_sc;
2551                         u8 reserved4[7];
2552                 } __packed excessive_buffer_overrun_ei;
2553
2554                 /* PortXmitConstraintErrorInfo */
2555                 struct {
2556                         u8 status;
2557                         u8 reserved5;
2558                         __be16 pkey;
2559                         __be32 slid;
2560                 } __packed port_xmit_constraint_ei;
2561
2562                 /* PortRcvConstraintErrorInfo */
2563                 struct {
2564                         u8 status;
2565                         u8 reserved6;
2566                         __be16 pkey;
2567                         __be32 slid;
2568                 } __packed port_rcv_constraint_ei;
2569
2570                 /* PortRcvSwitchRelayErrorInfo */
2571                 struct {
2572                         u8 status_and_code;
2573                         u8 reserved7[3];
2574                         __u32 error_info;
2575                 } __packed port_rcv_switch_relay_ei;
2576
2577                 /* UncorrectableErrorInfo */
2578                 struct {
2579                         u8 status_and_code;
2580                         u8 reserved8;
2581                 } __packed uncorrectable_ei;
2582
2583                 /* FMConfigErrorInfo */
2584                 struct {
2585                         u8 status_and_code;
2586                         u8 error_info;
2587                 } __packed fm_config_ei;
2588                 __u32 reserved9;
2589         } port[1]; /* actual array size defined by #ports in attr modifier */
2590 };
2591
2592 /* opa_port_error_info_msg error_info_select_mask bit definitions */
2593 enum error_info_selects {
2594         ES_PORT_RCV_ERROR_INFO                  = (1 << 31),
2595         ES_EXCESSIVE_BUFFER_OVERRUN_INFO        = (1 << 30),
2596         ES_PORT_XMIT_CONSTRAINT_ERROR_INFO      = (1 << 29),
2597         ES_PORT_RCV_CONSTRAINT_ERROR_INFO       = (1 << 28),
2598         ES_PORT_RCV_SWITCH_RELAY_ERROR_INFO     = (1 << 27),
2599         ES_UNCORRECTABLE_ERROR_INFO             = (1 << 26),
2600         ES_FM_CONFIG_ERROR_INFO                 = (1 << 25)
2601 };
2602
2603 static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
2604                                      struct ib_device *ibdev, u32 *resp_len)
2605 {
2606         struct opa_class_port_info *p =
2607                 (struct opa_class_port_info *)pmp->data;
2608
2609         memset(pmp->data, 0, sizeof(pmp->data));
2610
2611         if (pmp->mad_hdr.attr_mod != 0)
2612                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2613
2614         p->base_version = OPA_MGMT_BASE_VERSION;
2615         p->class_version = OPA_SM_CLASS_VERSION;
2616         /*
2617          * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2618          */
2619         p->cap_mask2_resp_time = cpu_to_be32(18);
2620
2621         if (resp_len)
2622                 *resp_len += sizeof(*p);
2623
2624         return reply((struct ib_mad_hdr *)pmp);
2625 }
2626
2627 static void a0_portstatus(struct hfi1_pportdata *ppd,
2628                           struct opa_port_status_rsp *rsp, u32 vl_select_mask)
2629 {
2630         if (!is_bx(ppd->dd)) {
2631                 unsigned long vl;
2632                 u64 sum_vl_xmit_wait = 0;
2633                 u32 vl_all_mask = VL_MASK_ALL;
2634
2635                 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2636                                  8 * sizeof(vl_all_mask)) {
2637                         u64 tmp = sum_vl_xmit_wait +
2638                                   read_port_cntr(ppd, C_TX_WAIT_VL,
2639                                                  idx_from_vl(vl));
2640                         if (tmp < sum_vl_xmit_wait) {
2641                                 /* we wrapped */
2642                                 sum_vl_xmit_wait = (u64)~0;
2643                                 break;
2644                         }
2645                         sum_vl_xmit_wait = tmp;
2646                 }
2647                 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2648                         rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2649         }
2650 }
2651
2652 /**
2653  * tx_link_width - convert link width bitmask to integer
2654  * value representing actual link width.
2655  * @link_width: width of active link
2656  * @return: return index of the bit set in link_width var
2657  *
2658  * The function convert and return the index of bit set
2659  * that indicate the current link width.
2660  */
2661 u16 tx_link_width(u16 link_width)
2662 {
2663         int n = LINK_WIDTH_DEFAULT;
2664         u16 tx_width = n;
2665
2666         while (link_width && n) {
2667                 if (link_width & (1 << (n - 1))) {
2668                         tx_width = n;
2669                         break;
2670                 }
2671                 n--;
2672         }
2673
2674         return tx_width;
2675 }
2676
2677 /**
2678  * get_xmit_wait_counters - Convert HFI 's SendWaitCnt/SendWaitVlCnt
2679  * counter in unit of TXE cycle times to flit times.
2680  * @ppd: info of physical Hfi port
2681  * @link_width: width of active link
2682  * @link_speed: speed of active link
2683  * @vl: represent VL0-VL7, VL15 for PortVLXmitWait counters request
2684  * and if vl value is C_VL_COUNT, it represent SendWaitCnt
2685  * counter request
2686  * @return: return SendWaitCnt/SendWaitVlCnt counter value per vl.
2687  *
2688  * Convert SendWaitCnt/SendWaitVlCnt counter from TXE cycle times to
2689  * flit times. Call this function to samples these counters. This
2690  * function will calculate for previous state transition and update
2691  * current state at end of function using ppd->prev_link_width and
2692  * ppd->port_vl_xmit_wait_last to port_vl_xmit_wait_curr and link_width.
2693  */
2694 u64 get_xmit_wait_counters(struct hfi1_pportdata *ppd,
2695                            u16 link_width, u16 link_speed, int vl)
2696 {
2697         u64 port_vl_xmit_wait_curr;
2698         u64 delta_vl_xmit_wait;
2699         u64 xmit_wait_val;
2700
2701         if (vl > C_VL_COUNT)
2702                 return  0;
2703         if (vl < C_VL_COUNT)
2704                 port_vl_xmit_wait_curr =
2705                         read_port_cntr(ppd, C_TX_WAIT_VL, vl);
2706         else
2707                 port_vl_xmit_wait_curr =
2708                         read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL);
2709
2710         xmit_wait_val =
2711                 port_vl_xmit_wait_curr -
2712                 ppd->port_vl_xmit_wait_last[vl];
2713         delta_vl_xmit_wait =
2714                 convert_xmit_counter(xmit_wait_val,
2715                                      ppd->prev_link_width,
2716                                      link_speed);
2717
2718         ppd->vl_xmit_flit_cnt[vl] += delta_vl_xmit_wait;
2719         ppd->port_vl_xmit_wait_last[vl] = port_vl_xmit_wait_curr;
2720         ppd->prev_link_width = link_width;
2721
2722         return ppd->vl_xmit_flit_cnt[vl];
2723 }
2724
2725 static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
2726                                   struct ib_device *ibdev,
2727                                   u8 port, u32 *resp_len)
2728 {
2729         struct opa_port_status_req *req =
2730                 (struct opa_port_status_req *)pmp->data;
2731         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2732         struct opa_port_status_rsp *rsp;
2733         u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2734         unsigned long vl;
2735         size_t response_data_size;
2736         u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2737         u8 port_num = req->port_num;
2738         u8 num_vls = hweight32(vl_select_mask);
2739         struct _vls_pctrs *vlinfo;
2740         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2741         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2742         int vfi;
2743         u64 tmp, tmp2;
2744         u16 link_width;
2745         u16 link_speed;
2746
2747         response_data_size = sizeof(struct opa_port_status_rsp) +
2748                                 num_vls * sizeof(struct _vls_pctrs);
2749         if (response_data_size > sizeof(pmp->data)) {
2750                 pmp->mad_hdr.status |= OPA_PM_STATUS_REQUEST_TOO_LARGE;
2751                 return reply((struct ib_mad_hdr *)pmp);
2752         }
2753
2754         if (nports != 1 || (port_num && port_num != port) ||
2755             num_vls > OPA_MAX_VLS || (vl_select_mask & ~VL_MASK_ALL)) {
2756                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2757                 return reply((struct ib_mad_hdr *)pmp);
2758         }
2759
2760         memset(pmp->data, 0, sizeof(pmp->data));
2761
2762         rsp = (struct opa_port_status_rsp *)pmp->data;
2763         if (port_num)
2764                 rsp->port_num = port_num;
2765         else
2766                 rsp->port_num = port;
2767
2768         rsp->port_rcv_constraint_errors =
2769                 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2770                                            CNTR_INVALID_VL));
2771
2772         hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
2773
2774         rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
2775         rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2776                                           CNTR_INVALID_VL));
2777         rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2778                                          CNTR_INVALID_VL));
2779         rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2780                                           CNTR_INVALID_VL));
2781         rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2782                                          CNTR_INVALID_VL));
2783         rsp->port_multicast_xmit_pkts =
2784                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2785                                           CNTR_INVALID_VL));
2786         rsp->port_multicast_rcv_pkts =
2787                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2788                                           CNTR_INVALID_VL));
2789         /*
2790          * Convert PortXmitWait counter from TXE cycle times
2791          * to flit times.
2792          */
2793         link_width =
2794                 tx_link_width(ppd->link_width_downgrade_tx_active);
2795         link_speed = get_link_speed(ppd->link_speed_active);
2796         rsp->port_xmit_wait =
2797                 cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
2798                                                    link_speed, C_VL_COUNT));
2799         rsp->port_rcv_fecn =
2800                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2801         rsp->port_rcv_becn =
2802                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2803         rsp->port_xmit_discards =
2804                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2805                                            CNTR_INVALID_VL));
2806         rsp->port_xmit_constraint_errors =
2807                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2808                                            CNTR_INVALID_VL));
2809         rsp->port_rcv_remote_physical_errors =
2810                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2811                                           CNTR_INVALID_VL));
2812         rsp->local_link_integrity_errors =
2813                 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2814                                           CNTR_INVALID_VL));
2815         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2816         tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2817                                    CNTR_INVALID_VL);
2818         if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2819                 /* overflow/wrapped */
2820                 rsp->link_error_recovery = cpu_to_be32(~0);
2821         } else {
2822                 rsp->link_error_recovery = cpu_to_be32(tmp2);
2823         }
2824         rsp->port_rcv_errors =
2825                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2826         rsp->excessive_buffer_overruns =
2827                 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2828         rsp->fm_config_errors =
2829                 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2830                                           CNTR_INVALID_VL));
2831         rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2832                                                       CNTR_INVALID_VL));
2833
2834         /* rsp->uncorrectable_errors is 8 bits wide, and it pegs at 0xff */
2835         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2836         rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2837
2838         vlinfo = &rsp->vls[0];
2839         vfi = 0;
2840         /* The vl_select_mask has been checked above, and we know
2841          * that it contains only entries which represent valid VLs.
2842          * So in the for_each_set_bit() loop below, we don't need
2843          * any additional checks for vl.
2844          */
2845         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2846                          8 * sizeof(vl_select_mask)) {
2847                 memset(vlinfo, 0, sizeof(*vlinfo));
2848
2849                 tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
2850                 rsp->vls[vfi].port_vl_rcv_data = cpu_to_be64(tmp);
2851
2852                 rsp->vls[vfi].port_vl_rcv_pkts =
2853                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2854                                                   idx_from_vl(vl)));
2855
2856                 rsp->vls[vfi].port_vl_xmit_data =
2857                         cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2858                                                    idx_from_vl(vl)));
2859
2860                 rsp->vls[vfi].port_vl_xmit_pkts =
2861                         cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2862                                                    idx_from_vl(vl)));
2863                 /*
2864                  * Convert PortVlXmitWait counter from TXE cycle
2865                  * times to flit times.
2866                  */
2867                 rsp->vls[vfi].port_vl_xmit_wait =
2868                         cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
2869                                                            link_speed,
2870                                                            idx_from_vl(vl)));
2871
2872                 rsp->vls[vfi].port_vl_rcv_fecn =
2873                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2874                                                   idx_from_vl(vl)));
2875
2876                 rsp->vls[vfi].port_vl_rcv_becn =
2877                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2878                                                   idx_from_vl(vl)));
2879
2880                 rsp->vls[vfi].port_vl_xmit_discards =
2881                         cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2882                                                    idx_from_vl(vl)));
2883                 vlinfo++;
2884                 vfi++;
2885         }
2886
2887         a0_portstatus(ppd, rsp, vl_select_mask);
2888
2889         if (resp_len)
2890                 *resp_len += response_data_size;
2891
2892         return reply((struct ib_mad_hdr *)pmp);
2893 }
2894
2895 static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
2896                                      u8 res_lli, u8 res_ler)
2897 {
2898         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2899         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2900         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2901         u64 error_counter_summary = 0, tmp;
2902
2903         error_counter_summary += read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2904                                                 CNTR_INVALID_VL);
2905         /* port_rcv_switch_relay_errors is 0 for HFIs */
2906         error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_DSCD,
2907                                                 CNTR_INVALID_VL);
2908         error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2909                                                 CNTR_INVALID_VL);
2910         error_counter_summary += read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2911                                                CNTR_INVALID_VL);
2912         /* local link integrity must be right-shifted by the lli resolution */
2913         error_counter_summary += (read_dev_cntr(dd, C_DC_RX_REPLAY,
2914                                                 CNTR_INVALID_VL) >> res_lli);
2915         /* link error recovery must b right-shifted by the ler resolution */
2916         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2917         tmp += read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL);
2918         error_counter_summary += (tmp >> res_ler);
2919         error_counter_summary += read_dev_cntr(dd, C_DC_RCV_ERR,
2920                                                CNTR_INVALID_VL);
2921         error_counter_summary += read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL);
2922         error_counter_summary += read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2923                                                CNTR_INVALID_VL);
2924         /* ppd->link_downed is a 32-bit value */
2925         error_counter_summary += read_port_cntr(ppd, C_SW_LINK_DOWN,
2926                                                 CNTR_INVALID_VL);
2927         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2928         /* this is an 8-bit quantity */
2929         error_counter_summary += tmp < 0x100 ? (tmp & 0xff) : 0xff;
2930
2931         return error_counter_summary;
2932 }
2933
2934 static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
2935                             u32 vl_select_mask)
2936 {
2937         if (!is_bx(ppd->dd)) {
2938                 unsigned long vl;
2939                 u64 sum_vl_xmit_wait = 0;
2940                 u32 vl_all_mask = VL_MASK_ALL;
2941
2942                 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2943                                  8 * sizeof(vl_all_mask)) {
2944                         u64 tmp = sum_vl_xmit_wait +
2945                                   read_port_cntr(ppd, C_TX_WAIT_VL,
2946                                                  idx_from_vl(vl));
2947                         if (tmp < sum_vl_xmit_wait) {
2948                                 /* we wrapped */
2949                                 sum_vl_xmit_wait = (u64)~0;
2950                                 break;
2951                         }
2952                         sum_vl_xmit_wait = tmp;
2953                 }
2954                 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2955                         rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2956         }
2957 }
2958
2959 static void pma_get_opa_port_dctrs(struct ib_device *ibdev,
2960                                    struct _port_dctrs *rsp)
2961 {
2962         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2963
2964         rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2965                                                 CNTR_INVALID_VL));
2966         rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2967                                                 CNTR_INVALID_VL));
2968         rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2969                                                 CNTR_INVALID_VL));
2970         rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2971                                                 CNTR_INVALID_VL));
2972         rsp->port_multicast_xmit_pkts =
2973                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2974                                           CNTR_INVALID_VL));
2975         rsp->port_multicast_rcv_pkts =
2976                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2977                                           CNTR_INVALID_VL));
2978 }
2979
2980 static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
2981                                     struct ib_device *ibdev,
2982                                     u8 port, u32 *resp_len)
2983 {
2984         struct opa_port_data_counters_msg *req =
2985                 (struct opa_port_data_counters_msg *)pmp->data;
2986         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2987         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2988         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2989         struct _port_dctrs *rsp;
2990         struct _vls_dctrs *vlinfo;
2991         size_t response_data_size;
2992         u32 num_ports;
2993         u8 lq, num_vls;
2994         u8 res_lli, res_ler;
2995         u64 port_mask;
2996         u8 port_num;
2997         unsigned long vl;
2998         u32 vl_select_mask;
2999         int vfi;
3000         u16 link_width;
3001         u16 link_speed;
3002
3003         num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3004         num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
3005         vl_select_mask = be32_to_cpu(req->vl_select_mask);
3006         res_lli = (u8)(be32_to_cpu(req->resolution) & MSK_LLI) >> MSK_LLI_SFT;
3007         res_lli = res_lli ? res_lli + ADD_LLI : 0;
3008         res_ler = (u8)(be32_to_cpu(req->resolution) & MSK_LER) >> MSK_LER_SFT;
3009         res_ler = res_ler ? res_ler + ADD_LER : 0;
3010
3011         if (num_ports != 1 || (vl_select_mask & ~VL_MASK_ALL)) {
3012                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3013                 return reply((struct ib_mad_hdr *)pmp);
3014         }
3015
3016         /* Sanity check */
3017         response_data_size = sizeof(struct opa_port_data_counters_msg) +
3018                                 num_vls * sizeof(struct _vls_dctrs);
3019
3020         if (response_data_size > sizeof(pmp->data)) {
3021                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3022                 return reply((struct ib_mad_hdr *)pmp);
3023         }
3024
3025         /*
3026          * The bit set in the mask needs to be consistent with the
3027          * port the request came in on.
3028          */
3029         port_mask = be64_to_cpu(req->port_select_mask[3]);
3030         port_num = find_first_bit((unsigned long *)&port_mask,
3031                                   sizeof(port_mask) * 8);
3032
3033         if (port_num != port) {
3034                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3035                 return reply((struct ib_mad_hdr *)pmp);
3036         }
3037
3038         rsp = &req->port[0];
3039         memset(rsp, 0, sizeof(*rsp));
3040
3041         rsp->port_number = port;
3042         /*
3043          * Note that link_quality_indicator is a 32 bit quantity in
3044          * 'datacounters' queries (as opposed to 'portinfo' queries,
3045          * where it's a byte).
3046          */
3047         hfi1_read_link_quality(dd, &lq);
3048         rsp->link_quality_indicator = cpu_to_be32((u32)lq);
3049         pma_get_opa_port_dctrs(ibdev, rsp);
3050
3051         /*
3052          * Convert PortXmitWait counter from TXE
3053          * cycle times to flit times.
3054          */
3055         link_width =
3056                 tx_link_width(ppd->link_width_downgrade_tx_active);
3057         link_speed = get_link_speed(ppd->link_speed_active);
3058         rsp->port_xmit_wait =
3059                 cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
3060                                                    link_speed, C_VL_COUNT));
3061         rsp->port_rcv_fecn =
3062                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
3063         rsp->port_rcv_becn =
3064                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
3065         rsp->port_error_counter_summary =
3066                 cpu_to_be64(get_error_counter_summary(ibdev, port,
3067                                                       res_lli, res_ler));
3068
3069         vlinfo = &rsp->vls[0];
3070         vfi = 0;
3071         /* The vl_select_mask has been checked above, and we know
3072          * that it contains only entries which represent valid VLs.
3073          * So in the for_each_set_bit() loop below, we don't need
3074          * any additional checks for vl.
3075          */
3076         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3077                          8 * sizeof(req->vl_select_mask)) {
3078                 memset(vlinfo, 0, sizeof(*vlinfo));
3079
3080                 rsp->vls[vfi].port_vl_xmit_data =
3081                         cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
3082                                                    idx_from_vl(vl)));
3083
3084                 rsp->vls[vfi].port_vl_rcv_data =
3085                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_FLIT_VL,
3086                                                   idx_from_vl(vl)));
3087
3088                 rsp->vls[vfi].port_vl_xmit_pkts =
3089                         cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
3090                                                    idx_from_vl(vl)));
3091
3092                 rsp->vls[vfi].port_vl_rcv_pkts =
3093                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
3094                                                   idx_from_vl(vl)));
3095
3096                 /*
3097                  * Convert PortVlXmitWait counter from TXE
3098                  * cycle times to flit times.
3099                  */
3100                 rsp->vls[vfi].port_vl_xmit_wait =
3101                         cpu_to_be64(get_xmit_wait_counters(ppd, link_width,
3102                                                            link_speed,
3103                                                            idx_from_vl(vl)));
3104
3105                 rsp->vls[vfi].port_vl_rcv_fecn =
3106                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
3107                                                   idx_from_vl(vl)));
3108                 rsp->vls[vfi].port_vl_rcv_becn =
3109                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
3110                                                   idx_from_vl(vl)));
3111
3112                 /* rsp->port_vl_xmit_time_cong is 0 for HFIs */
3113                 /* rsp->port_vl_xmit_wasted_bw ??? */
3114                 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ???
3115                  * does this differ from rsp->vls[vfi].port_vl_xmit_wait
3116                  */
3117                 /*rsp->vls[vfi].port_vl_mark_fecn =
3118                  *      cpu_to_be64(read_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT
3119                  *              + offset));
3120                  */
3121                 vlinfo++;
3122                 vfi++;
3123         }
3124
3125         a0_datacounters(ppd, rsp, vl_select_mask);
3126
3127         if (resp_len)
3128                 *resp_len += response_data_size;
3129
3130         return reply((struct ib_mad_hdr *)pmp);
3131 }
3132
3133 static int pma_get_ib_portcounters_ext(struct ib_pma_mad *pmp,
3134                                        struct ib_device *ibdev, u8 port)
3135 {
3136         struct ib_pma_portcounters_ext *p = (struct ib_pma_portcounters_ext *)
3137                                                 pmp->data;
3138         struct _port_dctrs rsp;
3139
3140         if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
3141                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3142                 goto bail;
3143         }
3144
3145         memset(&rsp, 0, sizeof(rsp));
3146         pma_get_opa_port_dctrs(ibdev, &rsp);
3147
3148         p->port_xmit_data = rsp.port_xmit_data;
3149         p->port_rcv_data = rsp.port_rcv_data;
3150         p->port_xmit_packets = rsp.port_xmit_pkts;
3151         p->port_rcv_packets = rsp.port_rcv_pkts;
3152         p->port_unicast_xmit_packets = 0;
3153         p->port_unicast_rcv_packets =  0;
3154         p->port_multicast_xmit_packets = rsp.port_multicast_xmit_pkts;
3155         p->port_multicast_rcv_packets = rsp.port_multicast_rcv_pkts;
3156
3157 bail:
3158         return reply((struct ib_mad_hdr *)pmp);
3159 }
3160
3161 static void pma_get_opa_port_ectrs(struct ib_device *ibdev,
3162                                    struct _port_ectrs *rsp, u8 port)
3163 {
3164         u64 tmp, tmp2;
3165         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3166         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3167         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3168
3169         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
3170         tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
3171                                         CNTR_INVALID_VL);
3172         if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
3173                 /* overflow/wrapped */
3174                 rsp->link_error_recovery = cpu_to_be32(~0);
3175         } else {
3176                 rsp->link_error_recovery = cpu_to_be32(tmp2);
3177         }
3178
3179         rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
3180                                                 CNTR_INVALID_VL));
3181         rsp->port_rcv_errors =
3182                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
3183         rsp->port_rcv_remote_physical_errors =
3184                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
3185                                           CNTR_INVALID_VL));
3186         rsp->port_rcv_switch_relay_errors = 0;
3187         rsp->port_xmit_discards =
3188                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
3189                                            CNTR_INVALID_VL));
3190         rsp->port_xmit_constraint_errors =
3191                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
3192                                            CNTR_INVALID_VL));
3193         rsp->port_rcv_constraint_errors =
3194                 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
3195                                            CNTR_INVALID_VL));
3196         rsp->local_link_integrity_errors =
3197                 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
3198                                           CNTR_INVALID_VL));
3199         rsp->excessive_buffer_overruns =
3200                 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
3201 }
3202
3203 static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
3204                                   struct ib_device *ibdev,
3205                                   u8 port, u32 *resp_len)
3206 {
3207         size_t response_data_size;
3208         struct _port_ectrs *rsp;
3209         u8 port_num;
3210         struct opa_port_error_counters64_msg *req;
3211         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3212         u32 num_ports;
3213         u8 num_pslm;
3214         u8 num_vls;
3215         struct hfi1_ibport *ibp;
3216         struct hfi1_pportdata *ppd;
3217         struct _vls_ectrs *vlinfo;
3218         unsigned long vl;
3219         u64 port_mask, tmp;
3220         u32 vl_select_mask;
3221         int vfi;
3222
3223         req = (struct opa_port_error_counters64_msg *)pmp->data;
3224
3225         num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3226
3227         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3228         num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
3229
3230         if (num_ports != 1 || num_ports != num_pslm) {
3231                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3232                 return reply((struct ib_mad_hdr *)pmp);
3233         }
3234
3235         response_data_size = sizeof(struct opa_port_error_counters64_msg) +
3236                                 num_vls * sizeof(struct _vls_ectrs);
3237
3238         if (response_data_size > sizeof(pmp->data)) {
3239                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3240                 return reply((struct ib_mad_hdr *)pmp);
3241         }
3242         /*
3243          * The bit set in the mask needs to be consistent with the
3244          * port the request came in on.
3245          */
3246         port_mask = be64_to_cpu(req->port_select_mask[3]);
3247         port_num = find_first_bit((unsigned long *)&port_mask,
3248                                   sizeof(port_mask) * 8);
3249
3250         if (port_num != port) {
3251                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3252                 return reply((struct ib_mad_hdr *)pmp);
3253         }
3254
3255         rsp = &req->port[0];
3256
3257         ibp = to_iport(ibdev, port_num);
3258         ppd = ppd_from_ibp(ibp);
3259
3260         memset(rsp, 0, sizeof(*rsp));
3261         rsp->port_number = port_num;
3262
3263         pma_get_opa_port_ectrs(ibdev, rsp, port_num);
3264
3265         rsp->port_rcv_remote_physical_errors =
3266                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
3267                                           CNTR_INVALID_VL));
3268         rsp->fm_config_errors =
3269                 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
3270                                           CNTR_INVALID_VL));
3271         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
3272
3273         rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
3274         rsp->port_rcv_errors =
3275                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
3276         vlinfo = &rsp->vls[0];
3277         vfi = 0;
3278         vl_select_mask = be32_to_cpu(req->vl_select_mask);
3279         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3280                          8 * sizeof(req->vl_select_mask)) {
3281                 memset(vlinfo, 0, sizeof(*vlinfo));
3282                 rsp->vls[vfi].port_vl_xmit_discards =
3283                         cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3284                                                    idx_from_vl(vl)));
3285                 vlinfo += 1;
3286                 vfi++;
3287         }
3288
3289         if (resp_len)
3290                 *resp_len += response_data_size;
3291
3292         return reply((struct ib_mad_hdr *)pmp);
3293 }
3294
3295 static int pma_get_ib_portcounters(struct ib_pma_mad *pmp,
3296                                    struct ib_device *ibdev, u8 port)
3297 {
3298         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
3299                 pmp->data;
3300         struct _port_ectrs rsp;
3301         u64 temp_link_overrun_errors;
3302         u64 temp_64;
3303         u32 temp_32;
3304
3305         memset(&rsp, 0, sizeof(rsp));
3306         pma_get_opa_port_ectrs(ibdev, &rsp, port);
3307
3308         if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
3309                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3310                 goto bail;
3311         }
3312
3313         p->symbol_error_counter = 0; /* N/A for OPA */
3314
3315         temp_32 = be32_to_cpu(rsp.link_error_recovery);
3316         if (temp_32 > 0xFFUL)
3317                 p->link_error_recovery_counter = 0xFF;
3318         else
3319                 p->link_error_recovery_counter = (u8)temp_32;
3320
3321         temp_32 = be32_to_cpu(rsp.link_downed);
3322         if (temp_32 > 0xFFUL)
3323                 p->link_downed_counter = 0xFF;
3324         else
3325                 p->link_downed_counter = (u8)temp_32;
3326
3327         temp_64 = be64_to_cpu(rsp.port_rcv_errors);
3328         if (temp_64 > 0xFFFFUL)
3329                 p->port_rcv_errors = cpu_to_be16(0xFFFF);
3330         else
3331                 p->port_rcv_errors = cpu_to_be16((u16)temp_64);
3332
3333         temp_64 = be64_to_cpu(rsp.port_rcv_remote_physical_errors);
3334         if (temp_64 > 0xFFFFUL)
3335                 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
3336         else
3337                 p->port_rcv_remphys_errors = cpu_to_be16((u16)temp_64);
3338
3339         temp_64 = be64_to_cpu(rsp.port_rcv_switch_relay_errors);
3340         p->port_rcv_switch_relay_errors = cpu_to_be16((u16)temp_64);
3341
3342         temp_64 = be64_to_cpu(rsp.port_xmit_discards);
3343         if (temp_64 > 0xFFFFUL)
3344                 p->port_xmit_discards = cpu_to_be16(0xFFFF);
3345         else
3346                 p->port_xmit_discards = cpu_to_be16((u16)temp_64);
3347
3348         temp_64 = be64_to_cpu(rsp.port_xmit_constraint_errors);
3349         if (temp_64 > 0xFFUL)
3350                 p->port_xmit_constraint_errors = 0xFF;
3351         else
3352                 p->port_xmit_constraint_errors = (u8)temp_64;
3353
3354         temp_64 = be64_to_cpu(rsp.port_rcv_constraint_errors);
3355         if (temp_64 > 0xFFUL)
3356                 p->port_rcv_constraint_errors = 0xFFUL;
3357         else
3358                 p->port_rcv_constraint_errors = (u8)temp_64;
3359
3360         /* LocalLink: 7:4, BufferOverrun: 3:0 */
3361         temp_64 = be64_to_cpu(rsp.local_link_integrity_errors);
3362         if (temp_64 > 0xFUL)
3363                 temp_64 = 0xFUL;
3364
3365         temp_link_overrun_errors = temp_64 << 4;
3366
3367         temp_64 = be64_to_cpu(rsp.excessive_buffer_overruns);
3368         if (temp_64 > 0xFUL)
3369                 temp_64 = 0xFUL;
3370         temp_link_overrun_errors |= temp_64;
3371
3372         p->link_overrun_errors = (u8)temp_link_overrun_errors;
3373
3374         p->vl15_dropped = 0; /* N/A for OPA */
3375
3376 bail:
3377         return reply((struct ib_mad_hdr *)pmp);
3378 }
3379
3380 static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
3381                                  struct ib_device *ibdev,
3382                                  u8 port, u32 *resp_len)
3383 {
3384         size_t response_data_size;
3385         struct _port_ei *rsp;
3386         struct opa_port_error_info_msg *req;
3387         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3388         u64 port_mask;
3389         u32 num_ports;
3390         u8 port_num;
3391         u8 num_pslm;
3392         u64 reg;
3393
3394         req = (struct opa_port_error_info_msg *)pmp->data;
3395         rsp = &req->port[0];
3396
3397         num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3398         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3399
3400         memset(rsp, 0, sizeof(*rsp));
3401
3402         if (num_ports != 1 || num_ports != num_pslm) {
3403                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3404                 return reply((struct ib_mad_hdr *)pmp);
3405         }
3406
3407         /* Sanity check */
3408         response_data_size = sizeof(struct opa_port_error_info_msg);
3409
3410         if (response_data_size > sizeof(pmp->data)) {
3411                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3412                 return reply((struct ib_mad_hdr *)pmp);
3413         }
3414
3415         /*
3416          * The bit set in the mask needs to be consistent with the port
3417          * the request came in on.
3418          */
3419         port_mask = be64_to_cpu(req->port_select_mask[3]);
3420         port_num = find_first_bit((unsigned long *)&port_mask,
3421                                   sizeof(port_mask) * 8);
3422
3423         if (port_num != port) {
3424                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3425                 return reply((struct ib_mad_hdr *)pmp);
3426         }
3427         rsp->port_number = port;
3428
3429         /* PortRcvErrorInfo */
3430         rsp->port_rcv_ei.status_and_code =
3431                 dd->err_info_rcvport.status_and_code;
3432         memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit1,
3433                &dd->err_info_rcvport.packet_flit1, sizeof(u64));
3434         memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit2,
3435                &dd->err_info_rcvport.packet_flit2, sizeof(u64));
3436
3437         /* ExcessiverBufferOverrunInfo */
3438         reg = read_csr(dd, RCV_ERR_INFO);
3439         if (reg & RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK) {
3440                 /*
3441                  * if the RcvExcessBufferOverrun bit is set, save SC of
3442                  * first pkt that encountered an excess buffer overrun
3443                  */
3444                 u8 tmp = (u8)reg;
3445
3446                 tmp &=  RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SC_SMASK;
3447                 tmp <<= 2;
3448                 rsp->excessive_buffer_overrun_ei.status_and_sc = tmp;
3449                 /* set the status bit */
3450                 rsp->excessive_buffer_overrun_ei.status_and_sc |= 0x80;
3451         }
3452
3453         rsp->port_xmit_constraint_ei.status =
3454                 dd->err_info_xmit_constraint.status;
3455         rsp->port_xmit_constraint_ei.pkey =
3456                 cpu_to_be16(dd->err_info_xmit_constraint.pkey);
3457         rsp->port_xmit_constraint_ei.slid =
3458                 cpu_to_be32(dd->err_info_xmit_constraint.slid);
3459
3460         rsp->port_rcv_constraint_ei.status =
3461                 dd->err_info_rcv_constraint.status;
3462         rsp->port_rcv_constraint_ei.pkey =
3463                 cpu_to_be16(dd->err_info_rcv_constraint.pkey);
3464         rsp->port_rcv_constraint_ei.slid =
3465                 cpu_to_be32(dd->err_info_rcv_constraint.slid);
3466
3467         /* UncorrectableErrorInfo */
3468         rsp->uncorrectable_ei.status_and_code = dd->err_info_uncorrectable;
3469
3470         /* FMConfigErrorInfo */
3471         rsp->fm_config_ei.status_and_code = dd->err_info_fmconfig;
3472
3473         if (resp_len)
3474                 *resp_len += response_data_size;
3475
3476         return reply((struct ib_mad_hdr *)pmp);
3477 }
3478
3479 static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
3480                                   struct ib_device *ibdev,
3481                                   u8 port, u32 *resp_len)
3482 {
3483         struct opa_clear_port_status *req =
3484                 (struct opa_clear_port_status *)pmp->data;
3485         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3486         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3487         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3488         u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3489         u64 portn = be64_to_cpu(req->port_select_mask[3]);
3490         u32 counter_select = be32_to_cpu(req->counter_select_mask);
3491         u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
3492         unsigned long vl;
3493
3494         if ((nports != 1) || (portn != 1 << port)) {
3495                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3496                 return reply((struct ib_mad_hdr *)pmp);
3497         }
3498         /*
3499          * only counters returned by pma_get_opa_portstatus() are
3500          * handled, so when pma_get_opa_portstatus() gets a fix,
3501          * the corresponding change should be made here as well.
3502          */
3503
3504         if (counter_select & CS_PORT_XMIT_DATA)
3505                 write_dev_cntr(dd, C_DC_XMIT_FLITS, CNTR_INVALID_VL, 0);
3506
3507         if (counter_select & CS_PORT_RCV_DATA)
3508                 write_dev_cntr(dd, C_DC_RCV_FLITS, CNTR_INVALID_VL, 0);
3509
3510         if (counter_select & CS_PORT_XMIT_PKTS)
3511                 write_dev_cntr(dd, C_DC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3512
3513         if (counter_select & CS_PORT_RCV_PKTS)
3514                 write_dev_cntr(dd, C_DC_RCV_PKTS, CNTR_INVALID_VL, 0);
3515
3516         if (counter_select & CS_PORT_MCAST_XMIT_PKTS)
3517                 write_dev_cntr(dd, C_DC_MC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3518
3519         if (counter_select & CS_PORT_MCAST_RCV_PKTS)
3520                 write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
3521
3522         if (counter_select & CS_PORT_XMIT_WAIT) {
3523                 write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
3524                 ppd->port_vl_xmit_wait_last[C_VL_COUNT] = 0;
3525                 ppd->vl_xmit_flit_cnt[C_VL_COUNT] = 0;
3526         }
3527         /* ignore cs_sw_portCongestion for HFIs */
3528
3529         if (counter_select & CS_PORT_RCV_FECN)
3530                 write_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL, 0);
3531
3532         if (counter_select & CS_PORT_RCV_BECN)
3533                 write_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL, 0);
3534
3535         /* ignore cs_port_xmit_time_cong for HFIs */
3536         /* ignore cs_port_xmit_wasted_bw for now */
3537         /* ignore cs_port_xmit_wait_data for now */
3538         if (counter_select & CS_PORT_RCV_BUBBLE)
3539                 write_dev_cntr(dd, C_DC_RCV_BBL, CNTR_INVALID_VL, 0);
3540
3541         /* Only applicable for switch */
3542         /* if (counter_select & CS_PORT_MARK_FECN)
3543          *      write_csr(dd, DCC_PRF_PORT_MARK_FECN_CNT, 0);
3544          */
3545
3546         if (counter_select & CS_PORT_RCV_CONSTRAINT_ERRORS)
3547                 write_port_cntr(ppd, C_SW_RCV_CSTR_ERR, CNTR_INVALID_VL, 0);
3548
3549         /* ignore cs_port_rcv_switch_relay_errors for HFIs */
3550         if (counter_select & CS_PORT_XMIT_DISCARDS)
3551                 write_port_cntr(ppd, C_SW_XMIT_DSCD, CNTR_INVALID_VL, 0);
3552
3553         if (counter_select & CS_PORT_XMIT_CONSTRAINT_ERRORS)
3554                 write_port_cntr(ppd, C_SW_XMIT_CSTR_ERR, CNTR_INVALID_VL, 0);
3555
3556         if (counter_select & CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS)
3557                 write_dev_cntr(dd, C_DC_RMT_PHY_ERR, CNTR_INVALID_VL, 0);
3558
3559         if (counter_select & CS_LOCAL_LINK_INTEGRITY_ERRORS)
3560                 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3561
3562         if (counter_select & CS_LINK_ERROR_RECOVERY) {
3563                 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3564                 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
3565                                CNTR_INVALID_VL, 0);
3566         }
3567
3568         if (counter_select & CS_PORT_RCV_ERRORS)
3569                 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3570
3571         if (counter_select & CS_EXCESSIVE_BUFFER_OVERRUNS) {
3572                 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3573                 dd->rcv_ovfl_cnt = 0;
3574         }
3575
3576         if (counter_select & CS_FM_CONFIG_ERRORS)
3577                 write_dev_cntr(dd, C_DC_FM_CFG_ERR, CNTR_INVALID_VL, 0);
3578
3579         if (counter_select & CS_LINK_DOWNED)
3580                 write_port_cntr(ppd, C_SW_LINK_DOWN, CNTR_INVALID_VL, 0);
3581
3582         if (counter_select & CS_UNCORRECTABLE_ERRORS)
3583                 write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
3584
3585         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3586                          8 * sizeof(vl_select_mask)) {
3587                 if (counter_select & CS_PORT_XMIT_DATA)
3588                         write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);
3589
3590                 if (counter_select & CS_PORT_RCV_DATA)
3591                         write_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl), 0);
3592
3593                 if (counter_select & CS_PORT_XMIT_PKTS)
3594                         write_port_cntr(ppd, C_TX_PKT_VL, idx_from_vl(vl), 0);
3595
3596                 if (counter_select & CS_PORT_RCV_PKTS)
3597                         write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
3598
3599                 if (counter_select & CS_PORT_XMIT_WAIT) {
3600                         write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
3601                         ppd->port_vl_xmit_wait_last[idx_from_vl(vl)] = 0;
3602                         ppd->vl_xmit_flit_cnt[idx_from_vl(vl)] = 0;
3603                 }
3604
3605                 /* sw_port_vl_congestion is 0 for HFIs */
3606                 if (counter_select & CS_PORT_RCV_FECN)
3607                         write_dev_cntr(dd, C_DC_RCV_FCN_VL, idx_from_vl(vl), 0);
3608
3609                 if (counter_select & CS_PORT_RCV_BECN)
3610                         write_dev_cntr(dd, C_DC_RCV_BCN_VL, idx_from_vl(vl), 0);
3611
3612                 /* port_vl_xmit_time_cong is 0 for HFIs */
3613                 /* port_vl_xmit_wasted_bw ??? */
3614                 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ??? */
3615                 if (counter_select & CS_PORT_RCV_BUBBLE)
3616                         write_dev_cntr(dd, C_DC_RCV_BBL_VL, idx_from_vl(vl), 0);
3617
3618                 /* if (counter_select & CS_PORT_MARK_FECN)
3619                  *     write_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT + offset, 0);
3620                  */
3621                 if (counter_select & C_SW_XMIT_DSCD_VL)
3622                         write_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3623                                         idx_from_vl(vl), 0);
3624         }
3625
3626         if (resp_len)
3627                 *resp_len += sizeof(*req);
3628
3629         return reply((struct ib_mad_hdr *)pmp);
3630 }
3631
3632 static int pma_set_opa_errorinfo(struct opa_pma_mad *pmp,
3633                                  struct ib_device *ibdev,
3634                                  u8 port, u32 *resp_len)
3635 {
3636         struct _port_ei *rsp;
3637         struct opa_port_error_info_msg *req;
3638         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3639         u64 port_mask;
3640         u32 num_ports;
3641         u8 port_num;
3642         u8 num_pslm;
3643         u32 error_info_select;
3644
3645         req = (struct opa_port_error_info_msg *)pmp->data;
3646         rsp = &req->port[0];
3647
3648         num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3649         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3650
3651         memset(rsp, 0, sizeof(*rsp));
3652
3653         if (num_ports != 1 || num_ports != num_pslm) {
3654                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3655                 return reply((struct ib_mad_hdr *)pmp);
3656         }
3657
3658         /*
3659          * The bit set in the mask needs to be consistent with the port
3660          * the request came in on.
3661          */
3662         port_mask = be64_to_cpu(req->port_select_mask[3]);
3663         port_num = find_first_bit((unsigned long *)&port_mask,
3664                                   sizeof(port_mask) * 8);
3665
3666         if (port_num != port) {
3667                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3668                 return reply((struct ib_mad_hdr *)pmp);
3669         }
3670
3671         error_info_select = be32_to_cpu(req->error_info_select_mask);
3672
3673         /* PortRcvErrorInfo */
3674         if (error_info_select & ES_PORT_RCV_ERROR_INFO)
3675                 /* turn off status bit */
3676                 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3677
3678         /* ExcessiverBufferOverrunInfo */
3679         if (error_info_select & ES_EXCESSIVE_BUFFER_OVERRUN_INFO)
3680                 /*
3681                  * status bit is essentially kept in the h/w - bit 5 of
3682                  * RCV_ERR_INFO
3683                  */
3684                 write_csr(dd, RCV_ERR_INFO,
3685                           RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK);
3686
3687         if (error_info_select & ES_PORT_XMIT_CONSTRAINT_ERROR_INFO)
3688                 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3689
3690         if (error_info_select & ES_PORT_RCV_CONSTRAINT_ERROR_INFO)
3691                 dd->err_info_rcv_constraint.status &= ~OPA_EI_STATUS_SMASK;
3692
3693         /* UncorrectableErrorInfo */
3694         if (error_info_select & ES_UNCORRECTABLE_ERROR_INFO)
3695                 /* turn off status bit */
3696                 dd->err_info_uncorrectable &= ~OPA_EI_STATUS_SMASK;
3697
3698         /* FMConfigErrorInfo */
3699         if (error_info_select & ES_FM_CONFIG_ERROR_INFO)
3700                 /* turn off status bit */
3701                 dd->err_info_fmconfig &= ~OPA_EI_STATUS_SMASK;
3702
3703         if (resp_len)
3704                 *resp_len += sizeof(*req);
3705
3706         return reply((struct ib_mad_hdr *)pmp);
3707 }
3708
3709 struct opa_congestion_info_attr {
3710         __be16 congestion_info;
3711         u8 control_table_cap;   /* Multiple of 64 entry unit CCTs */
3712         u8 congestion_log_length;
3713 } __packed;
3714
3715 static int __subn_get_opa_cong_info(struct opa_smp *smp, u32 am, u8 *data,
3716                                     struct ib_device *ibdev, u8 port,
3717                                     u32 *resp_len, u32 max_len)
3718 {
3719         struct opa_congestion_info_attr *p =
3720                 (struct opa_congestion_info_attr *)data;
3721         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3722         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3723
3724         if (smp_length_check(sizeof(*p), max_len)) {
3725                 smp->status |= IB_SMP_INVALID_FIELD;
3726                 return reply((struct ib_mad_hdr *)smp);
3727         }
3728
3729         p->congestion_info = 0;
3730         p->control_table_cap = ppd->cc_max_table_entries;
3731         p->congestion_log_length = OPA_CONG_LOG_ELEMS;
3732
3733         if (resp_len)
3734                 *resp_len += sizeof(*p);
3735
3736         return reply((struct ib_mad_hdr *)smp);
3737 }
3738
3739 static int __subn_get_opa_cong_setting(struct opa_smp *smp, u32 am,
3740                                        u8 *data, struct ib_device *ibdev,
3741                                        u8 port, u32 *resp_len, u32 max_len)
3742 {
3743         int i;
3744         struct opa_congestion_setting_attr *p =
3745                 (struct opa_congestion_setting_attr *)data;
3746         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3747         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3748         struct opa_congestion_setting_entry_shadow *entries;
3749         struct cc_state *cc_state;
3750
3751         if (smp_length_check(sizeof(*p), max_len)) {
3752                 smp->status |= IB_SMP_INVALID_FIELD;
3753                 return reply((struct ib_mad_hdr *)smp);
3754         }
3755
3756         rcu_read_lock();
3757
3758         cc_state = get_cc_state(ppd);
3759
3760         if (!cc_state) {
3761                 rcu_read_unlock();
3762                 return reply((struct ib_mad_hdr *)smp);
3763         }
3764
3765         entries = cc_state->cong_setting.entries;
3766         p->port_control = cpu_to_be16(cc_state->cong_setting.port_control);
3767         p->control_map = cpu_to_be32(cc_state->cong_setting.control_map);
3768         for (i = 0; i < OPA_MAX_SLS; i++) {
3769                 p->entries[i].ccti_increase = entries[i].ccti_increase;
3770                 p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
3771                 p->entries[i].trigger_threshold =
3772                         entries[i].trigger_threshold;
3773                 p->entries[i].ccti_min = entries[i].ccti_min;
3774         }
3775
3776         rcu_read_unlock();
3777
3778         if (resp_len)
3779                 *resp_len += sizeof(*p);
3780
3781         return reply((struct ib_mad_hdr *)smp);
3782 }
3783
3784 /*
3785  * Apply congestion control information stored in the ppd to the
3786  * active structure.
3787  */
3788 static void apply_cc_state(struct hfi1_pportdata *ppd)
3789 {
3790         struct cc_state *old_cc_state, *new_cc_state;
3791
3792         new_cc_state = kzalloc(sizeof(*new_cc_state), GFP_KERNEL);
3793         if (!new_cc_state)
3794                 return;
3795
3796         /*
3797          * Hold the lock for updating *and* to prevent ppd information
3798          * from changing during the update.
3799          */
3800         spin_lock(&ppd->cc_state_lock);
3801
3802         old_cc_state = get_cc_state_protected(ppd);
3803         if (!old_cc_state) {
3804                 /* never active, or shutting down */
3805                 spin_unlock(&ppd->cc_state_lock);
3806                 kfree(new_cc_state);
3807                 return;
3808         }
3809
3810         *new_cc_state = *old_cc_state;
3811
3812         if (ppd->total_cct_entry)
3813                 new_cc_state->cct.ccti_limit = ppd->total_cct_entry - 1;
3814         else
3815                 new_cc_state->cct.ccti_limit = 0;
3816
3817         memcpy(new_cc_state->cct.entries, ppd->ccti_entries,
3818                ppd->total_cct_entry * sizeof(struct ib_cc_table_entry));
3819
3820         new_cc_state->cong_setting.port_control = IB_CC_CCS_PC_SL_BASED;
3821         new_cc_state->cong_setting.control_map = ppd->cc_sl_control_map;
3822         memcpy(new_cc_state->cong_setting.entries, ppd->congestion_entries,
3823                OPA_MAX_SLS * sizeof(struct opa_congestion_setting_entry));
3824
3825         rcu_assign_pointer(ppd->cc_state, new_cc_state);
3826
3827         spin_unlock(&ppd->cc_state_lock);
3828
3829         kfree_rcu(old_cc_state, rcu);
3830 }
3831
3832 static int __subn_set_opa_cong_setting(struct opa_smp *smp, u32 am, u8 *data,
3833                                        struct ib_device *ibdev, u8 port,
3834                                        u32 *resp_len, u32 max_len)
3835 {
3836         struct opa_congestion_setting_attr *p =
3837                 (struct opa_congestion_setting_attr *)data;
3838         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3839         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3840         struct opa_congestion_setting_entry_shadow *entries;
3841         int i;
3842
3843         if (smp_length_check(sizeof(*p), max_len)) {
3844                 smp->status |= IB_SMP_INVALID_FIELD;
3845                 return reply((struct ib_mad_hdr *)smp);
3846         }
3847
3848         /*
3849          * Save details from packet into the ppd.  Hold the cc_state_lock so
3850          * our information is consistent with anyone trying to apply the state.
3851          */
3852         spin_lock(&ppd->cc_state_lock);
3853         ppd->cc_sl_control_map = be32_to_cpu(p->control_map);
3854
3855         entries = ppd->congestion_entries;
3856         for (i = 0; i < OPA_MAX_SLS; i++) {
3857                 entries[i].ccti_increase = p->entries[i].ccti_increase;
3858                 entries[i].ccti_timer = be16_to_cpu(p->entries[i].ccti_timer);
3859                 entries[i].trigger_threshold =
3860                         p->entries[i].trigger_threshold;
3861                 entries[i].ccti_min = p->entries[i].ccti_min;
3862         }
3863         spin_unlock(&ppd->cc_state_lock);
3864
3865         /* now apply the information */
3866         apply_cc_state(ppd);
3867
3868         return __subn_get_opa_cong_setting(smp, am, data, ibdev, port,
3869                                            resp_len, max_len);
3870 }
3871
3872 static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
3873                                         u8 *data, struct ib_device *ibdev,
3874                                         u8 port, u32 *resp_len, u32 max_len)
3875 {
3876         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3877         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3878         struct opa_hfi1_cong_log *cong_log = (struct opa_hfi1_cong_log *)data;
3879         u64 ts;
3880         int i;
3881
3882         if (am || smp_length_check(sizeof(*cong_log), max_len)) {
3883                 smp->status |= IB_SMP_INVALID_FIELD;
3884                 return reply((struct ib_mad_hdr *)smp);
3885         }
3886
3887         spin_lock_irq(&ppd->cc_log_lock);
3888
3889         cong_log->log_type = OPA_CC_LOG_TYPE_HFI;
3890         cong_log->congestion_flags = 0;
3891         cong_log->threshold_event_counter =
3892                 cpu_to_be16(ppd->threshold_event_counter);
3893         memcpy(cong_log->threshold_cong_event_map,
3894                ppd->threshold_cong_event_map,
3895                sizeof(cong_log->threshold_cong_event_map));
3896         /* keep timestamp in units of 1.024 usec */
3897         ts = ktime_get_ns() / 1024;
3898         cong_log->current_time_stamp = cpu_to_be32(ts);
3899         for (i = 0; i < OPA_CONG_LOG_ELEMS; i++) {
3900                 struct opa_hfi1_cong_log_event_internal *cce =
3901                         &ppd->cc_events[ppd->cc_mad_idx++];
3902                 if (ppd->cc_mad_idx == OPA_CONG_LOG_ELEMS)
3903                         ppd->cc_mad_idx = 0;
3904                 /*
3905                  * Entries which are older than twice the time
3906                  * required to wrap the counter are supposed to
3907                  * be zeroed (CA10-49 IBTA, release 1.2.1, V1).
3908                  */
3909                 if ((ts - cce->timestamp) / 2 > U32_MAX)
3910                         continue;
3911                 memcpy(cong_log->events[i].local_qp_cn_entry, &cce->lqpn, 3);
3912                 memcpy(cong_log->events[i].remote_qp_number_cn_entry,
3913                        &cce->rqpn, 3);
3914                 cong_log->events[i].sl_svc_type_cn_entry =
3915                         ((cce->sl & 0x1f) << 3) | (cce->svc_type & 0x7);
3916                 cong_log->events[i].remote_lid_cn_entry =
3917                         cpu_to_be32(cce->rlid);
3918                 cong_log->events[i].timestamp_cn_entry =
3919                         cpu_to_be32(cce->timestamp);
3920         }
3921
3922         /*
3923          * Reset threshold_cong_event_map, and threshold_event_counter
3924          * to 0 when log is read.
3925          */
3926         memset(ppd->threshold_cong_event_map, 0x0,
3927                sizeof(ppd->threshold_cong_event_map));
3928         ppd->threshold_event_counter = 0;
3929
3930         spin_unlock_irq(&ppd->cc_log_lock);
3931
3932         if (resp_len)
3933                 *resp_len += sizeof(struct opa_hfi1_cong_log);
3934
3935         return reply((struct ib_mad_hdr *)smp);
3936 }
3937
3938 static int __subn_get_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3939                                    struct ib_device *ibdev, u8 port,
3940                                    u32 *resp_len, u32 max_len)
3941 {
3942         struct ib_cc_table_attr *cc_table_attr =
3943                 (struct ib_cc_table_attr *)data;
3944         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3945         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3946         u32 start_block = OPA_AM_START_BLK(am);
3947         u32 n_blocks = OPA_AM_NBLK(am);
3948         struct ib_cc_table_entry_shadow *entries;
3949         int i, j;
3950         u32 sentry, eentry;
3951         struct cc_state *cc_state;
3952         u32 size = sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
3953
3954         /* sanity check n_blocks, start_block */
3955         if (n_blocks == 0 || smp_length_check(size, max_len) ||
3956             start_block + n_blocks > ppd->cc_max_table_entries) {
3957                 smp->status |= IB_SMP_INVALID_FIELD;
3958                 return reply((struct ib_mad_hdr *)smp);
3959         }
3960
3961         rcu_read_lock();
3962
3963         cc_state = get_cc_state(ppd);
3964
3965         if (!cc_state) {
3966                 rcu_read_unlock();
3967                 return reply((struct ib_mad_hdr *)smp);
3968         }
3969
3970         sentry = start_block * IB_CCT_ENTRIES;
3971         eentry = sentry + (IB_CCT_ENTRIES * n_blocks);
3972
3973         cc_table_attr->ccti_limit = cpu_to_be16(cc_state->cct.ccti_limit);
3974
3975         entries = cc_state->cct.entries;
3976
3977         /* return n_blocks, though the last block may not be full */
3978         for (j = 0, i = sentry; i < eentry; j++, i++)
3979                 cc_table_attr->ccti_entries[j].entry =
3980                         cpu_to_be16(entries[i].entry);
3981
3982         rcu_read_unlock();
3983
3984         if (resp_len)
3985                 *resp_len += size;
3986
3987         return reply((struct ib_mad_hdr *)smp);
3988 }
3989
3990 static int __subn_set_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3991                                    struct ib_device *ibdev, u8 port,
3992                                    u32 *resp_len, u32 max_len)
3993 {
3994         struct ib_cc_table_attr *p = (struct ib_cc_table_attr *)data;
3995         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3996         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3997         u32 start_block = OPA_AM_START_BLK(am);
3998         u32 n_blocks = OPA_AM_NBLK(am);
3999         struct ib_cc_table_entry_shadow *entries;
4000         int i, j;
4001         u32 sentry, eentry;
4002         u16 ccti_limit;
4003         u32 size = sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
4004
4005         /* sanity check n_blocks, start_block */
4006         if (n_blocks == 0 || smp_length_check(size, max_len) ||
4007             start_block + n_blocks > ppd->cc_max_table_entries) {
4008                 smp->status |= IB_SMP_INVALID_FIELD;
4009                 return reply((struct ib_mad_hdr *)smp);
4010         }
4011
4012         sentry = start_block * IB_CCT_ENTRIES;
4013         eentry = sentry + ((n_blocks - 1) * IB_CCT_ENTRIES) +
4014                  (be16_to_cpu(p->ccti_limit)) % IB_CCT_ENTRIES + 1;
4015
4016         /* sanity check ccti_limit */
4017         ccti_limit = be16_to_cpu(p->ccti_limit);
4018         if (ccti_limit + 1 > eentry) {
4019                 smp->status |= IB_SMP_INVALID_FIELD;
4020                 return reply((struct ib_mad_hdr *)smp);
4021         }
4022
4023         /*
4024          * Save details from packet into the ppd.  Hold the cc_state_lock so
4025          * our information is consistent with anyone trying to apply the state.
4026          */
4027         spin_lock(&ppd->cc_state_lock);
4028         ppd->total_cct_entry = ccti_limit + 1;
4029         entries = ppd->ccti_entries;
4030         for (j = 0, i = sentry; i < eentry; j++, i++)
4031                 entries[i].entry = be16_to_cpu(p->ccti_entries[j].entry);
4032         spin_unlock(&ppd->cc_state_lock);
4033
4034         /* now apply the information */
4035         apply_cc_state(ppd);
4036
4037         return __subn_get_opa_cc_table(smp, am, data, ibdev, port, resp_len,
4038                                        max_len);
4039 }
4040
4041 struct opa_led_info {
4042         __be32 rsvd_led_mask;
4043         __be32 rsvd;
4044 };
4045
4046 #define OPA_LED_SHIFT   31
4047 #define OPA_LED_MASK    BIT(OPA_LED_SHIFT)
4048
4049 static int __subn_get_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
4050                                    struct ib_device *ibdev, u8 port,
4051                                    u32 *resp_len, u32 max_len)
4052 {
4053         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
4054         struct hfi1_pportdata *ppd = dd->pport;
4055         struct opa_led_info *p = (struct opa_led_info *)data;
4056         u32 nport = OPA_AM_NPORT(am);
4057         u32 is_beaconing_active;
4058
4059         if (nport != 1 || smp_length_check(sizeof(*p), max_len)) {
4060                 smp->status |= IB_SMP_INVALID_FIELD;
4061                 return reply((struct ib_mad_hdr *)smp);
4062         }
4063
4064         /*
4065          * This pairs with the memory barrier in hfi1_start_led_override to
4066          * ensure that we read the correct state of LED beaconing represented
4067          * by led_override_timer_active
4068          */
4069         smp_rmb();
4070         is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
4071         p->rsvd_led_mask = cpu_to_be32(is_beaconing_active << OPA_LED_SHIFT);
4072
4073         if (resp_len)
4074                 *resp_len += sizeof(struct opa_led_info);
4075
4076         return reply((struct ib_mad_hdr *)smp);
4077 }
4078
4079 static int __subn_set_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
4080                                    struct ib_device *ibdev, u8 port,
4081                                    u32 *resp_len, u32 max_len)
4082 {
4083         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
4084         struct opa_led_info *p = (struct opa_led_info *)data;
4085         u32 nport = OPA_AM_NPORT(am);
4086         int on = !!(be32_to_cpu(p->rsvd_led_mask) & OPA_LED_MASK);
4087
4088         if (nport != 1 || smp_length_check(sizeof(*p), max_len)) {
4089                 smp->status |= IB_SMP_INVALID_FIELD;
4090                 return reply((struct ib_mad_hdr *)smp);
4091         }
4092
4093         if (on)
4094                 hfi1_start_led_override(dd->pport, 2000, 1500);
4095         else
4096                 shutdown_led_override(dd->pport);
4097
4098         return __subn_get_opa_led_info(smp, am, data, ibdev, port, resp_len,
4099                                        max_len);
4100 }
4101
4102 static int subn_get_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
4103                             u8 *data, struct ib_device *ibdev, u8 port,
4104                             u32 *resp_len, u32 max_len)
4105 {
4106         int ret;
4107         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4108
4109         switch (attr_id) {
4110         case IB_SMP_ATTR_NODE_DESC:
4111                 ret = __subn_get_opa_nodedesc(smp, am, data, ibdev, port,
4112                                               resp_len, max_len);
4113                 break;
4114         case IB_SMP_ATTR_NODE_INFO:
4115                 ret = __subn_get_opa_nodeinfo(smp, am, data, ibdev, port,
4116                                               resp_len, max_len);
4117                 break;
4118         case IB_SMP_ATTR_PORT_INFO:
4119                 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port,
4120                                               resp_len, max_len);
4121                 break;
4122         case IB_SMP_ATTR_PKEY_TABLE:
4123                 ret = __subn_get_opa_pkeytable(smp, am, data, ibdev, port,
4124                                                resp_len, max_len);
4125                 break;
4126         case OPA_ATTRIB_ID_SL_TO_SC_MAP:
4127                 ret = __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port,
4128                                               resp_len, max_len);
4129                 break;
4130         case OPA_ATTRIB_ID_SC_TO_SL_MAP:
4131                 ret = __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port,
4132                                               resp_len, max_len);
4133                 break;
4134         case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
4135                 ret = __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port,
4136                                                resp_len, max_len);
4137                 break;
4138         case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
4139                 ret = __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
4140                                                 resp_len, max_len);
4141                 break;
4142         case OPA_ATTRIB_ID_PORT_STATE_INFO:
4143                 ret = __subn_get_opa_psi(smp, am, data, ibdev, port,
4144                                          resp_len, max_len);
4145                 break;
4146         case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
4147                 ret = __subn_get_opa_bct(smp, am, data, ibdev, port,
4148                                          resp_len, max_len);
4149                 break;
4150         case OPA_ATTRIB_ID_CABLE_INFO:
4151                 ret = __subn_get_opa_cable_info(smp, am, data, ibdev, port,
4152                                                 resp_len, max_len);
4153                 break;
4154         case IB_SMP_ATTR_VL_ARB_TABLE:
4155                 ret = __subn_get_opa_vl_arb(smp, am, data, ibdev, port,
4156                                             resp_len, max_len);
4157                 break;
4158         case OPA_ATTRIB_ID_CONGESTION_INFO:
4159                 ret = __subn_get_opa_cong_info(smp, am, data, ibdev, port,
4160                                                resp_len, max_len);
4161                 break;
4162         case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
4163                 ret = __subn_get_opa_cong_setting(smp, am, data, ibdev,
4164                                                   port, resp_len, max_len);
4165                 break;
4166         case OPA_ATTRIB_ID_HFI_CONGESTION_LOG:
4167                 ret = __subn_get_opa_hfi1_cong_log(smp, am, data, ibdev,
4168                                                    port, resp_len, max_len);
4169                 break;
4170         case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
4171                 ret = __subn_get_opa_cc_table(smp, am, data, ibdev, port,
4172                                               resp_len, max_len);
4173                 break;
4174         case IB_SMP_ATTR_LED_INFO:
4175                 ret = __subn_get_opa_led_info(smp, am, data, ibdev, port,
4176                                               resp_len, max_len);
4177                 break;
4178         case IB_SMP_ATTR_SM_INFO:
4179                 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
4180                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4181                 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
4182                         return IB_MAD_RESULT_SUCCESS;
4183                 /* FALLTHROUGH */
4184         default:
4185                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4186                 ret = reply((struct ib_mad_hdr *)smp);
4187                 break;
4188         }
4189         return ret;
4190 }
4191
4192 static int subn_set_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
4193                             u8 *data, struct ib_device *ibdev, u8 port,
4194                             u32 *resp_len, u32 max_len, int local_mad)
4195 {
4196         int ret;
4197         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4198
4199         switch (attr_id) {
4200         case IB_SMP_ATTR_PORT_INFO:
4201                 ret = __subn_set_opa_portinfo(smp, am, data, ibdev, port,
4202                                               resp_len, max_len, local_mad);
4203                 break;
4204         case IB_SMP_ATTR_PKEY_TABLE:
4205                 ret = __subn_set_opa_pkeytable(smp, am, data, ibdev, port,
4206                                                resp_len, max_len);
4207                 break;
4208         case OPA_ATTRIB_ID_SL_TO_SC_MAP:
4209                 ret = __subn_set_opa_sl_to_sc(smp, am, data, ibdev, port,
4210                                               resp_len, max_len);
4211                 break;
4212         case OPA_ATTRIB_ID_SC_TO_SL_MAP:
4213                 ret = __subn_set_opa_sc_to_sl(smp, am, data, ibdev, port,
4214                                               resp_len, max_len);
4215                 break;
4216         case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
4217                 ret = __subn_set_opa_sc_to_vlt(smp, am, data, ibdev, port,
4218                                                resp_len, max_len);
4219                 break;
4220         case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
4221                 ret = __subn_set_opa_sc_to_vlnt(smp, am, data, ibdev, port,
4222                                                 resp_len, max_len);
4223                 break;
4224         case OPA_ATTRIB_ID_PORT_STATE_INFO:
4225                 ret = __subn_set_opa_psi(smp, am, data, ibdev, port,
4226                                          resp_len, max_len, local_mad);
4227                 break;
4228         case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
4229                 ret = __subn_set_opa_bct(smp, am, data, ibdev, port,
4230                                          resp_len, max_len);
4231                 break;
4232         case IB_SMP_ATTR_VL_ARB_TABLE:
4233                 ret = __subn_set_opa_vl_arb(smp, am, data, ibdev, port,
4234                                             resp_len, max_len);
4235                 break;
4236         case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
4237                 ret = __subn_set_opa_cong_setting(smp, am, data, ibdev,
4238                                                   port, resp_len, max_len);
4239                 break;
4240         case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
4241                 ret = __subn_set_opa_cc_table(smp, am, data, ibdev, port,
4242                                               resp_len, max_len);
4243                 break;
4244         case IB_SMP_ATTR_LED_INFO:
4245                 ret = __subn_set_opa_led_info(smp, am, data, ibdev, port,
4246                                               resp_len, max_len);
4247                 break;
4248         case IB_SMP_ATTR_SM_INFO:
4249                 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
4250                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
4251                 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
4252                         return IB_MAD_RESULT_SUCCESS;
4253                 /* FALLTHROUGH */
4254         default:
4255                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
4256                 ret = reply((struct ib_mad_hdr *)smp);
4257                 break;
4258         }
4259         return ret;
4260 }
4261
4262 static inline void set_aggr_error(struct opa_aggregate *ag)
4263 {
4264         ag->err_reqlength |= cpu_to_be16(0x8000);
4265 }
4266
4267 static int subn_get_opa_aggregate(struct opa_smp *smp,
4268                                   struct ib_device *ibdev, u8 port,
4269                                   u32 *resp_len)
4270 {
4271         int i;
4272         u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
4273         u8 *next_smp = opa_get_smp_data(smp);
4274
4275         if (num_attr < 1 || num_attr > 117) {
4276                 smp->status |= IB_SMP_INVALID_FIELD;
4277                 return reply((struct ib_mad_hdr *)smp);
4278         }
4279
4280         for (i = 0; i < num_attr; i++) {
4281                 struct opa_aggregate *agg;
4282                 size_t agg_data_len;
4283                 size_t agg_size;
4284                 u32 am;
4285
4286                 agg = (struct opa_aggregate *)next_smp;
4287                 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
4288                 agg_size = sizeof(*agg) + agg_data_len;
4289                 am = be32_to_cpu(agg->attr_mod);
4290
4291                 *resp_len += agg_size;
4292
4293                 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
4294                         smp->status |= IB_SMP_INVALID_FIELD;
4295                         return reply((struct ib_mad_hdr *)smp);
4296                 }
4297
4298                 /* zero the payload for this segment */
4299                 memset(next_smp + sizeof(*agg), 0, agg_data_len);
4300
4301                 (void)subn_get_opa_sma(agg->attr_id, smp, am, agg->data,
4302                                        ibdev, port, NULL, (u32)agg_data_len);
4303
4304                 if (smp->status & IB_SMP_INVALID_FIELD)
4305                         break;
4306                 if (smp->status & ~IB_SMP_DIRECTION) {
4307                         set_aggr_error(agg);
4308                         return reply((struct ib_mad_hdr *)smp);
4309                 }
4310                 next_smp += agg_size;
4311         }
4312
4313         return reply((struct ib_mad_hdr *)smp);
4314 }
4315
4316 static int subn_set_opa_aggregate(struct opa_smp *smp,
4317                                   struct ib_device *ibdev, u8 port,
4318                                   u32 *resp_len, int local_mad)
4319 {
4320         int i;
4321         u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
4322         u8 *next_smp = opa_get_smp_data(smp);
4323
4324         if (num_attr < 1 || num_attr > 117) {
4325                 smp->status |= IB_SMP_INVALID_FIELD;
4326                 return reply((struct ib_mad_hdr *)smp);
4327         }
4328
4329         for (i = 0; i < num_attr; i++) {
4330                 struct opa_aggregate *agg;
4331                 size_t agg_data_len;
4332                 size_t agg_size;
4333                 u32 am;
4334
4335                 agg = (struct opa_aggregate *)next_smp;
4336                 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
4337                 agg_size = sizeof(*agg) + agg_data_len;
4338                 am = be32_to_cpu(agg->attr_mod);
4339
4340                 *resp_len += agg_size;
4341
4342                 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
4343                         smp->status |= IB_SMP_INVALID_FIELD;
4344                         return reply((struct ib_mad_hdr *)smp);
4345                 }
4346
4347                 (void)subn_set_opa_sma(agg->attr_id, smp, am, agg->data,
4348                                        ibdev, port, NULL, (u32)agg_data_len,
4349                                        local_mad);
4350
4351                 if (smp->status & IB_SMP_INVALID_FIELD)
4352                         break;
4353                 if (smp->status & ~IB_SMP_DIRECTION) {
4354                         set_aggr_error(agg);
4355                         return reply((struct ib_mad_hdr *)smp);
4356                 }
4357                 next_smp += agg_size;
4358         }
4359
4360         return reply((struct ib_mad_hdr *)smp);
4361 }
4362
4363 /*
4364  * OPAv1 specifies that, on the transition to link up, these counters
4365  * are cleared:
4366  *   PortRcvErrors [*]
4367  *   LinkErrorRecovery
4368  *   LocalLinkIntegrityErrors
4369  *   ExcessiveBufferOverruns [*]
4370  *
4371  * [*] Error info associated with these counters is retained, but the
4372  * error info status is reset to 0.
4373  */
4374 void clear_linkup_counters(struct hfi1_devdata *dd)
4375 {
4376         /* PortRcvErrors */
4377         write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
4378         dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
4379         /* LinkErrorRecovery */
4380         write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
4381         write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL, 0);
4382         /* LocalLinkIntegrityErrors */
4383         write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
4384         /* ExcessiveBufferOverruns */
4385         write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
4386         dd->rcv_ovfl_cnt = 0;
4387         dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
4388 }
4389
4390 static int is_full_mgmt_pkey_in_table(struct hfi1_ibport *ibp)
4391 {
4392         unsigned int i;
4393         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
4394
4395         for (i = 0; i < ARRAY_SIZE(ppd->pkeys); ++i)
4396                 if (ppd->pkeys[i] == FULL_MGMT_P_KEY)
4397                         return 1;
4398
4399         return 0;
4400 }
4401
4402 /*
4403  * is_local_mad() returns 1 if 'mad' is sent from, and destined to the
4404  * local node, 0 otherwise.
4405  */
4406 static int is_local_mad(struct hfi1_ibport *ibp, const struct opa_mad *mad,
4407                         const struct ib_wc *in_wc)
4408 {
4409         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
4410         const struct opa_smp *smp = (const struct opa_smp *)mad;
4411
4412         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
4413                 return (smp->hop_cnt == 0 &&
4414                         smp->route.dr.dr_slid == OPA_LID_PERMISSIVE &&
4415                         smp->route.dr.dr_dlid == OPA_LID_PERMISSIVE);
4416         }
4417
4418         return (in_wc->slid == ppd->lid);
4419 }
4420
4421 /*
4422  * opa_local_smp_check() should only be called on MADs for which
4423  * is_local_mad() returns true. It applies the SMP checks that are
4424  * specific to SMPs which are sent from, and destined to this node.
4425  * opa_local_smp_check() returns 0 if the SMP passes its checks, 1
4426  * otherwise.
4427  *
4428  * SMPs which arrive from other nodes are instead checked by
4429  * opa_smp_check().
4430  */
4431 static int opa_local_smp_check(struct hfi1_ibport *ibp,
4432                                const struct ib_wc *in_wc)
4433 {
4434         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
4435         u16 pkey;
4436
4437         if (in_wc->pkey_index >= ARRAY_SIZE(ppd->pkeys))
4438                 return 1;
4439
4440         pkey = ppd->pkeys[in_wc->pkey_index];
4441         /*
4442          * We need to do the "node-local" checks specified in OPAv1,
4443          * rev 0.90, section 9.10.26, which are:
4444          *   - pkey is 0x7fff, or 0xffff
4445          *   - Source QPN == 0 || Destination QPN == 0
4446          *   - the MAD header's management class is either
4447          *     IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE or
4448          *     IB_MGMT_CLASS_SUBN_LID_ROUTED
4449          *   - SLID != 0
4450          *
4451          * However, we know (and so don't need to check again) that,
4452          * for local SMPs, the MAD stack passes MADs with:
4453          *   - Source QPN of 0
4454          *   - MAD mgmt_class is IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4455          *   - SLID is either: OPA_LID_PERMISSIVE (0xFFFFFFFF), or
4456          *     our own port's lid
4457          *
4458          */
4459         if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY)
4460                 return 0;
4461         ingress_pkey_table_fail(ppd, pkey, in_wc->slid);
4462         return 1;
4463 }
4464
4465 /**
4466  * hfi1_pkey_validation_pma - It validates PKEYs for incoming PMA MAD packets.
4467  * @ibp: IB port data
4468  * @in_mad: MAD packet with header and data
4469  * @in_wc: Work completion data such as source LID, port number, etc.
4470  *
4471  * These are all the possible logic rules for validating a pkey:
4472  *
4473  * a) If pkey neither FULL_MGMT_P_KEY nor LIM_MGMT_P_KEY,
4474  *    and NOT self-originated packet:
4475  *     Drop MAD packet as it should always be part of the
4476  *     management partition unless it's a self-originated packet.
4477  *
4478  * b) If pkey_index -> FULL_MGMT_P_KEY, and LIM_MGMT_P_KEY in pkey table:
4479  *     The packet is coming from a management node and the receiving node
4480  *     is also a management node, so it is safe for the packet to go through.
4481  *
4482  * c) If pkey_index -> FULL_MGMT_P_KEY, and LIM_MGMT_P_KEY is NOT in pkey table:
4483  *     Drop the packet as LIM_MGMT_P_KEY should always be in the pkey table.
4484  *     It could be an FM misconfiguration.
4485  *
4486  * d) If pkey_index -> LIM_MGMT_P_KEY and FULL_MGMT_P_KEY is NOT in pkey table:
4487  *     It is safe for the packet to go through since a non-management node is
4488  *     talking to another non-management node.
4489  *
4490  * e) If pkey_index -> LIM_MGMT_P_KEY and FULL_MGMT_P_KEY in pkey table:
4491  *     Drop the packet because a non-management node is talking to a
4492  *     management node, and it could be an attack.
4493  *
4494  * For the implementation, these rules can be simplied to only checking
4495  * for (a) and (e). There's no need to check for rule (b) as
4496  * the packet doesn't need to be dropped. Rule (c) is not possible in
4497  * the driver as LIM_MGMT_P_KEY is always in the pkey table.
4498  *
4499  * Return:
4500  * 0 - pkey is okay, -EINVAL it's a bad pkey
4501  */
4502 static int hfi1_pkey_validation_pma(struct hfi1_ibport *ibp,
4503                                     const struct opa_mad *in_mad,
4504                                     const struct ib_wc *in_wc)
4505 {
4506         u16 pkey_value = hfi1_lookup_pkey_value(ibp, in_wc->pkey_index);
4507
4508         /* Rule (a) from above */
4509         if (!is_local_mad(ibp, in_mad, in_wc) &&
4510             pkey_value != LIM_MGMT_P_KEY &&
4511             pkey_value != FULL_MGMT_P_KEY)
4512                 return -EINVAL;
4513
4514         /* Rule (e) from above */
4515         if (pkey_value == LIM_MGMT_P_KEY &&
4516             is_full_mgmt_pkey_in_table(ibp))
4517                 return -EINVAL;
4518
4519         return 0;
4520 }
4521
4522 static int process_subn_opa(struct ib_device *ibdev, int mad_flags,
4523                             u8 port, const struct opa_mad *in_mad,
4524                             struct opa_mad *out_mad,
4525                             u32 *resp_len, int local_mad)
4526 {
4527         struct opa_smp *smp = (struct opa_smp *)out_mad;
4528         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4529         u8 *data;
4530         u32 am, data_size;
4531         __be16 attr_id;
4532         int ret;
4533
4534         *out_mad = *in_mad;
4535         data = opa_get_smp_data(smp);
4536         data_size = (u32)opa_get_smp_data_size(smp);
4537
4538         am = be32_to_cpu(smp->attr_mod);
4539         attr_id = smp->attr_id;
4540         if (smp->class_version != OPA_SM_CLASS_VERSION) {
4541                 smp->status |= IB_SMP_UNSUP_VERSION;
4542                 ret = reply((struct ib_mad_hdr *)smp);
4543                 return ret;
4544         }
4545         ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags, smp->mkey,
4546                          smp->route.dr.dr_slid, smp->route.dr.return_path,
4547                          smp->hop_cnt);
4548         if (ret) {
4549                 u32 port_num = be32_to_cpu(smp->attr_mod);
4550
4551                 /*
4552                  * If this is a get/set portinfo, we already check the
4553                  * M_Key if the MAD is for another port and the M_Key
4554                  * is OK on the receiving port. This check is needed
4555                  * to increment the error counters when the M_Key
4556                  * fails to match on *both* ports.
4557                  */
4558                 if (attr_id == IB_SMP_ATTR_PORT_INFO &&
4559                     (smp->method == IB_MGMT_METHOD_GET ||
4560                      smp->method == IB_MGMT_METHOD_SET) &&
4561                     port_num && port_num <= ibdev->phys_port_cnt &&
4562                     port != port_num)
4563                         (void)check_mkey(to_iport(ibdev, port_num),
4564                                           (struct ib_mad_hdr *)smp, 0,
4565                                           smp->mkey, smp->route.dr.dr_slid,
4566                                           smp->route.dr.return_path,
4567                                           smp->hop_cnt);
4568                 ret = IB_MAD_RESULT_FAILURE;
4569                 return ret;
4570         }
4571
4572         *resp_len = opa_get_smp_header_size(smp);
4573
4574         switch (smp->method) {
4575         case IB_MGMT_METHOD_GET:
4576                 switch (attr_id) {
4577                 default:
4578                         clear_opa_smp_data(smp);
4579                         ret = subn_get_opa_sma(attr_id, smp, am, data,
4580                                                ibdev, port, resp_len,
4581                                                data_size);
4582                         break;
4583                 case OPA_ATTRIB_ID_AGGREGATE:
4584                         ret = subn_get_opa_aggregate(smp, ibdev, port,
4585                                                      resp_len);
4586                         break;
4587                 }
4588                 break;
4589         case IB_MGMT_METHOD_SET:
4590                 switch (attr_id) {
4591                 default:
4592                         ret = subn_set_opa_sma(attr_id, smp, am, data,
4593                                                ibdev, port, resp_len,
4594                                                data_size, local_mad);
4595                         break;
4596                 case OPA_ATTRIB_ID_AGGREGATE:
4597                         ret = subn_set_opa_aggregate(smp, ibdev, port,
4598                                                      resp_len, local_mad);
4599                         break;
4600                 }
4601                 break;
4602         case IB_MGMT_METHOD_TRAP:
4603         case IB_MGMT_METHOD_REPORT:
4604         case IB_MGMT_METHOD_REPORT_RESP:
4605         case IB_MGMT_METHOD_GET_RESP:
4606                 /*
4607                  * The ib_mad module will call us to process responses
4608                  * before checking for other consumers.
4609                  * Just tell the caller to process it normally.
4610                  */
4611                 ret = IB_MAD_RESULT_SUCCESS;
4612                 break;
4613         case IB_MGMT_METHOD_TRAP_REPRESS:
4614                 subn_handle_opa_trap_repress(ibp, smp);
4615                 /* Always successful */
4616                 ret = IB_MAD_RESULT_SUCCESS;
4617                 break;
4618         default:
4619                 smp->status |= IB_SMP_UNSUP_METHOD;
4620                 ret = reply((struct ib_mad_hdr *)smp);
4621                 break;
4622         }
4623
4624         return ret;
4625 }
4626
4627 static int process_subn(struct ib_device *ibdev, int mad_flags,
4628                         u8 port, const struct ib_mad *in_mad,
4629                         struct ib_mad *out_mad)
4630 {
4631         struct ib_smp *smp = (struct ib_smp *)out_mad;
4632         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4633         int ret;
4634
4635         *out_mad = *in_mad;
4636         if (smp->class_version != 1) {
4637                 smp->status |= IB_SMP_UNSUP_VERSION;
4638                 ret = reply((struct ib_mad_hdr *)smp);
4639                 return ret;
4640         }
4641
4642         ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags,
4643                          smp->mkey, (__force __be32)smp->dr_slid,
4644                          smp->return_path, smp->hop_cnt);
4645         if (ret) {
4646                 u32 port_num = be32_to_cpu(smp->attr_mod);
4647
4648                 /*
4649                  * If this is a get/set portinfo, we already check the
4650                  * M_Key if the MAD is for another port and the M_Key
4651                  * is OK on the receiving port. This check is needed
4652                  * to increment the error counters when the M_Key
4653                  * fails to match on *both* ports.
4654                  */
4655                 if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
4656                     (smp->method == IB_MGMT_METHOD_GET ||
4657                      smp->method == IB_MGMT_METHOD_SET) &&
4658                     port_num && port_num <= ibdev->phys_port_cnt &&
4659                     port != port_num)
4660                         (void)check_mkey(to_iport(ibdev, port_num),
4661                                          (struct ib_mad_hdr *)smp, 0,
4662                                          smp->mkey,
4663                                          (__force __be32)smp->dr_slid,
4664                                          smp->return_path, smp->hop_cnt);
4665                 ret = IB_MAD_RESULT_FAILURE;
4666                 return ret;
4667         }
4668
4669         switch (smp->method) {
4670         case IB_MGMT_METHOD_GET:
4671                 switch (smp->attr_id) {
4672                 case IB_SMP_ATTR_NODE_INFO:
4673                         ret = subn_get_nodeinfo(smp, ibdev, port);
4674                         break;
4675                 default:
4676                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
4677                         ret = reply((struct ib_mad_hdr *)smp);
4678                         break;
4679                 }
4680                 break;
4681         }
4682
4683         return ret;
4684 }
4685
4686 static int process_perf(struct ib_device *ibdev, u8 port,
4687                         const struct ib_mad *in_mad,
4688                         struct ib_mad *out_mad)
4689 {
4690         struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
4691         struct ib_class_port_info *cpi = (struct ib_class_port_info *)
4692                                                 &pmp->data;
4693         int ret = IB_MAD_RESULT_FAILURE;
4694
4695         *out_mad = *in_mad;
4696         if (pmp->mad_hdr.class_version != 1) {
4697                 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4698                 ret = reply((struct ib_mad_hdr *)pmp);
4699                 return ret;
4700         }
4701
4702         switch (pmp->mad_hdr.method) {
4703         case IB_MGMT_METHOD_GET:
4704                 switch (pmp->mad_hdr.attr_id) {
4705                 case IB_PMA_PORT_COUNTERS:
4706                         ret = pma_get_ib_portcounters(pmp, ibdev, port);
4707                         break;
4708                 case IB_PMA_PORT_COUNTERS_EXT:
4709                         ret = pma_get_ib_portcounters_ext(pmp, ibdev, port);
4710                         break;
4711                 case IB_PMA_CLASS_PORT_INFO:
4712                         cpi->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
4713                         ret = reply((struct ib_mad_hdr *)pmp);
4714                         break;
4715                 default:
4716                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4717                         ret = reply((struct ib_mad_hdr *)pmp);
4718                         break;
4719                 }
4720                 break;
4721
4722         case IB_MGMT_METHOD_SET:
4723                 if (pmp->mad_hdr.attr_id) {
4724                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4725                         ret = reply((struct ib_mad_hdr *)pmp);
4726                 }
4727                 break;
4728
4729         case IB_MGMT_METHOD_TRAP:
4730         case IB_MGMT_METHOD_GET_RESP:
4731                 /*
4732                  * The ib_mad module will call us to process responses
4733                  * before checking for other consumers.
4734                  * Just tell the caller to process it normally.
4735                  */
4736                 ret = IB_MAD_RESULT_SUCCESS;
4737                 break;
4738
4739         default:
4740                 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4741                 ret = reply((struct ib_mad_hdr *)pmp);
4742                 break;
4743         }
4744
4745         return ret;
4746 }
4747
4748 static int process_perf_opa(struct ib_device *ibdev, u8 port,
4749                             const struct opa_mad *in_mad,
4750                             struct opa_mad *out_mad, u32 *resp_len)
4751 {
4752         struct opa_pma_mad *pmp = (struct opa_pma_mad *)out_mad;
4753         int ret;
4754
4755         *out_mad = *in_mad;
4756
4757         if (pmp->mad_hdr.class_version != OPA_SM_CLASS_VERSION) {
4758                 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4759                 return reply((struct ib_mad_hdr *)pmp);
4760         }
4761
4762         *resp_len = sizeof(pmp->mad_hdr);
4763
4764         switch (pmp->mad_hdr.method) {
4765         case IB_MGMT_METHOD_GET:
4766                 switch (pmp->mad_hdr.attr_id) {
4767                 case IB_PMA_CLASS_PORT_INFO:
4768                         ret = pma_get_opa_classportinfo(pmp, ibdev, resp_len);
4769                         break;
4770                 case OPA_PM_ATTRIB_ID_PORT_STATUS:
4771                         ret = pma_get_opa_portstatus(pmp, ibdev, port,
4772                                                      resp_len);
4773                         break;
4774                 case OPA_PM_ATTRIB_ID_DATA_PORT_COUNTERS:
4775                         ret = pma_get_opa_datacounters(pmp, ibdev, port,
4776                                                        resp_len);
4777                         break;
4778                 case OPA_PM_ATTRIB_ID_ERROR_PORT_COUNTERS:
4779                         ret = pma_get_opa_porterrors(pmp, ibdev, port,
4780                                                      resp_len);
4781                         break;
4782                 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4783                         ret = pma_get_opa_errorinfo(pmp, ibdev, port,
4784                                                     resp_len);
4785                         break;
4786                 default:
4787                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4788                         ret = reply((struct ib_mad_hdr *)pmp);
4789                         break;
4790                 }
4791                 break;
4792
4793         case IB_MGMT_METHOD_SET:
4794                 switch (pmp->mad_hdr.attr_id) {
4795                 case OPA_PM_ATTRIB_ID_CLEAR_PORT_STATUS:
4796                         ret = pma_set_opa_portstatus(pmp, ibdev, port,
4797                                                      resp_len);
4798                         break;
4799                 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4800                         ret = pma_set_opa_errorinfo(pmp, ibdev, port,
4801                                                     resp_len);
4802                         break;
4803                 default:
4804                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4805                         ret = reply((struct ib_mad_hdr *)pmp);
4806                         break;
4807                 }
4808                 break;
4809
4810         case IB_MGMT_METHOD_TRAP:
4811         case IB_MGMT_METHOD_GET_RESP:
4812                 /*
4813                  * The ib_mad module will call us to process responses
4814                  * before checking for other consumers.
4815                  * Just tell the caller to process it normally.
4816                  */
4817                 ret = IB_MAD_RESULT_SUCCESS;
4818                 break;
4819
4820         default:
4821                 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4822                 ret = reply((struct ib_mad_hdr *)pmp);
4823                 break;
4824         }
4825
4826         return ret;
4827 }
4828
4829 static int hfi1_process_opa_mad(struct ib_device *ibdev, int mad_flags,
4830                                 u8 port, const struct ib_wc *in_wc,
4831                                 const struct ib_grh *in_grh,
4832                                 const struct opa_mad *in_mad,
4833                                 struct opa_mad *out_mad, size_t *out_mad_size,
4834                                 u16 *out_mad_pkey_index)
4835 {
4836         int ret;
4837         int pkey_idx;
4838         int local_mad = 0;
4839         u32 resp_len = in_wc->byte_len - sizeof(*in_grh);
4840         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4841
4842         pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
4843         if (pkey_idx < 0) {
4844                 pr_warn("failed to find limited mgmt pkey, defaulting 0x%x\n",
4845                         hfi1_get_pkey(ibp, 1));
4846                 pkey_idx = 1;
4847         }
4848         *out_mad_pkey_index = (u16)pkey_idx;
4849
4850         switch (in_mad->mad_hdr.mgmt_class) {
4851         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4852         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4853                 local_mad = is_local_mad(ibp, in_mad, in_wc);
4854                 if (local_mad) {
4855                         ret = opa_local_smp_check(ibp, in_wc);
4856                         if (ret)
4857                                 return IB_MAD_RESULT_FAILURE;
4858                 }
4859                 ret = process_subn_opa(ibdev, mad_flags, port, in_mad,
4860                                        out_mad, &resp_len, local_mad);
4861                 goto bail;
4862         case IB_MGMT_CLASS_PERF_MGMT:
4863                 ret = hfi1_pkey_validation_pma(ibp, in_mad, in_wc);
4864                 if (ret)
4865                         return IB_MAD_RESULT_FAILURE;
4866
4867                 ret = process_perf_opa(ibdev, port, in_mad, out_mad, &resp_len);
4868                 goto bail;
4869
4870         default:
4871                 ret = IB_MAD_RESULT_SUCCESS;
4872         }
4873
4874 bail:
4875         if (ret & IB_MAD_RESULT_REPLY)
4876                 *out_mad_size = round_up(resp_len, 8);
4877         else if (ret & IB_MAD_RESULT_SUCCESS)
4878                 *out_mad_size = in_wc->byte_len - sizeof(struct ib_grh);
4879
4880         return ret;
4881 }
4882
4883 static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4884                                const struct ib_wc *in_wc,
4885                                const struct ib_grh *in_grh,
4886                                const struct ib_mad *in_mad,
4887                                struct ib_mad *out_mad)
4888 {
4889         int ret;
4890
4891         switch (in_mad->mad_hdr.mgmt_class) {
4892         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4893         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4894                 ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
4895                 break;
4896         case IB_MGMT_CLASS_PERF_MGMT:
4897                 ret = process_perf(ibdev, port, in_mad, out_mad);
4898                 break;
4899         default:
4900                 ret = IB_MAD_RESULT_SUCCESS;
4901                 break;
4902         }
4903
4904         return ret;
4905 }
4906
4907 /**
4908  * hfi1_process_mad - process an incoming MAD packet
4909  * @ibdev: the infiniband device this packet came in on
4910  * @mad_flags: MAD flags
4911  * @port: the port number this packet came in on
4912  * @in_wc: the work completion entry for this packet
4913  * @in_grh: the global route header for this packet
4914  * @in_mad: the incoming MAD
4915  * @out_mad: any outgoing MAD reply
4916  *
4917  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
4918  * interested in processing.
4919  *
4920  * Note that the verbs framework has already done the MAD sanity checks,
4921  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4922  * MADs.
4923  *
4924  * This is called by the ib_mad module.
4925  */
4926 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4927                      const struct ib_wc *in_wc, const struct ib_grh *in_grh,
4928                      const struct ib_mad_hdr *in_mad, size_t in_mad_size,
4929                      struct ib_mad_hdr *out_mad, size_t *out_mad_size,
4930                      u16 *out_mad_pkey_index)
4931 {
4932         switch (in_mad->base_version) {
4933         case OPA_MGMT_BASE_VERSION:
4934                 if (unlikely(in_mad_size != sizeof(struct opa_mad))) {
4935                         dev_err(ibdev->dev.parent, "invalid in_mad_size\n");
4936                         return IB_MAD_RESULT_FAILURE;
4937                 }
4938                 return hfi1_process_opa_mad(ibdev, mad_flags, port,
4939                                             in_wc, in_grh,
4940                                             (struct opa_mad *)in_mad,
4941                                             (struct opa_mad *)out_mad,
4942                                             out_mad_size,
4943                                             out_mad_pkey_index);
4944         case IB_MGMT_BASE_VERSION:
4945                 return hfi1_process_ib_mad(ibdev, mad_flags, port,
4946                                           in_wc, in_grh,
4947                                           (const struct ib_mad *)in_mad,
4948                                           (struct ib_mad *)out_mad);
4949         default:
4950                 break;
4951         }
4952
4953         return IB_MAD_RESULT_FAILURE;
4954 }