drivers/infiniband/hw/hfi1/mad.c

   1 /*
   2  * Copyright(c) 2015, 2016 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 #define OPA_NUM_PKEY_BLOCKS_PER_SMP (OPA_SMP_DR_DATA_SIZE \
  50                         / (OPA_PARTITION_TABLE_BLK_SIZE * sizeof(u16)))
  51
  52 #include "hfi.h"
  53 #include "mad.h"
  54 #include "trace.h"
  55 #include "qp.h"
  56
  57 /* the reset value from the FM is supposed to be 0xffff, handle both */
  58 #define OPA_LINK_WIDTH_RESET_OLD 0x0fff
  59 #define OPA_LINK_WIDTH_RESET 0xffff
  60
  61 static int reply(struct ib_mad_hdr *smp)
  62 {
  63         /*
  64          * The verbs framework will handle the directed/LID route
  65          * packet changes.
  66          */
  67         smp->method = IB_MGMT_METHOD_GET_RESP;
  68         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
  69                 smp->status |= IB_SMP_DIRECTION;
  70         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
  71 }
  72
  73 static inline void clear_opa_smp_data(struct opa_smp *smp)
  74 {
  75         void *data = opa_get_smp_data(smp);
  76         size_t size = opa_get_smp_data_size(smp);
  77
  78         memset(data, 0, size);
  79 }
  80
  81 void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port)
  82 {
  83         struct ib_event event;
  84
  85         event.event = IB_EVENT_PKEY_CHANGE;
  86         event.device = &dd->verbs_dev.rdi.ibdev;
  87         event.element.port_num = port;
  88         ib_dispatch_event(&event);
  89 }
  90
  91 static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len)
  92 {
  93         struct ib_mad_send_buf *send_buf;
  94         struct ib_mad_agent *agent;
  95         struct opa_smp *smp;
  96         int ret;
  97         unsigned long flags;
  98         unsigned long timeout;
  99         int pkey_idx;
 100         u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp;
 101
 102         agent = ibp->rvp.send_agent;
 103         if (!agent)
 104                 return;
 105
 106         /* o14-3.2.1 */
 107         if (ppd_from_ibp(ibp)->lstate != IB_PORT_ACTIVE)
 108                 return;
 109
 110         /* o14-2 */
 111         if (ibp->rvp.trap_timeout && time_before(jiffies,
 112                                                  ibp->rvp.trap_timeout))
 113                 return;
 114
 115         pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
 116         if (pkey_idx < 0) {
 117                 pr_warn("%s: failed to find limited mgmt pkey, defaulting 0x%x\n",
 118                         __func__, hfi1_get_pkey(ibp, 1));
 119                 pkey_idx = 1;
 120         }
 121
 122         send_buf = ib_create_send_mad(agent, qpn, pkey_idx, 0,
 123                                       IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
 124                                       GFP_ATOMIC, IB_MGMT_BASE_VERSION);
 125         if (IS_ERR(send_buf))
 126                 return;
 127
 128         smp = send_buf->mad;
 129         smp->base_version = OPA_MGMT_BASE_VERSION;
 130         smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
 131         smp->class_version = OPA_SMI_CLASS_VERSION;
 132         smp->method = IB_MGMT_METHOD_TRAP;
 133         ibp->rvp.tid++;
 134         smp->tid = cpu_to_be64(ibp->rvp.tid);
 135         smp->attr_id = IB_SMP_ATTR_NOTICE;
 136         /* o14-1: smp->mkey = 0; */
 137         memcpy(smp->route.lid.data, data, len);
 138
 139         spin_lock_irqsave(&ibp->rvp.lock, flags);
 140         if (!ibp->rvp.sm_ah) {
 141                 if (ibp->rvp.sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
 142                         struct ib_ah *ah;
 143
 144                         ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid);
 145                         if (IS_ERR(ah)) {
 146                                 ret = PTR_ERR(ah);
 147                         } else {
 148                                 send_buf->ah = ah;
 149                                 ibp->rvp.sm_ah = ibah_to_rvtah(ah);
 150                                 ret = 0;
 151                         }
 152                 } else {
 153                         ret = -EINVAL;
 154                 }
 155         } else {
 156                 send_buf->ah = &ibp->rvp.sm_ah->ibah;
 157                 ret = 0;
 158         }
 159         spin_unlock_irqrestore(&ibp->rvp.lock, flags);
 160
 161         if (!ret)
 162                 ret = ib_post_send_mad(send_buf, NULL);
 163         if (!ret) {
 164                 /* 4.096 usec. */
 165                 timeout = (4096 * (1UL << ibp->rvp.subnet_timeout)) / 1000;
 166                 ibp->rvp.trap_timeout = jiffies + usecs_to_jiffies(timeout);
 167         } else {
 168                 ib_free_send_mad(send_buf);
 169                 ibp->rvp.trap_timeout = 0;
 170         }
 171 }
 172
 173 /*
 174  * Send a bad [PQ]_Key trap (ch. 14.3.8).
 175  */
 176 void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
 177                     u32 qp1, u32 qp2, u16 lid1, u16 lid2)
 178 {
 179         struct opa_mad_notice_attr data;
 180         u32 lid = ppd_from_ibp(ibp)->lid;
 181         u32 _lid1 = lid1;
 182         u32 _lid2 = lid2;
 183
 184         memset(&data, 0, sizeof(data));
 185
 186         if (trap_num == OPA_TRAP_BAD_P_KEY)
 187                 ibp->rvp.pkey_violations++;
 188         else
 189                 ibp->rvp.qkey_violations++;
 190         ibp->rvp.n_pkt_drops++;
 191
 192         /* Send violation trap */
 193         data.generic_type = IB_NOTICE_TYPE_SECURITY;
 194         data.prod_type_lsb = IB_NOTICE_PROD_CA;
 195         data.trap_num = trap_num;
 196         data.issuer_lid = cpu_to_be32(lid);
 197         data.ntc_257_258.lid1 = cpu_to_be32(_lid1);
 198         data.ntc_257_258.lid2 = cpu_to_be32(_lid2);
 199         data.ntc_257_258.key = cpu_to_be32(key);
 200         data.ntc_257_258.sl = sl << 3;
 201         data.ntc_257_258.qp1 = cpu_to_be32(qp1);
 202         data.ntc_257_258.qp2 = cpu_to_be32(qp2);
 203
 204         send_trap(ibp, &data, sizeof(data));
 205 }
 206
 207 /*
 208  * Send a bad M_Key trap (ch. 14.3.9).
 209  */
 210 static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
 211                      __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt)
 212 {
 213         struct opa_mad_notice_attr data;
 214         u32 lid = ppd_from_ibp(ibp)->lid;
 215
 216         memset(&data, 0, sizeof(data));
 217         /* Send violation trap */
 218         data.generic_type = IB_NOTICE_TYPE_SECURITY;
 219         data.prod_type_lsb = IB_NOTICE_PROD_CA;
 220         data.trap_num = OPA_TRAP_BAD_M_KEY;
 221         data.issuer_lid = cpu_to_be32(lid);
 222         data.ntc_256.lid = data.issuer_lid;
 223         data.ntc_256.method = mad->method;
 224         data.ntc_256.attr_id = mad->attr_id;
 225         data.ntc_256.attr_mod = mad->attr_mod;
 226         data.ntc_256.mkey = mkey;
 227         if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
 228                 data.ntc_256.dr_slid = dr_slid;
 229                 data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
 230                 if (hop_cnt > ARRAY_SIZE(data.ntc_256.dr_rtn_path)) {
 231                         data.ntc_256.dr_trunc_hop |=
 232                                 IB_NOTICE_TRAP_DR_TRUNC;
 233                         hop_cnt = ARRAY_SIZE(data.ntc_256.dr_rtn_path);
 234                 }
 235                 data.ntc_256.dr_trunc_hop |= hop_cnt;
 236                 memcpy(data.ntc_256.dr_rtn_path, return_path,
 237                        hop_cnt);
 238         }
 239
 240         send_trap(ibp, &data, sizeof(data));
 241 }
 242
 243 /*
 244  * Send a Port Capability Mask Changed trap (ch. 14.3.11).
 245  */
 246 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num)
 247 {
 248         struct opa_mad_notice_attr data;
 249         struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi);
 250         struct hfi1_devdata *dd = dd_from_dev(verbs_dev);
 251         struct hfi1_ibport *ibp = &dd->pport[port_num - 1].ibport_data;
 252         u32 lid = ppd_from_ibp(ibp)->lid;
 253
 254         memset(&data, 0, sizeof(data));
 255
 256         data.generic_type = IB_NOTICE_TYPE_INFO;
 257         data.prod_type_lsb = IB_NOTICE_PROD_CA;
 258         data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
 259         data.issuer_lid = cpu_to_be32(lid);
 260         data.ntc_144.lid = data.issuer_lid;
 261         data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
 262
 263         send_trap(ibp, &data, sizeof(data));
 264 }
 265
 266 /*
 267  * Send a System Image GUID Changed trap (ch. 14.3.12).
 268  */
 269 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp)
 270 {
 271         struct opa_mad_notice_attr data;
 272         u32 lid = ppd_from_ibp(ibp)->lid;
 273
 274         memset(&data, 0, sizeof(data));
 275
 276         data.generic_type = IB_NOTICE_TYPE_INFO;
 277         data.prod_type_lsb = IB_NOTICE_PROD_CA;
 278         data.trap_num = OPA_TRAP_CHANGE_SYSGUID;
 279         data.issuer_lid = cpu_to_be32(lid);
 280         data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid;
 281         data.ntc_145.lid = data.issuer_lid;
 282
 283         send_trap(ibp, &data, sizeof(data));
 284 }
 285
 286 /*
 287  * Send a Node Description Changed trap (ch. 14.3.13).
 288  */
 289 void hfi1_node_desc_chg(struct hfi1_ibport *ibp)
 290 {
 291         struct opa_mad_notice_attr data;
 292         u32 lid = ppd_from_ibp(ibp)->lid;
 293
 294         memset(&data, 0, sizeof(data));
 295
 296         data.generic_type = IB_NOTICE_TYPE_INFO;
 297         data.prod_type_lsb = IB_NOTICE_PROD_CA;
 298         data.trap_num = OPA_TRAP_CHANGE_CAPABILITY;
 299         data.issuer_lid = cpu_to_be32(lid);
 300         data.ntc_144.lid = data.issuer_lid;
 301         data.ntc_144.change_flags =
 302                 cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG);
 303
 304         send_trap(ibp, &data, sizeof(data));
 305 }
 306
 307 static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am,
 308                                    u8 *data, struct ib_device *ibdev,
 309                                    u8 port, u32 *resp_len)
 310 {
 311         struct opa_node_description *nd;
 312
 313         if (am) {
 314                 smp->status |= IB_SMP_INVALID_FIELD;
 315                 return reply((struct ib_mad_hdr *)smp);
 316         }
 317
 318         nd = (struct opa_node_description *)data;
 319
 320         memcpy(nd->data, ibdev->node_desc, sizeof(nd->data));
 321
 322         if (resp_len)
 323                 *resp_len += sizeof(*nd);
 324
 325         return reply((struct ib_mad_hdr *)smp);
 326 }
 327
 328 static int __subn_get_opa_nodeinfo(struct opa_smp *smp, u32 am, u8 *data,
 329                                    struct ib_device *ibdev, u8 port,
 330                                    u32 *resp_len)
 331 {
 332         struct opa_node_info *ni;
 333         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 334         unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
 335
 336         ni = (struct opa_node_info *)data;
 337
 338         /* GUID 0 is illegal */
 339         if (am || pidx >= dd->num_pports || dd->pport[pidx].guid == 0) {
 340                 smp->status |= IB_SMP_INVALID_FIELD;
 341                 return reply((struct ib_mad_hdr *)smp);
 342         }
 343
 344         ni->port_guid = cpu_to_be64(dd->pport[pidx].guid);
 345         ni->base_version = OPA_MGMT_BASE_VERSION;
 346         ni->class_version = OPA_SMI_CLASS_VERSION;
 347         ni->node_type = 1;     /* channel adapter */
 348         ni->num_ports = ibdev->phys_port_cnt;
 349         /* This is already in network order */
 350         ni->system_image_guid = ib_hfi1_sys_image_guid;
 351         /* Use first-port GUID as node */
 352         ni->node_guid = cpu_to_be64(dd->pport->guid);
 353         ni->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
 354         ni->device_id = cpu_to_be16(dd->pcidev->device);
 355         ni->revision = cpu_to_be32(dd->minrev);
 356         ni->local_port_num = port;
 357         ni->vendor_id[0] = dd->oui1;
 358         ni->vendor_id[1] = dd->oui2;
 359         ni->vendor_id[2] = dd->oui3;
 360
 361         if (resp_len)
 362                 *resp_len += sizeof(*ni);
 363
 364         return reply((struct ib_mad_hdr *)smp);
 365 }
 366
 367 static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
 368                              u8 port)
 369 {
 370         struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
 371         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 372         unsigned pidx = port - 1; /* IB number port from 1, hw from 0 */
 373
 374         /* GUID 0 is illegal */
 375         if (smp->attr_mod || pidx >= dd->num_pports ||
 376             dd->pport[pidx].guid == 0)
 377                 smp->status |= IB_SMP_INVALID_FIELD;
 378         else
 379                 nip->port_guid = cpu_to_be64(dd->pport[pidx].guid);
 380
 381         nip->base_version = OPA_MGMT_BASE_VERSION;
 382         nip->class_version = OPA_SMI_CLASS_VERSION;
 383         nip->node_type = 1;     /* channel adapter */
 384         nip->num_ports = ibdev->phys_port_cnt;
 385         /* This is already in network order */
 386         nip->sys_guid = ib_hfi1_sys_image_guid;
 387          /* Use first-port GUID as node */
 388         nip->node_guid = cpu_to_be64(dd->pport->guid);
 389         nip->partition_cap = cpu_to_be16(hfi1_get_npkeys(dd));
 390         nip->device_id = cpu_to_be16(dd->pcidev->device);
 391         nip->revision = cpu_to_be32(dd->minrev);
 392         nip->local_port_num = port;
 393         nip->vendor_id[0] = dd->oui1;
 394         nip->vendor_id[1] = dd->oui2;
 395         nip->vendor_id[2] = dd->oui3;
 396
 397         return reply((struct ib_mad_hdr *)smp);
 398 }
 399
 400 static void set_link_width_enabled(struct hfi1_pportdata *ppd, u32 w)
 401 {
 402         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_ENB, w);
 403 }
 404
 405 static void set_link_width_downgrade_enabled(struct hfi1_pportdata *ppd, u32 w)
 406 {
 407         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LWID_DG_ENB, w);
 408 }
 409
 410 static void set_link_speed_enabled(struct hfi1_pportdata *ppd, u32 s)
 411 {
 412         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_SPD_ENB, s);
 413 }
 414
 415 static int check_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad,
 416                       int mad_flags, __be64 mkey, __be32 dr_slid,
 417                       u8 return_path[], u8 hop_cnt)
 418 {
 419         int valid_mkey = 0;
 420         int ret = 0;
 421
 422         /* Is the mkey in the process of expiring? */
 423         if (ibp->rvp.mkey_lease_timeout &&
 424             time_after_eq(jiffies, ibp->rvp.mkey_lease_timeout)) {
 425                 /* Clear timeout and mkey protection field. */
 426                 ibp->rvp.mkey_lease_timeout = 0;
 427                 ibp->rvp.mkeyprot = 0;
 428         }
 429
 430         if ((mad_flags & IB_MAD_IGNORE_MKEY) ||  ibp->rvp.mkey == 0 ||
 431             ibp->rvp.mkey == mkey)
 432                 valid_mkey = 1;
 433
 434         /* Unset lease timeout on any valid Get/Set/TrapRepress */
 435         if (valid_mkey && ibp->rvp.mkey_lease_timeout &&
 436             (mad->method == IB_MGMT_METHOD_GET ||
 437              mad->method == IB_MGMT_METHOD_SET ||
 438              mad->method == IB_MGMT_METHOD_TRAP_REPRESS))
 439                 ibp->rvp.mkey_lease_timeout = 0;
 440
 441         if (!valid_mkey) {
 442                 switch (mad->method) {
 443                 case IB_MGMT_METHOD_GET:
 444                         /* Bad mkey not a violation below level 2 */
 445                         if (ibp->rvp.mkeyprot < 2)
 446                                 break;
 447                 case IB_MGMT_METHOD_SET:
 448                 case IB_MGMT_METHOD_TRAP_REPRESS:
 449                         if (ibp->rvp.mkey_violations != 0xFFFF)
 450                                 ++ibp->rvp.mkey_violations;
 451                         if (!ibp->rvp.mkey_lease_timeout &&
 452                             ibp->rvp.mkey_lease_period)
 453                                 ibp->rvp.mkey_lease_timeout = jiffies +
 454                                         ibp->rvp.mkey_lease_period * HZ;
 455                         /* Generate a trap notice. */
 456                         bad_mkey(ibp, mad, mkey, dr_slid, return_path,
 457                                  hop_cnt);
 458                         ret = 1;
 459                 }
 460         }
 461
 462         return ret;
 463 }
 464
 465 /*
 466  * The SMA caches reads from LCB registers in case the LCB is unavailable.
 467  * (The LCB is unavailable in certain link states, for example.)
 468  */
 469 struct lcb_datum {
 470         u32 off;
 471         u64 val;
 472 };
 473
 474 static struct lcb_datum lcb_cache[] = {
 475         { DC_LCB_STS_ROUND_TRIP_LTP_CNT, 0 },
 476 };
 477
 478 static int write_lcb_cache(u32 off, u64 val)
 479 {
 480         int i;
 481
 482         for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
 483                 if (lcb_cache[i].off == off) {
 484                         lcb_cache[i].val = val;
 485                         return 0;
 486                 }
 487         }
 488
 489         pr_warn("%s bad offset 0x%x\n", __func__, off);
 490         return -1;
 491 }
 492
 493 static int read_lcb_cache(u32 off, u64 *val)
 494 {
 495         int i;
 496
 497         for (i = 0; i < ARRAY_SIZE(lcb_cache); i++) {
 498                 if (lcb_cache[i].off == off) {
 499                         *val = lcb_cache[i].val;
 500                         return 0;
 501                 }
 502         }
 503
 504         pr_warn("%s bad offset 0x%x\n", __func__, off);
 505         return -1;
 506 }
 507
 508 void read_ltp_rtt(struct hfi1_devdata *dd)
 509 {
 510         u64 reg;
 511
 512         if (read_lcb_csr(dd, DC_LCB_STS_ROUND_TRIP_LTP_CNT, &reg))
 513                 dd_dev_err(dd, "%s: unable to read LTP RTT\n", __func__);
 514         else
 515                 write_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, reg);
 516 }
 517
 518 static int __subn_get_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
 519                                    struct ib_device *ibdev, u8 port,
 520                                    u32 *resp_len)
 521 {
 522         int i;
 523         struct hfi1_devdata *dd;
 524         struct hfi1_pportdata *ppd;
 525         struct hfi1_ibport *ibp;
 526         struct opa_port_info *pi = (struct opa_port_info *)data;
 527         u8 mtu;
 528         u8 credit_rate;
 529         u8 is_beaconing_active;
 530         u32 state;
 531         u32 num_ports = OPA_AM_NPORT(am);
 532         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
 533         u32 buffer_units;
 534         u64 tmp = 0;
 535
 536         if (num_ports != 1) {
 537                 smp->status |= IB_SMP_INVALID_FIELD;
 538                 return reply((struct ib_mad_hdr *)smp);
 539         }
 540
 541         dd = dd_from_ibdev(ibdev);
 542         /* IB numbers ports from 1, hw from 0 */
 543         ppd = dd->pport + (port - 1);
 544         ibp = &ppd->ibport_data;
 545
 546         if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
 547             ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
 548                 smp->status |= IB_SMP_INVALID_FIELD;
 549                 return reply((struct ib_mad_hdr *)smp);
 550         }
 551
 552         pi->lid = cpu_to_be32(ppd->lid);
 553
 554         /* Only return the mkey if the protection field allows it. */
 555         if (!(smp->method == IB_MGMT_METHOD_GET &&
 556               ibp->rvp.mkey != smp->mkey &&
 557               ibp->rvp.mkeyprot == 1))
 558                 pi->mkey = ibp->rvp.mkey;
 559
 560         pi->subnet_prefix = ibp->rvp.gid_prefix;
 561         pi->sm_lid = cpu_to_be32(ibp->rvp.sm_lid);
 562         pi->ib_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags);
 563         pi->mkey_lease_period = cpu_to_be16(ibp->rvp.mkey_lease_period);
 564         pi->sm_trap_qp = cpu_to_be32(ppd->sm_trap_qp);
 565         pi->sa_qp = cpu_to_be32(ppd->sa_qp);
 566
 567         pi->link_width.enabled = cpu_to_be16(ppd->link_width_enabled);
 568         pi->link_width.supported = cpu_to_be16(ppd->link_width_supported);
 569         pi->link_width.active = cpu_to_be16(ppd->link_width_active);
 570
 571         pi->link_width_downgrade.supported =
 572                         cpu_to_be16(ppd->link_width_downgrade_supported);
 573         pi->link_width_downgrade.enabled =
 574                         cpu_to_be16(ppd->link_width_downgrade_enabled);
 575         pi->link_width_downgrade.tx_active =
 576                         cpu_to_be16(ppd->link_width_downgrade_tx_active);
 577         pi->link_width_downgrade.rx_active =
 578                         cpu_to_be16(ppd->link_width_downgrade_rx_active);
 579
 580         pi->link_speed.supported = cpu_to_be16(ppd->link_speed_supported);
 581         pi->link_speed.active = cpu_to_be16(ppd->link_speed_active);
 582         pi->link_speed.enabled = cpu_to_be16(ppd->link_speed_enabled);
 583
 584         state = driver_lstate(ppd);
 585
 586         if (start_of_sm_config && (state == IB_PORT_INIT))
 587                 ppd->is_sm_config_started = 1;
 588
 589         pi->port_phys_conf = (ppd->port_type & 0xf);
 590
 591         pi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
 592         pi->port_states.ledenable_offlinereason |=
 593                 ppd->is_sm_config_started << 5;
 594         /*
 595          * This pairs with the memory barrier in hfi1_start_led_override to
 596          * ensure that we read the correct state of LED beaconing represented
 597          * by led_override_timer_active
 598          */
 599         smp_rmb();
 600         is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
 601         pi->port_states.ledenable_offlinereason |= is_beaconing_active << 6;
 602         pi->port_states.ledenable_offlinereason |=
 603                 ppd->offline_disabled_reason;
 604
 605         pi->port_states.portphysstate_portstate =
 606                 (hfi1_ibphys_portstate(ppd) << 4) | state;
 607
 608         pi->mkeyprotect_lmc = (ibp->rvp.mkeyprot << 6) | ppd->lmc;
 609
 610         memset(pi->neigh_mtu.pvlx_to_mtu, 0, sizeof(pi->neigh_mtu.pvlx_to_mtu));
 611         for (i = 0; i < ppd->vls_supported; i++) {
 612                 mtu = mtu_to_enum(dd->vld[i].mtu, HFI1_DEFAULT_ACTIVE_MTU);
 613                 if ((i % 2) == 0)
 614                         pi->neigh_mtu.pvlx_to_mtu[i / 2] |= (mtu << 4);
 615                 else
 616                         pi->neigh_mtu.pvlx_to_mtu[i / 2] |= mtu;
 617         }
 618         /* don't forget VL 15 */
 619         mtu = mtu_to_enum(dd->vld[15].mtu, 2048);
 620         pi->neigh_mtu.pvlx_to_mtu[15 / 2] |= mtu;
 621         pi->smsl = ibp->rvp.sm_sl & OPA_PI_MASK_SMSL;
 622         pi->operational_vls = hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS);
 623         pi->partenforce_filterraw |=
 624                 (ppd->linkinit_reason & OPA_PI_MASK_LINKINIT_REASON);
 625         if (ppd->part_enforce & HFI1_PART_ENFORCE_IN)
 626                 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_IN;
 627         if (ppd->part_enforce & HFI1_PART_ENFORCE_OUT)
 628                 pi->partenforce_filterraw |= OPA_PI_MASK_PARTITION_ENFORCE_OUT;
 629         pi->mkey_violations = cpu_to_be16(ibp->rvp.mkey_violations);
 630         /* P_KeyViolations are counted by hardware. */
 631         pi->pkey_violations = cpu_to_be16(ibp->rvp.pkey_violations);
 632         pi->qkey_violations = cpu_to_be16(ibp->rvp.qkey_violations);
 633
 634         pi->vl.cap = ppd->vls_supported;
 635         pi->vl.high_limit = cpu_to_be16(ibp->rvp.vl_high_limit);
 636         pi->vl.arb_high_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_CAP);
 637         pi->vl.arb_low_cap = (u8)hfi1_get_ib_cfg(ppd, HFI1_IB_CFG_VL_LOW_CAP);
 638
 639         pi->clientrereg_subnettimeout = ibp->rvp.subnet_timeout;
 640
 641         pi->port_link_mode  = cpu_to_be16(OPA_PORT_LINK_MODE_OPA << 10 |
 642                                           OPA_PORT_LINK_MODE_OPA << 5 |
 643                                           OPA_PORT_LINK_MODE_OPA);
 644
 645         pi->port_ltp_crc_mode = cpu_to_be16(ppd->port_ltp_crc_mode);
 646
 647         pi->port_mode = cpu_to_be16(
 648                                 ppd->is_active_optimize_enabled ?
 649                                         OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE : 0);
 650
 651         pi->port_packet_format.supported =
 652                 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
 653         pi->port_packet_format.enabled =
 654                 cpu_to_be16(OPA_PORT_PACKET_FORMAT_9B);
 655
 656         /* flit_control.interleave is (OPA V1, version .76):
 657          * bits         use
 658          * ----         ---
 659          * 2            res
 660          * 2            DistanceSupported
 661          * 2            DistanceEnabled
 662          * 5            MaxNextLevelTxEnabled
 663          * 5            MaxNestLevelRxSupported
 664          *
 665          * HFI supports only "distance mode 1" (see OPA V1, version .76,
 666          * section 9.6.2), so set DistanceSupported, DistanceEnabled
 667          * to 0x1.
 668          */
 669         pi->flit_control.interleave = cpu_to_be16(0x1400);
 670
 671         pi->link_down_reason = ppd->local_link_down_reason.sma;
 672         pi->neigh_link_down_reason = ppd->neigh_link_down_reason.sma;
 673         pi->port_error_action = cpu_to_be32(ppd->port_error_action);
 674         pi->mtucap = mtu_to_enum(hfi1_max_mtu, IB_MTU_4096);
 675
 676         /* 32.768 usec. response time (guessing) */
 677         pi->resptimevalue = 3;
 678
 679         pi->local_port_num = port;
 680
 681         /* buffer info for FM */
 682         pi->overall_buffer_space = cpu_to_be16(dd->link_credits);
 683
 684         pi->neigh_node_guid = cpu_to_be64(ppd->neighbor_guid);
 685         pi->neigh_port_num = ppd->neighbor_port_number;
 686         pi->port_neigh_mode =
 687                 (ppd->neighbor_type & OPA_PI_MASK_NEIGH_NODE_TYPE) |
 688                 (ppd->mgmt_allowed ? OPA_PI_MASK_NEIGH_MGMT_ALLOWED : 0) |
 689                 (ppd->neighbor_fm_security ?
 690                         OPA_PI_MASK_NEIGH_FW_AUTH_BYPASS : 0);
 691
 692         /* HFIs shall always return VL15 credits to their
 693          * neighbor in a timely manner, without any credit return pacing.
 694          */
 695         credit_rate = 0;
 696         buffer_units  = (dd->vau) & OPA_PI_MASK_BUF_UNIT_BUF_ALLOC;
 697         buffer_units |= (dd->vcu << 3) & OPA_PI_MASK_BUF_UNIT_CREDIT_ACK;
 698         buffer_units |= (credit_rate << 6) &
 699                                 OPA_PI_MASK_BUF_UNIT_VL15_CREDIT_RATE;
 700         buffer_units |= (dd->vl15_init << 11) & OPA_PI_MASK_BUF_UNIT_VL15_INIT;
 701         pi->buffer_units = cpu_to_be32(buffer_units);
 702
 703         pi->opa_cap_mask = cpu_to_be16(OPA_CAP_MASK3_IsSharedSpaceSupported);
 704
 705         /* HFI supports a replay buffer 128 LTPs in size */
 706         pi->replay_depth.buffer = 0x80;
 707         /* read the cached value of DC_LCB_STS_ROUND_TRIP_LTP_CNT */
 708         read_lcb_cache(DC_LCB_STS_ROUND_TRIP_LTP_CNT, &tmp);
 709
 710         /*
 711          * this counter is 16 bits wide, but the replay_depth.wire
 712          * variable is only 8 bits
 713          */
 714         if (tmp > 0xff)
 715                 tmp = 0xff;
 716         pi->replay_depth.wire = tmp;
 717
 718         if (resp_len)
 719                 *resp_len += sizeof(struct opa_port_info);
 720
 721         return reply((struct ib_mad_hdr *)smp);
 722 }
 723
 724 /**
 725  * get_pkeys - return the PKEY table
 726  * @dd: the hfi1_ib device
 727  * @port: the IB port number
 728  * @pkeys: the pkey table is placed here
 729  */
 730 static int get_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
 731 {
 732         struct hfi1_pportdata *ppd = dd->pport + port - 1;
 733
 734         memcpy(pkeys, ppd->pkeys, sizeof(ppd->pkeys));
 735
 736         return 0;
 737 }
 738
 739 static int __subn_get_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
 740                                     struct ib_device *ibdev, u8 port,
 741                                     u32 *resp_len)
 742 {
 743         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
 744         u32 n_blocks_req = OPA_AM_NBLK(am);
 745         u32 start_block = am & 0x7ff;
 746         __be16 *p;
 747         u16 *q;
 748         int i;
 749         u16 n_blocks_avail;
 750         unsigned npkeys = hfi1_get_npkeys(dd);
 751         size_t size;
 752
 753         if (n_blocks_req == 0) {
 754                 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
 755                         port, start_block, n_blocks_req);
 756                 smp->status |= IB_SMP_INVALID_FIELD;
 757                 return reply((struct ib_mad_hdr *)smp);
 758         }
 759
 760         n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
 761
 762         size = (n_blocks_req * OPA_PARTITION_TABLE_BLK_SIZE) * sizeof(u16);
 763
 764         if (start_block + n_blocks_req > n_blocks_avail ||
 765             n_blocks_req > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
 766                 pr_warn("OPA Get PKey AM Invalid : s 0x%x; req 0x%x; "
 767                         "avail 0x%x; blk/smp 0x%lx\n",
 768                         start_block, n_blocks_req, n_blocks_avail,
 769                         OPA_NUM_PKEY_BLOCKS_PER_SMP);
 770                 smp->status |= IB_SMP_INVALID_FIELD;
 771                 return reply((struct ib_mad_hdr *)smp);
 772         }
 773
 774         p = (__be16 *)data;
 775         q = (u16 *)data;
 776         /* get the real pkeys if we are requesting the first block */
 777         if (start_block == 0) {
 778                 get_pkeys(dd, port, q);
 779                 for (i = 0; i < npkeys; i++)
 780                         p[i] = cpu_to_be16(q[i]);
 781                 if (resp_len)
 782                         *resp_len += size;
 783         } else {
 784                 smp->status |= IB_SMP_INVALID_FIELD;
 785         }
 786         return reply((struct ib_mad_hdr *)smp);
 787 }
 788
 789 enum {
 790         HFI_TRANSITION_DISALLOWED,
 791         HFI_TRANSITION_IGNORED,
 792         HFI_TRANSITION_ALLOWED,
 793         HFI_TRANSITION_UNDEFINED,
 794 };
 795
 796 /*
 797  * Use shortened names to improve readability of
 798  * {logical,physical}_state_transitions
 799  */
 800 enum {
 801         __D = HFI_TRANSITION_DISALLOWED,
 802         __I = HFI_TRANSITION_IGNORED,
 803         __A = HFI_TRANSITION_ALLOWED,
 804         __U = HFI_TRANSITION_UNDEFINED,
 805 };
 806
 807 /*
 808  * IB_PORTPHYSSTATE_POLLING (2) through OPA_PORTPHYSSTATE_MAX (11) are
 809  * represented in physical_state_transitions.
 810  */
 811 #define __N_PHYSTATES (OPA_PORTPHYSSTATE_MAX - IB_PORTPHYSSTATE_POLLING + 1)
 812
 813 /*
 814  * Within physical_state_transitions, rows represent "old" states,
 815  * columns "new" states, and physical_state_transitions.allowed[old][new]
 816  * indicates if the transition from old state to new state is legal (see
 817  * OPAg1v1, Table 6-4).
 818  */
 819 static const struct {
 820         u8 allowed[__N_PHYSTATES][__N_PHYSTATES];
 821 } physical_state_transitions = {
 822         {
 823                 /* 2    3    4    5    6    7    8    9   10   11 */
 824         /* 2 */ { __A, __A, __D, __D, __D, __D, __D, __D, __D, __D },
 825         /* 3 */ { __A, __I, __D, __D, __D, __D, __D, __D, __D, __A },
 826         /* 4 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
 827         /* 5 */ { __A, __A, __D, __I, __D, __D, __D, __D, __D, __D },
 828         /* 6 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
 829         /* 7 */ { __D, __A, __D, __D, __D, __I, __D, __D, __D, __D },
 830         /* 8 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
 831         /* 9 */ { __I, __A, __D, __D, __D, __D, __D, __I, __D, __D },
 832         /*10 */ { __U, __U, __U, __U, __U, __U, __U, __U, __U, __U },
 833         /*11 */ { __D, __A, __D, __D, __D, __D, __D, __D, __D, __I },
 834         }
 835 };
 836
 837 /*
 838  * IB_PORT_DOWN (1) through IB_PORT_ACTIVE_DEFER (5) are represented
 839  * logical_state_transitions
 840  */
 841
 842 #define __N_LOGICAL_STATES (IB_PORT_ACTIVE_DEFER - IB_PORT_DOWN + 1)
 843
 844 /*
 845  * Within logical_state_transitions rows represent "old" states,
 846  * columns "new" states, and logical_state_transitions.allowed[old][new]
 847  * indicates if the transition from old state to new state is legal (see
 848  * OPAg1v1, Table 9-12).
 849  */
 850 static const struct {
 851         u8 allowed[__N_LOGICAL_STATES][__N_LOGICAL_STATES];
 852 } logical_state_transitions = {
 853         {
 854                 /* 1    2    3    4    5 */
 855         /* 1 */ { __I, __D, __D, __D, __U},
 856         /* 2 */ { __D, __I, __A, __D, __U},
 857         /* 3 */ { __D, __D, __I, __A, __U},
 858         /* 4 */ { __D, __D, __I, __I, __U},
 859         /* 5 */ { __U, __U, __U, __U, __U},
 860         }
 861 };
 862
 863 static int logical_transition_allowed(int old, int new)
 864 {
 865         if (old < IB_PORT_NOP || old > IB_PORT_ACTIVE_DEFER ||
 866             new < IB_PORT_NOP || new > IB_PORT_ACTIVE_DEFER) {
 867                 pr_warn("invalid logical state(s) (old %d new %d)\n",
 868                         old, new);
 869                 return HFI_TRANSITION_UNDEFINED;
 870         }
 871
 872         if (new == IB_PORT_NOP)
 873                 return HFI_TRANSITION_ALLOWED; /* always allowed */
 874
 875         /* adjust states for indexing into logical_state_transitions */
 876         old -= IB_PORT_DOWN;
 877         new -= IB_PORT_DOWN;
 878
 879         if (old < 0 || new < 0)
 880                 return HFI_TRANSITION_UNDEFINED;
 881         return logical_state_transitions.allowed[old][new];
 882 }
 883
 884 static int physical_transition_allowed(int old, int new)
 885 {
 886         if (old < IB_PORTPHYSSTATE_NOP || old > OPA_PORTPHYSSTATE_MAX ||
 887             new < IB_PORTPHYSSTATE_NOP || new > OPA_PORTPHYSSTATE_MAX) {
 888                 pr_warn("invalid physical state(s) (old %d new %d)\n",
 889                         old, new);
 890                 return HFI_TRANSITION_UNDEFINED;
 891         }
 892
 893         if (new == IB_PORTPHYSSTATE_NOP)
 894                 return HFI_TRANSITION_ALLOWED; /* always allowed */
 895
 896         /* adjust states for indexing into physical_state_transitions */
 897         old -= IB_PORTPHYSSTATE_POLLING;
 898         new -= IB_PORTPHYSSTATE_POLLING;
 899
 900         if (old < 0 || new < 0)
 901                 return HFI_TRANSITION_UNDEFINED;
 902         return physical_state_transitions.allowed[old][new];
 903 }
 904
 905 static int port_states_transition_allowed(struct hfi1_pportdata *ppd,
 906                                           u32 logical_new, u32 physical_new)
 907 {
 908         u32 physical_old = driver_physical_state(ppd);
 909         u32 logical_old = driver_logical_state(ppd);
 910         int ret, logical_allowed, physical_allowed;
 911
 912         ret = logical_transition_allowed(logical_old, logical_new);
 913         logical_allowed = ret;
 914
 915         if (ret == HFI_TRANSITION_DISALLOWED ||
 916             ret == HFI_TRANSITION_UNDEFINED) {
 917                 pr_warn("invalid logical state transition %s -> %s\n",
 918                         opa_lstate_name(logical_old),
 919                         opa_lstate_name(logical_new));
 920                 return ret;
 921         }
 922
 923         ret = physical_transition_allowed(physical_old, physical_new);
 924         physical_allowed = ret;
 925
 926         if (ret == HFI_TRANSITION_DISALLOWED ||
 927             ret == HFI_TRANSITION_UNDEFINED) {
 928                 pr_warn("invalid physical state transition %s -> %s\n",
 929                         opa_pstate_name(physical_old),
 930                         opa_pstate_name(physical_new));
 931                 return ret;
 932         }
 933
 934         if (logical_allowed == HFI_TRANSITION_IGNORED &&
 935             physical_allowed == HFI_TRANSITION_IGNORED)
 936                 return HFI_TRANSITION_IGNORED;
 937
 938         /*
 939          * A change request of Physical Port State from
 940          * 'Offline' to 'Polling' should be ignored.
 941          */
 942         if ((physical_old == OPA_PORTPHYSSTATE_OFFLINE) &&
 943             (physical_new == IB_PORTPHYSSTATE_POLLING))
 944                 return HFI_TRANSITION_IGNORED;
 945
 946         /*
 947          * Either physical_allowed or logical_allowed is
 948          * HFI_TRANSITION_ALLOWED.
 949          */
 950         return HFI_TRANSITION_ALLOWED;
 951 }
 952
 953 static int set_port_states(struct hfi1_pportdata *ppd, struct opa_smp *smp,
 954                            u32 logical_state, u32 phys_state,
 955                            int suppress_idle_sma)
 956 {
 957         struct hfi1_devdata *dd = ppd->dd;
 958         u32 link_state;
 959         int ret;
 960
 961         ret = port_states_transition_allowed(ppd, logical_state, phys_state);
 962         if (ret == HFI_TRANSITION_DISALLOWED ||
 963             ret == HFI_TRANSITION_UNDEFINED) {
 964                 /* error message emitted above */
 965                 smp->status |= IB_SMP_INVALID_FIELD;
 966                 return 0;
 967         }
 968
 969         if (ret == HFI_TRANSITION_IGNORED)
 970                 return 0;
 971
 972         if ((phys_state != IB_PORTPHYSSTATE_NOP) &&
 973             !(logical_state == IB_PORT_DOWN ||
 974               logical_state == IB_PORT_NOP)){
 975                 pr_warn("SubnSet(OPA_PortInfo) port state invalid: logical_state 0x%x physical_state 0x%x\n",
 976                         logical_state, phys_state);
 977                 smp->status |= IB_SMP_INVALID_FIELD;
 978         }
 979
 980         /*
 981          * Logical state changes are summarized in OPAv1g1 spec.,
 982          * Table 9-12; physical state changes are summarized in
 983          * OPAv1g1 spec., Table 6.4.
 984          */
 985         switch (logical_state) {
 986         case IB_PORT_NOP:
 987                 if (phys_state == IB_PORTPHYSSTATE_NOP)
 988                         break;
 989                 /* FALLTHROUGH */
 990         case IB_PORT_DOWN:
 991                 if (phys_state == IB_PORTPHYSSTATE_NOP) {
 992                         link_state = HLS_DN_DOWNDEF;
 993                 } else if (phys_state == IB_PORTPHYSSTATE_POLLING) {
 994                         link_state = HLS_DN_POLL;
 995                         set_link_down_reason(ppd, OPA_LINKDOWN_REASON_FM_BOUNCE,
 996                                              0, OPA_LINKDOWN_REASON_FM_BOUNCE);
 997                 } else if (phys_state == IB_PORTPHYSSTATE_DISABLED) {
 998                         link_state = HLS_DN_DISABLE;
 999                 } else {
1000                         pr_warn("SubnSet(OPA_PortInfo) invalid physical state 0x%x\n",
1001                                 phys_state);
1002                         smp->status |= IB_SMP_INVALID_FIELD;
1003                         break;
1004                 }
1005
1006                 if ((link_state == HLS_DN_POLL ||
1007                      link_state == HLS_DN_DOWNDEF)) {
1008                         /*
1009                          * Going to poll.  No matter what the current state,
1010                          * always move offline first, then tune and start the
1011                          * link.  This correctly handles a FM link bounce and
1012                          * a link enable.  Going offline is a no-op if already
1013                          * offline.
1014                          */
1015                         set_link_state(ppd, HLS_DN_OFFLINE);
1016                         tune_serdes(ppd);
1017                         start_link(ppd);
1018                 } else {
1019                         set_link_state(ppd, link_state);
1020                 }
1021                 if (link_state == HLS_DN_DISABLE &&
1022                     (ppd->offline_disabled_reason >
1023                      HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED) ||
1024                      ppd->offline_disabled_reason ==
1025                      HFI1_ODR_MASK(OPA_LINKDOWN_REASON_NONE)))
1026                         ppd->offline_disabled_reason =
1027                         HFI1_ODR_MASK(OPA_LINKDOWN_REASON_SMA_DISABLED);
1028                 /*
1029                  * Don't send a reply if the response would be sent
1030                  * through the disabled port.
1031                  */
1032                 if (link_state == HLS_DN_DISABLE && smp->hop_cnt)
1033                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1034                 break;
1035         case IB_PORT_ARMED:
1036                 ret = set_link_state(ppd, HLS_UP_ARMED);
1037                 if ((ret == 0) && (suppress_idle_sma == 0))
1038                         send_idle_sma(dd, SMA_IDLE_ARM);
1039                 break;
1040         case IB_PORT_ACTIVE:
1041                 if (ppd->neighbor_normal) {
1042                         ret = set_link_state(ppd, HLS_UP_ACTIVE);
1043                         if (ret == 0)
1044                                 send_idle_sma(dd, SMA_IDLE_ACTIVE);
1045                 } else {
1046                         pr_warn("SubnSet(OPA_PortInfo) Cannot move to Active with NeighborNormal 0\n");
1047                         smp->status |= IB_SMP_INVALID_FIELD;
1048                 }
1049                 break;
1050         default:
1051                 pr_warn("SubnSet(OPA_PortInfo) invalid logical state 0x%x\n",
1052                         logical_state);
1053                 smp->status |= IB_SMP_INVALID_FIELD;
1054         }
1055
1056         return 0;
1057 }
1058
1059 /**
1060  * subn_set_opa_portinfo - set port information
1061  * @smp: the incoming SM packet
1062  * @ibdev: the infiniband device
1063  * @port: the port on the device
1064  *
1065  */
1066 static int __subn_set_opa_portinfo(struct opa_smp *smp, u32 am, u8 *data,
1067                                    struct ib_device *ibdev, u8 port,
1068                                    u32 *resp_len)
1069 {
1070         struct opa_port_info *pi = (struct opa_port_info *)data;
1071         struct ib_event event;
1072         struct hfi1_devdata *dd;
1073         struct hfi1_pportdata *ppd;
1074         struct hfi1_ibport *ibp;
1075         u8 clientrereg;
1076         unsigned long flags;
1077         u32 smlid, opa_lid; /* tmp vars to hold LID values */
1078         u16 lid;
1079         u8 ls_old, ls_new, ps_new;
1080         u8 vls;
1081         u8 msl;
1082         u8 crc_enabled;
1083         u16 lse, lwe, mtu;
1084         u32 num_ports = OPA_AM_NPORT(am);
1085         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1086         int ret, i, invalid = 0, call_set_mtu = 0;
1087         int call_link_downgrade_policy = 0;
1088
1089         if (num_ports != 1) {
1090                 smp->status |= IB_SMP_INVALID_FIELD;
1091                 return reply((struct ib_mad_hdr *)smp);
1092         }
1093
1094         opa_lid = be32_to_cpu(pi->lid);
1095         if (opa_lid & 0xFFFF0000) {
1096                 pr_warn("OPA_PortInfo lid out of range: %X\n", opa_lid);
1097                 smp->status |= IB_SMP_INVALID_FIELD;
1098                 goto get_only;
1099         }
1100
1101         lid = (u16)(opa_lid & 0x0000FFFF);
1102
1103         smlid = be32_to_cpu(pi->sm_lid);
1104         if (smlid & 0xFFFF0000) {
1105                 pr_warn("OPA_PortInfo SM lid out of range: %X\n", smlid);
1106                 smp->status |= IB_SMP_INVALID_FIELD;
1107                 goto get_only;
1108         }
1109         smlid &= 0x0000FFFF;
1110
1111         clientrereg = (pi->clientrereg_subnettimeout &
1112                         OPA_PI_MASK_CLIENT_REREGISTER);
1113
1114         dd = dd_from_ibdev(ibdev);
1115         /* IB numbers ports from 1, hw from 0 */
1116         ppd = dd->pport + (port - 1);
1117         ibp = &ppd->ibport_data;
1118         event.device = ibdev;
1119         event.element.port_num = port;
1120
1121         ls_old = driver_lstate(ppd);
1122
1123         ibp->rvp.mkey = pi->mkey;
1124         ibp->rvp.gid_prefix = pi->subnet_prefix;
1125         ibp->rvp.mkey_lease_period = be16_to_cpu(pi->mkey_lease_period);
1126
1127         /* Must be a valid unicast LID address. */
1128         if ((lid == 0 && ls_old > IB_PORT_INIT) ||
1129             lid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
1130                 smp->status |= IB_SMP_INVALID_FIELD;
1131                 pr_warn("SubnSet(OPA_PortInfo) lid invalid 0x%x\n",
1132                         lid);
1133         } else if (ppd->lid != lid ||
1134                  ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC)) {
1135                 if (ppd->lid != lid)
1136                         hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LID_CHANGE_BIT);
1137                 if (ppd->lmc != (pi->mkeyprotect_lmc & OPA_PI_MASK_LMC))
1138                         hfi1_set_uevent_bits(ppd, _HFI1_EVENT_LMC_CHANGE_BIT);
1139                 hfi1_set_lid(ppd, lid, pi->mkeyprotect_lmc & OPA_PI_MASK_LMC);
1140                 event.event = IB_EVENT_LID_CHANGE;
1141                 ib_dispatch_event(&event);
1142         }
1143
1144         msl = pi->smsl & OPA_PI_MASK_SMSL;
1145         if (pi->partenforce_filterraw & OPA_PI_MASK_LINKINIT_REASON)
1146                 ppd->linkinit_reason =
1147                         (pi->partenforce_filterraw &
1148                          OPA_PI_MASK_LINKINIT_REASON);
1149         /* enable/disable SW pkey checking as per FM control */
1150         if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_IN)
1151                 ppd->part_enforce |= HFI1_PART_ENFORCE_IN;
1152         else
1153                 ppd->part_enforce &= ~HFI1_PART_ENFORCE_IN;
1154
1155         if (pi->partenforce_filterraw & OPA_PI_MASK_PARTITION_ENFORCE_OUT)
1156                 ppd->part_enforce |= HFI1_PART_ENFORCE_OUT;
1157         else
1158                 ppd->part_enforce &= ~HFI1_PART_ENFORCE_OUT;
1159
1160         /* Must be a valid unicast LID address. */
1161         if ((smlid == 0 && ls_old > IB_PORT_INIT) ||
1162             smlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) {
1163                 smp->status |= IB_SMP_INVALID_FIELD;
1164                 pr_warn("SubnSet(OPA_PortInfo) smlid invalid 0x%x\n", smlid);
1165         } else if (smlid != ibp->rvp.sm_lid || msl != ibp->rvp.sm_sl) {
1166                 pr_warn("SubnSet(OPA_PortInfo) smlid 0x%x\n", smlid);
1167                 spin_lock_irqsave(&ibp->rvp.lock, flags);
1168                 if (ibp->rvp.sm_ah) {
1169                         if (smlid != ibp->rvp.sm_lid)
1170                                 ibp->rvp.sm_ah->attr.dlid = smlid;
1171                         if (msl != ibp->rvp.sm_sl)
1172                                 ibp->rvp.sm_ah->attr.sl = msl;
1173                 }
1174                 spin_unlock_irqrestore(&ibp->rvp.lock, flags);
1175                 if (smlid != ibp->rvp.sm_lid)
1176                         ibp->rvp.sm_lid = smlid;
1177                 if (msl != ibp->rvp.sm_sl)
1178                         ibp->rvp.sm_sl = msl;
1179                 event.event = IB_EVENT_SM_CHANGE;
1180                 ib_dispatch_event(&event);
1181         }
1182
1183         if (pi->link_down_reason == 0) {
1184                 ppd->local_link_down_reason.sma = 0;
1185                 ppd->local_link_down_reason.latest = 0;
1186         }
1187
1188         if (pi->neigh_link_down_reason == 0) {
1189                 ppd->neigh_link_down_reason.sma = 0;
1190                 ppd->neigh_link_down_reason.latest = 0;
1191         }
1192
1193         ppd->sm_trap_qp = be32_to_cpu(pi->sm_trap_qp);
1194         ppd->sa_qp = be32_to_cpu(pi->sa_qp);
1195
1196         ppd->port_error_action = be32_to_cpu(pi->port_error_action);
1197         lwe = be16_to_cpu(pi->link_width.enabled);
1198         if (lwe) {
1199                 if (lwe == OPA_LINK_WIDTH_RESET ||
1200                     lwe == OPA_LINK_WIDTH_RESET_OLD)
1201                         set_link_width_enabled(ppd, ppd->link_width_supported);
1202                 else if ((lwe & ~ppd->link_width_supported) == 0)
1203                         set_link_width_enabled(ppd, lwe);
1204                 else
1205                         smp->status |= IB_SMP_INVALID_FIELD;
1206         }
1207         lwe = be16_to_cpu(pi->link_width_downgrade.enabled);
1208         /* LWD.E is always applied - 0 means "disabled" */
1209         if (lwe == OPA_LINK_WIDTH_RESET ||
1210             lwe == OPA_LINK_WIDTH_RESET_OLD) {
1211                 set_link_width_downgrade_enabled(ppd,
1212                                                  ppd->
1213                                                  link_width_downgrade_supported
1214                                                  );
1215         } else if ((lwe & ~ppd->link_width_downgrade_supported) == 0) {
1216                 /* only set and apply if something changed */
1217                 if (lwe != ppd->link_width_downgrade_enabled) {
1218                         set_link_width_downgrade_enabled(ppd, lwe);
1219                         call_link_downgrade_policy = 1;
1220                 }
1221         } else {
1222                 smp->status |= IB_SMP_INVALID_FIELD;
1223         }
1224         lse = be16_to_cpu(pi->link_speed.enabled);
1225         if (lse) {
1226                 if (lse & be16_to_cpu(pi->link_speed.supported))
1227                         set_link_speed_enabled(ppd, lse);
1228                 else
1229                         smp->status |= IB_SMP_INVALID_FIELD;
1230         }
1231
1232         ibp->rvp.mkeyprot =
1233                 (pi->mkeyprotect_lmc & OPA_PI_MASK_MKEY_PROT_BIT) >> 6;
1234         ibp->rvp.vl_high_limit = be16_to_cpu(pi->vl.high_limit) & 0xFF;
1235         (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_VL_HIGH_LIMIT,
1236                                     ibp->rvp.vl_high_limit);
1237
1238         if (ppd->vls_supported / 2 > ARRAY_SIZE(pi->neigh_mtu.pvlx_to_mtu) ||
1239             ppd->vls_supported > ARRAY_SIZE(dd->vld)) {
1240                 smp->status |= IB_SMP_INVALID_FIELD;
1241                 return reply((struct ib_mad_hdr *)smp);
1242         }
1243         for (i = 0; i < ppd->vls_supported; i++) {
1244                 if ((i % 2) == 0)
1245                         mtu = enum_to_mtu((pi->neigh_mtu.pvlx_to_mtu[i / 2] >>
1246                                            4) & 0xF);
1247                 else
1248                         mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[i / 2] &
1249                                           0xF);
1250                 if (mtu == 0xffff) {
1251                         pr_warn("SubnSet(OPA_PortInfo) mtu invalid %d (0x%x)\n",
1252                                 mtu,
1253                                 (pi->neigh_mtu.pvlx_to_mtu[0] >> 4) & 0xF);
1254                         smp->status |= IB_SMP_INVALID_FIELD;
1255                         mtu = hfi1_max_mtu; /* use a valid MTU */
1256                 }
1257                 if (dd->vld[i].mtu != mtu) {
1258                         dd_dev_info(dd,
1259                                     "MTU change on vl %d from %d to %d\n",
1260                                     i, dd->vld[i].mtu, mtu);
1261                         dd->vld[i].mtu = mtu;
1262                         call_set_mtu++;
1263                 }
1264         }
1265         /* As per OPAV1 spec: VL15 must support and be configured
1266          * for operation with a 2048 or larger MTU.
1267          */
1268         mtu = enum_to_mtu(pi->neigh_mtu.pvlx_to_mtu[15 / 2] & 0xF);
1269         if (mtu < 2048 || mtu == 0xffff)
1270                 mtu = 2048;
1271         if (dd->vld[15].mtu != mtu) {
1272                 dd_dev_info(dd,
1273                             "MTU change on vl 15 from %d to %d\n",
1274                             dd->vld[15].mtu, mtu);
1275                 dd->vld[15].mtu = mtu;
1276                 call_set_mtu++;
1277         }
1278         if (call_set_mtu)
1279                 set_mtu(ppd);
1280
1281         /* Set operational VLs */
1282         vls = pi->operational_vls & OPA_PI_MASK_OPERATIONAL_VL;
1283         if (vls) {
1284                 if (vls > ppd->vls_supported) {
1285                         pr_warn("SubnSet(OPA_PortInfo) VL's supported invalid %d\n",
1286                                 pi->operational_vls);
1287                         smp->status |= IB_SMP_INVALID_FIELD;
1288                 } else {
1289                         if (hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_OP_VLS,
1290                                             vls) == -EINVAL)
1291                                 smp->status |= IB_SMP_INVALID_FIELD;
1292                 }
1293         }
1294
1295         if (pi->mkey_violations == 0)
1296                 ibp->rvp.mkey_violations = 0;
1297
1298         if (pi->pkey_violations == 0)
1299                 ibp->rvp.pkey_violations = 0;
1300
1301         if (pi->qkey_violations == 0)
1302                 ibp->rvp.qkey_violations = 0;
1303
1304         ibp->rvp.subnet_timeout =
1305                 pi->clientrereg_subnettimeout & OPA_PI_MASK_SUBNET_TIMEOUT;
1306
1307         crc_enabled = be16_to_cpu(pi->port_ltp_crc_mode);
1308         crc_enabled >>= 4;
1309         crc_enabled &= 0xf;
1310
1311         if (crc_enabled != 0)
1312                 ppd->port_crc_mode_enabled = port_ltp_to_cap(crc_enabled);
1313
1314         ppd->is_active_optimize_enabled =
1315                         !!(be16_to_cpu(pi->port_mode)
1316                                         & OPA_PI_MASK_PORT_ACTIVE_OPTOMIZE);
1317
1318         ls_new = pi->port_states.portphysstate_portstate &
1319                         OPA_PI_MASK_PORT_STATE;
1320         ps_new = (pi->port_states.portphysstate_portstate &
1321                         OPA_PI_MASK_PORT_PHYSICAL_STATE) >> 4;
1322
1323         if (ls_old == IB_PORT_INIT) {
1324                 if (start_of_sm_config) {
1325                         if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1326                                 ppd->is_sm_config_started = 1;
1327                 } else if (ls_new == IB_PORT_ARMED) {
1328                         if (ppd->is_sm_config_started == 0)
1329                                 invalid = 1;
1330                 }
1331         }
1332
1333         /* Handle CLIENT_REREGISTER event b/c SM asked us for it */
1334         if (clientrereg) {
1335                 event.event = IB_EVENT_CLIENT_REREGISTER;
1336                 ib_dispatch_event(&event);
1337         }
1338
1339         /*
1340          * Do the port state change now that the other link parameters
1341          * have been set.
1342          * Changing the port physical state only makes sense if the link
1343          * is down or is being set to down.
1344          */
1345
1346         ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1347         if (ret)
1348                 return ret;
1349
1350         ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1351
1352         /* restore re-reg bit per o14-12.2.1 */
1353         pi->clientrereg_subnettimeout |= clientrereg;
1354
1355         /*
1356          * Apply the new link downgrade policy.  This may result in a link
1357          * bounce.  Do this after everything else so things are settled.
1358          * Possible problem: if setting the port state above fails, then
1359          * the policy change is not applied.
1360          */
1361         if (call_link_downgrade_policy)
1362                 apply_link_downgrade_policy(ppd, 0);
1363
1364         return ret;
1365
1366 get_only:
1367         return __subn_get_opa_portinfo(smp, am, data, ibdev, port, resp_len);
1368 }
1369
1370 /**
1371  * set_pkeys - set the PKEY table for ctxt 0
1372  * @dd: the hfi1_ib device
1373  * @port: the IB port number
1374  * @pkeys: the PKEY table
1375  */
1376 static int set_pkeys(struct hfi1_devdata *dd, u8 port, u16 *pkeys)
1377 {
1378         struct hfi1_pportdata *ppd;
1379         int i;
1380         int changed = 0;
1381         int update_includes_mgmt_partition = 0;
1382
1383         /*
1384          * IB port one/two always maps to context zero/one,
1385          * always a kernel context, no locking needed
1386          * If we get here with ppd setup, no need to check
1387          * that rcd is valid.
1388          */
1389         ppd = dd->pport + (port - 1);
1390         /*
1391          * If the update does not include the management pkey, don't do it.
1392          */
1393         for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1394                 if (pkeys[i] == LIM_MGMT_P_KEY) {
1395                         update_includes_mgmt_partition = 1;
1396                         break;
1397                 }
1398         }
1399
1400         if (!update_includes_mgmt_partition)
1401                 return 1;
1402
1403         for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
1404                 u16 key = pkeys[i];
1405                 u16 okey = ppd->pkeys[i];
1406
1407                 if (key == okey)
1408                         continue;
1409                 /*
1410                  * Don't update pkeys[2], if an HFI port without MgmtAllowed
1411                  * by neighbor is a switch.
1412                  */
1413                 if (i == 2 && !ppd->mgmt_allowed && ppd->neighbor_type == 1)
1414                         continue;
1415                 /*
1416                  * The SM gives us the complete PKey table. We have
1417                  * to ensure that we put the PKeys in the matching
1418                  * slots.
1419                  */
1420                 ppd->pkeys[i] = key;
1421                 changed = 1;
1422         }
1423
1424         if (changed) {
1425                 (void)hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_PKEYS, 0);
1426                 hfi1_event_pkey_change(dd, port);
1427         }
1428
1429         return 0;
1430 }
1431
1432 static int __subn_set_opa_pkeytable(struct opa_smp *smp, u32 am, u8 *data,
1433                                     struct ib_device *ibdev, u8 port,
1434                                     u32 *resp_len)
1435 {
1436         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1437         u32 n_blocks_sent = OPA_AM_NBLK(am);
1438         u32 start_block = am & 0x7ff;
1439         u16 *p = (u16 *)data;
1440         __be16 *q = (__be16 *)data;
1441         int i;
1442         u16 n_blocks_avail;
1443         unsigned npkeys = hfi1_get_npkeys(dd);
1444
1445         if (n_blocks_sent == 0) {
1446                 pr_warn("OPA Get PKey AM Invalid : P = %d; B = 0x%x; N = 0x%x\n",
1447                         port, start_block, n_blocks_sent);
1448                 smp->status |= IB_SMP_INVALID_FIELD;
1449                 return reply((struct ib_mad_hdr *)smp);
1450         }
1451
1452         n_blocks_avail = (u16)(npkeys / OPA_PARTITION_TABLE_BLK_SIZE) + 1;
1453
1454         if (start_block + n_blocks_sent > n_blocks_avail ||
1455             n_blocks_sent > OPA_NUM_PKEY_BLOCKS_PER_SMP) {
1456                 pr_warn("OPA Set PKey AM Invalid : s 0x%x; req 0x%x; avail 0x%x; blk/smp 0x%lx\n",
1457                         start_block, n_blocks_sent, n_blocks_avail,
1458                         OPA_NUM_PKEY_BLOCKS_PER_SMP);
1459                 smp->status |= IB_SMP_INVALID_FIELD;
1460                 return reply((struct ib_mad_hdr *)smp);
1461         }
1462
1463         for (i = 0; i < n_blocks_sent * OPA_PARTITION_TABLE_BLK_SIZE; i++)
1464                 p[i] = be16_to_cpu(q[i]);
1465
1466         if (start_block == 0 && set_pkeys(dd, port, p) != 0) {
1467                 smp->status |= IB_SMP_INVALID_FIELD;
1468                 return reply((struct ib_mad_hdr *)smp);
1469         }
1470
1471         return __subn_get_opa_pkeytable(smp, am, data, ibdev, port, resp_len);
1472 }
1473
1474 static int get_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1475 {
1476         u64 *val = data;
1477
1478         *val++ = read_csr(dd, SEND_SC2VLT0);
1479         *val++ = read_csr(dd, SEND_SC2VLT1);
1480         *val++ = read_csr(dd, SEND_SC2VLT2);
1481         *val++ = read_csr(dd, SEND_SC2VLT3);
1482         return 0;
1483 }
1484
1485 #define ILLEGAL_VL 12
1486 /*
1487  * filter_sc2vlt changes mappings to VL15 to ILLEGAL_VL (except
1488  * for SC15, which must map to VL15). If we don't remap things this
1489  * way it is possible for VL15 counters to increment when we try to
1490  * send on a SC which is mapped to an invalid VL.
1491  */
1492 static void filter_sc2vlt(void *data)
1493 {
1494         int i;
1495         u8 *pd = data;
1496
1497         for (i = 0; i < OPA_MAX_SCS; i++) {
1498                 if (i == 15)
1499                         continue;
1500                 if ((pd[i] & 0x1f) == 0xf)
1501                         pd[i] = ILLEGAL_VL;
1502         }
1503 }
1504
1505 static int set_sc2vlt_tables(struct hfi1_devdata *dd, void *data)
1506 {
1507         u64 *val = data;
1508
1509         filter_sc2vlt(data);
1510
1511         write_csr(dd, SEND_SC2VLT0, *val++);
1512         write_csr(dd, SEND_SC2VLT1, *val++);
1513         write_csr(dd, SEND_SC2VLT2, *val++);
1514         write_csr(dd, SEND_SC2VLT3, *val++);
1515         write_seqlock_irq(&dd->sc2vl_lock);
1516         memcpy(dd->sc2vl, data, sizeof(dd->sc2vl));
1517         write_sequnlock_irq(&dd->sc2vl_lock);
1518         return 0;
1519 }
1520
1521 static int __subn_get_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1522                                    struct ib_device *ibdev, u8 port,
1523                                    u32 *resp_len)
1524 {
1525         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1526         u8 *p = data;
1527         size_t size = ARRAY_SIZE(ibp->sl_to_sc); /* == 32 */
1528         unsigned i;
1529
1530         if (am) {
1531                 smp->status |= IB_SMP_INVALID_FIELD;
1532                 return reply((struct ib_mad_hdr *)smp);
1533         }
1534
1535         for (i = 0; i < ARRAY_SIZE(ibp->sl_to_sc); i++)
1536                 *p++ = ibp->sl_to_sc[i];
1537
1538         if (resp_len)
1539                 *resp_len += size;
1540
1541         return reply((struct ib_mad_hdr *)smp);
1542 }
1543
1544 static int __subn_set_opa_sl_to_sc(struct opa_smp *smp, u32 am, u8 *data,
1545                                    struct ib_device *ibdev, u8 port,
1546                                    u32 *resp_len)
1547 {
1548         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1549         u8 *p = data;
1550         int i;
1551         u8 sc;
1552
1553         if (am) {
1554                 smp->status |= IB_SMP_INVALID_FIELD;
1555                 return reply((struct ib_mad_hdr *)smp);
1556         }
1557
1558         for (i = 0; i <  ARRAY_SIZE(ibp->sl_to_sc); i++) {
1559                 sc = *p++;
1560                 if (ibp->sl_to_sc[i] != sc) {
1561                         ibp->sl_to_sc[i] = sc;
1562
1563                         /* Put all stale qps into error state */
1564                         hfi1_error_port_qps(ibp, i);
1565                 }
1566         }
1567
1568         return __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port, resp_len);
1569 }
1570
1571 static int __subn_get_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1572                                    struct ib_device *ibdev, u8 port,
1573                                    u32 *resp_len)
1574 {
1575         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1576         u8 *p = data;
1577         size_t size = ARRAY_SIZE(ibp->sc_to_sl); /* == 32 */
1578         unsigned i;
1579
1580         if (am) {
1581                 smp->status |= IB_SMP_INVALID_FIELD;
1582                 return reply((struct ib_mad_hdr *)smp);
1583         }
1584
1585         for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1586                 *p++ = ibp->sc_to_sl[i];
1587
1588         if (resp_len)
1589                 *resp_len += size;
1590
1591         return reply((struct ib_mad_hdr *)smp);
1592 }
1593
1594 static int __subn_set_opa_sc_to_sl(struct opa_smp *smp, u32 am, u8 *data,
1595                                    struct ib_device *ibdev, u8 port,
1596                                    u32 *resp_len)
1597 {
1598         struct hfi1_ibport *ibp = to_iport(ibdev, port);
1599         u8 *p = data;
1600         int i;
1601
1602         if (am) {
1603                 smp->status |= IB_SMP_INVALID_FIELD;
1604                 return reply((struct ib_mad_hdr *)smp);
1605         }
1606
1607         for (i = 0; i < ARRAY_SIZE(ibp->sc_to_sl); i++)
1608                 ibp->sc_to_sl[i] = *p++;
1609
1610         return __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port, resp_len);
1611 }
1612
1613 static int __subn_get_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1614                                     struct ib_device *ibdev, u8 port,
1615                                     u32 *resp_len)
1616 {
1617         u32 n_blocks = OPA_AM_NBLK(am);
1618         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1619         void *vp = (void *)data;
1620         size_t size = 4 * sizeof(u64);
1621
1622         if (n_blocks != 1) {
1623                 smp->status |= IB_SMP_INVALID_FIELD;
1624                 return reply((struct ib_mad_hdr *)smp);
1625         }
1626
1627         get_sc2vlt_tables(dd, vp);
1628
1629         if (resp_len)
1630                 *resp_len += size;
1631
1632         return reply((struct ib_mad_hdr *)smp);
1633 }
1634
1635 static int __subn_set_opa_sc_to_vlt(struct opa_smp *smp, u32 am, u8 *data,
1636                                     struct ib_device *ibdev, u8 port,
1637                                     u32 *resp_len)
1638 {
1639         u32 n_blocks = OPA_AM_NBLK(am);
1640         int async_update = OPA_AM_ASYNC(am);
1641         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1642         void *vp = (void *)data;
1643         struct hfi1_pportdata *ppd;
1644         int lstate;
1645
1646         if (n_blocks != 1 || async_update) {
1647                 smp->status |= IB_SMP_INVALID_FIELD;
1648                 return reply((struct ib_mad_hdr *)smp);
1649         }
1650
1651         /* IB numbers ports from 1, hw from 0 */
1652         ppd = dd->pport + (port - 1);
1653         lstate = driver_lstate(ppd);
1654         /*
1655          * it's known that async_update is 0 by this point, but include
1656          * the explicit check for clarity
1657          */
1658         if (!async_update &&
1659             (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE)) {
1660                 smp->status |= IB_SMP_INVALID_FIELD;
1661                 return reply((struct ib_mad_hdr *)smp);
1662         }
1663
1664         set_sc2vlt_tables(dd, vp);
1665
1666         return __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port, resp_len);
1667 }
1668
1669 static int __subn_get_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1670                                      struct ib_device *ibdev, u8 port,
1671                                      u32 *resp_len)
1672 {
1673         u32 n_blocks = OPA_AM_NPORT(am);
1674         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1675         struct hfi1_pportdata *ppd;
1676         void *vp = (void *)data;
1677         int size;
1678
1679         if (n_blocks != 1) {
1680                 smp->status |= IB_SMP_INVALID_FIELD;
1681                 return reply((struct ib_mad_hdr *)smp);
1682         }
1683
1684         ppd = dd->pport + (port - 1);
1685
1686         size = fm_get_table(ppd, FM_TBL_SC2VLNT, vp);
1687
1688         if (resp_len)
1689                 *resp_len += size;
1690
1691         return reply((struct ib_mad_hdr *)smp);
1692 }
1693
1694 static int __subn_set_opa_sc_to_vlnt(struct opa_smp *smp, u32 am, u8 *data,
1695                                      struct ib_device *ibdev, u8 port,
1696                                      u32 *resp_len)
1697 {
1698         u32 n_blocks = OPA_AM_NPORT(am);
1699         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1700         struct hfi1_pportdata *ppd;
1701         void *vp = (void *)data;
1702         int lstate;
1703
1704         if (n_blocks != 1) {
1705                 smp->status |= IB_SMP_INVALID_FIELD;
1706                 return reply((struct ib_mad_hdr *)smp);
1707         }
1708
1709         /* IB numbers ports from 1, hw from 0 */
1710         ppd = dd->pport + (port - 1);
1711         lstate = driver_lstate(ppd);
1712         if (lstate == IB_PORT_ARMED || lstate == IB_PORT_ACTIVE) {
1713                 smp->status |= IB_SMP_INVALID_FIELD;
1714                 return reply((struct ib_mad_hdr *)smp);
1715         }
1716
1717         ppd = dd->pport + (port - 1);
1718
1719         fm_set_table(ppd, FM_TBL_SC2VLNT, vp);
1720
1721         return __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
1722                                          resp_len);
1723 }
1724
1725 static int __subn_get_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1726                               struct ib_device *ibdev, u8 port,
1727                               u32 *resp_len)
1728 {
1729         u32 nports = OPA_AM_NPORT(am);
1730         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1731         u32 lstate;
1732         struct hfi1_ibport *ibp;
1733         struct hfi1_pportdata *ppd;
1734         struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
1735
1736         if (nports != 1) {
1737                 smp->status |= IB_SMP_INVALID_FIELD;
1738                 return reply((struct ib_mad_hdr *)smp);
1739         }
1740
1741         ibp = to_iport(ibdev, port);
1742         ppd = ppd_from_ibp(ibp);
1743
1744         lstate = driver_lstate(ppd);
1745
1746         if (start_of_sm_config && (lstate == IB_PORT_INIT))
1747                 ppd->is_sm_config_started = 1;
1748
1749         psi->port_states.ledenable_offlinereason = ppd->neighbor_normal << 4;
1750         psi->port_states.ledenable_offlinereason |=
1751                 ppd->is_sm_config_started << 5;
1752         psi->port_states.ledenable_offlinereason |=
1753                 ppd->offline_disabled_reason;
1754
1755         psi->port_states.portphysstate_portstate =
1756                 (hfi1_ibphys_portstate(ppd) << 4) | (lstate & 0xf);
1757         psi->link_width_downgrade_tx_active =
1758                 cpu_to_be16(ppd->link_width_downgrade_tx_active);
1759         psi->link_width_downgrade_rx_active =
1760                 cpu_to_be16(ppd->link_width_downgrade_rx_active);
1761         if (resp_len)
1762                 *resp_len += sizeof(struct opa_port_state_info);
1763
1764         return reply((struct ib_mad_hdr *)smp);
1765 }
1766
1767 static int __subn_set_opa_psi(struct opa_smp *smp, u32 am, u8 *data,
1768                               struct ib_device *ibdev, u8 port,
1769                               u32 *resp_len)
1770 {
1771         u32 nports = OPA_AM_NPORT(am);
1772         u32 start_of_sm_config = OPA_AM_START_SM_CFG(am);
1773         u32 ls_old;
1774         u8 ls_new, ps_new;
1775         struct hfi1_ibport *ibp;
1776         struct hfi1_pportdata *ppd;
1777         struct opa_port_state_info *psi = (struct opa_port_state_info *)data;
1778         int ret, invalid = 0;
1779
1780         if (nports != 1) {
1781                 smp->status |= IB_SMP_INVALID_FIELD;
1782                 return reply((struct ib_mad_hdr *)smp);
1783         }
1784
1785         ibp = to_iport(ibdev, port);
1786         ppd = ppd_from_ibp(ibp);
1787
1788         ls_old = driver_lstate(ppd);
1789
1790         ls_new = port_states_to_logical_state(&psi->port_states);
1791         ps_new = port_states_to_phys_state(&psi->port_states);
1792
1793         if (ls_old == IB_PORT_INIT) {
1794                 if (start_of_sm_config) {
1795                         if (ls_new == ls_old || (ls_new == IB_PORT_ARMED))
1796                                 ppd->is_sm_config_started = 1;
1797                 } else if (ls_new == IB_PORT_ARMED) {
1798                         if (ppd->is_sm_config_started == 0)
1799                                 invalid = 1;
1800                 }
1801         }
1802
1803         ret = set_port_states(ppd, smp, ls_new, ps_new, invalid);
1804         if (ret)
1805                 return ret;
1806
1807         if (invalid)
1808                 smp->status |= IB_SMP_INVALID_FIELD;
1809
1810         return __subn_get_opa_psi(smp, am, data, ibdev, port, resp_len);
1811 }
1812
1813 static int __subn_get_opa_cable_info(struct opa_smp *smp, u32 am, u8 *data,
1814                                      struct ib_device *ibdev, u8 port,
1815                                      u32 *resp_len)
1816 {
1817         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1818         u32 addr = OPA_AM_CI_ADDR(am);
1819         u32 len = OPA_AM_CI_LEN(am) + 1;
1820         int ret;
1821
1822         if (dd->pport->port_type != PORT_TYPE_QSFP) {
1823                 smp->status |= IB_SMP_INVALID_FIELD;
1824                 return reply((struct ib_mad_hdr *)smp);
1825         }
1826
1827 #define __CI_PAGE_SIZE BIT(7) /* 128 bytes */
1828 #define __CI_PAGE_MASK ~(__CI_PAGE_SIZE - 1)
1829 #define __CI_PAGE_NUM(a) ((a) & __CI_PAGE_MASK)
1830
1831         /*
1832          * check that addr is within spec, and
1833          * addr and (addr + len - 1) are on the same "page"
1834          */
1835         if (addr >= 4096 ||
1836             (__CI_PAGE_NUM(addr) != __CI_PAGE_NUM(addr + len - 1))) {
1837                 smp->status |= IB_SMP_INVALID_FIELD;
1838                 return reply((struct ib_mad_hdr *)smp);
1839         }
1840
1841         ret = get_cable_info(dd, port, addr, len, data);
1842
1843         if (ret == -ENODEV) {
1844                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1845                 return reply((struct ib_mad_hdr *)smp);
1846         }
1847
1848         /* The address range for the CableInfo SMA query is wider than the
1849          * memory available on the QSFP cable. We want to return a valid
1850          * response, albeit zeroed out, for address ranges beyond available
1851          * memory but that are within the CableInfo query spec
1852          */
1853         if (ret < 0 && ret != -ERANGE) {
1854                 smp->status |= IB_SMP_INVALID_FIELD;
1855                 return reply((struct ib_mad_hdr *)smp);
1856         }
1857
1858         if (resp_len)
1859                 *resp_len += len;
1860
1861         return reply((struct ib_mad_hdr *)smp);
1862 }
1863
1864 static int __subn_get_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1865                               struct ib_device *ibdev, u8 port, u32 *resp_len)
1866 {
1867         u32 num_ports = OPA_AM_NPORT(am);
1868         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1869         struct hfi1_pportdata *ppd;
1870         struct buffer_control *p = (struct buffer_control *)data;
1871         int size;
1872
1873         if (num_ports != 1) {
1874                 smp->status |= IB_SMP_INVALID_FIELD;
1875                 return reply((struct ib_mad_hdr *)smp);
1876         }
1877
1878         ppd = dd->pport + (port - 1);
1879         size = fm_get_table(ppd, FM_TBL_BUFFER_CONTROL, p);
1880         trace_bct_get(dd, p);
1881         if (resp_len)
1882                 *resp_len += size;
1883
1884         return reply((struct ib_mad_hdr *)smp);
1885 }
1886
1887 static int __subn_set_opa_bct(struct opa_smp *smp, u32 am, u8 *data,
1888                               struct ib_device *ibdev, u8 port, u32 *resp_len)
1889 {
1890         u32 num_ports = OPA_AM_NPORT(am);
1891         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1892         struct hfi1_pportdata *ppd;
1893         struct buffer_control *p = (struct buffer_control *)data;
1894
1895         if (num_ports != 1) {
1896                 smp->status |= IB_SMP_INVALID_FIELD;
1897                 return reply((struct ib_mad_hdr *)smp);
1898         }
1899         ppd = dd->pport + (port - 1);
1900         trace_bct_set(dd, p);
1901         if (fm_set_table(ppd, FM_TBL_BUFFER_CONTROL, p) < 0) {
1902                 smp->status |= IB_SMP_INVALID_FIELD;
1903                 return reply((struct ib_mad_hdr *)smp);
1904         }
1905
1906         return __subn_get_opa_bct(smp, am, data, ibdev, port, resp_len);
1907 }
1908
1909 static int __subn_get_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1910                                  struct ib_device *ibdev, u8 port,
1911                                  u32 *resp_len)
1912 {
1913         struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1914         u32 num_ports = OPA_AM_NPORT(am);
1915         u8 section = (am & 0x00ff0000) >> 16;
1916         u8 *p = data;
1917         int size = 0;
1918
1919         if (num_ports != 1) {
1920                 smp->status |= IB_SMP_INVALID_FIELD;
1921                 return reply((struct ib_mad_hdr *)smp);
1922         }
1923
1924         switch (section) {
1925         case OPA_VLARB_LOW_ELEMENTS:
1926                 size = fm_get_table(ppd, FM_TBL_VL_LOW_ARB, p);
1927                 break;
1928         case OPA_VLARB_HIGH_ELEMENTS:
1929                 size = fm_get_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1930                 break;
1931         case OPA_VLARB_PREEMPT_ELEMENTS:
1932                 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_ELEMS, p);
1933                 break;
1934         case OPA_VLARB_PREEMPT_MATRIX:
1935                 size = fm_get_table(ppd, FM_TBL_VL_PREEMPT_MATRIX, p);
1936                 break;
1937         default:
1938                 pr_warn("OPA SubnGet(VL Arb) AM Invalid : 0x%x\n",
1939                         be32_to_cpu(smp->attr_mod));
1940                 smp->status |= IB_SMP_INVALID_FIELD;
1941                 break;
1942         }
1943
1944         if (size > 0 && resp_len)
1945                 *resp_len += size;
1946
1947         return reply((struct ib_mad_hdr *)smp);
1948 }
1949
1950 static int __subn_set_opa_vl_arb(struct opa_smp *smp, u32 am, u8 *data,
1951                                  struct ib_device *ibdev, u8 port,
1952                                  u32 *resp_len)
1953 {
1954         struct hfi1_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
1955         u32 num_ports = OPA_AM_NPORT(am);
1956         u8 section = (am & 0x00ff0000) >> 16;
1957         u8 *p = data;
1958
1959         if (num_ports != 1) {
1960                 smp->status |= IB_SMP_INVALID_FIELD;
1961                 return reply((struct ib_mad_hdr *)smp);
1962         }
1963
1964         switch (section) {
1965         case OPA_VLARB_LOW_ELEMENTS:
1966                 (void)fm_set_table(ppd, FM_TBL_VL_LOW_ARB, p);
1967                 break;
1968         case OPA_VLARB_HIGH_ELEMENTS:
1969                 (void)fm_set_table(ppd, FM_TBL_VL_HIGH_ARB, p);
1970                 break;
1971         /*
1972          * neither OPA_VLARB_PREEMPT_ELEMENTS, or OPA_VLARB_PREEMPT_MATRIX
1973          * can be changed from the default values
1974          */
1975         case OPA_VLARB_PREEMPT_ELEMENTS:
1976                 /* FALLTHROUGH */
1977         case OPA_VLARB_PREEMPT_MATRIX:
1978                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
1979                 break;
1980         default:
1981                 pr_warn("OPA SubnSet(VL Arb) AM Invalid : 0x%x\n",
1982                         be32_to_cpu(smp->attr_mod));
1983                 smp->status |= IB_SMP_INVALID_FIELD;
1984                 break;
1985         }
1986
1987         return __subn_get_opa_vl_arb(smp, am, data, ibdev, port, resp_len);
1988 }
1989
1990 struct opa_pma_mad {
1991         struct ib_mad_hdr mad_hdr;
1992         u8 data[2024];
1993 } __packed;
1994
1995 struct opa_class_port_info {
1996         u8 base_version;
1997         u8 class_version;
1998         __be16 cap_mask;
1999         __be32 cap_mask2_resp_time;
2000
2001         u8 redirect_gid[16];
2002         __be32 redirect_tc_fl;
2003         __be32 redirect_lid;
2004         __be32 redirect_sl_qp;
2005         __be32 redirect_qkey;
2006
2007         u8 trap_gid[16];
2008         __be32 trap_tc_fl;
2009         __be32 trap_lid;
2010         __be32 trap_hl_qp;
2011         __be32 trap_qkey;
2012
2013         __be16 trap_pkey;
2014         __be16 redirect_pkey;
2015
2016         u8 trap_sl_rsvd;
2017         u8 reserved[3];
2018 } __packed;
2019
2020 struct opa_port_status_req {
2021         __u8 port_num;
2022         __u8 reserved[3];
2023         __be32 vl_select_mask;
2024 };
2025
2026 #define VL_MASK_ALL             0x000080ff
2027
2028 struct opa_port_status_rsp {
2029         __u8 port_num;
2030         __u8 reserved[3];
2031         __be32  vl_select_mask;
2032
2033         /* Data counters */
2034         __be64 port_xmit_data;
2035         __be64 port_rcv_data;
2036         __be64 port_xmit_pkts;
2037         __be64 port_rcv_pkts;
2038         __be64 port_multicast_xmit_pkts;
2039         __be64 port_multicast_rcv_pkts;
2040         __be64 port_xmit_wait;
2041         __be64 sw_port_congestion;
2042         __be64 port_rcv_fecn;
2043         __be64 port_rcv_becn;
2044         __be64 port_xmit_time_cong;
2045         __be64 port_xmit_wasted_bw;
2046         __be64 port_xmit_wait_data;
2047         __be64 port_rcv_bubble;
2048         __be64 port_mark_fecn;
2049         /* Error counters */
2050         __be64 port_rcv_constraint_errors;
2051         __be64 port_rcv_switch_relay_errors;
2052         __be64 port_xmit_discards;
2053         __be64 port_xmit_constraint_errors;
2054         __be64 port_rcv_remote_physical_errors;
2055         __be64 local_link_integrity_errors;
2056         __be64 port_rcv_errors;
2057         __be64 excessive_buffer_overruns;
2058         __be64 fm_config_errors;
2059         __be32 link_error_recovery;
2060         __be32 link_downed;
2061         u8 uncorrectable_errors;
2062
2063         u8 link_quality_indicator; /* 5res, 3bit */
2064         u8 res2[6];
2065         struct _vls_pctrs {
2066                 /* per-VL Data counters */
2067                 __be64 port_vl_xmit_data;
2068                 __be64 port_vl_rcv_data;
2069                 __be64 port_vl_xmit_pkts;
2070                 __be64 port_vl_rcv_pkts;
2071                 __be64 port_vl_xmit_wait;
2072                 __be64 sw_port_vl_congestion;
2073                 __be64 port_vl_rcv_fecn;
2074                 __be64 port_vl_rcv_becn;
2075                 __be64 port_xmit_time_cong;
2076                 __be64 port_vl_xmit_wasted_bw;
2077                 __be64 port_vl_xmit_wait_data;
2078                 __be64 port_vl_rcv_bubble;
2079                 __be64 port_vl_mark_fecn;
2080                 __be64 port_vl_xmit_discards;
2081         } vls[0]; /* real array size defined by # bits set in vl_select_mask */
2082 };
2083
2084 enum counter_selects {
2085         CS_PORT_XMIT_DATA                       = (1 << 31),
2086         CS_PORT_RCV_DATA                        = (1 << 30),
2087         CS_PORT_XMIT_PKTS                       = (1 << 29),
2088         CS_PORT_RCV_PKTS                        = (1 << 28),
2089         CS_PORT_MCAST_XMIT_PKTS                 = (1 << 27),
2090         CS_PORT_MCAST_RCV_PKTS                  = (1 << 26),
2091         CS_PORT_XMIT_WAIT                       = (1 << 25),
2092         CS_SW_PORT_CONGESTION                   = (1 << 24),
2093         CS_PORT_RCV_FECN                        = (1 << 23),
2094         CS_PORT_RCV_BECN                        = (1 << 22),
2095         CS_PORT_XMIT_TIME_CONG                  = (1 << 21),
2096         CS_PORT_XMIT_WASTED_BW                  = (1 << 20),
2097         CS_PORT_XMIT_WAIT_DATA                  = (1 << 19),
2098         CS_PORT_RCV_BUBBLE                      = (1 << 18),
2099         CS_PORT_MARK_FECN                       = (1 << 17),
2100         CS_PORT_RCV_CONSTRAINT_ERRORS           = (1 << 16),
2101         CS_PORT_RCV_SWITCH_RELAY_ERRORS         = (1 << 15),
2102         CS_PORT_XMIT_DISCARDS                   = (1 << 14),
2103         CS_PORT_XMIT_CONSTRAINT_ERRORS          = (1 << 13),
2104         CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS      = (1 << 12),
2105         CS_LOCAL_LINK_INTEGRITY_ERRORS          = (1 << 11),
2106         CS_PORT_RCV_ERRORS                      = (1 << 10),
2107         CS_EXCESSIVE_BUFFER_OVERRUNS            = (1 << 9),
2108         CS_FM_CONFIG_ERRORS                     = (1 << 8),
2109         CS_LINK_ERROR_RECOVERY                  = (1 << 7),
2110         CS_LINK_DOWNED                          = (1 << 6),
2111         CS_UNCORRECTABLE_ERRORS                 = (1 << 5),
2112 };
2113
2114 struct opa_clear_port_status {
2115         __be64 port_select_mask[4];
2116         __be32 counter_select_mask;
2117 };
2118
2119 struct opa_aggregate {
2120         __be16 attr_id;
2121         __be16 err_reqlength;   /* 1 bit, 8 res, 7 bit */
2122         __be32 attr_mod;
2123         u8 data[0];
2124 };
2125
2126 #define MSK_LLI 0x000000f0
2127 #define MSK_LLI_SFT 4
2128 #define MSK_LER 0x0000000f
2129 #define MSK_LER_SFT 0
2130 #define ADD_LLI 8
2131 #define ADD_LER 2
2132
2133 /* Request contains first three fields, response contains those plus the rest */
2134 struct opa_port_data_counters_msg {
2135         __be64 port_select_mask[4];
2136         __be32 vl_select_mask;
2137         __be32 resolution;
2138
2139         /* Response fields follow */
2140         struct _port_dctrs {
2141                 u8 port_number;
2142                 u8 reserved2[3];
2143                 __be32 link_quality_indicator; /* 29res, 3bit */
2144
2145                 /* Data counters */
2146                 __be64 port_xmit_data;
2147                 __be64 port_rcv_data;
2148                 __be64 port_xmit_pkts;
2149                 __be64 port_rcv_pkts;
2150                 __be64 port_multicast_xmit_pkts;
2151                 __be64 port_multicast_rcv_pkts;
2152                 __be64 port_xmit_wait;
2153                 __be64 sw_port_congestion;
2154                 __be64 port_rcv_fecn;
2155                 __be64 port_rcv_becn;
2156                 __be64 port_xmit_time_cong;
2157                 __be64 port_xmit_wasted_bw;
2158                 __be64 port_xmit_wait_data;
2159                 __be64 port_rcv_bubble;
2160                 __be64 port_mark_fecn;
2161
2162                 __be64 port_error_counter_summary;
2163                 /* Sum of error counts/port */
2164
2165                 struct _vls_dctrs {
2166                         /* per-VL Data counters */
2167                         __be64 port_vl_xmit_data;
2168                         __be64 port_vl_rcv_data;
2169                         __be64 port_vl_xmit_pkts;
2170                         __be64 port_vl_rcv_pkts;
2171                         __be64 port_vl_xmit_wait;
2172                         __be64 sw_port_vl_congestion;
2173                         __be64 port_vl_rcv_fecn;
2174                         __be64 port_vl_rcv_becn;
2175                         __be64 port_xmit_time_cong;
2176                         __be64 port_vl_xmit_wasted_bw;
2177                         __be64 port_vl_xmit_wait_data;
2178                         __be64 port_vl_rcv_bubble;
2179                         __be64 port_vl_mark_fecn;
2180                 } vls[0];
2181                 /* array size defined by #bits set in vl_select_mask*/
2182         } port[1]; /* array size defined by  #ports in attribute modifier */
2183 };
2184
2185 struct opa_port_error_counters64_msg {
2186         /*
2187          * Request contains first two fields, response contains the
2188          * whole magilla
2189          */
2190         __be64 port_select_mask[4];
2191         __be32 vl_select_mask;
2192
2193         /* Response-only fields follow */
2194         __be32 reserved1;
2195         struct _port_ectrs {
2196                 u8 port_number;
2197                 u8 reserved2[7];
2198                 __be64 port_rcv_constraint_errors;
2199                 __be64 port_rcv_switch_relay_errors;
2200                 __be64 port_xmit_discards;
2201                 __be64 port_xmit_constraint_errors;
2202                 __be64 port_rcv_remote_physical_errors;
2203                 __be64 local_link_integrity_errors;
2204                 __be64 port_rcv_errors;
2205                 __be64 excessive_buffer_overruns;
2206                 __be64 fm_config_errors;
2207                 __be32 link_error_recovery;
2208                 __be32 link_downed;
2209                 u8 uncorrectable_errors;
2210                 u8 reserved3[7];
2211                 struct _vls_ectrs {
2212                         __be64 port_vl_xmit_discards;
2213                 } vls[0];
2214                 /* array size defined by #bits set in vl_select_mask */
2215         } port[1]; /* array size defined by #ports in attribute modifier */
2216 };
2217
2218 struct opa_port_error_info_msg {
2219         __be64 port_select_mask[4];
2220         __be32 error_info_select_mask;
2221         __be32 reserved1;
2222         struct _port_ei {
2223                 u8 port_number;
2224                 u8 reserved2[7];
2225
2226                 /* PortRcvErrorInfo */
2227                 struct {
2228                         u8 status_and_code;
2229                         union {
2230                                 u8 raw[17];
2231                                 struct {
2232                                         /* EI1to12 format */
2233                                         u8 packet_flit1[8];
2234                                         u8 packet_flit2[8];
2235                                         u8 remaining_flit_bits12;
2236                                 } ei1to12;
2237                                 struct {
2238                                         u8 packet_bytes[8];
2239                                         u8 remaining_flit_bits;
2240                                 } ei13;
2241                         } ei;
2242                         u8 reserved3[6];
2243                 } __packed port_rcv_ei;
2244
2245                 /* ExcessiveBufferOverrunInfo */
2246                 struct {
2247                         u8 status_and_sc;
2248                         u8 reserved4[7];
2249                 } __packed excessive_buffer_overrun_ei;
2250
2251                 /* PortXmitConstraintErrorInfo */
2252                 struct {
2253                         u8 status;
2254                         u8 reserved5;
2255                         __be16 pkey;
2256                         __be32 slid;
2257                 } __packed port_xmit_constraint_ei;
2258
2259                 /* PortRcvConstraintErrorInfo */
2260                 struct {
2261                         u8 status;
2262                         u8 reserved6;
2263                         __be16 pkey;
2264                         __be32 slid;
2265                 } __packed port_rcv_constraint_ei;
2266
2267                 /* PortRcvSwitchRelayErrorInfo */
2268                 struct {
2269                         u8 status_and_code;
2270                         u8 reserved7[3];
2271                         __u32 error_info;
2272                 } __packed port_rcv_switch_relay_ei;
2273
2274                 /* UncorrectableErrorInfo */
2275                 struct {
2276                         u8 status_and_code;
2277                         u8 reserved8;
2278                 } __packed uncorrectable_ei;
2279
2280                 /* FMConfigErrorInfo */
2281                 struct {
2282                         u8 status_and_code;
2283                         u8 error_info;
2284                 } __packed fm_config_ei;
2285                 __u32 reserved9;
2286         } port[1]; /* actual array size defined by #ports in attr modifier */
2287 };
2288
2289 /* opa_port_error_info_msg error_info_select_mask bit definitions */
2290 enum error_info_selects {
2291         ES_PORT_RCV_ERROR_INFO                  = (1 << 31),
2292         ES_EXCESSIVE_BUFFER_OVERRUN_INFO        = (1 << 30),
2293         ES_PORT_XMIT_CONSTRAINT_ERROR_INFO      = (1 << 29),
2294         ES_PORT_RCV_CONSTRAINT_ERROR_INFO       = (1 << 28),
2295         ES_PORT_RCV_SWITCH_RELAY_ERROR_INFO     = (1 << 27),
2296         ES_UNCORRECTABLE_ERROR_INFO             = (1 << 26),
2297         ES_FM_CONFIG_ERROR_INFO                 = (1 << 25)
2298 };
2299
2300 static int pma_get_opa_classportinfo(struct opa_pma_mad *pmp,
2301                                      struct ib_device *ibdev, u32 *resp_len)
2302 {
2303         struct opa_class_port_info *p =
2304                 (struct opa_class_port_info *)pmp->data;
2305
2306         memset(pmp->data, 0, sizeof(pmp->data));
2307
2308         if (pmp->mad_hdr.attr_mod != 0)
2309                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2310
2311         p->base_version = OPA_MGMT_BASE_VERSION;
2312         p->class_version = OPA_SMI_CLASS_VERSION;
2313         /*
2314          * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
2315          */
2316         p->cap_mask2_resp_time = cpu_to_be32(18);
2317
2318         if (resp_len)
2319                 *resp_len += sizeof(*p);
2320
2321         return reply((struct ib_mad_hdr *)pmp);
2322 }
2323
2324 static void a0_portstatus(struct hfi1_pportdata *ppd,
2325                           struct opa_port_status_rsp *rsp, u32 vl_select_mask)
2326 {
2327         if (!is_bx(ppd->dd)) {
2328                 unsigned long vl;
2329                 u64 sum_vl_xmit_wait = 0;
2330                 u32 vl_all_mask = VL_MASK_ALL;
2331
2332                 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2333                                  8 * sizeof(vl_all_mask)) {
2334                         u64 tmp = sum_vl_xmit_wait +
2335                                   read_port_cntr(ppd, C_TX_WAIT_VL,
2336                                                  idx_from_vl(vl));
2337                         if (tmp < sum_vl_xmit_wait) {
2338                                 /* we wrapped */
2339                                 sum_vl_xmit_wait = (u64)~0;
2340                                 break;
2341                         }
2342                         sum_vl_xmit_wait = tmp;
2343                 }
2344                 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2345                         rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2346         }
2347 }
2348
2349 static int pma_get_opa_portstatus(struct opa_pma_mad *pmp,
2350                                   struct ib_device *ibdev,
2351                                   u8 port, u32 *resp_len)
2352 {
2353         struct opa_port_status_req *req =
2354                 (struct opa_port_status_req *)pmp->data;
2355         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2356         struct opa_port_status_rsp *rsp;
2357         u32 vl_select_mask = be32_to_cpu(req->vl_select_mask);
2358         unsigned long vl;
2359         size_t response_data_size;
2360         u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2361         u8 port_num = req->port_num;
2362         u8 num_vls = hweight32(vl_select_mask);
2363         struct _vls_pctrs *vlinfo;
2364         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2365         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2366         int vfi;
2367         u64 tmp, tmp2;
2368
2369         response_data_size = sizeof(struct opa_port_status_rsp) +
2370                                 num_vls * sizeof(struct _vls_pctrs);
2371         if (response_data_size > sizeof(pmp->data)) {
2372                 pmp->mad_hdr.status |= OPA_PM_STATUS_REQUEST_TOO_LARGE;
2373                 return reply((struct ib_mad_hdr *)pmp);
2374         }
2375
2376         if (nports != 1 || (port_num && port_num != port) ||
2377             num_vls > OPA_MAX_VLS || (vl_select_mask & ~VL_MASK_ALL)) {
2378                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2379                 return reply((struct ib_mad_hdr *)pmp);
2380         }
2381
2382         memset(pmp->data, 0, sizeof(pmp->data));
2383
2384         rsp = (struct opa_port_status_rsp *)pmp->data;
2385         if (port_num)
2386                 rsp->port_num = port_num;
2387         else
2388                 rsp->port_num = port;
2389
2390         rsp->port_rcv_constraint_errors =
2391                 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2392                                            CNTR_INVALID_VL));
2393
2394         hfi1_read_link_quality(dd, &rsp->link_quality_indicator);
2395
2396         rsp->vl_select_mask = cpu_to_be32(vl_select_mask);
2397         rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2398                                           CNTR_INVALID_VL));
2399         rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2400                                          CNTR_INVALID_VL));
2401         rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2402                                           CNTR_INVALID_VL));
2403         rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2404                                          CNTR_INVALID_VL));
2405         rsp->port_multicast_xmit_pkts =
2406                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2407                                           CNTR_INVALID_VL));
2408         rsp->port_multicast_rcv_pkts =
2409                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2410                                           CNTR_INVALID_VL));
2411         rsp->port_xmit_wait =
2412                 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2413         rsp->port_rcv_fecn =
2414                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2415         rsp->port_rcv_becn =
2416                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2417         rsp->port_xmit_discards =
2418                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2419                                            CNTR_INVALID_VL));
2420         rsp->port_xmit_constraint_errors =
2421                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2422                                            CNTR_INVALID_VL));
2423         rsp->port_rcv_remote_physical_errors =
2424                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2425                                           CNTR_INVALID_VL));
2426         rsp->local_link_integrity_errors =
2427                 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2428                                           CNTR_INVALID_VL));
2429         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2430         tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2431                                    CNTR_INVALID_VL);
2432         if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2433                 /* overflow/wrapped */
2434                 rsp->link_error_recovery = cpu_to_be32(~0);
2435         } else {
2436                 rsp->link_error_recovery = cpu_to_be32(tmp2);
2437         }
2438         rsp->port_rcv_errors =
2439                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2440         rsp->excessive_buffer_overruns =
2441                 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2442         rsp->fm_config_errors =
2443                 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2444                                           CNTR_INVALID_VL));
2445         rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2446                                                       CNTR_INVALID_VL));
2447
2448         /* rsp->uncorrectable_errors is 8 bits wide, and it pegs at 0xff */
2449         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2450         rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2451
2452         vlinfo = &rsp->vls[0];
2453         vfi = 0;
2454         /* The vl_select_mask has been checked above, and we know
2455          * that it contains only entries which represent valid VLs.
2456          * So in the for_each_set_bit() loop below, we don't need
2457          * any additional checks for vl.
2458          */
2459         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2460                          8 * sizeof(vl_select_mask)) {
2461                 memset(vlinfo, 0, sizeof(*vlinfo));
2462
2463                 tmp = read_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl));
2464                 rsp->vls[vfi].port_vl_rcv_data = cpu_to_be64(tmp);
2465
2466                 rsp->vls[vfi].port_vl_rcv_pkts =
2467                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2468                                                   idx_from_vl(vl)));
2469
2470                 rsp->vls[vfi].port_vl_xmit_data =
2471                         cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2472                                                    idx_from_vl(vl)));
2473
2474                 rsp->vls[vfi].port_vl_xmit_pkts =
2475                         cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2476                                                    idx_from_vl(vl)));
2477
2478                 rsp->vls[vfi].port_vl_xmit_wait =
2479                         cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2480                                                    idx_from_vl(vl)));
2481
2482                 rsp->vls[vfi].port_vl_rcv_fecn =
2483                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2484                                                   idx_from_vl(vl)));
2485
2486                 rsp->vls[vfi].port_vl_rcv_becn =
2487                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2488                                                   idx_from_vl(vl)));
2489
2490                 rsp->vls[vfi].port_vl_xmit_discards =
2491                         cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2492                                                    idx_from_vl(vl)));
2493                 vlinfo++;
2494                 vfi++;
2495         }
2496
2497         a0_portstatus(ppd, rsp, vl_select_mask);
2498
2499         if (resp_len)
2500                 *resp_len += response_data_size;
2501
2502         return reply((struct ib_mad_hdr *)pmp);
2503 }
2504
2505 static u64 get_error_counter_summary(struct ib_device *ibdev, u8 port,
2506                                      u8 res_lli, u8 res_ler)
2507 {
2508         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2509         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2510         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2511         u64 error_counter_summary = 0, tmp;
2512
2513         error_counter_summary += read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2514                                                 CNTR_INVALID_VL);
2515         /* port_rcv_switch_relay_errors is 0 for HFIs */
2516         error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_DSCD,
2517                                                 CNTR_INVALID_VL);
2518         error_counter_summary += read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2519                                                 CNTR_INVALID_VL);
2520         error_counter_summary += read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2521                                                CNTR_INVALID_VL);
2522         /* local link integrity must be right-shifted by the lli resolution */
2523         error_counter_summary += (read_dev_cntr(dd, C_DC_RX_REPLAY,
2524                                                 CNTR_INVALID_VL) >> res_lli);
2525         /* link error recovery must b right-shifted by the ler resolution */
2526         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2527         tmp += read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL);
2528         error_counter_summary += (tmp >> res_ler);
2529         error_counter_summary += read_dev_cntr(dd, C_DC_RCV_ERR,
2530                                                CNTR_INVALID_VL);
2531         error_counter_summary += read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL);
2532         error_counter_summary += read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2533                                                CNTR_INVALID_VL);
2534         /* ppd->link_downed is a 32-bit value */
2535         error_counter_summary += read_port_cntr(ppd, C_SW_LINK_DOWN,
2536                                                 CNTR_INVALID_VL);
2537         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2538         /* this is an 8-bit quantity */
2539         error_counter_summary += tmp < 0x100 ? (tmp & 0xff) : 0xff;
2540
2541         return error_counter_summary;
2542 }
2543
2544 static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
2545                             u32 vl_select_mask)
2546 {
2547         if (!is_bx(ppd->dd)) {
2548                 unsigned long vl;
2549                 u64 sum_vl_xmit_wait = 0;
2550                 u32 vl_all_mask = VL_MASK_ALL;
2551
2552                 for_each_set_bit(vl, (unsigned long *)&(vl_all_mask),
2553                                  8 * sizeof(vl_all_mask)) {
2554                         u64 tmp = sum_vl_xmit_wait +
2555                                   read_port_cntr(ppd, C_TX_WAIT_VL,
2556                                                  idx_from_vl(vl));
2557                         if (tmp < sum_vl_xmit_wait) {
2558                                 /* we wrapped */
2559                                 sum_vl_xmit_wait = (u64)~0;
2560                                 break;
2561                         }
2562                         sum_vl_xmit_wait = tmp;
2563                 }
2564                 if (be64_to_cpu(rsp->port_xmit_wait) > sum_vl_xmit_wait)
2565                         rsp->port_xmit_wait = cpu_to_be64(sum_vl_xmit_wait);
2566         }
2567 }
2568
2569 static void pma_get_opa_port_dctrs(struct ib_device *ibdev,
2570                                    struct _port_dctrs *rsp)
2571 {
2572         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2573
2574         rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
2575                                                 CNTR_INVALID_VL));
2576         rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
2577                                                 CNTR_INVALID_VL));
2578         rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
2579                                                 CNTR_INVALID_VL));
2580         rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
2581                                                 CNTR_INVALID_VL));
2582         rsp->port_multicast_xmit_pkts =
2583                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
2584                                           CNTR_INVALID_VL));
2585         rsp->port_multicast_rcv_pkts =
2586                 cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
2587                                           CNTR_INVALID_VL));
2588 }
2589
2590 static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
2591                                     struct ib_device *ibdev,
2592                                     u8 port, u32 *resp_len)
2593 {
2594         struct opa_port_data_counters_msg *req =
2595                 (struct opa_port_data_counters_msg *)pmp->data;
2596         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2597         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2598         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2599         struct _port_dctrs *rsp;
2600         struct _vls_dctrs *vlinfo;
2601         size_t response_data_size;
2602         u32 num_ports;
2603         u8 num_pslm;
2604         u8 lq, num_vls;
2605         u8 res_lli, res_ler;
2606         u64 port_mask;
2607         unsigned long port_num;
2608         unsigned long vl;
2609         u32 vl_select_mask;
2610         int vfi;
2611
2612         num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2613         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2614         num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2615         vl_select_mask = be32_to_cpu(req->vl_select_mask);
2616         res_lli = (u8)(be32_to_cpu(req->resolution) & MSK_LLI) >> MSK_LLI_SFT;
2617         res_lli = res_lli ? res_lli + ADD_LLI : 0;
2618         res_ler = (u8)(be32_to_cpu(req->resolution) & MSK_LER) >> MSK_LER_SFT;
2619         res_ler = res_ler ? res_ler + ADD_LER : 0;
2620
2621         if (num_ports != 1 || (vl_select_mask & ~VL_MASK_ALL)) {
2622                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2623                 return reply((struct ib_mad_hdr *)pmp);
2624         }
2625
2626         /* Sanity check */
2627         response_data_size = sizeof(struct opa_port_data_counters_msg) +
2628                                 num_vls * sizeof(struct _vls_dctrs);
2629
2630         if (response_data_size > sizeof(pmp->data)) {
2631                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2632                 return reply((struct ib_mad_hdr *)pmp);
2633         }
2634
2635         /*
2636          * The bit set in the mask needs to be consistent with the
2637          * port the request came in on.
2638          */
2639         port_mask = be64_to_cpu(req->port_select_mask[3]);
2640         port_num = find_first_bit((unsigned long *)&port_mask,
2641                                   sizeof(port_mask));
2642
2643         if ((u8)port_num != port) {
2644                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2645                 return reply((struct ib_mad_hdr *)pmp);
2646         }
2647
2648         rsp = &req->port[0];
2649         memset(rsp, 0, sizeof(*rsp));
2650
2651         rsp->port_number = port;
2652         /*
2653          * Note that link_quality_indicator is a 32 bit quantity in
2654          * 'datacounters' queries (as opposed to 'portinfo' queries,
2655          * where it's a byte).
2656          */
2657         hfi1_read_link_quality(dd, &lq);
2658         rsp->link_quality_indicator = cpu_to_be32((u32)lq);
2659         pma_get_opa_port_dctrs(ibdev, rsp);
2660
2661         rsp->port_xmit_wait =
2662                 cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
2663         rsp->port_rcv_fecn =
2664                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
2665         rsp->port_rcv_becn =
2666                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
2667         rsp->port_error_counter_summary =
2668                 cpu_to_be64(get_error_counter_summary(ibdev, port,
2669                                                       res_lli, res_ler));
2670
2671         vlinfo = &rsp->vls[0];
2672         vfi = 0;
2673         /* The vl_select_mask has been checked above, and we know
2674          * that it contains only entries which represent valid VLs.
2675          * So in the for_each_set_bit() loop below, we don't need
2676          * any additional checks for vl.
2677          */
2678         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2679                          8 * sizeof(req->vl_select_mask)) {
2680                 memset(vlinfo, 0, sizeof(*vlinfo));
2681
2682                 rsp->vls[vfi].port_vl_xmit_data =
2683                         cpu_to_be64(read_port_cntr(ppd, C_TX_FLIT_VL,
2684                                                    idx_from_vl(vl)));
2685
2686                 rsp->vls[vfi].port_vl_rcv_data =
2687                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_FLIT_VL,
2688                                                   idx_from_vl(vl)));
2689
2690                 rsp->vls[vfi].port_vl_xmit_pkts =
2691                         cpu_to_be64(read_port_cntr(ppd, C_TX_PKT_VL,
2692                                                    idx_from_vl(vl)));
2693
2694                 rsp->vls[vfi].port_vl_rcv_pkts =
2695                         cpu_to_be64(read_dev_cntr(dd, C_DC_RX_PKT_VL,
2696                                                   idx_from_vl(vl)));
2697
2698                 rsp->vls[vfi].port_vl_xmit_wait =
2699                         cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT_VL,
2700                                                    idx_from_vl(vl)));
2701
2702                 rsp->vls[vfi].port_vl_rcv_fecn =
2703                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN_VL,
2704                                                   idx_from_vl(vl)));
2705                 rsp->vls[vfi].port_vl_rcv_becn =
2706                         cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN_VL,
2707                                                   idx_from_vl(vl)));
2708
2709                 /* rsp->port_vl_xmit_time_cong is 0 for HFIs */
2710                 /* rsp->port_vl_xmit_wasted_bw ??? */
2711                 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ???
2712                  * does this differ from rsp->vls[vfi].port_vl_xmit_wait
2713                  */
2714                 /*rsp->vls[vfi].port_vl_mark_fecn =
2715                  *      cpu_to_be64(read_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT
2716                  *              + offset));
2717                  */
2718                 vlinfo++;
2719                 vfi++;
2720         }
2721
2722         a0_datacounters(ppd, rsp, vl_select_mask);
2723
2724         if (resp_len)
2725                 *resp_len += response_data_size;
2726
2727         return reply((struct ib_mad_hdr *)pmp);
2728 }
2729
2730 static int pma_get_ib_portcounters_ext(struct ib_pma_mad *pmp,
2731                                        struct ib_device *ibdev, u8 port)
2732 {
2733         struct ib_pma_portcounters_ext *p = (struct ib_pma_portcounters_ext *)
2734                                                 pmp->data;
2735         struct _port_dctrs rsp;
2736
2737         if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
2738                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2739                 goto bail;
2740         }
2741
2742         memset(&rsp, 0, sizeof(rsp));
2743         pma_get_opa_port_dctrs(ibdev, &rsp);
2744
2745         p->port_xmit_data = rsp.port_xmit_data;
2746         p->port_rcv_data = rsp.port_rcv_data;
2747         p->port_xmit_packets = rsp.port_xmit_pkts;
2748         p->port_rcv_packets = rsp.port_rcv_pkts;
2749         p->port_unicast_xmit_packets = 0;
2750         p->port_unicast_rcv_packets =  0;
2751         p->port_multicast_xmit_packets = rsp.port_multicast_xmit_pkts;
2752         p->port_multicast_rcv_packets = rsp.port_multicast_rcv_pkts;
2753
2754 bail:
2755         return reply((struct ib_mad_hdr *)pmp);
2756 }
2757
2758 static void pma_get_opa_port_ectrs(struct ib_device *ibdev,
2759                                    struct _port_ectrs *rsp, u8 port)
2760 {
2761         u64 tmp, tmp2;
2762         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2763         struct hfi1_ibport *ibp = to_iport(ibdev, port);
2764         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2765
2766         tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
2767         tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
2768                                         CNTR_INVALID_VL);
2769         if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
2770                 /* overflow/wrapped */
2771                 rsp->link_error_recovery = cpu_to_be32(~0);
2772         } else {
2773                 rsp->link_error_recovery = cpu_to_be32(tmp2);
2774         }
2775
2776         rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
2777                                                 CNTR_INVALID_VL));
2778         rsp->port_rcv_errors =
2779                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2780         rsp->port_rcv_remote_physical_errors =
2781                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2782                                           CNTR_INVALID_VL));
2783         rsp->port_rcv_switch_relay_errors = 0;
2784         rsp->port_xmit_discards =
2785                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
2786                                            CNTR_INVALID_VL));
2787         rsp->port_xmit_constraint_errors =
2788                 cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
2789                                            CNTR_INVALID_VL));
2790         rsp->port_rcv_constraint_errors =
2791                 cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
2792                                            CNTR_INVALID_VL));
2793         rsp->local_link_integrity_errors =
2794                 cpu_to_be64(read_dev_cntr(dd, C_DC_RX_REPLAY,
2795                                           CNTR_INVALID_VL));
2796         rsp->excessive_buffer_overruns =
2797                 cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
2798 }
2799
2800 static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
2801                                   struct ib_device *ibdev,
2802                                   u8 port, u32 *resp_len)
2803 {
2804         size_t response_data_size;
2805         struct _port_ectrs *rsp;
2806         u8 port_num;
2807         struct opa_port_error_counters64_msg *req;
2808         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2809         u32 num_ports;
2810         u8 num_pslm;
2811         u8 num_vls;
2812         struct hfi1_ibport *ibp;
2813         struct hfi1_pportdata *ppd;
2814         struct _vls_ectrs *vlinfo;
2815         unsigned long vl;
2816         u64 port_mask, tmp;
2817         u32 vl_select_mask;
2818         int vfi;
2819
2820         req = (struct opa_port_error_counters64_msg *)pmp->data;
2821
2822         num_ports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
2823
2824         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2825         num_vls = hweight32(be32_to_cpu(req->vl_select_mask));
2826
2827         if (num_ports != 1 || num_ports != num_pslm) {
2828                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2829                 return reply((struct ib_mad_hdr *)pmp);
2830         }
2831
2832         response_data_size = sizeof(struct opa_port_error_counters64_msg) +
2833                                 num_vls * sizeof(struct _vls_ectrs);
2834
2835         if (response_data_size > sizeof(pmp->data)) {
2836                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2837                 return reply((struct ib_mad_hdr *)pmp);
2838         }
2839         /*
2840          * The bit set in the mask needs to be consistent with the
2841          * port the request came in on.
2842          */
2843         port_mask = be64_to_cpu(req->port_select_mask[3]);
2844         port_num = find_first_bit((unsigned long *)&port_mask,
2845                                   sizeof(port_mask));
2846
2847         if (port_num != port) {
2848                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2849                 return reply((struct ib_mad_hdr *)pmp);
2850         }
2851
2852         rsp = &req->port[0];
2853
2854         ibp = to_iport(ibdev, port_num);
2855         ppd = ppd_from_ibp(ibp);
2856
2857         memset(rsp, 0, sizeof(*rsp));
2858         rsp->port_number = port_num;
2859
2860         pma_get_opa_port_ectrs(ibdev, rsp, port_num);
2861
2862         rsp->port_rcv_remote_physical_errors =
2863                 cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
2864                                           CNTR_INVALID_VL));
2865         rsp->fm_config_errors =
2866                 cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
2867                                           CNTR_INVALID_VL));
2868         tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
2869
2870         rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
2871         rsp->port_rcv_errors =
2872                 cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
2873         vlinfo = &rsp->vls[0];
2874         vfi = 0;
2875         vl_select_mask = be32_to_cpu(req->vl_select_mask);
2876         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
2877                          8 * sizeof(req->vl_select_mask)) {
2878                 memset(vlinfo, 0, sizeof(*vlinfo));
2879                 rsp->vls[vfi].port_vl_xmit_discards =
2880                         cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
2881                                                    idx_from_vl(vl)));
2882                 vlinfo += 1;
2883                 vfi++;
2884         }
2885
2886         if (resp_len)
2887                 *resp_len += response_data_size;
2888
2889         return reply((struct ib_mad_hdr *)pmp);
2890 }
2891
2892 static int pma_get_ib_portcounters(struct ib_pma_mad *pmp,
2893                                    struct ib_device *ibdev, u8 port)
2894 {
2895         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
2896                 pmp->data;
2897         struct _port_ectrs rsp;
2898         u64 temp_link_overrun_errors;
2899         u64 temp_64;
2900         u32 temp_32;
2901
2902         memset(&rsp, 0, sizeof(rsp));
2903         pma_get_opa_port_ectrs(ibdev, &rsp, port);
2904
2905         if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
2906                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
2907                 goto bail;
2908         }
2909
2910         p->symbol_error_counter = 0; /* N/A for OPA */
2911
2912         temp_32 = be32_to_cpu(rsp.link_error_recovery);
2913         if (temp_32 > 0xFFUL)
2914                 p->link_error_recovery_counter = 0xFF;
2915         else
2916                 p->link_error_recovery_counter = (u8)temp_32;
2917
2918         temp_32 = be32_to_cpu(rsp.link_downed);
2919         if (temp_32 > 0xFFUL)
2920                 p->link_downed_counter = 0xFF;
2921         else
2922                 p->link_downed_counter = (u8)temp_32;
2923
2924         temp_64 = be64_to_cpu(rsp.port_rcv_errors);
2925         if (temp_64 > 0xFFFFUL)
2926                 p->port_rcv_errors = cpu_to_be16(0xFFFF);
2927         else
2928                 p->port_rcv_errors = cpu_to_be16((u16)temp_64);
2929
2930         temp_64 = be64_to_cpu(rsp.port_rcv_remote_physical_errors);
2931         if (temp_64 > 0xFFFFUL)
2932                 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
2933         else
2934                 p->port_rcv_remphys_errors = cpu_to_be16((u16)temp_64);
2935
2936         temp_64 = be64_to_cpu(rsp.port_rcv_switch_relay_errors);
2937         p->port_rcv_switch_relay_errors = cpu_to_be16((u16)temp_64);
2938
2939         temp_64 = be64_to_cpu(rsp.port_xmit_discards);
2940         if (temp_64 > 0xFFFFUL)
2941                 p->port_xmit_discards = cpu_to_be16(0xFFFF);
2942         else
2943                 p->port_xmit_discards = cpu_to_be16((u16)temp_64);
2944
2945         temp_64 = be64_to_cpu(rsp.port_xmit_constraint_errors);
2946         if (temp_64 > 0xFFUL)
2947                 p->port_xmit_constraint_errors = 0xFF;
2948         else
2949                 p->port_xmit_constraint_errors = (u8)temp_64;
2950
2951         temp_64 = be64_to_cpu(rsp.port_rcv_constraint_errors);
2952         if (temp_64 > 0xFFUL)
2953                 p->port_rcv_constraint_errors = 0xFFUL;
2954         else
2955                 p->port_rcv_constraint_errors = (u8)temp_64;
2956
2957         /* LocalLink: 7:4, BufferOverrun: 3:0 */
2958         temp_64 = be64_to_cpu(rsp.local_link_integrity_errors);
2959         if (temp_64 > 0xFUL)
2960                 temp_64 = 0xFUL;
2961
2962         temp_link_overrun_errors = temp_64 << 4;
2963
2964         temp_64 = be64_to_cpu(rsp.excessive_buffer_overruns);
2965         if (temp_64 > 0xFUL)
2966                 temp_64 = 0xFUL;
2967         temp_link_overrun_errors |= temp_64;
2968
2969         p->link_overrun_errors = (u8)temp_link_overrun_errors;
2970
2971         p->vl15_dropped = 0; /* N/A for OPA */
2972
2973 bail:
2974         return reply((struct ib_mad_hdr *)pmp);
2975 }
2976
2977 static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
2978                                  struct ib_device *ibdev,
2979                                  u8 port, u32 *resp_len)
2980 {
2981         size_t response_data_size;
2982         struct _port_ei *rsp;
2983         struct opa_port_error_info_msg *req;
2984         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
2985         u64 port_mask;
2986         u32 num_ports;
2987         u8 port_num;
2988         u8 num_pslm;
2989         u64 reg;
2990
2991         req = (struct opa_port_error_info_msg *)pmp->data;
2992         rsp = &req->port[0];
2993
2994         num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
2995         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
2996
2997         memset(rsp, 0, sizeof(*rsp));
2998
2999         if (num_ports != 1 || num_ports != num_pslm) {
3000                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3001                 return reply((struct ib_mad_hdr *)pmp);
3002         }
3003
3004         /* Sanity check */
3005         response_data_size = sizeof(struct opa_port_error_info_msg);
3006
3007         if (response_data_size > sizeof(pmp->data)) {
3008                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3009                 return reply((struct ib_mad_hdr *)pmp);
3010         }
3011
3012         /*
3013          * The bit set in the mask needs to be consistent with the port
3014          * the request came in on.
3015          */
3016         port_mask = be64_to_cpu(req->port_select_mask[3]);
3017         port_num = find_first_bit((unsigned long *)&port_mask,
3018                                   sizeof(port_mask));
3019
3020         if (port_num != port) {
3021                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3022                 return reply((struct ib_mad_hdr *)pmp);
3023         }
3024
3025         /* PortRcvErrorInfo */
3026         rsp->port_rcv_ei.status_and_code =
3027                 dd->err_info_rcvport.status_and_code;
3028         memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit1,
3029                &dd->err_info_rcvport.packet_flit1, sizeof(u64));
3030         memcpy(&rsp->port_rcv_ei.ei.ei1to12.packet_flit2,
3031                &dd->err_info_rcvport.packet_flit2, sizeof(u64));
3032
3033         /* ExcessiverBufferOverrunInfo */
3034         reg = read_csr(dd, RCV_ERR_INFO);
3035         if (reg & RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK) {
3036                 /*
3037                  * if the RcvExcessBufferOverrun bit is set, save SC of
3038                  * first pkt that encountered an excess buffer overrun
3039                  */
3040                 u8 tmp = (u8)reg;
3041
3042                 tmp &=  RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SC_SMASK;
3043                 tmp <<= 2;
3044                 rsp->excessive_buffer_overrun_ei.status_and_sc = tmp;
3045                 /* set the status bit */
3046                 rsp->excessive_buffer_overrun_ei.status_and_sc |= 0x80;
3047         }
3048
3049         rsp->port_xmit_constraint_ei.status =
3050                 dd->err_info_xmit_constraint.status;
3051         rsp->port_xmit_constraint_ei.pkey =
3052                 cpu_to_be16(dd->err_info_xmit_constraint.pkey);
3053         rsp->port_xmit_constraint_ei.slid =
3054                 cpu_to_be32(dd->err_info_xmit_constraint.slid);
3055
3056         rsp->port_rcv_constraint_ei.status =
3057                 dd->err_info_rcv_constraint.status;
3058         rsp->port_rcv_constraint_ei.pkey =
3059                 cpu_to_be16(dd->err_info_rcv_constraint.pkey);
3060         rsp->port_rcv_constraint_ei.slid =
3061                 cpu_to_be32(dd->err_info_rcv_constraint.slid);
3062
3063         /* UncorrectableErrorInfo */
3064         rsp->uncorrectable_ei.status_and_code = dd->err_info_uncorrectable;
3065
3066         /* FMConfigErrorInfo */
3067         rsp->fm_config_ei.status_and_code = dd->err_info_fmconfig;
3068
3069         if (resp_len)
3070                 *resp_len += response_data_size;
3071
3072         return reply((struct ib_mad_hdr *)pmp);
3073 }
3074
3075 static int pma_set_opa_portstatus(struct opa_pma_mad *pmp,
3076                                   struct ib_device *ibdev,
3077                                   u8 port, u32 *resp_len)
3078 {
3079         struct opa_clear_port_status *req =
3080                 (struct opa_clear_port_status *)pmp->data;
3081         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3082         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3083         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3084         u32 nports = be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24;
3085         u64 portn = be64_to_cpu(req->port_select_mask[3]);
3086         u32 counter_select = be32_to_cpu(req->counter_select_mask);
3087         u32 vl_select_mask = VL_MASK_ALL; /* clear all per-vl cnts */
3088         unsigned long vl;
3089
3090         if ((nports != 1) || (portn != 1 << port)) {
3091                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3092                 return reply((struct ib_mad_hdr *)pmp);
3093         }
3094         /*
3095          * only counters returned by pma_get_opa_portstatus() are
3096          * handled, so when pma_get_opa_portstatus() gets a fix,
3097          * the corresponding change should be made here as well.
3098          */
3099
3100         if (counter_select & CS_PORT_XMIT_DATA)
3101                 write_dev_cntr(dd, C_DC_XMIT_FLITS, CNTR_INVALID_VL, 0);
3102
3103         if (counter_select & CS_PORT_RCV_DATA)
3104                 write_dev_cntr(dd, C_DC_RCV_FLITS, CNTR_INVALID_VL, 0);
3105
3106         if (counter_select & CS_PORT_XMIT_PKTS)
3107                 write_dev_cntr(dd, C_DC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3108
3109         if (counter_select & CS_PORT_RCV_PKTS)
3110                 write_dev_cntr(dd, C_DC_RCV_PKTS, CNTR_INVALID_VL, 0);
3111
3112         if (counter_select & CS_PORT_MCAST_XMIT_PKTS)
3113                 write_dev_cntr(dd, C_DC_MC_XMIT_PKTS, CNTR_INVALID_VL, 0);
3114
3115         if (counter_select & CS_PORT_MCAST_RCV_PKTS)
3116                 write_dev_cntr(dd, C_DC_MC_RCV_PKTS, CNTR_INVALID_VL, 0);
3117
3118         if (counter_select & CS_PORT_XMIT_WAIT)
3119                 write_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL, 0);
3120
3121         /* ignore cs_sw_portCongestion for HFIs */
3122
3123         if (counter_select & CS_PORT_RCV_FECN)
3124                 write_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL, 0);
3125
3126         if (counter_select & CS_PORT_RCV_BECN)
3127                 write_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL, 0);
3128
3129         /* ignore cs_port_xmit_time_cong for HFIs */
3130         /* ignore cs_port_xmit_wasted_bw for now */
3131         /* ignore cs_port_xmit_wait_data for now */
3132         if (counter_select & CS_PORT_RCV_BUBBLE)
3133                 write_dev_cntr(dd, C_DC_RCV_BBL, CNTR_INVALID_VL, 0);
3134
3135         /* Only applicable for switch */
3136         /* if (counter_select & CS_PORT_MARK_FECN)
3137          *      write_csr(dd, DCC_PRF_PORT_MARK_FECN_CNT, 0);
3138          */
3139
3140         if (counter_select & CS_PORT_RCV_CONSTRAINT_ERRORS)
3141                 write_port_cntr(ppd, C_SW_RCV_CSTR_ERR, CNTR_INVALID_VL, 0);
3142
3143         /* ignore cs_port_rcv_switch_relay_errors for HFIs */
3144         if (counter_select & CS_PORT_XMIT_DISCARDS)
3145                 write_port_cntr(ppd, C_SW_XMIT_DSCD, CNTR_INVALID_VL, 0);
3146
3147         if (counter_select & CS_PORT_XMIT_CONSTRAINT_ERRORS)
3148                 write_port_cntr(ppd, C_SW_XMIT_CSTR_ERR, CNTR_INVALID_VL, 0);
3149
3150         if (counter_select & CS_PORT_RCV_REMOTE_PHYSICAL_ERRORS)
3151                 write_dev_cntr(dd, C_DC_RMT_PHY_ERR, CNTR_INVALID_VL, 0);
3152
3153         if (counter_select & CS_LOCAL_LINK_INTEGRITY_ERRORS)
3154                 write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3155
3156         if (counter_select & CS_LINK_ERROR_RECOVERY) {
3157                 write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3158                 write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
3159                                CNTR_INVALID_VL, 0);
3160         }
3161
3162         if (counter_select & CS_PORT_RCV_ERRORS)
3163                 write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3164
3165         if (counter_select & CS_EXCESSIVE_BUFFER_OVERRUNS) {
3166                 write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3167                 dd->rcv_ovfl_cnt = 0;
3168         }
3169
3170         if (counter_select & CS_FM_CONFIG_ERRORS)
3171                 write_dev_cntr(dd, C_DC_FM_CFG_ERR, CNTR_INVALID_VL, 0);
3172
3173         if (counter_select & CS_LINK_DOWNED)
3174                 write_port_cntr(ppd, C_SW_LINK_DOWN, CNTR_INVALID_VL, 0);
3175
3176         if (counter_select & CS_UNCORRECTABLE_ERRORS)
3177                 write_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL, 0);
3178
3179         for_each_set_bit(vl, (unsigned long *)&(vl_select_mask),
3180                          8 * sizeof(vl_select_mask)) {
3181                 if (counter_select & CS_PORT_XMIT_DATA)
3182                         write_port_cntr(ppd, C_TX_FLIT_VL, idx_from_vl(vl), 0);
3183
3184                 if (counter_select & CS_PORT_RCV_DATA)
3185                         write_dev_cntr(dd, C_DC_RX_FLIT_VL, idx_from_vl(vl), 0);
3186
3187                 if (counter_select & CS_PORT_XMIT_PKTS)
3188                         write_port_cntr(ppd, C_TX_PKT_VL, idx_from_vl(vl), 0);
3189
3190                 if (counter_select & CS_PORT_RCV_PKTS)
3191                         write_dev_cntr(dd, C_DC_RX_PKT_VL, idx_from_vl(vl), 0);
3192
3193                 if (counter_select & CS_PORT_XMIT_WAIT)
3194                         write_port_cntr(ppd, C_TX_WAIT_VL, idx_from_vl(vl), 0);
3195
3196                 /* sw_port_vl_congestion is 0 for HFIs */
3197                 if (counter_select & CS_PORT_RCV_FECN)
3198                         write_dev_cntr(dd, C_DC_RCV_FCN_VL, idx_from_vl(vl), 0);
3199
3200                 if (counter_select & CS_PORT_RCV_BECN)
3201                         write_dev_cntr(dd, C_DC_RCV_BCN_VL, idx_from_vl(vl), 0);
3202
3203                 /* port_vl_xmit_time_cong is 0 for HFIs */
3204                 /* port_vl_xmit_wasted_bw ??? */
3205                 /* port_vl_xmit_wait_data - TXE (table 13-9 HFI spec) ??? */
3206                 if (counter_select & CS_PORT_RCV_BUBBLE)
3207                         write_dev_cntr(dd, C_DC_RCV_BBL_VL, idx_from_vl(vl), 0);
3208
3209                 /* if (counter_select & CS_PORT_MARK_FECN)
3210                  *     write_csr(dd, DCC_PRF_PORT_VL_MARK_FECN_CNT + offset, 0);
3211                  */
3212                 if (counter_select & C_SW_XMIT_DSCD_VL)
3213                         write_port_cntr(ppd, C_SW_XMIT_DSCD_VL,
3214                                         idx_from_vl(vl), 0);
3215         }
3216
3217         if (resp_len)
3218                 *resp_len += sizeof(*req);
3219
3220         return reply((struct ib_mad_hdr *)pmp);
3221 }
3222
3223 static int pma_set_opa_errorinfo(struct opa_pma_mad *pmp,
3224                                  struct ib_device *ibdev,
3225                                  u8 port, u32 *resp_len)
3226 {
3227         struct _port_ei *rsp;
3228         struct opa_port_error_info_msg *req;
3229         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3230         u64 port_mask;
3231         u32 num_ports;
3232         u8 port_num;
3233         u8 num_pslm;
3234         u32 error_info_select;
3235
3236         req = (struct opa_port_error_info_msg *)pmp->data;
3237         rsp = &req->port[0];
3238
3239         num_ports = OPA_AM_NPORT(be32_to_cpu(pmp->mad_hdr.attr_mod));
3240         num_pslm = hweight64(be64_to_cpu(req->port_select_mask[3]));
3241
3242         memset(rsp, 0, sizeof(*rsp));
3243
3244         if (num_ports != 1 || num_ports != num_pslm) {
3245                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3246                 return reply((struct ib_mad_hdr *)pmp);
3247         }
3248
3249         /*
3250          * The bit set in the mask needs to be consistent with the port
3251          * the request came in on.
3252          */
3253         port_mask = be64_to_cpu(req->port_select_mask[3]);
3254         port_num = find_first_bit((unsigned long *)&port_mask,
3255                                   sizeof(port_mask));
3256
3257         if (port_num != port) {
3258                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
3259                 return reply((struct ib_mad_hdr *)pmp);
3260         }
3261
3262         error_info_select = be32_to_cpu(req->error_info_select_mask);
3263
3264         /* PortRcvErrorInfo */
3265         if (error_info_select & ES_PORT_RCV_ERROR_INFO)
3266                 /* turn off status bit */
3267                 dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3268
3269         /* ExcessiverBufferOverrunInfo */
3270         if (error_info_select & ES_EXCESSIVE_BUFFER_OVERRUN_INFO)
3271                 /*
3272                  * status bit is essentially kept in the h/w - bit 5 of
3273                  * RCV_ERR_INFO
3274                  */
3275                 write_csr(dd, RCV_ERR_INFO,
3276                           RCV_ERR_INFO_RCV_EXCESS_BUFFER_OVERRUN_SMASK);
3277
3278         if (error_info_select & ES_PORT_XMIT_CONSTRAINT_ERROR_INFO)
3279                 dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3280
3281         if (error_info_select & ES_PORT_RCV_CONSTRAINT_ERROR_INFO)
3282                 dd->err_info_rcv_constraint.status &= ~OPA_EI_STATUS_SMASK;
3283
3284         /* UncorrectableErrorInfo */
3285         if (error_info_select & ES_UNCORRECTABLE_ERROR_INFO)
3286                 /* turn off status bit */
3287                 dd->err_info_uncorrectable &= ~OPA_EI_STATUS_SMASK;
3288
3289         /* FMConfigErrorInfo */
3290         if (error_info_select & ES_FM_CONFIG_ERROR_INFO)
3291                 /* turn off status bit */
3292                 dd->err_info_fmconfig &= ~OPA_EI_STATUS_SMASK;
3293
3294         if (resp_len)
3295                 *resp_len += sizeof(*req);
3296
3297         return reply((struct ib_mad_hdr *)pmp);
3298 }
3299
3300 struct opa_congestion_info_attr {
3301         __be16 congestion_info;
3302         u8 control_table_cap;   /* Multiple of 64 entry unit CCTs */
3303         u8 congestion_log_length;
3304 } __packed;
3305
3306 static int __subn_get_opa_cong_info(struct opa_smp *smp, u32 am, u8 *data,
3307                                     struct ib_device *ibdev, u8 port,
3308                                     u32 *resp_len)
3309 {
3310         struct opa_congestion_info_attr *p =
3311                 (struct opa_congestion_info_attr *)data;
3312         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3313         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3314
3315         p->congestion_info = 0;
3316         p->control_table_cap = ppd->cc_max_table_entries;
3317         p->congestion_log_length = OPA_CONG_LOG_ELEMS;
3318
3319         if (resp_len)
3320                 *resp_len += sizeof(*p);
3321
3322         return reply((struct ib_mad_hdr *)smp);
3323 }
3324
3325 static int __subn_get_opa_cong_setting(struct opa_smp *smp, u32 am,
3326                                        u8 *data, struct ib_device *ibdev,
3327                                        u8 port, u32 *resp_len)
3328 {
3329         int i;
3330         struct opa_congestion_setting_attr *p =
3331                 (struct opa_congestion_setting_attr *)data;
3332         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3333         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3334         struct opa_congestion_setting_entry_shadow *entries;
3335         struct cc_state *cc_state;
3336
3337         rcu_read_lock();
3338
3339         cc_state = get_cc_state(ppd);
3340
3341         if (!cc_state) {
3342                 rcu_read_unlock();
3343                 return reply((struct ib_mad_hdr *)smp);
3344         }
3345
3346         entries = cc_state->cong_setting.entries;
3347         p->port_control = cpu_to_be16(cc_state->cong_setting.port_control);
3348         p->control_map = cpu_to_be32(cc_state->cong_setting.control_map);
3349         for (i = 0; i < OPA_MAX_SLS; i++) {
3350                 p->entries[i].ccti_increase = entries[i].ccti_increase;
3351                 p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
3352                 p->entries[i].trigger_threshold =
3353                         entries[i].trigger_threshold;
3354                 p->entries[i].ccti_min = entries[i].ccti_min;
3355         }
3356
3357         rcu_read_unlock();
3358
3359         if (resp_len)
3360                 *resp_len += sizeof(*p);
3361
3362         return reply((struct ib_mad_hdr *)smp);
3363 }
3364
3365 /*
3366  * Apply congestion control information stored in the ppd to the
3367  * active structure.
3368  */
3369 static void apply_cc_state(struct hfi1_pportdata *ppd)
3370 {
3371         struct cc_state *old_cc_state, *new_cc_state;
3372
3373         new_cc_state = kzalloc(sizeof(*new_cc_state), GFP_KERNEL);
3374         if (!new_cc_state)
3375                 return;
3376
3377         /*
3378          * Hold the lock for updating *and* to prevent ppd information
3379          * from changing during the update.
3380          */
3381         spin_lock(&ppd->cc_state_lock);
3382
3383         old_cc_state = get_cc_state_protected(ppd);
3384         if (!old_cc_state) {
3385                 /* never active, or shutting down */
3386                 spin_unlock(&ppd->cc_state_lock);
3387                 kfree(new_cc_state);
3388                 return;
3389         }
3390
3391         *new_cc_state = *old_cc_state;
3392
3393         new_cc_state->cct.ccti_limit = ppd->total_cct_entry - 1;
3394         memcpy(new_cc_state->cct.entries, ppd->ccti_entries,
3395                ppd->total_cct_entry * sizeof(struct ib_cc_table_entry));
3396
3397         new_cc_state->cong_setting.port_control = IB_CC_CCS_PC_SL_BASED;
3398         new_cc_state->cong_setting.control_map = ppd->cc_sl_control_map;
3399         memcpy(new_cc_state->cong_setting.entries, ppd->congestion_entries,
3400                OPA_MAX_SLS * sizeof(struct opa_congestion_setting_entry));
3401
3402         rcu_assign_pointer(ppd->cc_state, new_cc_state);
3403
3404         spin_unlock(&ppd->cc_state_lock);
3405
3406         kfree_rcu(old_cc_state, rcu);
3407 }
3408
3409 static int __subn_set_opa_cong_setting(struct opa_smp *smp, u32 am, u8 *data,
3410                                        struct ib_device *ibdev, u8 port,
3411                                        u32 *resp_len)
3412 {
3413         struct opa_congestion_setting_attr *p =
3414                 (struct opa_congestion_setting_attr *)data;
3415         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3416         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3417         struct opa_congestion_setting_entry_shadow *entries;
3418         int i;
3419
3420         /*
3421          * Save details from packet into the ppd.  Hold the cc_state_lock so
3422          * our information is consistent with anyone trying to apply the state.
3423          */
3424         spin_lock(&ppd->cc_state_lock);
3425         ppd->cc_sl_control_map = be32_to_cpu(p->control_map);
3426
3427         entries = ppd->congestion_entries;
3428         for (i = 0; i < OPA_MAX_SLS; i++) {
3429                 entries[i].ccti_increase = p->entries[i].ccti_increase;
3430                 entries[i].ccti_timer = be16_to_cpu(p->entries[i].ccti_timer);
3431                 entries[i].trigger_threshold =
3432                         p->entries[i].trigger_threshold;
3433                 entries[i].ccti_min = p->entries[i].ccti_min;
3434         }
3435         spin_unlock(&ppd->cc_state_lock);
3436
3437         /* now apply the information */
3438         apply_cc_state(ppd);
3439
3440         return __subn_get_opa_cong_setting(smp, am, data, ibdev, port,
3441                                            resp_len);
3442 }
3443
3444 static int __subn_get_opa_hfi1_cong_log(struct opa_smp *smp, u32 am,
3445                                         u8 *data, struct ib_device *ibdev,
3446                                         u8 port, u32 *resp_len)
3447 {
3448         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3449         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3450         struct opa_hfi1_cong_log *cong_log = (struct opa_hfi1_cong_log *)data;
3451         s64 ts;
3452         int i;
3453
3454         if (am != 0) {
3455                 smp->status |= IB_SMP_INVALID_FIELD;
3456                 return reply((struct ib_mad_hdr *)smp);
3457         }
3458
3459         spin_lock_irq(&ppd->cc_log_lock);
3460
3461         cong_log->log_type = OPA_CC_LOG_TYPE_HFI;
3462         cong_log->congestion_flags = 0;
3463         cong_log->threshold_event_counter =
3464                 cpu_to_be16(ppd->threshold_event_counter);
3465         memcpy(cong_log->threshold_cong_event_map,
3466                ppd->threshold_cong_event_map,
3467                sizeof(cong_log->threshold_cong_event_map));
3468         /* keep timestamp in units of 1.024 usec */
3469         ts = ktime_to_ns(ktime_get()) / 1024;
3470         cong_log->current_time_stamp = cpu_to_be32(ts);
3471         for (i = 0; i < OPA_CONG_LOG_ELEMS; i++) {
3472                 struct opa_hfi1_cong_log_event_internal *cce =
3473                         &ppd->cc_events[ppd->cc_mad_idx++];
3474                 if (ppd->cc_mad_idx == OPA_CONG_LOG_ELEMS)
3475                         ppd->cc_mad_idx = 0;
3476                 /*
3477                  * Entries which are older than twice the time
3478                  * required to wrap the counter are supposed to
3479                  * be zeroed (CA10-49 IBTA, release 1.2.1, V1).
3480                  */
3481                 if ((u64)(ts - cce->timestamp) > (2 * UINT_MAX))
3482                         continue;
3483                 memcpy(cong_log->events[i].local_qp_cn_entry, &cce->lqpn, 3);
3484                 memcpy(cong_log->events[i].remote_qp_number_cn_entry,
3485                        &cce->rqpn, 3);
3486                 cong_log->events[i].sl_svc_type_cn_entry =
3487                         ((cce->sl & 0x1f) << 3) | (cce->svc_type & 0x7);
3488                 cong_log->events[i].remote_lid_cn_entry =
3489                         cpu_to_be32(cce->rlid);
3490                 cong_log->events[i].timestamp_cn_entry =
3491                         cpu_to_be32(cce->timestamp);
3492         }
3493
3494         /*
3495          * Reset threshold_cong_event_map, and threshold_event_counter
3496          * to 0 when log is read.
3497          */
3498         memset(ppd->threshold_cong_event_map, 0x0,
3499                sizeof(ppd->threshold_cong_event_map));
3500         ppd->threshold_event_counter = 0;
3501
3502         spin_unlock_irq(&ppd->cc_log_lock);
3503
3504         if (resp_len)
3505                 *resp_len += sizeof(struct opa_hfi1_cong_log);
3506
3507         return reply((struct ib_mad_hdr *)smp);
3508 }
3509
3510 static int __subn_get_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3511                                    struct ib_device *ibdev, u8 port,
3512                                    u32 *resp_len)
3513 {
3514         struct ib_cc_table_attr *cc_table_attr =
3515                 (struct ib_cc_table_attr *)data;
3516         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3517         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3518         u32 start_block = OPA_AM_START_BLK(am);
3519         u32 n_blocks = OPA_AM_NBLK(am);
3520         struct ib_cc_table_entry_shadow *entries;
3521         int i, j;
3522         u32 sentry, eentry;
3523         struct cc_state *cc_state;
3524
3525         /* sanity check n_blocks, start_block */
3526         if (n_blocks == 0 ||
3527             start_block + n_blocks > ppd->cc_max_table_entries) {
3528                 smp->status |= IB_SMP_INVALID_FIELD;
3529                 return reply((struct ib_mad_hdr *)smp);
3530         }
3531
3532         rcu_read_lock();
3533
3534         cc_state = get_cc_state(ppd);
3535
3536         if (!cc_state) {
3537                 rcu_read_unlock();
3538                 return reply((struct ib_mad_hdr *)smp);
3539         }
3540
3541         sentry = start_block * IB_CCT_ENTRIES;
3542         eentry = sentry + (IB_CCT_ENTRIES * n_blocks);
3543
3544         cc_table_attr->ccti_limit = cpu_to_be16(cc_state->cct.ccti_limit);
3545
3546         entries = cc_state->cct.entries;
3547
3548         /* return n_blocks, though the last block may not be full */
3549         for (j = 0, i = sentry; i < eentry; j++, i++)
3550                 cc_table_attr->ccti_entries[j].entry =
3551                         cpu_to_be16(entries[i].entry);
3552
3553         rcu_read_unlock();
3554
3555         if (resp_len)
3556                 *resp_len += sizeof(u16) * (IB_CCT_ENTRIES * n_blocks + 1);
3557
3558         return reply((struct ib_mad_hdr *)smp);
3559 }
3560
3561 static int __subn_set_opa_cc_table(struct opa_smp *smp, u32 am, u8 *data,
3562                                    struct ib_device *ibdev, u8 port,
3563                                    u32 *resp_len)
3564 {
3565         struct ib_cc_table_attr *p = (struct ib_cc_table_attr *)data;
3566         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3567         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3568         u32 start_block = OPA_AM_START_BLK(am);
3569         u32 n_blocks = OPA_AM_NBLK(am);
3570         struct ib_cc_table_entry_shadow *entries;
3571         int i, j;
3572         u32 sentry, eentry;
3573         u16 ccti_limit;
3574
3575         /* sanity check n_blocks, start_block */
3576         if (n_blocks == 0 ||
3577             start_block + n_blocks > ppd->cc_max_table_entries) {
3578                 smp->status |= IB_SMP_INVALID_FIELD;
3579                 return reply((struct ib_mad_hdr *)smp);
3580         }
3581
3582         sentry = start_block * IB_CCT_ENTRIES;
3583         eentry = sentry + ((n_blocks - 1) * IB_CCT_ENTRIES) +
3584                  (be16_to_cpu(p->ccti_limit)) % IB_CCT_ENTRIES + 1;
3585
3586         /* sanity check ccti_limit */
3587         ccti_limit = be16_to_cpu(p->ccti_limit);
3588         if (ccti_limit + 1 > eentry) {
3589                 smp->status |= IB_SMP_INVALID_FIELD;
3590                 return reply((struct ib_mad_hdr *)smp);
3591         }
3592
3593         /*
3594          * Save details from packet into the ppd.  Hold the cc_state_lock so
3595          * our information is consistent with anyone trying to apply the state.
3596          */
3597         spin_lock(&ppd->cc_state_lock);
3598         ppd->total_cct_entry = ccti_limit + 1;
3599         entries = ppd->ccti_entries;
3600         for (j = 0, i = sentry; i < eentry; j++, i++)
3601                 entries[i].entry = be16_to_cpu(p->ccti_entries[j].entry);
3602         spin_unlock(&ppd->cc_state_lock);
3603
3604         /* now apply the information */
3605         apply_cc_state(ppd);
3606
3607         return __subn_get_opa_cc_table(smp, am, data, ibdev, port, resp_len);
3608 }
3609
3610 struct opa_led_info {
3611         __be32 rsvd_led_mask;
3612         __be32 rsvd;
3613 };
3614
3615 #define OPA_LED_SHIFT   31
3616 #define OPA_LED_MASK    BIT(OPA_LED_SHIFT)
3617
3618 static int __subn_get_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3619                                    struct ib_device *ibdev, u8 port,
3620                                    u32 *resp_len)
3621 {
3622         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3623         struct hfi1_pportdata *ppd = dd->pport;
3624         struct opa_led_info *p = (struct opa_led_info *)data;
3625         u32 nport = OPA_AM_NPORT(am);
3626         u32 is_beaconing_active;
3627
3628         if (nport != 1) {
3629                 smp->status |= IB_SMP_INVALID_FIELD;
3630                 return reply((struct ib_mad_hdr *)smp);
3631         }
3632
3633         /*
3634          * This pairs with the memory barrier in hfi1_start_led_override to
3635          * ensure that we read the correct state of LED beaconing represented
3636          * by led_override_timer_active
3637          */
3638         smp_rmb();
3639         is_beaconing_active = !!atomic_read(&ppd->led_override_timer_active);
3640         p->rsvd_led_mask = cpu_to_be32(is_beaconing_active << OPA_LED_SHIFT);
3641
3642         if (resp_len)
3643                 *resp_len += sizeof(struct opa_led_info);
3644
3645         return reply((struct ib_mad_hdr *)smp);
3646 }
3647
3648 static int __subn_set_opa_led_info(struct opa_smp *smp, u32 am, u8 *data,
3649                                    struct ib_device *ibdev, u8 port,
3650                                    u32 *resp_len)
3651 {
3652         struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
3653         struct opa_led_info *p = (struct opa_led_info *)data;
3654         u32 nport = OPA_AM_NPORT(am);
3655         int on = !!(be32_to_cpu(p->rsvd_led_mask) & OPA_LED_MASK);
3656
3657         if (nport != 1) {
3658                 smp->status |= IB_SMP_INVALID_FIELD;
3659                 return reply((struct ib_mad_hdr *)smp);
3660         }
3661
3662         if (on)
3663                 hfi1_start_led_override(dd->pport, 2000, 1500);
3664         else
3665                 shutdown_led_override(dd->pport);
3666
3667         return __subn_get_opa_led_info(smp, am, data, ibdev, port, resp_len);
3668 }
3669
3670 static int subn_get_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3671                             u8 *data, struct ib_device *ibdev, u8 port,
3672                             u32 *resp_len)
3673 {
3674         int ret;
3675         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3676
3677         switch (attr_id) {
3678         case IB_SMP_ATTR_NODE_DESC:
3679                 ret = __subn_get_opa_nodedesc(smp, am, data, ibdev, port,
3680                                               resp_len);
3681                 break;
3682         case IB_SMP_ATTR_NODE_INFO:
3683                 ret = __subn_get_opa_nodeinfo(smp, am, data, ibdev, port,
3684                                               resp_len);
3685                 break;
3686         case IB_SMP_ATTR_PORT_INFO:
3687                 ret = __subn_get_opa_portinfo(smp, am, data, ibdev, port,
3688                                               resp_len);
3689                 break;
3690         case IB_SMP_ATTR_PKEY_TABLE:
3691                 ret = __subn_get_opa_pkeytable(smp, am, data, ibdev, port,
3692                                                resp_len);
3693                 break;
3694         case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3695                 ret = __subn_get_opa_sl_to_sc(smp, am, data, ibdev, port,
3696                                               resp_len);
3697                 break;
3698         case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3699                 ret = __subn_get_opa_sc_to_sl(smp, am, data, ibdev, port,
3700                                               resp_len);
3701                 break;
3702         case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3703                 ret = __subn_get_opa_sc_to_vlt(smp, am, data, ibdev, port,
3704                                                resp_len);
3705                 break;
3706         case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3707                 ret = __subn_get_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3708                                                 resp_len);
3709                 break;
3710         case OPA_ATTRIB_ID_PORT_STATE_INFO:
3711                 ret = __subn_get_opa_psi(smp, am, data, ibdev, port,
3712                                          resp_len);
3713                 break;
3714         case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3715                 ret = __subn_get_opa_bct(smp, am, data, ibdev, port,
3716                                          resp_len);
3717                 break;
3718         case OPA_ATTRIB_ID_CABLE_INFO:
3719                 ret = __subn_get_opa_cable_info(smp, am, data, ibdev, port,
3720                                                 resp_len);
3721                 break;
3722         case IB_SMP_ATTR_VL_ARB_TABLE:
3723                 ret = __subn_get_opa_vl_arb(smp, am, data, ibdev, port,
3724                                             resp_len);
3725                 break;
3726         case OPA_ATTRIB_ID_CONGESTION_INFO:
3727                 ret = __subn_get_opa_cong_info(smp, am, data, ibdev, port,
3728                                                resp_len);
3729                 break;
3730         case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3731                 ret = __subn_get_opa_cong_setting(smp, am, data, ibdev,
3732                                                   port, resp_len);
3733                 break;
3734         case OPA_ATTRIB_ID_HFI_CONGESTION_LOG:
3735                 ret = __subn_get_opa_hfi1_cong_log(smp, am, data, ibdev,
3736                                                    port, resp_len);
3737                 break;
3738         case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3739                 ret = __subn_get_opa_cc_table(smp, am, data, ibdev, port,
3740                                               resp_len);
3741                 break;
3742         case IB_SMP_ATTR_LED_INFO:
3743                 ret = __subn_get_opa_led_info(smp, am, data, ibdev, port,
3744                                               resp_len);
3745                 break;
3746         case IB_SMP_ATTR_SM_INFO:
3747                 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
3748                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
3749                 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
3750                         return IB_MAD_RESULT_SUCCESS;
3751                 /* FALLTHROUGH */
3752         default:
3753                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3754                 ret = reply((struct ib_mad_hdr *)smp);
3755                 break;
3756         }
3757         return ret;
3758 }
3759
3760 static int subn_set_opa_sma(__be16 attr_id, struct opa_smp *smp, u32 am,
3761                             u8 *data, struct ib_device *ibdev, u8 port,
3762                             u32 *resp_len)
3763 {
3764         int ret;
3765         struct hfi1_ibport *ibp = to_iport(ibdev, port);
3766
3767         switch (attr_id) {
3768         case IB_SMP_ATTR_PORT_INFO:
3769                 ret = __subn_set_opa_portinfo(smp, am, data, ibdev, port,
3770                                               resp_len);
3771                 break;
3772         case IB_SMP_ATTR_PKEY_TABLE:
3773                 ret = __subn_set_opa_pkeytable(smp, am, data, ibdev, port,
3774                                                resp_len);
3775                 break;
3776         case OPA_ATTRIB_ID_SL_TO_SC_MAP:
3777                 ret = __subn_set_opa_sl_to_sc(smp, am, data, ibdev, port,
3778                                               resp_len);
3779                 break;
3780         case OPA_ATTRIB_ID_SC_TO_SL_MAP:
3781                 ret = __subn_set_opa_sc_to_sl(smp, am, data, ibdev, port,
3782                                               resp_len);
3783                 break;
3784         case OPA_ATTRIB_ID_SC_TO_VLT_MAP:
3785                 ret = __subn_set_opa_sc_to_vlt(smp, am, data, ibdev, port,
3786                                                resp_len);
3787                 break;
3788         case OPA_ATTRIB_ID_SC_TO_VLNT_MAP:
3789                 ret = __subn_set_opa_sc_to_vlnt(smp, am, data, ibdev, port,
3790                                                 resp_len);
3791                 break;
3792         case OPA_ATTRIB_ID_PORT_STATE_INFO:
3793                 ret = __subn_set_opa_psi(smp, am, data, ibdev, port,
3794                                          resp_len);
3795                 break;
3796         case OPA_ATTRIB_ID_BUFFER_CONTROL_TABLE:
3797                 ret = __subn_set_opa_bct(smp, am, data, ibdev, port,
3798                                          resp_len);
3799                 break;
3800         case IB_SMP_ATTR_VL_ARB_TABLE:
3801                 ret = __subn_set_opa_vl_arb(smp, am, data, ibdev, port,
3802                                             resp_len);
3803                 break;
3804         case OPA_ATTRIB_ID_HFI_CONGESTION_SETTING:
3805                 ret = __subn_set_opa_cong_setting(smp, am, data, ibdev,
3806                                                   port, resp_len);
3807                 break;
3808         case OPA_ATTRIB_ID_CONGESTION_CONTROL_TABLE:
3809                 ret = __subn_set_opa_cc_table(smp, am, data, ibdev, port,
3810                                               resp_len);
3811                 break;
3812         case IB_SMP_ATTR_LED_INFO:
3813                 ret = __subn_set_opa_led_info(smp, am, data, ibdev, port,
3814                                               resp_len);
3815                 break;
3816         case IB_SMP_ATTR_SM_INFO:
3817                 if (ibp->rvp.port_cap_flags & IB_PORT_SM_DISABLED)
3818                         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
3819                 if (ibp->rvp.port_cap_flags & IB_PORT_SM)
3820                         return IB_MAD_RESULT_SUCCESS;
3821                 /* FALLTHROUGH */
3822         default:
3823                 smp->status |= IB_SMP_UNSUP_METH_ATTR;
3824                 ret = reply((struct ib_mad_hdr *)smp);
3825                 break;
3826         }
3827         return ret;
3828 }
3829
3830 static inline void set_aggr_error(struct opa_aggregate *ag)
3831 {
3832         ag->err_reqlength |= cpu_to_be16(0x8000);
3833 }
3834
3835 static int subn_get_opa_aggregate(struct opa_smp *smp,
3836                                   struct ib_device *ibdev, u8 port,
3837                                   u32 *resp_len)
3838 {
3839         int i;
3840         u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3841         u8 *next_smp = opa_get_smp_data(smp);
3842
3843         if (num_attr < 1 || num_attr > 117) {
3844                 smp->status |= IB_SMP_INVALID_FIELD;
3845                 return reply((struct ib_mad_hdr *)smp);
3846         }
3847
3848         for (i = 0; i < num_attr; i++) {
3849                 struct opa_aggregate *agg;
3850                 size_t agg_data_len;
3851                 size_t agg_size;
3852                 u32 am;
3853
3854                 agg = (struct opa_aggregate *)next_smp;
3855                 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3856                 agg_size = sizeof(*agg) + agg_data_len;
3857                 am = be32_to_cpu(agg->attr_mod);
3858
3859                 *resp_len += agg_size;
3860
3861                 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3862                         smp->status |= IB_SMP_INVALID_FIELD;
3863                         return reply((struct ib_mad_hdr *)smp);
3864                 }
3865
3866                 /* zero the payload for this segment */
3867                 memset(next_smp + sizeof(*agg), 0, agg_data_len);
3868
3869                 (void)subn_get_opa_sma(agg->attr_id, smp, am, agg->data,
3870                                         ibdev, port, NULL);
3871                 if (smp->status & ~IB_SMP_DIRECTION) {
3872                         set_aggr_error(agg);
3873                         return reply((struct ib_mad_hdr *)smp);
3874                 }
3875                 next_smp += agg_size;
3876         }
3877
3878         return reply((struct ib_mad_hdr *)smp);
3879 }
3880
3881 static int subn_set_opa_aggregate(struct opa_smp *smp,
3882                                   struct ib_device *ibdev, u8 port,
3883                                   u32 *resp_len)
3884 {
3885         int i;
3886         u32 num_attr = be32_to_cpu(smp->attr_mod) & 0x000000ff;
3887         u8 *next_smp = opa_get_smp_data(smp);
3888
3889         if (num_attr < 1 || num_attr > 117) {
3890                 smp->status |= IB_SMP_INVALID_FIELD;
3891                 return reply((struct ib_mad_hdr *)smp);
3892         }
3893
3894         for (i = 0; i < num_attr; i++) {
3895                 struct opa_aggregate *agg;
3896                 size_t agg_data_len;
3897                 size_t agg_size;
3898                 u32 am;
3899
3900                 agg = (struct opa_aggregate *)next_smp;
3901                 agg_data_len = (be16_to_cpu(agg->err_reqlength) & 0x007f) * 8;
3902                 agg_size = sizeof(*agg) + agg_data_len;
3903                 am = be32_to_cpu(agg->attr_mod);
3904
3905                 *resp_len += agg_size;
3906
3907                 if (next_smp + agg_size > ((u8 *)smp) + sizeof(*smp)) {
3908                         smp->status |= IB_SMP_INVALID_FIELD;
3909                         return reply((struct ib_mad_hdr *)smp);
3910                 }
3911
3912                 (void)subn_set_opa_sma(agg->attr_id, smp, am, agg->data,
3913                                         ibdev, port, NULL);
3914                 if (smp->status & ~IB_SMP_DIRECTION) {
3915                         set_aggr_error(agg);
3916                         return reply((struct ib_mad_hdr *)smp);
3917                 }
3918                 next_smp += agg_size;
3919         }
3920
3921         return reply((struct ib_mad_hdr *)smp);
3922 }
3923
3924 /*
3925  * OPAv1 specifies that, on the transition to link up, these counters
3926  * are cleared:
3927  *   PortRcvErrors [*]
3928  *   LinkErrorRecovery
3929  *   LocalLinkIntegrityErrors
3930  *   ExcessiveBufferOverruns [*]
3931  *
3932  * [*] Error info associated with these counters is retained, but the
3933  * error info status is reset to 0.
3934  */
3935 void clear_linkup_counters(struct hfi1_devdata *dd)
3936 {
3937         /* PortRcvErrors */
3938         write_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL, 0);
3939         dd->err_info_rcvport.status_and_code &= ~OPA_EI_STATUS_SMASK;
3940         /* LinkErrorRecovery */
3941         write_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL, 0);
3942         write_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT, CNTR_INVALID_VL, 0);
3943         /* LocalLinkIntegrityErrors */
3944         write_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL, 0);
3945         /* ExcessiveBufferOverruns */
3946         write_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL, 0);
3947         dd->rcv_ovfl_cnt = 0;
3948         dd->err_info_xmit_constraint.status &= ~OPA_EI_STATUS_SMASK;
3949 }
3950
3951 /*
3952  * is_local_mad() returns 1 if 'mad' is sent from, and destined to the
3953  * local node, 0 otherwise.
3954  */
3955 static int is_local_mad(struct hfi1_ibport *ibp, const struct opa_mad *mad,
3956                         const struct ib_wc *in_wc)
3957 {
3958         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3959         const struct opa_smp *smp = (const struct opa_smp *)mad;
3960
3961         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
3962                 return (smp->hop_cnt == 0 &&
3963                         smp->route.dr.dr_slid == OPA_LID_PERMISSIVE &&
3964                         smp->route.dr.dr_dlid == OPA_LID_PERMISSIVE);
3965         }
3966
3967         return (in_wc->slid == ppd->lid);
3968 }
3969
3970 /*
3971  * opa_local_smp_check() should only be called on MADs for which
3972  * is_local_mad() returns true. It applies the SMP checks that are
3973  * specific to SMPs which are sent from, and destined to this node.
3974  * opa_local_smp_check() returns 0 if the SMP passes its checks, 1
3975  * otherwise.
3976  *
3977  * SMPs which arrive from other nodes are instead checked by
3978  * opa_smp_check().
3979  */
3980 static int opa_local_smp_check(struct hfi1_ibport *ibp,
3981                                const struct ib_wc *in_wc)
3982 {
3983         struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
3984         u16 slid = in_wc->slid;
3985         u16 pkey;
3986
3987         if (in_wc->pkey_index >= ARRAY_SIZE(ppd->pkeys))
3988                 return 1;
3989
3990         pkey = ppd->pkeys[in_wc->pkey_index];
3991         /*
3992          * We need to do the "node-local" checks specified in OPAv1,
3993          * rev 0.90, section 9.10.26, which are:
3994          *   - pkey is 0x7fff, or 0xffff
3995          *   - Source QPN == 0 || Destination QPN == 0
3996          *   - the MAD header's management class is either
3997          *     IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE or
3998          *     IB_MGMT_CLASS_SUBN_LID_ROUTED
3999          *   - SLID != 0
4000          *
4001          * However, we know (and so don't need to check again) that,
4002          * for local SMPs, the MAD stack passes MADs with:
4003          *   - Source QPN of 0
4004          *   - MAD mgmt_class is IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4005          *   - SLID is either: OPA_LID_PERMISSIVE (0xFFFFFFFF), or
4006          *     our own port's lid
4007          *
4008          */
4009         if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY)
4010                 return 0;
4011         ingress_pkey_table_fail(ppd, pkey, slid);
4012         return 1;
4013 }
4014
4015 static int process_subn_opa(struct ib_device *ibdev, int mad_flags,
4016                             u8 port, const struct opa_mad *in_mad,
4017                             struct opa_mad *out_mad,
4018                             u32 *resp_len)
4019 {
4020         struct opa_smp *smp = (struct opa_smp *)out_mad;
4021         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4022         u8 *data;
4023         u32 am;
4024         __be16 attr_id;
4025         int ret;
4026
4027         *out_mad = *in_mad;
4028         data = opa_get_smp_data(smp);
4029
4030         am = be32_to_cpu(smp->attr_mod);
4031         attr_id = smp->attr_id;
4032         if (smp->class_version != OPA_SMI_CLASS_VERSION) {
4033                 smp->status |= IB_SMP_UNSUP_VERSION;
4034                 ret = reply((struct ib_mad_hdr *)smp);
4035                 return ret;
4036         }
4037         ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags, smp->mkey,
4038                          smp->route.dr.dr_slid, smp->route.dr.return_path,
4039                          smp->hop_cnt);
4040         if (ret) {
4041                 u32 port_num = be32_to_cpu(smp->attr_mod);
4042
4043                 /*
4044                  * If this is a get/set portinfo, we already check the
4045                  * M_Key if the MAD is for another port and the M_Key
4046                  * is OK on the receiving port. This check is needed
4047                  * to increment the error counters when the M_Key
4048                  * fails to match on *both* ports.
4049                  */
4050                 if (attr_id == IB_SMP_ATTR_PORT_INFO &&
4051                     (smp->method == IB_MGMT_METHOD_GET ||
4052                      smp->method == IB_MGMT_METHOD_SET) &&
4053                     port_num && port_num <= ibdev->phys_port_cnt &&
4054                     port != port_num)
4055                         (void)check_mkey(to_iport(ibdev, port_num),
4056                                           (struct ib_mad_hdr *)smp, 0,
4057                                           smp->mkey, smp->route.dr.dr_slid,
4058                                           smp->route.dr.return_path,
4059                                           smp->hop_cnt);
4060                 ret = IB_MAD_RESULT_FAILURE;
4061                 return ret;
4062         }
4063
4064         *resp_len = opa_get_smp_header_size(smp);
4065
4066         switch (smp->method) {
4067         case IB_MGMT_METHOD_GET:
4068                 switch (attr_id) {
4069                 default:
4070                         clear_opa_smp_data(smp);
4071                         ret = subn_get_opa_sma(attr_id, smp, am, data,
4072                                                ibdev, port, resp_len);
4073                         break;
4074                 case OPA_ATTRIB_ID_AGGREGATE:
4075                         ret = subn_get_opa_aggregate(smp, ibdev, port,
4076                                                      resp_len);
4077                         break;
4078                 }
4079                 break;
4080         case IB_MGMT_METHOD_SET:
4081                 switch (attr_id) {
4082                 default:
4083                         ret = subn_set_opa_sma(attr_id, smp, am, data,
4084                                                ibdev, port, resp_len);
4085                         break;
4086                 case OPA_ATTRIB_ID_AGGREGATE:
4087                         ret = subn_set_opa_aggregate(smp, ibdev, port,
4088                                                      resp_len);
4089                         break;
4090                 }
4091                 break;
4092         case IB_MGMT_METHOD_TRAP:
4093         case IB_MGMT_METHOD_REPORT:
4094         case IB_MGMT_METHOD_REPORT_RESP:
4095         case IB_MGMT_METHOD_GET_RESP:
4096                 /*
4097                  * The ib_mad module will call us to process responses
4098                  * before checking for other consumers.
4099                  * Just tell the caller to process it normally.
4100                  */
4101                 ret = IB_MAD_RESULT_SUCCESS;
4102                 break;
4103         default:
4104                 smp->status |= IB_SMP_UNSUP_METHOD;
4105                 ret = reply((struct ib_mad_hdr *)smp);
4106                 break;
4107         }
4108
4109         return ret;
4110 }
4111
4112 static int process_subn(struct ib_device *ibdev, int mad_flags,
4113                         u8 port, const struct ib_mad *in_mad,
4114                         struct ib_mad *out_mad)
4115 {
4116         struct ib_smp *smp = (struct ib_smp *)out_mad;
4117         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4118         int ret;
4119
4120         *out_mad = *in_mad;
4121         if (smp->class_version != 1) {
4122                 smp->status |= IB_SMP_UNSUP_VERSION;
4123                 ret = reply((struct ib_mad_hdr *)smp);
4124                 return ret;
4125         }
4126
4127         ret = check_mkey(ibp, (struct ib_mad_hdr *)smp, mad_flags,
4128                          smp->mkey, (__force __be32)smp->dr_slid,
4129                          smp->return_path, smp->hop_cnt);
4130         if (ret) {
4131                 u32 port_num = be32_to_cpu(smp->attr_mod);
4132
4133                 /*
4134                  * If this is a get/set portinfo, we already check the
4135                  * M_Key if the MAD is for another port and the M_Key
4136                  * is OK on the receiving port. This check is needed
4137                  * to increment the error counters when the M_Key
4138                  * fails to match on *both* ports.
4139                  */
4140                 if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
4141                     (smp->method == IB_MGMT_METHOD_GET ||
4142                      smp->method == IB_MGMT_METHOD_SET) &&
4143                     port_num && port_num <= ibdev->phys_port_cnt &&
4144                     port != port_num)
4145                         (void)check_mkey(to_iport(ibdev, port_num),
4146                                          (struct ib_mad_hdr *)smp, 0,
4147                                          smp->mkey,
4148                                          (__force __be32)smp->dr_slid,
4149                                          smp->return_path, smp->hop_cnt);
4150                 ret = IB_MAD_RESULT_FAILURE;
4151                 return ret;
4152         }
4153
4154         switch (smp->method) {
4155         case IB_MGMT_METHOD_GET:
4156                 switch (smp->attr_id) {
4157                 case IB_SMP_ATTR_NODE_INFO:
4158                         ret = subn_get_nodeinfo(smp, ibdev, port);
4159                         break;
4160                 default:
4161                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
4162                         ret = reply((struct ib_mad_hdr *)smp);
4163                         break;
4164                 }
4165                 break;
4166         }
4167
4168         return ret;
4169 }
4170
4171 static int process_perf(struct ib_device *ibdev, u8 port,
4172                         const struct ib_mad *in_mad,
4173                         struct ib_mad *out_mad)
4174 {
4175         struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
4176         struct ib_class_port_info *cpi = (struct ib_class_port_info *)
4177                                                 &pmp->data;
4178         int ret = IB_MAD_RESULT_FAILURE;
4179
4180         *out_mad = *in_mad;
4181         if (pmp->mad_hdr.class_version != 1) {
4182                 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4183                 ret = reply((struct ib_mad_hdr *)pmp);
4184                 return ret;
4185         }
4186
4187         switch (pmp->mad_hdr.method) {
4188         case IB_MGMT_METHOD_GET:
4189                 switch (pmp->mad_hdr.attr_id) {
4190                 case IB_PMA_PORT_COUNTERS:
4191                         ret = pma_get_ib_portcounters(pmp, ibdev, port);
4192                         break;
4193                 case IB_PMA_PORT_COUNTERS_EXT:
4194                         ret = pma_get_ib_portcounters_ext(pmp, ibdev, port);
4195                         break;
4196                 case IB_PMA_CLASS_PORT_INFO:
4197                         cpi->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
4198                         ret = reply((struct ib_mad_hdr *)pmp);
4199                         break;
4200                 default:
4201                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4202                         ret = reply((struct ib_mad_hdr *)pmp);
4203                         break;
4204                 }
4205                 break;
4206
4207         case IB_MGMT_METHOD_SET:
4208                 if (pmp->mad_hdr.attr_id) {
4209                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4210                         ret = reply((struct ib_mad_hdr *)pmp);
4211                 }
4212                 break;
4213
4214         case IB_MGMT_METHOD_TRAP:
4215         case IB_MGMT_METHOD_GET_RESP:
4216                 /*
4217                  * The ib_mad module will call us to process responses
4218                  * before checking for other consumers.
4219                  * Just tell the caller to process it normally.
4220                  */
4221                 ret = IB_MAD_RESULT_SUCCESS;
4222                 break;
4223
4224         default:
4225                 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4226                 ret = reply((struct ib_mad_hdr *)pmp);
4227                 break;
4228         }
4229
4230         return ret;
4231 }
4232
4233 static int process_perf_opa(struct ib_device *ibdev, u8 port,
4234                             const struct opa_mad *in_mad,
4235                             struct opa_mad *out_mad, u32 *resp_len)
4236 {
4237         struct opa_pma_mad *pmp = (struct opa_pma_mad *)out_mad;
4238         int ret;
4239
4240         *out_mad = *in_mad;
4241
4242         if (pmp->mad_hdr.class_version != OPA_SMI_CLASS_VERSION) {
4243                 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
4244                 return reply((struct ib_mad_hdr *)pmp);
4245         }
4246
4247         *resp_len = sizeof(pmp->mad_hdr);
4248
4249         switch (pmp->mad_hdr.method) {
4250         case IB_MGMT_METHOD_GET:
4251                 switch (pmp->mad_hdr.attr_id) {
4252                 case IB_PMA_CLASS_PORT_INFO:
4253                         ret = pma_get_opa_classportinfo(pmp, ibdev, resp_len);
4254                         break;
4255                 case OPA_PM_ATTRIB_ID_PORT_STATUS:
4256                         ret = pma_get_opa_portstatus(pmp, ibdev, port,
4257                                                      resp_len);
4258                         break;
4259                 case OPA_PM_ATTRIB_ID_DATA_PORT_COUNTERS:
4260                         ret = pma_get_opa_datacounters(pmp, ibdev, port,
4261                                                        resp_len);
4262                         break;
4263                 case OPA_PM_ATTRIB_ID_ERROR_PORT_COUNTERS:
4264                         ret = pma_get_opa_porterrors(pmp, ibdev, port,
4265                                                      resp_len);
4266                         break;
4267                 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4268                         ret = pma_get_opa_errorinfo(pmp, ibdev, port,
4269                                                     resp_len);
4270                         break;
4271                 default:
4272                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4273                         ret = reply((struct ib_mad_hdr *)pmp);
4274                         break;
4275                 }
4276                 break;
4277
4278         case IB_MGMT_METHOD_SET:
4279                 switch (pmp->mad_hdr.attr_id) {
4280                 case OPA_PM_ATTRIB_ID_CLEAR_PORT_STATUS:
4281                         ret = pma_set_opa_portstatus(pmp, ibdev, port,
4282                                                      resp_len);
4283                         break;
4284                 case OPA_PM_ATTRIB_ID_ERROR_INFO:
4285                         ret = pma_set_opa_errorinfo(pmp, ibdev, port,
4286                                                     resp_len);
4287                         break;
4288                 default:
4289                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
4290                         ret = reply((struct ib_mad_hdr *)pmp);
4291                         break;
4292                 }
4293                 break;
4294
4295         case IB_MGMT_METHOD_TRAP:
4296         case IB_MGMT_METHOD_GET_RESP:
4297                 /*
4298                  * The ib_mad module will call us to process responses
4299                  * before checking for other consumers.
4300                  * Just tell the caller to process it normally.
4301                  */
4302                 ret = IB_MAD_RESULT_SUCCESS;
4303                 break;
4304
4305         default:
4306                 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
4307                 ret = reply((struct ib_mad_hdr *)pmp);
4308                 break;
4309         }
4310
4311         return ret;
4312 }
4313
4314 static int hfi1_process_opa_mad(struct ib_device *ibdev, int mad_flags,
4315                                 u8 port, const struct ib_wc *in_wc,
4316                                 const struct ib_grh *in_grh,
4317                                 const struct opa_mad *in_mad,
4318                                 struct opa_mad *out_mad, size_t *out_mad_size,
4319                                 u16 *out_mad_pkey_index)
4320 {
4321         int ret;
4322         int pkey_idx;
4323         u32 resp_len = 0;
4324         struct hfi1_ibport *ibp = to_iport(ibdev, port);
4325
4326         pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY);
4327         if (pkey_idx < 0) {
4328                 pr_warn("failed to find limited mgmt pkey, defaulting 0x%x\n",
4329                         hfi1_get_pkey(ibp, 1));
4330                 pkey_idx = 1;
4331         }
4332         *out_mad_pkey_index = (u16)pkey_idx;
4333
4334         switch (in_mad->mad_hdr.mgmt_class) {
4335         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4336         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4337                 if (is_local_mad(ibp, in_mad, in_wc)) {
4338                         ret = opa_local_smp_check(ibp, in_wc);
4339                         if (ret)
4340                                 return IB_MAD_RESULT_FAILURE;
4341                 }
4342                 ret = process_subn_opa(ibdev, mad_flags, port, in_mad,
4343                                        out_mad, &resp_len);
4344                 goto bail;
4345         case IB_MGMT_CLASS_PERF_MGMT:
4346                 ret = process_perf_opa(ibdev, port, in_mad, out_mad,
4347                                        &resp_len);
4348                 goto bail;
4349
4350         default:
4351                 ret = IB_MAD_RESULT_SUCCESS;
4352         }
4353
4354 bail:
4355         if (ret & IB_MAD_RESULT_REPLY)
4356                 *out_mad_size = round_up(resp_len, 8);
4357         else if (ret & IB_MAD_RESULT_SUCCESS)
4358                 *out_mad_size = in_wc->byte_len - sizeof(struct ib_grh);
4359
4360         return ret;
4361 }
4362
4363 static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4364                                const struct ib_wc *in_wc,
4365                                const struct ib_grh *in_grh,
4366                                const struct ib_mad *in_mad,
4367                                struct ib_mad *out_mad)
4368 {
4369         int ret;
4370
4371         switch (in_mad->mad_hdr.mgmt_class) {
4372         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
4373         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
4374                 ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
4375                 break;
4376         case IB_MGMT_CLASS_PERF_MGMT:
4377                 ret = process_perf(ibdev, port, in_mad, out_mad);
4378                 break;
4379         default:
4380                 ret = IB_MAD_RESULT_SUCCESS;
4381                 break;
4382         }
4383
4384         return ret;
4385 }
4386
4387 /**
4388  * hfi1_process_mad - process an incoming MAD packet
4389  * @ibdev: the infiniband device this packet came in on
4390  * @mad_flags: MAD flags
4391  * @port: the port number this packet came in on
4392  * @in_wc: the work completion entry for this packet
4393  * @in_grh: the global route header for this packet
4394  * @in_mad: the incoming MAD
4395  * @out_mad: any outgoing MAD reply
4396  *
4397  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
4398  * interested in processing.
4399  *
4400  * Note that the verbs framework has already done the MAD sanity checks,
4401  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
4402  * MADs.
4403  *
4404  * This is called by the ib_mad module.
4405  */
4406 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
4407                      const struct ib_wc *in_wc, const struct ib_grh *in_grh,
4408                      const struct ib_mad_hdr *in_mad, size_t in_mad_size,
4409                      struct ib_mad_hdr *out_mad, size_t *out_mad_size,
4410                      u16 *out_mad_pkey_index)
4411 {
4412         switch (in_mad->base_version) {
4413         case OPA_MGMT_BASE_VERSION:
4414                 if (unlikely(in_mad_size != sizeof(struct opa_mad))) {
4415                         dev_err(ibdev->dma_device, "invalid in_mad_size\n");
4416                         return IB_MAD_RESULT_FAILURE;
4417                 }
4418                 return hfi1_process_opa_mad(ibdev, mad_flags, port,
4419                                             in_wc, in_grh,
4420                                             (struct opa_mad *)in_mad,
4421                                             (struct opa_mad *)out_mad,
4422                                             out_mad_size,
4423                                             out_mad_pkey_index);
4424         case IB_MGMT_BASE_VERSION:
4425                 return hfi1_process_ib_mad(ibdev, mad_flags, port,
4426                                           in_wc, in_grh,
4427                                           (const struct ib_mad *)in_mad,
4428                                           (struct ib_mad *)out_mad);
4429         default:
4430                 break;
4431         }
4432
4433         return IB_MAD_RESULT_FAILURE;
4434 }