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