drivers/infiniband/hw/ipath/ipath_mad.c

   1 /*
   2  * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
   3  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <rdma/ib_smi.h>
  35
  36 #include "ipath_kernel.h"
  37 #include "ipath_verbs.h"
  38 #include "ipath_common.h"
  39
  40 #define IB_SMP_UNSUP_VERSION    __constant_htons(0x0004)
  41 #define IB_SMP_UNSUP_METHOD     __constant_htons(0x0008)
  42 #define IB_SMP_UNSUP_METH_ATTR  __constant_htons(0x000C)
  43 #define IB_SMP_INVALID_FIELD    __constant_htons(0x001C)
  44
  45 static int reply(struct ib_smp *smp)
  46 {
  47         /*
  48          * The verbs framework will handle the directed/LID route
  49          * packet changes.
  50          */
  51         smp->method = IB_MGMT_METHOD_GET_RESP;
  52         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
  53                 smp->status |= IB_SMP_DIRECTION;
  54         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
  55 }
  56
  57 static int recv_subn_get_nodedescription(struct ib_smp *smp,
  58                                          struct ib_device *ibdev)
  59 {
  60         if (smp->attr_mod)
  61                 smp->status |= IB_SMP_INVALID_FIELD;
  62
  63         strncpy(smp->data, ibdev->node_desc, sizeof(smp->data));
  64
  65         return reply(smp);
  66 }
  67
  68 struct nodeinfo {
  69         u8 base_version;
  70         u8 class_version;
  71         u8 node_type;
  72         u8 num_ports;
  73         __be64 sys_guid;
  74         __be64 node_guid;
  75         __be64 port_guid;
  76         __be16 partition_cap;
  77         __be16 device_id;
  78         __be32 revision;
  79         u8 local_port_num;
  80         u8 vendor_id[3];
  81 } __attribute__ ((packed));
  82
  83 static int recv_subn_get_nodeinfo(struct ib_smp *smp,
  84                                   struct ib_device *ibdev, u8 port)
  85 {
  86         struct nodeinfo *nip = (struct nodeinfo *)&smp->data;
  87         struct ipath_devdata *dd = to_idev(ibdev)->dd;
  88         u32 vendor, majrev, minrev;
  89
  90         /* GUID 0 is illegal */
  91         if (smp->attr_mod || (dd->ipath_guid == 0))
  92                 smp->status |= IB_SMP_INVALID_FIELD;
  93
  94         nip->base_version = 1;
  95         nip->class_version = 1;
  96         nip->node_type = 1;     /* channel adapter */
  97         /*
  98          * XXX The num_ports value will need a layer function to get
  99          * the value if we ever have more than one IB port on a chip.
 100          * We will also need to get the GUID for the port.
 101          */
 102         nip->num_ports = ibdev->phys_port_cnt;
 103         /* This is already in network order */
 104         nip->sys_guid = to_idev(ibdev)->sys_image_guid;
 105         nip->node_guid = dd->ipath_guid;
 106         nip->port_guid = dd->ipath_guid;
 107         nip->partition_cap = cpu_to_be16(ipath_get_npkeys(dd));
 108         nip->device_id = cpu_to_be16(dd->ipath_deviceid);
 109         majrev = dd->ipath_majrev;
 110         minrev = dd->ipath_minrev;
 111         nip->revision = cpu_to_be32((majrev << 16) | minrev);
 112         nip->local_port_num = port;
 113         vendor = dd->ipath_vendorid;
 114         nip->vendor_id[0] = 0;
 115         nip->vendor_id[1] = vendor >> 8;
 116         nip->vendor_id[2] = vendor;
 117
 118         return reply(smp);
 119 }
 120
 121 static int recv_subn_get_guidinfo(struct ib_smp *smp,
 122                                   struct ib_device *ibdev)
 123 {
 124         u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
 125         __be64 *p = (__be64 *) smp->data;
 126
 127         /* 32 blocks of 8 64-bit GUIDs per block */
 128
 129         memset(smp->data, 0, sizeof(smp->data));
 130
 131         /*
 132          * We only support one GUID for now.  If this changes, the
 133          * portinfo.guid_cap field needs to be updated too.
 134          */
 135         if (startgx == 0) {
 136                 __be64 g = to_idev(ibdev)->dd->ipath_guid;
 137                 if (g == 0)
 138                         /* GUID 0 is illegal */
 139                         smp->status |= IB_SMP_INVALID_FIELD;
 140                 else
 141                         /* The first is a copy of the read-only HW GUID. */
 142                         *p = g;
 143         } else
 144                 smp->status |= IB_SMP_INVALID_FIELD;
 145
 146         return reply(smp);
 147 }
 148
 149
 150 static int get_overrunthreshold(struct ipath_devdata *dd)
 151 {
 152         return (dd->ipath_ibcctrl >>
 153                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 154                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 155 }
 156
 157 /**
 158  * set_overrunthreshold - set the overrun threshold
 159  * @dd: the infinipath device
 160  * @n: the new threshold
 161  *
 162  * Note that this will only take effect when the link state changes.
 163  */
 164 static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
 165 {
 166         unsigned v;
 167
 168         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 169                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 170         if (v != n) {
 171                 dd->ipath_ibcctrl &=
 172                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
 173                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
 174                 dd->ipath_ibcctrl |=
 175                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
 176                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 177                                  dd->ipath_ibcctrl);
 178         }
 179         return 0;
 180 }
 181
 182 static int get_phyerrthreshold(struct ipath_devdata *dd)
 183 {
 184         return (dd->ipath_ibcctrl >>
 185                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 186                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 187 }
 188
 189 /**
 190  * set_phyerrthreshold - set the physical error threshold
 191  * @dd: the infinipath device
 192  * @n: the new threshold
 193  *
 194  * Note that this will only take effect when the link state changes.
 195  */
 196 static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
 197 {
 198         unsigned v;
 199
 200         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 201                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 202         if (v != n) {
 203                 dd->ipath_ibcctrl &=
 204                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
 205                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
 206                 dd->ipath_ibcctrl |=
 207                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
 208                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 209                                  dd->ipath_ibcctrl);
 210         }
 211         return 0;
 212 }
 213
 214 /**
 215  * get_linkdowndefaultstate - get the default linkdown state
 216  * @dd: the infinipath device
 217  *
 218  * Returns zero if the default is POLL, 1 if the default is SLEEP.
 219  */
 220 static int get_linkdowndefaultstate(struct ipath_devdata *dd)
 221 {
 222         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
 223 }
 224
 225 static int recv_subn_get_portinfo(struct ib_smp *smp,
 226                                   struct ib_device *ibdev, u8 port)
 227 {
 228         struct ipath_ibdev *dev;
 229         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 230         u16 lid;
 231         u8 ibcstat;
 232         u8 mtu;
 233         int ret;
 234
 235         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) {
 236                 smp->status |= IB_SMP_INVALID_FIELD;
 237                 ret = reply(smp);
 238                 goto bail;
 239         }
 240
 241         dev = to_idev(ibdev);
 242
 243         /* Clear all fields.  Only set the non-zero fields. */
 244         memset(smp->data, 0, sizeof(smp->data));
 245
 246         /* Only return the mkey if the protection field allows it. */
 247         if (smp->method == IB_MGMT_METHOD_SET || dev->mkey == smp->mkey ||
 248             dev->mkeyprot == 0)
 249                 pip->mkey = dev->mkey;
 250         pip->gid_prefix = dev->gid_prefix;
 251         lid = dev->dd->ipath_lid;
 252         pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
 253         pip->sm_lid = cpu_to_be16(dev->sm_lid);
 254         pip->cap_mask = cpu_to_be32(dev->port_cap_flags);
 255         /* pip->diag_code; */
 256         pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period);
 257         pip->local_port_num = port;
 258         pip->link_width_enabled = dev->link_width_enabled;
 259         pip->link_width_supported = 3;  /* 1x or 4x */
 260         pip->link_width_active = 2;     /* 4x */
 261         pip->linkspeed_portstate = 0x10;        /* 2.5Gbps */
 262         ibcstat = dev->dd->ipath_lastibcstat;
 263         pip->linkspeed_portstate |= ((ibcstat >> 4) & 0x3) + 1;
 264         pip->portphysstate_linkdown =
 265                 (ipath_cvt_physportstate[ibcstat & 0xf] << 4) |
 266                 (get_linkdowndefaultstate(dev->dd) ? 1 : 2);
 267         pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dev->dd->ipath_lmc;
 268         pip->linkspeedactive_enabled = 0x11;    /* 2.5Gbps, 2.5Gbps */
 269         switch (dev->dd->ipath_ibmtu) {
 270         case 4096:
 271                 mtu = IB_MTU_4096;
 272                 break;
 273         case 2048:
 274                 mtu = IB_MTU_2048;
 275                 break;
 276         case 1024:
 277                 mtu = IB_MTU_1024;
 278                 break;
 279         case 512:
 280                 mtu = IB_MTU_512;
 281                 break;
 282         case 256:
 283                 mtu = IB_MTU_256;
 284                 break;
 285         default:                /* oops, something is wrong */
 286                 mtu = IB_MTU_2048;
 287                 break;
 288         }
 289         pip->neighbormtu_mastersmsl = (mtu << 4) | dev->sm_sl;
 290         pip->vlcap_inittype = 0x10;     /* VLCap = VL0, InitType = 0 */
 291         pip->vl_high_limit = dev->vl_high_limit;
 292         /* pip->vl_arb_high_cap; // only one VL */
 293         /* pip->vl_arb_low_cap; // only one VL */
 294         /* InitTypeReply = 0 */
 295         /*
 296          * Note: the chips support a maximum MTU of 4096, but the driver
 297          * hasn't implemented this feature yet, so set the maximum value
 298          * to 2048.
 299          */
 300         pip->inittypereply_mtucap = IB_MTU_2048;
 301         // HCAs ignore VLStallCount and HOQLife
 302         /* pip->vlstallcnt_hoqlife; */
 303         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
 304         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
 305         /* P_KeyViolations are counted by hardware. */
 306         pip->pkey_violations =
 307                 cpu_to_be16((ipath_get_cr_errpkey(dev->dd) -
 308                              dev->z_pkey_violations) & 0xFFFF);
 309         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
 310         /* Only the hardware GUID is supported for now */
 311         pip->guid_cap = 1;
 312         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
 313         /* 32.768 usec. response time (guessing) */
 314         pip->resv_resptimevalue = 3;
 315         pip->localphyerrors_overrunerrors =
 316                 (get_phyerrthreshold(dev->dd) << 4) |
 317                 get_overrunthreshold(dev->dd);
 318         /* pip->max_credit_hint; */
 319         /* pip->link_roundtrip_latency[3]; */
 320
 321         ret = reply(smp);
 322
 323 bail:
 324         return ret;
 325 }
 326
 327 /**
 328  * get_pkeys - return the PKEY table for port 0
 329  * @dd: the infinipath device
 330  * @pkeys: the pkey table is placed here
 331  */
 332 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
 333 {
 334         struct ipath_portdata *pd = dd->ipath_pd[0];
 335
 336         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
 337
 338         return 0;
 339 }
 340
 341 static int recv_subn_get_pkeytable(struct ib_smp *smp,
 342                                    struct ib_device *ibdev)
 343 {
 344         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 345         u16 *p = (u16 *) smp->data;
 346         __be16 *q = (__be16 *) smp->data;
 347
 348         /* 64 blocks of 32 16-bit P_Key entries */
 349
 350         memset(smp->data, 0, sizeof(smp->data));
 351         if (startpx == 0) {
 352                 struct ipath_ibdev *dev = to_idev(ibdev);
 353                 unsigned i, n = ipath_get_npkeys(dev->dd);
 354
 355                 get_pkeys(dev->dd, p);
 356
 357                 for (i = 0; i < n; i++)
 358                         q[i] = cpu_to_be16(p[i]);
 359         } else
 360                 smp->status |= IB_SMP_INVALID_FIELD;
 361
 362         return reply(smp);
 363 }
 364
 365 static int recv_subn_set_guidinfo(struct ib_smp *smp,
 366                                   struct ib_device *ibdev)
 367 {
 368         /* The only GUID we support is the first read-only entry. */
 369         return recv_subn_get_guidinfo(smp, ibdev);
 370 }
 371
 372 /**
 373  * set_linkdowndefaultstate - set the default linkdown state
 374  * @dd: the infinipath device
 375  * @sleep: the new state
 376  *
 377  * Note that this will only take effect when the link state changes.
 378  */
 379 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
 380 {
 381         if (sleep)
 382                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 383         else
 384                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 385         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 386                          dd->ipath_ibcctrl);
 387         return 0;
 388 }
 389
 390 /**
 391  * recv_subn_set_portinfo - set port information
 392  * @smp: the incoming SM packet
 393  * @ibdev: the infiniband device
 394  * @port: the port on the device
 395  *
 396  * Set Portinfo (see ch. 14.2.5.6).
 397  */
 398 static int recv_subn_set_portinfo(struct ib_smp *smp,
 399                                   struct ib_device *ibdev, u8 port)
 400 {
 401         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 402         struct ib_event event;
 403         struct ipath_ibdev *dev;
 404         struct ipath_devdata *dd;
 405         u32 flags;
 406         char clientrereg = 0;
 407         u16 lid, smlid;
 408         u8 lwe;
 409         u8 lse;
 410         u8 state;
 411         u16 lstate;
 412         u32 mtu;
 413         int ret, ore;
 414
 415         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
 416                 goto err;
 417
 418         dev = to_idev(ibdev);
 419         dd = dev->dd;
 420         event.device = ibdev;
 421         event.element.port_num = port;
 422
 423         dev->mkey = pip->mkey;
 424         dev->gid_prefix = pip->gid_prefix;
 425         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
 426
 427         lid = be16_to_cpu(pip->lid);
 428         if (dd->ipath_lid != lid ||
 429             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
 430                 /* Must be a valid unicast LID address. */
 431                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
 432                         goto err;
 433                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
 434                 event.event = IB_EVENT_LID_CHANGE;
 435                 ib_dispatch_event(&event);
 436         }
 437
 438         smlid = be16_to_cpu(pip->sm_lid);
 439         if (smlid != dev->sm_lid) {
 440                 /* Must be a valid unicast LID address. */
 441                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
 442                         goto err;
 443                 dev->sm_lid = smlid;
 444                 event.event = IB_EVENT_SM_CHANGE;
 445                 ib_dispatch_event(&event);
 446         }
 447
 448         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
 449         lwe = pip->link_width_enabled;
 450         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
 451                 goto err;
 452         if (lwe == 0xFF)
 453                 dev->link_width_enabled = 3;    /* 1x or 4x */
 454         else if (lwe)
 455                 dev->link_width_enabled = lwe;
 456
 457         /* Only 2.5 Gbs supported. */
 458         lse = pip->linkspeedactive_enabled & 0xF;
 459         if (lse >= 2 && lse <= 0xE)
 460                 goto err;
 461
 462         /* Set link down default state. */
 463         switch (pip->portphysstate_linkdown & 0xF) {
 464         case 0: /* NOP */
 465                 break;
 466         case 1: /* SLEEP */
 467                 if (set_linkdowndefaultstate(dd, 1))
 468                         goto err;
 469                 break;
 470         case 2: /* POLL */
 471                 if (set_linkdowndefaultstate(dd, 0))
 472                         goto err;
 473                 break;
 474         default:
 475                 goto err;
 476         }
 477
 478         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
 479         dev->vl_high_limit = pip->vl_high_limit;
 480
 481         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
 482         case IB_MTU_256:
 483                 mtu = 256;
 484                 break;
 485         case IB_MTU_512:
 486                 mtu = 512;
 487                 break;
 488         case IB_MTU_1024:
 489                 mtu = 1024;
 490                 break;
 491         case IB_MTU_2048:
 492                 mtu = 2048;
 493                 break;
 494         case IB_MTU_4096:
 495                 mtu = 4096;
 496                 break;
 497         default:
 498                 /* XXX We have already partially updated our state! */
 499                 goto err;
 500         }
 501         ipath_set_mtu(dd, mtu);
 502
 503         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
 504
 505         /* We only support VL0 */
 506         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
 507                 goto err;
 508
 509         if (pip->mkey_violations == 0)
 510                 dev->mkey_violations = 0;
 511
 512         /*
 513          * Hardware counter can't be reset so snapshot and subtract
 514          * later.
 515          */
 516         if (pip->pkey_violations == 0)
 517                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
 518
 519         if (pip->qkey_violations == 0)
 520                 dev->qkey_violations = 0;
 521
 522         ore = pip->localphyerrors_overrunerrors;
 523         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
 524                 goto err;
 525
 526         if (set_overrunthreshold(dd, (ore & 0xF)))
 527                 goto err;
 528
 529         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 530
 531         if (pip->clientrereg_resv_subnetto & 0x80) {
 532                 clientrereg = 1;
 533                 event.event = IB_EVENT_CLIENT_REREGISTER;
 534                 ib_dispatch_event(&event);
 535         }
 536
 537         /*
 538          * Do the port state change now that the other link parameters
 539          * have been set.
 540          * Changing the port physical state only makes sense if the link
 541          * is down or is being set to down.
 542          */
 543         state = pip->linkspeed_portstate & 0xF;
 544         flags = dd->ipath_flags;
 545         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 546         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
 547                 goto err;
 548
 549         /*
 550          * Only state changes of DOWN, ARM, and ACTIVE are valid
 551          * and must be in the correct state to take effect (see 7.2.6).
 552          */
 553         switch (state) {
 554         case IB_PORT_NOP:
 555                 if (lstate == 0)
 556                         break;
 557                 /* FALLTHROUGH */
 558         case IB_PORT_DOWN:
 559                 if (lstate == 0)
 560                         if (get_linkdowndefaultstate(dd))
 561                                 lstate = IPATH_IB_LINKDOWN_SLEEP;
 562                         else
 563                                 lstate = IPATH_IB_LINKDOWN;
 564                 else if (lstate == 1)
 565                         lstate = IPATH_IB_LINKDOWN_SLEEP;
 566                 else if (lstate == 2)
 567                         lstate = IPATH_IB_LINKDOWN;
 568                 else if (lstate == 3)
 569                         lstate = IPATH_IB_LINKDOWN_DISABLE;
 570                 else
 571                         goto err;
 572                 ipath_set_linkstate(dd, lstate);
 573                 break;
 574         case IB_PORT_ARMED:
 575                 if (!(flags & (IPATH_LINKINIT | IPATH_LINKACTIVE)))
 576                         break;
 577                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
 578                 break;
 579         case IB_PORT_ACTIVE:
 580                 if (!(flags & IPATH_LINKARMED))
 581                         break;
 582                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
 583                 break;
 584         default:
 585                 /* XXX We have already partially updated our state! */
 586                 goto err;
 587         }
 588
 589         ret = recv_subn_get_portinfo(smp, ibdev, port);
 590
 591         if (clientrereg)
 592                 pip->clientrereg_resv_subnetto |= 0x80;
 593
 594         goto done;
 595
 596 err:
 597         smp->status |= IB_SMP_INVALID_FIELD;
 598         ret = recv_subn_get_portinfo(smp, ibdev, port);
 599
 600 done:
 601         return ret;
 602 }
 603
 604 /**
 605  * rm_pkey - decrecment the reference count for the given PKEY
 606  * @dd: the infinipath device
 607  * @key: the PKEY index
 608  *
 609  * Return true if this was the last reference and the hardware table entry
 610  * needs to be changed.
 611  */
 612 static int rm_pkey(struct ipath_devdata *dd, u16 key)
 613 {
 614         int i;
 615         int ret;
 616
 617         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 618                 if (dd->ipath_pkeys[i] != key)
 619                         continue;
 620                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
 621                         dd->ipath_pkeys[i] = 0;
 622                         ret = 1;
 623                         goto bail;
 624                 }
 625                 break;
 626         }
 627
 628         ret = 0;
 629
 630 bail:
 631         return ret;
 632 }
 633
 634 /**
 635  * add_pkey - add the given PKEY to the hardware table
 636  * @dd: the infinipath device
 637  * @key: the PKEY
 638  *
 639  * Return an error code if unable to add the entry, zero if no change,
 640  * or 1 if the hardware PKEY register needs to be updated.
 641  */
 642 static int add_pkey(struct ipath_devdata *dd, u16 key)
 643 {
 644         int i;
 645         u16 lkey = key & 0x7FFF;
 646         int any = 0;
 647         int ret;
 648
 649         if (lkey == 0x7FFF) {
 650                 ret = 0;
 651                 goto bail;
 652         }
 653
 654         /* Look for an empty slot or a matching PKEY. */
 655         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 656                 if (!dd->ipath_pkeys[i]) {
 657                         any++;
 658                         continue;
 659                 }
 660                 /* If it matches exactly, try to increment the ref count */
 661                 if (dd->ipath_pkeys[i] == key) {
 662                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
 663                                 ret = 0;
 664                                 goto bail;
 665                         }
 666                         /* Lost the race. Look for an empty slot below. */
 667                         atomic_dec(&dd->ipath_pkeyrefs[i]);
 668                         any++;
 669                 }
 670                 /*
 671                  * It makes no sense to have both the limited and unlimited
 672                  * PKEY set at the same time since the unlimited one will
 673                  * disable the limited one.
 674                  */
 675                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
 676                         ret = -EEXIST;
 677                         goto bail;
 678                 }
 679         }
 680         if (!any) {
 681                 ret = -EBUSY;
 682                 goto bail;
 683         }
 684         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 685                 if (!dd->ipath_pkeys[i] &&
 686                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
 687                         /* for ipathstats, etc. */
 688                         ipath_stats.sps_pkeys[i] = lkey;
 689                         dd->ipath_pkeys[i] = key;
 690                         ret = 1;
 691                         goto bail;
 692                 }
 693         }
 694         ret = -EBUSY;
 695
 696 bail:
 697         return ret;
 698 }
 699
 700 /**
 701  * set_pkeys - set the PKEY table for port 0
 702  * @dd: the infinipath device
 703  * @pkeys: the PKEY table
 704  */
 705 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 706 {
 707         struct ipath_portdata *pd;
 708         int i;
 709         int changed = 0;
 710
 711         pd = dd->ipath_pd[0];
 712
 713         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
 714                 u16 key = pkeys[i];
 715                 u16 okey = pd->port_pkeys[i];
 716
 717                 if (key == okey)
 718                         continue;
 719                 /*
 720                  * The value of this PKEY table entry is changing.
 721                  * Remove the old entry in the hardware's array of PKEYs.
 722                  */
 723                 if (okey & 0x7FFF)
 724                         changed |= rm_pkey(dd, okey);
 725                 if (key & 0x7FFF) {
 726                         int ret = add_pkey(dd, key);
 727
 728                         if (ret < 0)
 729                                 key = 0;
 730                         else
 731                                 changed |= ret;
 732                 }
 733                 pd->port_pkeys[i] = key;
 734         }
 735         if (changed) {
 736                 u64 pkey;
 737
 738                 pkey = (u64) dd->ipath_pkeys[0] |
 739                         ((u64) dd->ipath_pkeys[1] << 16) |
 740                         ((u64) dd->ipath_pkeys[2] << 32) |
 741                         ((u64) dd->ipath_pkeys[3] << 48);
 742                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
 743                            (unsigned long long) pkey);
 744                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
 745                                  pkey);
 746         }
 747         return 0;
 748 }
 749
 750 static int recv_subn_set_pkeytable(struct ib_smp *smp,
 751                                    struct ib_device *ibdev)
 752 {
 753         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 754         __be16 *p = (__be16 *) smp->data;
 755         u16 *q = (u16 *) smp->data;
 756         struct ipath_ibdev *dev = to_idev(ibdev);
 757         unsigned i, n = ipath_get_npkeys(dev->dd);
 758
 759         for (i = 0; i < n; i++)
 760                 q[i] = be16_to_cpu(p[i]);
 761
 762         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
 763                 smp->status |= IB_SMP_INVALID_FIELD;
 764
 765         return recv_subn_get_pkeytable(smp, ibdev);
 766 }
 767
 768 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
 769 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
 770 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
 771 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
 772 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
 773 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
 774
 775 struct ib_perf {
 776         u8 base_version;
 777         u8 mgmt_class;
 778         u8 class_version;
 779         u8 method;
 780         __be16 status;
 781         __be16 unused;
 782         __be64 tid;
 783         __be16 attr_id;
 784         __be16 resv;
 785         __be32 attr_mod;
 786         u8 reserved[40];
 787         u8 data[192];
 788 } __attribute__ ((packed));
 789
 790 struct ib_pma_classportinfo {
 791         u8 base_version;
 792         u8 class_version;
 793         __be16 cap_mask;
 794         u8 reserved[3];
 795         u8 resp_time_value;     /* only lower 5 bits */
 796         union ib_gid redirect_gid;
 797         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
 798         __be16 redirect_lid;
 799         __be16 redirect_pkey;
 800         __be32 redirect_qp;     /* only lower 24 bits */
 801         __be32 redirect_qkey;
 802         union ib_gid trap_gid;
 803         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
 804         __be16 trap_lid;
 805         __be16 trap_pkey;
 806         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
 807         __be32 trap_qkey;
 808 } __attribute__ ((packed));
 809
 810 struct ib_pma_portsamplescontrol {
 811         u8 opcode;
 812         u8 port_select;
 813         u8 tick;
 814         u8 counter_width;       /* only lower 3 bits */
 815         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
 816         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
 817         u8 sample_mechanisms;
 818         u8 sample_status;       /* only lower 2 bits */
 819         __be64 option_mask;
 820         __be64 vendor_mask;
 821         __be32 sample_start;
 822         __be32 sample_interval;
 823         __be16 tag;
 824         __be16 counter_select[15];
 825 } __attribute__ ((packed));
 826
 827 struct ib_pma_portsamplesresult {
 828         __be16 tag;
 829         __be16 sample_status;   /* only lower 2 bits */
 830         __be32 counter[15];
 831 } __attribute__ ((packed));
 832
 833 struct ib_pma_portsamplesresult_ext {
 834         __be16 tag;
 835         __be16 sample_status;   /* only lower 2 bits */
 836         __be32 extended_width;  /* only upper 2 bits */
 837         __be64 counter[15];
 838 } __attribute__ ((packed));
 839
 840 struct ib_pma_portcounters {
 841         u8 reserved;
 842         u8 port_select;
 843         __be16 counter_select;
 844         __be16 symbol_error_counter;
 845         u8 link_error_recovery_counter;
 846         u8 link_downed_counter;
 847         __be16 port_rcv_errors;
 848         __be16 port_rcv_remphys_errors;
 849         __be16 port_rcv_switch_relay_errors;
 850         __be16 port_xmit_discards;
 851         u8 port_xmit_constraint_errors;
 852         u8 port_rcv_constraint_errors;
 853         u8 reserved1;
 854         u8 lli_ebor_errors;     /* 4, 4, bits */
 855         __be16 reserved2;
 856         __be16 vl15_dropped;
 857         __be32 port_xmit_data;
 858         __be32 port_rcv_data;
 859         __be32 port_xmit_packets;
 860         __be32 port_rcv_packets;
 861 } __attribute__ ((packed));
 862
 863 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
 864 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
 865 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
 866 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
 867 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
 868 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
 869 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
 870 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
 871 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
 872 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
 873 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
 874 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
 875 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
 876
 877 struct ib_pma_portcounters_ext {
 878         u8 reserved;
 879         u8 port_select;
 880         __be16 counter_select;
 881         __be32 reserved1;
 882         __be64 port_xmit_data;
 883         __be64 port_rcv_data;
 884         __be64 port_xmit_packets;
 885         __be64 port_rcv_packets;
 886         __be64 port_unicast_xmit_packets;
 887         __be64 port_unicast_rcv_packets;
 888         __be64 port_multicast_xmit_packets;
 889         __be64 port_multicast_rcv_packets;
 890 } __attribute__ ((packed));
 891
 892 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
 893 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
 894 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
 895 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
 896 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
 897 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
 898 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
 899 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
 900
 901 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
 902 {
 903         struct ib_pma_classportinfo *p =
 904                 (struct ib_pma_classportinfo *)pmp->data;
 905
 906         memset(pmp->data, 0, sizeof(pmp->data));
 907
 908         if (pmp->attr_mod != 0)
 909                 pmp->status |= IB_SMP_INVALID_FIELD;
 910
 911         /* Indicate AllPortSelect is valid (only one port anyway) */
 912         p->cap_mask = __constant_cpu_to_be16(1 << 8);
 913         p->base_version = 1;
 914         p->class_version = 1;
 915         /*
 916          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
 917          * sec.
 918          */
 919         p->resp_time_value = 18;
 920
 921         return reply((struct ib_smp *) pmp);
 922 }
 923
 924 /*
 925  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
 926  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
 927  * We support 5 counters which only count the mandatory quantities.
 928  */
 929 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
 930 #define COUNTER_MASK0_9 \
 931         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
 932                                COUNTER_MASK(1, 1) | \
 933                                COUNTER_MASK(1, 2) | \
 934                                COUNTER_MASK(1, 3) | \
 935                                COUNTER_MASK(1, 4))
 936
 937 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 938                                            struct ib_device *ibdev, u8 port)
 939 {
 940         struct ib_pma_portsamplescontrol *p =
 941                 (struct ib_pma_portsamplescontrol *)pmp->data;
 942         struct ipath_ibdev *dev = to_idev(ibdev);
 943         unsigned long flags;
 944         u8 port_select = p->port_select;
 945
 946         memset(pmp->data, 0, sizeof(pmp->data));
 947
 948         p->port_select = port_select;
 949         if (pmp->attr_mod != 0 ||
 950             (port_select != port && port_select != 0xFF))
 951                 pmp->status |= IB_SMP_INVALID_FIELD;
 952         /*
 953          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 954          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
 955          * intervals are counted in ticks.  Since we use Linux timers, that
 956          * count in jiffies, we can't sample for less than 1000 ticks if HZ
 957          * == 1000 (4000 ticks if HZ is 250).
 958          */
 959         /* XXX This is WRONG. */
 960         p->tick = 250;          /* 1 usec. */
 961         p->counter_width = 4;   /* 32 bit counters */
 962         p->counter_mask0_9 = COUNTER_MASK0_9;
 963         spin_lock_irqsave(&dev->pending_lock, flags);
 964         p->sample_status = dev->pma_sample_status;
 965         p->sample_start = cpu_to_be32(dev->pma_sample_start);
 966         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
 967         p->tag = cpu_to_be16(dev->pma_tag);
 968         p->counter_select[0] = dev->pma_counter_select[0];
 969         p->counter_select[1] = dev->pma_counter_select[1];
 970         p->counter_select[2] = dev->pma_counter_select[2];
 971         p->counter_select[3] = dev->pma_counter_select[3];
 972         p->counter_select[4] = dev->pma_counter_select[4];
 973         spin_unlock_irqrestore(&dev->pending_lock, flags);
 974
 975         return reply((struct ib_smp *) pmp);
 976 }
 977
 978 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
 979                                            struct ib_device *ibdev, u8 port)
 980 {
 981         struct ib_pma_portsamplescontrol *p =
 982                 (struct ib_pma_portsamplescontrol *)pmp->data;
 983         struct ipath_ibdev *dev = to_idev(ibdev);
 984         unsigned long flags;
 985         u32 start;
 986         int ret;
 987
 988         if (pmp->attr_mod != 0 ||
 989             (p->port_select != port && p->port_select != 0xFF)) {
 990                 pmp->status |= IB_SMP_INVALID_FIELD;
 991                 ret = reply((struct ib_smp *) pmp);
 992                 goto bail;
 993         }
 994
 995         start = be32_to_cpu(p->sample_start);
 996         if (start != 0) {
 997                 spin_lock_irqsave(&dev->pending_lock, flags);
 998                 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_DONE) {
 999                         dev->pma_sample_status =
1000                                 IB_PMA_SAMPLE_STATUS_STARTED;
1001                         dev->pma_sample_start = start;
1002                         dev->pma_sample_interval =
1003                                 be32_to_cpu(p->sample_interval);
1004                         dev->pma_tag = be16_to_cpu(p->tag);
1005                         if (p->counter_select[0])
1006                                 dev->pma_counter_select[0] =
1007                                         p->counter_select[0];
1008                         if (p->counter_select[1])
1009                                 dev->pma_counter_select[1] =
1010                                         p->counter_select[1];
1011                         if (p->counter_select[2])
1012                                 dev->pma_counter_select[2] =
1013                                         p->counter_select[2];
1014                         if (p->counter_select[3])
1015                                 dev->pma_counter_select[3] =
1016                                         p->counter_select[3];
1017                         if (p->counter_select[4])
1018                                 dev->pma_counter_select[4] =
1019                                         p->counter_select[4];
1020                 }
1021                 spin_unlock_irqrestore(&dev->pending_lock, flags);
1022         }
1023         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1024
1025 bail:
1026         return ret;
1027 }
1028
1029 static u64 get_counter(struct ipath_ibdev *dev, __be16 sel)
1030 {
1031         u64 ret;
1032
1033         switch (sel) {
1034         case IB_PMA_PORT_XMIT_DATA:
1035                 ret = dev->ipath_sword;
1036                 break;
1037         case IB_PMA_PORT_RCV_DATA:
1038                 ret = dev->ipath_rword;
1039                 break;
1040         case IB_PMA_PORT_XMIT_PKTS:
1041                 ret = dev->ipath_spkts;
1042                 break;
1043         case IB_PMA_PORT_RCV_PKTS:
1044                 ret = dev->ipath_rpkts;
1045                 break;
1046         case IB_PMA_PORT_XMIT_WAIT:
1047                 ret = dev->ipath_xmit_wait;
1048                 break;
1049         default:
1050                 ret = 0;
1051         }
1052
1053         return ret;
1054 }
1055
1056 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1057                                           struct ib_device *ibdev)
1058 {
1059         struct ib_pma_portsamplesresult *p =
1060                 (struct ib_pma_portsamplesresult *)pmp->data;
1061         struct ipath_ibdev *dev = to_idev(ibdev);
1062         int i;
1063
1064         memset(pmp->data, 0, sizeof(pmp->data));
1065         p->tag = cpu_to_be16(dev->pma_tag);
1066         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1067         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1068                 p->counter[i] = cpu_to_be32(
1069                         get_counter(dev, dev->pma_counter_select[i]));
1070
1071         return reply((struct ib_smp *) pmp);
1072 }
1073
1074 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1075                                               struct ib_device *ibdev)
1076 {
1077         struct ib_pma_portsamplesresult_ext *p =
1078                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1079         struct ipath_ibdev *dev = to_idev(ibdev);
1080         int i;
1081
1082         memset(pmp->data, 0, sizeof(pmp->data));
1083         p->tag = cpu_to_be16(dev->pma_tag);
1084         p->sample_status = cpu_to_be16(dev->pma_sample_status);
1085         /* 64 bits */
1086         p->extended_width = __constant_cpu_to_be32(0x80000000);
1087         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1088                 p->counter[i] = cpu_to_be64(
1089                         get_counter(dev, dev->pma_counter_select[i]));
1090
1091         return reply((struct ib_smp *) pmp);
1092 }
1093
1094 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1095                                      struct ib_device *ibdev, u8 port)
1096 {
1097         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1098                 pmp->data;
1099         struct ipath_ibdev *dev = to_idev(ibdev);
1100         struct ipath_verbs_counters cntrs;
1101         u8 port_select = p->port_select;
1102
1103         ipath_get_counters(dev->dd, &cntrs);
1104
1105         /* Adjust counters for any resets done. */
1106         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1107         cntrs.link_error_recovery_counter -=
1108                 dev->z_link_error_recovery_counter;
1109         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1110         cntrs.port_rcv_errors += dev->rcv_errors;
1111         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1112         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1113         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1114         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1115         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1116         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1117         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1118         cntrs.local_link_integrity_errors -=
1119                 dev->z_local_link_integrity_errors;
1120         cntrs.excessive_buffer_overrun_errors -=
1121                 dev->z_excessive_buffer_overrun_errors;
1122
1123         memset(pmp->data, 0, sizeof(pmp->data));
1124
1125         p->port_select = port_select;
1126         if (pmp->attr_mod != 0 ||
1127             (port_select != port && port_select != 0xFF))
1128                 pmp->status |= IB_SMP_INVALID_FIELD;
1129
1130         if (cntrs.symbol_error_counter > 0xFFFFUL)
1131                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1132         else
1133                 p->symbol_error_counter =
1134                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1135         if (cntrs.link_error_recovery_counter > 0xFFUL)
1136                 p->link_error_recovery_counter = 0xFF;
1137         else
1138                 p->link_error_recovery_counter =
1139                         (u8)cntrs.link_error_recovery_counter;
1140         if (cntrs.link_downed_counter > 0xFFUL)
1141                 p->link_downed_counter = 0xFF;
1142         else
1143                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1144         if (cntrs.port_rcv_errors > 0xFFFFUL)
1145                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1146         else
1147                 p->port_rcv_errors =
1148                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1149         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1150                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1151         else
1152                 p->port_rcv_remphys_errors =
1153                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1154         if (cntrs.port_xmit_discards > 0xFFFFUL)
1155                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1156         else
1157                 p->port_xmit_discards =
1158                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1159         if (cntrs.local_link_integrity_errors > 0xFUL)
1160                 cntrs.local_link_integrity_errors = 0xFUL;
1161         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1162                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1163         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1164                 cntrs.excessive_buffer_overrun_errors;
1165         if (dev->n_vl15_dropped > 0xFFFFUL)
1166                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1167         else
1168                 p->vl15_dropped = cpu_to_be16((u16)dev->n_vl15_dropped);
1169         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1170                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1171         else
1172                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1173         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1174                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1175         else
1176                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1177         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1178                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1179         else
1180                 p->port_xmit_packets =
1181                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1182         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1183                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1184         else
1185                 p->port_rcv_packets =
1186                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1187
1188         return reply((struct ib_smp *) pmp);
1189 }
1190
1191 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1192                                          struct ib_device *ibdev, u8 port)
1193 {
1194         struct ib_pma_portcounters_ext *p =
1195                 (struct ib_pma_portcounters_ext *)pmp->data;
1196         struct ipath_ibdev *dev = to_idev(ibdev);
1197         u64 swords, rwords, spkts, rpkts, xwait;
1198         u8 port_select = p->port_select;
1199
1200         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1201                                 &rpkts, &xwait);
1202
1203         /* Adjust counters for any resets done. */
1204         swords -= dev->z_port_xmit_data;
1205         rwords -= dev->z_port_rcv_data;
1206         spkts -= dev->z_port_xmit_packets;
1207         rpkts -= dev->z_port_rcv_packets;
1208
1209         memset(pmp->data, 0, sizeof(pmp->data));
1210
1211         p->port_select = port_select;
1212         if (pmp->attr_mod != 0 ||
1213             (port_select != port && port_select != 0xFF))
1214                 pmp->status |= IB_SMP_INVALID_FIELD;
1215
1216         p->port_xmit_data = cpu_to_be64(swords);
1217         p->port_rcv_data = cpu_to_be64(rwords);
1218         p->port_xmit_packets = cpu_to_be64(spkts);
1219         p->port_rcv_packets = cpu_to_be64(rpkts);
1220         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1221         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1222         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1223         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1224
1225         return reply((struct ib_smp *) pmp);
1226 }
1227
1228 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1229                                      struct ib_device *ibdev, u8 port)
1230 {
1231         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1232                 pmp->data;
1233         struct ipath_ibdev *dev = to_idev(ibdev);
1234         struct ipath_verbs_counters cntrs;
1235
1236         /*
1237          * Since the HW doesn't support clearing counters, we save the
1238          * current count and subtract it from future responses.
1239          */
1240         ipath_get_counters(dev->dd, &cntrs);
1241
1242         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1243                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1244
1245         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1246                 dev->z_link_error_recovery_counter =
1247                         cntrs.link_error_recovery_counter;
1248
1249         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1250                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1251
1252         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1253                 dev->z_port_rcv_errors =
1254                         cntrs.port_rcv_errors + dev->rcv_errors;
1255
1256         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1257                 dev->z_port_rcv_remphys_errors =
1258                         cntrs.port_rcv_remphys_errors;
1259
1260         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1261                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1262
1263         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1264                 dev->z_local_link_integrity_errors =
1265                         cntrs.local_link_integrity_errors;
1266
1267         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1268                 dev->z_excessive_buffer_overrun_errors =
1269                         cntrs.excessive_buffer_overrun_errors;
1270
1271         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED)
1272                 dev->n_vl15_dropped = 0;
1273
1274         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1275                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1276
1277         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1278                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1279
1280         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1281                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1282
1283         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1284                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1285
1286         return recv_pma_get_portcounters(pmp, ibdev, port);
1287 }
1288
1289 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1290                                          struct ib_device *ibdev, u8 port)
1291 {
1292         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1293                 pmp->data;
1294         struct ipath_ibdev *dev = to_idev(ibdev);
1295         u64 swords, rwords, spkts, rpkts, xwait;
1296
1297         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1298                                 &rpkts, &xwait);
1299
1300         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1301                 dev->z_port_xmit_data = swords;
1302
1303         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1304                 dev->z_port_rcv_data = rwords;
1305
1306         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1307                 dev->z_port_xmit_packets = spkts;
1308
1309         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1310                 dev->z_port_rcv_packets = rpkts;
1311
1312         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1313                 dev->n_unicast_xmit = 0;
1314
1315         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1316                 dev->n_unicast_rcv = 0;
1317
1318         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1319                 dev->n_multicast_xmit = 0;
1320
1321         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1322                 dev->n_multicast_rcv = 0;
1323
1324         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1325 }
1326
1327 static int process_subn(struct ib_device *ibdev, int mad_flags,
1328                         u8 port_num, struct ib_mad *in_mad,
1329                         struct ib_mad *out_mad)
1330 {
1331         struct ib_smp *smp = (struct ib_smp *)out_mad;
1332         struct ipath_ibdev *dev = to_idev(ibdev);
1333         int ret;
1334
1335         *out_mad = *in_mad;
1336         if (smp->class_version != 1) {
1337                 smp->status |= IB_SMP_UNSUP_VERSION;
1338                 ret = reply(smp);
1339                 goto bail;
1340         }
1341
1342         /* Is the mkey in the process of expiring? */
1343         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1344                 /* Clear timeout and mkey protection field. */
1345                 dev->mkey_lease_timeout = 0;
1346                 dev->mkeyprot = 0;
1347         }
1348
1349         /*
1350          * M_Key checking depends on
1351          * Portinfo:M_Key_protect_bits
1352          */
1353         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1354             dev->mkey != smp->mkey &&
1355             (smp->method == IB_MGMT_METHOD_SET ||
1356              (smp->method == IB_MGMT_METHOD_GET &&
1357               dev->mkeyprot >= 2))) {
1358                 if (dev->mkey_violations != 0xFFFF)
1359                         ++dev->mkey_violations;
1360                 if (dev->mkey_lease_timeout ||
1361                     dev->mkey_lease_period == 0) {
1362                         ret = IB_MAD_RESULT_SUCCESS |
1363                                 IB_MAD_RESULT_CONSUMED;
1364                         goto bail;
1365                 }
1366                 dev->mkey_lease_timeout = jiffies +
1367                         dev->mkey_lease_period * HZ;
1368                 /* Future: Generate a trap notice. */
1369                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1370                 goto bail;
1371         } else if (dev->mkey_lease_timeout)
1372                 dev->mkey_lease_timeout = 0;
1373
1374         switch (smp->method) {
1375         case IB_MGMT_METHOD_GET:
1376                 switch (smp->attr_id) {
1377                 case IB_SMP_ATTR_NODE_DESC:
1378                         ret = recv_subn_get_nodedescription(smp, ibdev);
1379                         goto bail;
1380                 case IB_SMP_ATTR_NODE_INFO:
1381                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1382                         goto bail;
1383                 case IB_SMP_ATTR_GUID_INFO:
1384                         ret = recv_subn_get_guidinfo(smp, ibdev);
1385                         goto bail;
1386                 case IB_SMP_ATTR_PORT_INFO:
1387                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1388                         goto bail;
1389                 case IB_SMP_ATTR_PKEY_TABLE:
1390                         ret = recv_subn_get_pkeytable(smp, ibdev);
1391                         goto bail;
1392                 case IB_SMP_ATTR_SM_INFO:
1393                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1394                                 ret = IB_MAD_RESULT_SUCCESS |
1395                                         IB_MAD_RESULT_CONSUMED;
1396                                 goto bail;
1397                         }
1398                         if (dev->port_cap_flags & IB_PORT_SM) {
1399                                 ret = IB_MAD_RESULT_SUCCESS;
1400                                 goto bail;
1401                         }
1402                         /* FALLTHROUGH */
1403                 default:
1404                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1405                         ret = reply(smp);
1406                         goto bail;
1407                 }
1408
1409         case IB_MGMT_METHOD_SET:
1410                 switch (smp->attr_id) {
1411                 case IB_SMP_ATTR_GUID_INFO:
1412                         ret = recv_subn_set_guidinfo(smp, ibdev);
1413                         goto bail;
1414                 case IB_SMP_ATTR_PORT_INFO:
1415                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1416                         goto bail;
1417                 case IB_SMP_ATTR_PKEY_TABLE:
1418                         ret = recv_subn_set_pkeytable(smp, ibdev);
1419                         goto bail;
1420                 case IB_SMP_ATTR_SM_INFO:
1421                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1422                                 ret = IB_MAD_RESULT_SUCCESS |
1423                                         IB_MAD_RESULT_CONSUMED;
1424                                 goto bail;
1425                         }
1426                         if (dev->port_cap_flags & IB_PORT_SM) {
1427                                 ret = IB_MAD_RESULT_SUCCESS;
1428                                 goto bail;
1429                         }
1430                         /* FALLTHROUGH */
1431                 default:
1432                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1433                         ret = reply(smp);
1434                         goto bail;
1435                 }
1436
1437         case IB_MGMT_METHOD_GET_RESP:
1438                 /*
1439                  * The ib_mad module will call us to process responses
1440                  * before checking for other consumers.
1441                  * Just tell the caller to process it normally.
1442                  */
1443                 ret = IB_MAD_RESULT_FAILURE;
1444                 goto bail;
1445         default:
1446                 smp->status |= IB_SMP_UNSUP_METHOD;
1447                 ret = reply(smp);
1448         }
1449
1450 bail:
1451         return ret;
1452 }
1453
1454 static int process_perf(struct ib_device *ibdev, u8 port_num,
1455                         struct ib_mad *in_mad,
1456                         struct ib_mad *out_mad)
1457 {
1458         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1459         int ret;
1460
1461         *out_mad = *in_mad;
1462         if (pmp->class_version != 1) {
1463                 pmp->status |= IB_SMP_UNSUP_VERSION;
1464                 ret = reply((struct ib_smp *) pmp);
1465                 goto bail;
1466         }
1467
1468         switch (pmp->method) {
1469         case IB_MGMT_METHOD_GET:
1470                 switch (pmp->attr_id) {
1471                 case IB_PMA_CLASS_PORT_INFO:
1472                         ret = recv_pma_get_classportinfo(pmp);
1473                         goto bail;
1474                 case IB_PMA_PORT_SAMPLES_CONTROL:
1475                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1476                                                               port_num);
1477                         goto bail;
1478                 case IB_PMA_PORT_SAMPLES_RESULT:
1479                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1480                         goto bail;
1481                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1482                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1483                                                                  ibdev);
1484                         goto bail;
1485                 case IB_PMA_PORT_COUNTERS:
1486                         ret = recv_pma_get_portcounters(pmp, ibdev,
1487                                                         port_num);
1488                         goto bail;
1489                 case IB_PMA_PORT_COUNTERS_EXT:
1490                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1491                                                             port_num);
1492                         goto bail;
1493                 default:
1494                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1495                         ret = reply((struct ib_smp *) pmp);
1496                         goto bail;
1497                 }
1498
1499         case IB_MGMT_METHOD_SET:
1500                 switch (pmp->attr_id) {
1501                 case IB_PMA_PORT_SAMPLES_CONTROL:
1502                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1503                                                               port_num);
1504                         goto bail;
1505                 case IB_PMA_PORT_COUNTERS:
1506                         ret = recv_pma_set_portcounters(pmp, ibdev,
1507                                                         port_num);
1508                         goto bail;
1509                 case IB_PMA_PORT_COUNTERS_EXT:
1510                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1511                                                             port_num);
1512                         goto bail;
1513                 default:
1514                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1515                         ret = reply((struct ib_smp *) pmp);
1516                         goto bail;
1517                 }
1518
1519         case IB_MGMT_METHOD_GET_RESP:
1520                 /*
1521                  * The ib_mad module will call us to process responses
1522                  * before checking for other consumers.
1523                  * Just tell the caller to process it normally.
1524                  */
1525                 ret = IB_MAD_RESULT_FAILURE;
1526                 goto bail;
1527         default:
1528                 pmp->status |= IB_SMP_UNSUP_METHOD;
1529                 ret = reply((struct ib_smp *) pmp);
1530         }
1531
1532 bail:
1533         return ret;
1534 }
1535
1536 /**
1537  * ipath_process_mad - process an incoming MAD packet
1538  * @ibdev: the infiniband device this packet came in on
1539  * @mad_flags: MAD flags
1540  * @port_num: the port number this packet came in on
1541  * @in_wc: the work completion entry for this packet
1542  * @in_grh: the global route header for this packet
1543  * @in_mad: the incoming MAD
1544  * @out_mad: any outgoing MAD reply
1545  *
1546  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1547  * interested in processing.
1548  *
1549  * Note that the verbs framework has already done the MAD sanity checks,
1550  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1551  * MADs.
1552  *
1553  * This is called by the ib_mad module.
1554  */
1555 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1556                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1557                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1558 {
1559         int ret;
1560
1561         switch (in_mad->mad_hdr.mgmt_class) {
1562         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1563         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1564                 ret = process_subn(ibdev, mad_flags, port_num,
1565                                    in_mad, out_mad);
1566                 goto bail;
1567         case IB_MGMT_CLASS_PERF_MGMT:
1568                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1569                 goto bail;
1570         default:
1571                 ret = IB_MAD_RESULT_SUCCESS;
1572         }
1573
1574 bail:
1575         return ret;
1576 }