err.no Git - linux-2.6/blob - 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         /* our mtu cap depends on whether 4K MTU enabled or not */
 296         pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
 297         /* HCAs ignore VLStallCount and HOQLife */
 298         /* pip->vlstallcnt_hoqlife; */
 299         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
 300         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
 301         /* P_KeyViolations are counted by hardware. */
 302         pip->pkey_violations =
 303                 cpu_to_be16((ipath_get_cr_errpkey(dev->dd) -
 304                              dev->z_pkey_violations) & 0xFFFF);
 305         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
 306         /* Only the hardware GUID is supported for now */
 307         pip->guid_cap = 1;
 308         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
 309         /* 32.768 usec. response time (guessing) */
 310         pip->resv_resptimevalue = 3;
 311         pip->localphyerrors_overrunerrors =
 312                 (get_phyerrthreshold(dev->dd) << 4) |
 313                 get_overrunthreshold(dev->dd);
 314         /* pip->max_credit_hint; */
 315         /* pip->link_roundtrip_latency[3]; */
 316
 317         ret = reply(smp);
 318
 319 bail:
 320         return ret;
 321 }
 322
 323 /**
 324  * get_pkeys - return the PKEY table for port 0
 325  * @dd: the infinipath device
 326  * @pkeys: the pkey table is placed here
 327  */
 328 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
 329 {
 330         struct ipath_portdata *pd = dd->ipath_pd[0];
 331
 332         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
 333
 334         return 0;
 335 }
 336
 337 static int recv_subn_get_pkeytable(struct ib_smp *smp,
 338                                    struct ib_device *ibdev)
 339 {
 340         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 341         u16 *p = (u16 *) smp->data;
 342         __be16 *q = (__be16 *) smp->data;
 343
 344         /* 64 blocks of 32 16-bit P_Key entries */
 345
 346         memset(smp->data, 0, sizeof(smp->data));
 347         if (startpx == 0) {
 348                 struct ipath_ibdev *dev = to_idev(ibdev);
 349                 unsigned i, n = ipath_get_npkeys(dev->dd);
 350
 351                 get_pkeys(dev->dd, p);
 352
 353                 for (i = 0; i < n; i++)
 354                         q[i] = cpu_to_be16(p[i]);
 355         } else
 356                 smp->status |= IB_SMP_INVALID_FIELD;
 357
 358         return reply(smp);
 359 }
 360
 361 static int recv_subn_set_guidinfo(struct ib_smp *smp,
 362                                   struct ib_device *ibdev)
 363 {
 364         /* The only GUID we support is the first read-only entry. */
 365         return recv_subn_get_guidinfo(smp, ibdev);
 366 }
 367
 368 /**
 369  * set_linkdowndefaultstate - set the default linkdown state
 370  * @dd: the infinipath device
 371  * @sleep: the new state
 372  *
 373  * Note that this will only take effect when the link state changes.
 374  */
 375 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
 376 {
 377         if (sleep)
 378                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 379         else
 380                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 381         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 382                          dd->ipath_ibcctrl);
 383         return 0;
 384 }
 385
 386 /**
 387  * recv_subn_set_portinfo - set port information
 388  * @smp: the incoming SM packet
 389  * @ibdev: the infiniband device
 390  * @port: the port on the device
 391  *
 392  * Set Portinfo (see ch. 14.2.5.6).
 393  */
 394 static int recv_subn_set_portinfo(struct ib_smp *smp,
 395                                   struct ib_device *ibdev, u8 port)
 396 {
 397         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 398         struct ib_event event;
 399         struct ipath_ibdev *dev;
 400         struct ipath_devdata *dd;
 401         char clientrereg = 0;
 402         u16 lid, smlid;
 403         u8 lwe;
 404         u8 lse;
 405         u8 state;
 406         u16 lstate;
 407         u32 mtu;
 408         int ret, ore;
 409
 410         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
 411                 goto err;
 412
 413         dev = to_idev(ibdev);
 414         dd = dev->dd;
 415         event.device = ibdev;
 416         event.element.port_num = port;
 417
 418         dev->mkey = pip->mkey;
 419         dev->gid_prefix = pip->gid_prefix;
 420         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
 421
 422         lid = be16_to_cpu(pip->lid);
 423         if (dd->ipath_lid != lid ||
 424             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
 425                 /* Must be a valid unicast LID address. */
 426                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
 427                         goto err;
 428                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
 429                 event.event = IB_EVENT_LID_CHANGE;
 430                 ib_dispatch_event(&event);
 431         }
 432
 433         smlid = be16_to_cpu(pip->sm_lid);
 434         if (smlid != dev->sm_lid) {
 435                 /* Must be a valid unicast LID address. */
 436                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
 437                         goto err;
 438                 dev->sm_lid = smlid;
 439                 event.event = IB_EVENT_SM_CHANGE;
 440                 ib_dispatch_event(&event);
 441         }
 442
 443         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
 444         lwe = pip->link_width_enabled;
 445         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
 446                 goto err;
 447         if (lwe == 0xFF)
 448                 dev->link_width_enabled = 3;    /* 1x or 4x */
 449         else if (lwe)
 450                 dev->link_width_enabled = lwe;
 451
 452         /* Only 2.5 Gbs supported. */
 453         lse = pip->linkspeedactive_enabled & 0xF;
 454         if (lse >= 2 && lse <= 0xE)
 455                 goto err;
 456
 457         /* Set link down default state. */
 458         switch (pip->portphysstate_linkdown & 0xF) {
 459         case 0: /* NOP */
 460                 break;
 461         case 1: /* SLEEP */
 462                 if (set_linkdowndefaultstate(dd, 1))
 463                         goto err;
 464                 break;
 465         case 2: /* POLL */
 466                 if (set_linkdowndefaultstate(dd, 0))
 467                         goto err;
 468                 break;
 469         default:
 470                 goto err;
 471         }
 472
 473         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
 474         dev->vl_high_limit = pip->vl_high_limit;
 475
 476         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
 477         case IB_MTU_256:
 478                 mtu = 256;
 479                 break;
 480         case IB_MTU_512:
 481                 mtu = 512;
 482                 break;
 483         case IB_MTU_1024:
 484                 mtu = 1024;
 485                 break;
 486         case IB_MTU_2048:
 487                 mtu = 2048;
 488                 break;
 489         case IB_MTU_4096:
 490                 if (!ipath_mtu4096)
 491                         goto err;
 492                 mtu = 4096;
 493                 break;
 494         default:
 495                 /* XXX We have already partially updated our state! */
 496                 goto err;
 497         }
 498         ipath_set_mtu(dd, mtu);
 499
 500         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
 501
 502         /* We only support VL0 */
 503         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
 504                 goto err;
 505
 506         if (pip->mkey_violations == 0)
 507                 dev->mkey_violations = 0;
 508
 509         /*
 510          * Hardware counter can't be reset so snapshot and subtract
 511          * later.
 512          */
 513         if (pip->pkey_violations == 0)
 514                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
 515
 516         if (pip->qkey_violations == 0)
 517                 dev->qkey_violations = 0;
 518
 519         ore = pip->localphyerrors_overrunerrors;
 520         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
 521                 goto err;
 522
 523         if (set_overrunthreshold(dd, (ore & 0xF)))
 524                 goto err;
 525
 526         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 527
 528         if (pip->clientrereg_resv_subnetto & 0x80) {
 529                 clientrereg = 1;
 530                 event.event = IB_EVENT_CLIENT_REREGISTER;
 531                 ib_dispatch_event(&event);
 532         }
 533
 534         /*
 535          * Do the port state change now that the other link parameters
 536          * have been set.
 537          * Changing the port physical state only makes sense if the link
 538          * is down or is being set to down.
 539          */
 540         state = pip->linkspeed_portstate & 0xF;
 541         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 542         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
 543                 goto err;
 544
 545         /*
 546          * Only state changes of DOWN, ARM, and ACTIVE are valid
 547          * and must be in the correct state to take effect (see 7.2.6).
 548          */
 549         switch (state) {
 550         case IB_PORT_NOP:
 551                 if (lstate == 0)
 552                         break;
 553                 /* FALLTHROUGH */
 554         case IB_PORT_DOWN:
 555                 if (lstate == 0)
 556                         lstate = IPATH_IB_LINKDOWN_ONLY;
 557                 else if (lstate == 1)
 558                         lstate = IPATH_IB_LINKDOWN_SLEEP;
 559                 else if (lstate == 2)
 560                         lstate = IPATH_IB_LINKDOWN;
 561                 else if (lstate == 3)
 562                         lstate = IPATH_IB_LINKDOWN_DISABLE;
 563                 else
 564                         goto err;
 565                 ipath_set_linkstate(dd, lstate);
 566                 ipath_wait_linkstate(dd, IPATH_LINKINIT | IPATH_LINKARMED |
 567                                 IPATH_LINKACTIVE, 1000);
 568                 break;
 569         case IB_PORT_ARMED:
 570                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
 571                 break;
 572         case IB_PORT_ACTIVE:
 573                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
 574                 break;
 575         default:
 576                 /* XXX We have already partially updated our state! */
 577                 goto err;
 578         }
 579
 580         ret = recv_subn_get_portinfo(smp, ibdev, port);
 581
 582         if (clientrereg)
 583                 pip->clientrereg_resv_subnetto |= 0x80;
 584
 585         goto done;
 586
 587 err:
 588         smp->status |= IB_SMP_INVALID_FIELD;
 589         ret = recv_subn_get_portinfo(smp, ibdev, port);
 590
 591 done:
 592         return ret;
 593 }
 594
 595 /**
 596  * rm_pkey - decrecment the reference count for the given PKEY
 597  * @dd: the infinipath device
 598  * @key: the PKEY index
 599  *
 600  * Return true if this was the last reference and the hardware table entry
 601  * needs to be changed.
 602  */
 603 static int rm_pkey(struct ipath_devdata *dd, u16 key)
 604 {
 605         int i;
 606         int ret;
 607
 608         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 609                 if (dd->ipath_pkeys[i] != key)
 610                         continue;
 611                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
 612                         dd->ipath_pkeys[i] = 0;
 613                         ret = 1;
 614                         goto bail;
 615                 }
 616                 break;
 617         }
 618
 619         ret = 0;
 620
 621 bail:
 622         return ret;
 623 }
 624
 625 /**
 626  * add_pkey - add the given PKEY to the hardware table
 627  * @dd: the infinipath device
 628  * @key: the PKEY
 629  *
 630  * Return an error code if unable to add the entry, zero if no change,
 631  * or 1 if the hardware PKEY register needs to be updated.
 632  */
 633 static int add_pkey(struct ipath_devdata *dd, u16 key)
 634 {
 635         int i;
 636         u16 lkey = key & 0x7FFF;
 637         int any = 0;
 638         int ret;
 639
 640         if (lkey == 0x7FFF) {
 641                 ret = 0;
 642                 goto bail;
 643         }
 644
 645         /* Look for an empty slot or a matching PKEY. */
 646         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 647                 if (!dd->ipath_pkeys[i]) {
 648                         any++;
 649                         continue;
 650                 }
 651                 /* If it matches exactly, try to increment the ref count */
 652                 if (dd->ipath_pkeys[i] == key) {
 653                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
 654                                 ret = 0;
 655                                 goto bail;
 656                         }
 657                         /* Lost the race. Look for an empty slot below. */
 658                         atomic_dec(&dd->ipath_pkeyrefs[i]);
 659                         any++;
 660                 }
 661                 /*
 662                  * It makes no sense to have both the limited and unlimited
 663                  * PKEY set at the same time since the unlimited one will
 664                  * disable the limited one.
 665                  */
 666                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
 667                         ret = -EEXIST;
 668                         goto bail;
 669                 }
 670         }
 671         if (!any) {
 672                 ret = -EBUSY;
 673                 goto bail;
 674         }
 675         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 676                 if (!dd->ipath_pkeys[i] &&
 677                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
 678                         /* for ipathstats, etc. */
 679                         ipath_stats.sps_pkeys[i] = lkey;
 680                         dd->ipath_pkeys[i] = key;
 681                         ret = 1;
 682                         goto bail;
 683                 }
 684         }
 685         ret = -EBUSY;
 686
 687 bail:
 688         return ret;
 689 }
 690
 691 /**
 692  * set_pkeys - set the PKEY table for port 0
 693  * @dd: the infinipath device
 694  * @pkeys: the PKEY table
 695  */
 696 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 697 {
 698         struct ipath_portdata *pd;
 699         int i;
 700         int changed = 0;
 701
 702         pd = dd->ipath_pd[0];
 703
 704         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
 705                 u16 key = pkeys[i];
 706                 u16 okey = pd->port_pkeys[i];
 707
 708                 if (key == okey)
 709                         continue;
 710                 /*
 711                  * The value of this PKEY table entry is changing.
 712                  * Remove the old entry in the hardware's array of PKEYs.
 713                  */
 714                 if (okey & 0x7FFF)
 715                         changed |= rm_pkey(dd, okey);
 716                 if (key & 0x7FFF) {
 717                         int ret = add_pkey(dd, key);
 718
 719                         if (ret < 0)
 720                                 key = 0;
 721                         else
 722                                 changed |= ret;
 723                 }
 724                 pd->port_pkeys[i] = key;
 725         }
 726         if (changed) {
 727                 u64 pkey;
 728
 729                 pkey = (u64) dd->ipath_pkeys[0] |
 730                         ((u64) dd->ipath_pkeys[1] << 16) |
 731                         ((u64) dd->ipath_pkeys[2] << 32) |
 732                         ((u64) dd->ipath_pkeys[3] << 48);
 733                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
 734                            (unsigned long long) pkey);
 735                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
 736                                  pkey);
 737         }
 738         return 0;
 739 }
 740
 741 static int recv_subn_set_pkeytable(struct ib_smp *smp,
 742                                    struct ib_device *ibdev)
 743 {
 744         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 745         __be16 *p = (__be16 *) smp->data;
 746         u16 *q = (u16 *) smp->data;
 747         struct ipath_ibdev *dev = to_idev(ibdev);
 748         unsigned i, n = ipath_get_npkeys(dev->dd);
 749
 750         for (i = 0; i < n; i++)
 751                 q[i] = be16_to_cpu(p[i]);
 752
 753         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
 754                 smp->status |= IB_SMP_INVALID_FIELD;
 755
 756         return recv_subn_get_pkeytable(smp, ibdev);
 757 }
 758
 759 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
 760 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
 761 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
 762 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
 763 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
 764 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
 765
 766 struct ib_perf {
 767         u8 base_version;
 768         u8 mgmt_class;
 769         u8 class_version;
 770         u8 method;
 771         __be16 status;
 772         __be16 unused;
 773         __be64 tid;
 774         __be16 attr_id;
 775         __be16 resv;
 776         __be32 attr_mod;
 777         u8 reserved[40];
 778         u8 data[192];
 779 } __attribute__ ((packed));
 780
 781 struct ib_pma_classportinfo {
 782         u8 base_version;
 783         u8 class_version;
 784         __be16 cap_mask;
 785         u8 reserved[3];
 786         u8 resp_time_value;     /* only lower 5 bits */
 787         union ib_gid redirect_gid;
 788         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
 789         __be16 redirect_lid;
 790         __be16 redirect_pkey;
 791         __be32 redirect_qp;     /* only lower 24 bits */
 792         __be32 redirect_qkey;
 793         union ib_gid trap_gid;
 794         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
 795         __be16 trap_lid;
 796         __be16 trap_pkey;
 797         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
 798         __be32 trap_qkey;
 799 } __attribute__ ((packed));
 800
 801 struct ib_pma_portsamplescontrol {
 802         u8 opcode;
 803         u8 port_select;
 804         u8 tick;
 805         u8 counter_width;       /* only lower 3 bits */
 806         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
 807         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
 808         u8 sample_mechanisms;
 809         u8 sample_status;       /* only lower 2 bits */
 810         __be64 option_mask;
 811         __be64 vendor_mask;
 812         __be32 sample_start;
 813         __be32 sample_interval;
 814         __be16 tag;
 815         __be16 counter_select[15];
 816 } __attribute__ ((packed));
 817
 818 struct ib_pma_portsamplesresult {
 819         __be16 tag;
 820         __be16 sample_status;   /* only lower 2 bits */
 821         __be32 counter[15];
 822 } __attribute__ ((packed));
 823
 824 struct ib_pma_portsamplesresult_ext {
 825         __be16 tag;
 826         __be16 sample_status;   /* only lower 2 bits */
 827         __be32 extended_width;  /* only upper 2 bits */
 828         __be64 counter[15];
 829 } __attribute__ ((packed));
 830
 831 struct ib_pma_portcounters {
 832         u8 reserved;
 833         u8 port_select;
 834         __be16 counter_select;
 835         __be16 symbol_error_counter;
 836         u8 link_error_recovery_counter;
 837         u8 link_downed_counter;
 838         __be16 port_rcv_errors;
 839         __be16 port_rcv_remphys_errors;
 840         __be16 port_rcv_switch_relay_errors;
 841         __be16 port_xmit_discards;
 842         u8 port_xmit_constraint_errors;
 843         u8 port_rcv_constraint_errors;
 844         u8 reserved1;
 845         u8 lli_ebor_errors;     /* 4, 4, bits */
 846         __be16 reserved2;
 847         __be16 vl15_dropped;
 848         __be32 port_xmit_data;
 849         __be32 port_rcv_data;
 850         __be32 port_xmit_packets;
 851         __be32 port_rcv_packets;
 852 } __attribute__ ((packed));
 853
 854 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
 855 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
 856 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
 857 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
 858 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
 859 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
 860 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
 861 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
 862 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
 863 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
 864 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
 865 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
 866 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
 867
 868 struct ib_pma_portcounters_ext {
 869         u8 reserved;
 870         u8 port_select;
 871         __be16 counter_select;
 872         __be32 reserved1;
 873         __be64 port_xmit_data;
 874         __be64 port_rcv_data;
 875         __be64 port_xmit_packets;
 876         __be64 port_rcv_packets;
 877         __be64 port_unicast_xmit_packets;
 878         __be64 port_unicast_rcv_packets;
 879         __be64 port_multicast_xmit_packets;
 880         __be64 port_multicast_rcv_packets;
 881 } __attribute__ ((packed));
 882
 883 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
 884 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
 885 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
 886 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
 887 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
 888 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
 889 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
 890 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
 891
 892 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
 893 {
 894         struct ib_pma_classportinfo *p =
 895                 (struct ib_pma_classportinfo *)pmp->data;
 896
 897         memset(pmp->data, 0, sizeof(pmp->data));
 898
 899         if (pmp->attr_mod != 0)
 900                 pmp->status |= IB_SMP_INVALID_FIELD;
 901
 902         /* Indicate AllPortSelect is valid (only one port anyway) */
 903         p->cap_mask = __constant_cpu_to_be16(1 << 8);
 904         p->base_version = 1;
 905         p->class_version = 1;
 906         /*
 907          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
 908          * sec.
 909          */
 910         p->resp_time_value = 18;
 911
 912         return reply((struct ib_smp *) pmp);
 913 }
 914
 915 /*
 916  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
 917  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
 918  * We support 5 counters which only count the mandatory quantities.
 919  */
 920 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
 921 #define COUNTER_MASK0_9 \
 922         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
 923                                COUNTER_MASK(1, 1) | \
 924                                COUNTER_MASK(1, 2) | \
 925                                COUNTER_MASK(1, 3) | \
 926                                COUNTER_MASK(1, 4))
 927
 928 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 929                                            struct ib_device *ibdev, u8 port)
 930 {
 931         struct ib_pma_portsamplescontrol *p =
 932                 (struct ib_pma_portsamplescontrol *)pmp->data;
 933         struct ipath_ibdev *dev = to_idev(ibdev);
 934         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 935         unsigned long flags;
 936         u8 port_select = p->port_select;
 937
 938         memset(pmp->data, 0, sizeof(pmp->data));
 939
 940         p->port_select = port_select;
 941         if (pmp->attr_mod != 0 ||
 942             (port_select != port && port_select != 0xFF))
 943                 pmp->status |= IB_SMP_INVALID_FIELD;
 944         /*
 945          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 946          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
 947          * intervals are counted in ticks.  Since we use Linux timers, that
 948          * count in jiffies, we can't sample for less than 1000 ticks if HZ
 949          * == 1000 (4000 ticks if HZ is 250).
 950          */
 951         /* XXX This is WRONG. */
 952         p->tick = 250;          /* 1 usec. */
 953         p->counter_width = 4;   /* 32 bit counters */
 954         p->counter_mask0_9 = COUNTER_MASK0_9;
 955         spin_lock_irqsave(&dev->pending_lock, flags);
 956         if (crp->cr_psstat)
 957                 p->sample_status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 958         else
 959                 p->sample_status = dev->pma_sample_status;
 960         p->sample_start = cpu_to_be32(dev->pma_sample_start);
 961         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
 962         p->tag = cpu_to_be16(dev->pma_tag);
 963         p->counter_select[0] = dev->pma_counter_select[0];
 964         p->counter_select[1] = dev->pma_counter_select[1];
 965         p->counter_select[2] = dev->pma_counter_select[2];
 966         p->counter_select[3] = dev->pma_counter_select[3];
 967         p->counter_select[4] = dev->pma_counter_select[4];
 968         spin_unlock_irqrestore(&dev->pending_lock, flags);
 969
 970         return reply((struct ib_smp *) pmp);
 971 }
 972
 973 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
 974                                            struct ib_device *ibdev, u8 port)
 975 {
 976         struct ib_pma_portsamplescontrol *p =
 977                 (struct ib_pma_portsamplescontrol *)pmp->data;
 978         struct ipath_ibdev *dev = to_idev(ibdev);
 979         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 980         unsigned long flags;
 981         u8 status;
 982         int ret;
 983
 984         if (pmp->attr_mod != 0 ||
 985             (p->port_select != port && p->port_select != 0xFF)) {
 986                 pmp->status |= IB_SMP_INVALID_FIELD;
 987                 ret = reply((struct ib_smp *) pmp);
 988                 goto bail;
 989         }
 990
 991         spin_lock_irqsave(&dev->pending_lock, flags);
 992         if (crp->cr_psstat)
 993                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 994         else
 995                 status = dev->pma_sample_status;
 996         if (status == IB_PMA_SAMPLE_STATUS_DONE) {
 997                 dev->pma_sample_start = be32_to_cpu(p->sample_start);
 998                 dev->pma_sample_interval = be32_to_cpu(p->sample_interval);
 999                 dev->pma_tag = be16_to_cpu(p->tag);
1000                 dev->pma_counter_select[0] = p->counter_select[0];
1001                 dev->pma_counter_select[1] = p->counter_select[1];
1002                 dev->pma_counter_select[2] = p->counter_select[2];
1003                 dev->pma_counter_select[3] = p->counter_select[3];
1004                 dev->pma_counter_select[4] = p->counter_select[4];
1005                 if (crp->cr_psstat) {
1006                         ipath_write_creg(dev->dd, crp->cr_psinterval,
1007                                          dev->pma_sample_interval);
1008                         ipath_write_creg(dev->dd, crp->cr_psstart,
1009                                          dev->pma_sample_start);
1010                 } else
1011                         dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
1012         }
1013         spin_unlock_irqrestore(&dev->pending_lock, flags);
1014
1015         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1016
1017 bail:
1018         return ret;
1019 }
1020
1021 static u64 get_counter(struct ipath_ibdev *dev,
1022                        struct ipath_cregs const *crp,
1023                        __be16 sel)
1024 {
1025         u64 ret;
1026
1027         switch (sel) {
1028         case IB_PMA_PORT_XMIT_DATA:
1029                 ret = (crp->cr_psxmitdatacount) ?
1030                         ipath_read_creg32(dev->dd, crp->cr_psxmitdatacount) :
1031                         dev->ipath_sword;
1032                 break;
1033         case IB_PMA_PORT_RCV_DATA:
1034                 ret = (crp->cr_psrcvdatacount) ?
1035                         ipath_read_creg32(dev->dd, crp->cr_psrcvdatacount) :
1036                         dev->ipath_rword;
1037                 break;
1038         case IB_PMA_PORT_XMIT_PKTS:
1039                 ret = (crp->cr_psxmitpktscount) ?
1040                         ipath_read_creg32(dev->dd, crp->cr_psxmitpktscount) :
1041                         dev->ipath_spkts;
1042                 break;
1043         case IB_PMA_PORT_RCV_PKTS:
1044                 ret = (crp->cr_psrcvpktscount) ?
1045                         ipath_read_creg32(dev->dd, crp->cr_psrcvpktscount) :
1046                         dev->ipath_rpkts;
1047                 break;
1048         case IB_PMA_PORT_XMIT_WAIT:
1049                 ret = (crp->cr_psxmitwaitcount) ?
1050                         ipath_read_creg32(dev->dd, crp->cr_psxmitwaitcount) :
1051                         dev->ipath_xmit_wait;
1052                 break;
1053         default:
1054                 ret = 0;
1055         }
1056
1057         return ret;
1058 }
1059
1060 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1061                                           struct ib_device *ibdev)
1062 {
1063         struct ib_pma_portsamplesresult *p =
1064                 (struct ib_pma_portsamplesresult *)pmp->data;
1065         struct ipath_ibdev *dev = to_idev(ibdev);
1066         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1067         u8 status;
1068         int i;
1069
1070         memset(pmp->data, 0, sizeof(pmp->data));
1071         p->tag = cpu_to_be16(dev->pma_tag);
1072         if (crp->cr_psstat)
1073                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1074         else
1075                 status = dev->pma_sample_status;
1076         p->sample_status = cpu_to_be16(status);
1077         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1078                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1079                     cpu_to_be32(
1080                         get_counter(dev, crp, dev->pma_counter_select[i]));
1081
1082         return reply((struct ib_smp *) pmp);
1083 }
1084
1085 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1086                                               struct ib_device *ibdev)
1087 {
1088         struct ib_pma_portsamplesresult_ext *p =
1089                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1090         struct ipath_ibdev *dev = to_idev(ibdev);
1091         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1092         u8 status;
1093         int i;
1094
1095         memset(pmp->data, 0, sizeof(pmp->data));
1096         p->tag = cpu_to_be16(dev->pma_tag);
1097         if (crp->cr_psstat)
1098                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1099         else
1100                 status = dev->pma_sample_status;
1101         p->sample_status = cpu_to_be16(status);
1102         /* 64 bits */
1103         p->extended_width = __constant_cpu_to_be32(0x80000000);
1104         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1105                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1106                     cpu_to_be64(
1107                         get_counter(dev, crp, dev->pma_counter_select[i]));
1108
1109         return reply((struct ib_smp *) pmp);
1110 }
1111
1112 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1113                                      struct ib_device *ibdev, u8 port)
1114 {
1115         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1116                 pmp->data;
1117         struct ipath_ibdev *dev = to_idev(ibdev);
1118         struct ipath_verbs_counters cntrs;
1119         u8 port_select = p->port_select;
1120
1121         ipath_get_counters(dev->dd, &cntrs);
1122
1123         /* Adjust counters for any resets done. */
1124         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1125         cntrs.link_error_recovery_counter -=
1126                 dev->z_link_error_recovery_counter;
1127         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1128         cntrs.port_rcv_errors += dev->rcv_errors;
1129         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1130         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1131         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1132         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1133         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1134         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1135         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1136         cntrs.local_link_integrity_errors -=
1137                 dev->z_local_link_integrity_errors;
1138         cntrs.excessive_buffer_overrun_errors -=
1139                 dev->z_excessive_buffer_overrun_errors;
1140         cntrs.vl15_dropped -= dev->z_vl15_dropped;
1141         cntrs.vl15_dropped += dev->n_vl15_dropped;
1142
1143         memset(pmp->data, 0, sizeof(pmp->data));
1144
1145         p->port_select = port_select;
1146         if (pmp->attr_mod != 0 ||
1147             (port_select != port && port_select != 0xFF))
1148                 pmp->status |= IB_SMP_INVALID_FIELD;
1149
1150         if (cntrs.symbol_error_counter > 0xFFFFUL)
1151                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1152         else
1153                 p->symbol_error_counter =
1154                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1155         if (cntrs.link_error_recovery_counter > 0xFFUL)
1156                 p->link_error_recovery_counter = 0xFF;
1157         else
1158                 p->link_error_recovery_counter =
1159                         (u8)cntrs.link_error_recovery_counter;
1160         if (cntrs.link_downed_counter > 0xFFUL)
1161                 p->link_downed_counter = 0xFF;
1162         else
1163                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1164         if (cntrs.port_rcv_errors > 0xFFFFUL)
1165                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1166         else
1167                 p->port_rcv_errors =
1168                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1169         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1170                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1171         else
1172                 p->port_rcv_remphys_errors =
1173                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1174         if (cntrs.port_xmit_discards > 0xFFFFUL)
1175                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1176         else
1177                 p->port_xmit_discards =
1178                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1179         if (cntrs.local_link_integrity_errors > 0xFUL)
1180                 cntrs.local_link_integrity_errors = 0xFUL;
1181         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1182                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1183         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1184                 cntrs.excessive_buffer_overrun_errors;
1185         if (cntrs.vl15_dropped > 0xFFFFUL)
1186                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1187         else
1188                 p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1189         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1190                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1191         else
1192                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1193         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1194                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1195         else
1196                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1197         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1198                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1199         else
1200                 p->port_xmit_packets =
1201                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1202         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1203                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1204         else
1205                 p->port_rcv_packets =
1206                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1207
1208         return reply((struct ib_smp *) pmp);
1209 }
1210
1211 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1212                                          struct ib_device *ibdev, u8 port)
1213 {
1214         struct ib_pma_portcounters_ext *p =
1215                 (struct ib_pma_portcounters_ext *)pmp->data;
1216         struct ipath_ibdev *dev = to_idev(ibdev);
1217         u64 swords, rwords, spkts, rpkts, xwait;
1218         u8 port_select = p->port_select;
1219
1220         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1221                                 &rpkts, &xwait);
1222
1223         /* Adjust counters for any resets done. */
1224         swords -= dev->z_port_xmit_data;
1225         rwords -= dev->z_port_rcv_data;
1226         spkts -= dev->z_port_xmit_packets;
1227         rpkts -= dev->z_port_rcv_packets;
1228
1229         memset(pmp->data, 0, sizeof(pmp->data));
1230
1231         p->port_select = port_select;
1232         if (pmp->attr_mod != 0 ||
1233             (port_select != port && port_select != 0xFF))
1234                 pmp->status |= IB_SMP_INVALID_FIELD;
1235
1236         p->port_xmit_data = cpu_to_be64(swords);
1237         p->port_rcv_data = cpu_to_be64(rwords);
1238         p->port_xmit_packets = cpu_to_be64(spkts);
1239         p->port_rcv_packets = cpu_to_be64(rpkts);
1240         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1241         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1242         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1243         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1244
1245         return reply((struct ib_smp *) pmp);
1246 }
1247
1248 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1249                                      struct ib_device *ibdev, u8 port)
1250 {
1251         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1252                 pmp->data;
1253         struct ipath_ibdev *dev = to_idev(ibdev);
1254         struct ipath_verbs_counters cntrs;
1255
1256         /*
1257          * Since the HW doesn't support clearing counters, we save the
1258          * current count and subtract it from future responses.
1259          */
1260         ipath_get_counters(dev->dd, &cntrs);
1261
1262         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1263                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1264
1265         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1266                 dev->z_link_error_recovery_counter =
1267                         cntrs.link_error_recovery_counter;
1268
1269         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1270                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1271
1272         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1273                 dev->z_port_rcv_errors =
1274                         cntrs.port_rcv_errors + dev->rcv_errors;
1275
1276         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1277                 dev->z_port_rcv_remphys_errors =
1278                         cntrs.port_rcv_remphys_errors;
1279
1280         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1281                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1282
1283         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1284                 dev->z_local_link_integrity_errors =
1285                         cntrs.local_link_integrity_errors;
1286
1287         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1288                 dev->z_excessive_buffer_overrun_errors =
1289                         cntrs.excessive_buffer_overrun_errors;
1290
1291         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1292                 dev->n_vl15_dropped = 0;
1293                 dev->z_vl15_dropped = cntrs.vl15_dropped;
1294         }
1295
1296         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1297                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1298
1299         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1300                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1301
1302         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1303                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1304
1305         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1306                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1307
1308         return recv_pma_get_portcounters(pmp, ibdev, port);
1309 }
1310
1311 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1312                                          struct ib_device *ibdev, u8 port)
1313 {
1314         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1315                 pmp->data;
1316         struct ipath_ibdev *dev = to_idev(ibdev);
1317         u64 swords, rwords, spkts, rpkts, xwait;
1318
1319         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1320                                 &rpkts, &xwait);
1321
1322         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1323                 dev->z_port_xmit_data = swords;
1324
1325         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1326                 dev->z_port_rcv_data = rwords;
1327
1328         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1329                 dev->z_port_xmit_packets = spkts;
1330
1331         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1332                 dev->z_port_rcv_packets = rpkts;
1333
1334         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1335                 dev->n_unicast_xmit = 0;
1336
1337         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1338                 dev->n_unicast_rcv = 0;
1339
1340         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1341                 dev->n_multicast_xmit = 0;
1342
1343         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1344                 dev->n_multicast_rcv = 0;
1345
1346         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1347 }
1348
1349 static int process_subn(struct ib_device *ibdev, int mad_flags,
1350                         u8 port_num, struct ib_mad *in_mad,
1351                         struct ib_mad *out_mad)
1352 {
1353         struct ib_smp *smp = (struct ib_smp *)out_mad;
1354         struct ipath_ibdev *dev = to_idev(ibdev);
1355         int ret;
1356
1357         *out_mad = *in_mad;
1358         if (smp->class_version != 1) {
1359                 smp->status |= IB_SMP_UNSUP_VERSION;
1360                 ret = reply(smp);
1361                 goto bail;
1362         }
1363
1364         /* Is the mkey in the process of expiring? */
1365         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1366                 /* Clear timeout and mkey protection field. */
1367                 dev->mkey_lease_timeout = 0;
1368                 dev->mkeyprot = 0;
1369         }
1370
1371         /*
1372          * M_Key checking depends on
1373          * Portinfo:M_Key_protect_bits
1374          */
1375         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1376             dev->mkey != smp->mkey &&
1377             (smp->method == IB_MGMT_METHOD_SET ||
1378              (smp->method == IB_MGMT_METHOD_GET &&
1379               dev->mkeyprot >= 2))) {
1380                 if (dev->mkey_violations != 0xFFFF)
1381                         ++dev->mkey_violations;
1382                 if (dev->mkey_lease_timeout ||
1383                     dev->mkey_lease_period == 0) {
1384                         ret = IB_MAD_RESULT_SUCCESS |
1385                                 IB_MAD_RESULT_CONSUMED;
1386                         goto bail;
1387                 }
1388                 dev->mkey_lease_timeout = jiffies +
1389                         dev->mkey_lease_period * HZ;
1390                 /* Future: Generate a trap notice. */
1391                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1392                 goto bail;
1393         } else if (dev->mkey_lease_timeout)
1394                 dev->mkey_lease_timeout = 0;
1395
1396         switch (smp->method) {
1397         case IB_MGMT_METHOD_GET:
1398                 switch (smp->attr_id) {
1399                 case IB_SMP_ATTR_NODE_DESC:
1400                         ret = recv_subn_get_nodedescription(smp, ibdev);
1401                         goto bail;
1402                 case IB_SMP_ATTR_NODE_INFO:
1403                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1404                         goto bail;
1405                 case IB_SMP_ATTR_GUID_INFO:
1406                         ret = recv_subn_get_guidinfo(smp, ibdev);
1407                         goto bail;
1408                 case IB_SMP_ATTR_PORT_INFO:
1409                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1410                         goto bail;
1411                 case IB_SMP_ATTR_PKEY_TABLE:
1412                         ret = recv_subn_get_pkeytable(smp, ibdev);
1413                         goto bail;
1414                 case IB_SMP_ATTR_SM_INFO:
1415                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1416                                 ret = IB_MAD_RESULT_SUCCESS |
1417                                         IB_MAD_RESULT_CONSUMED;
1418                                 goto bail;
1419                         }
1420                         if (dev->port_cap_flags & IB_PORT_SM) {
1421                                 ret = IB_MAD_RESULT_SUCCESS;
1422                                 goto bail;
1423                         }
1424                         /* FALLTHROUGH */
1425                 default:
1426                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1427                         ret = reply(smp);
1428                         goto bail;
1429                 }
1430
1431         case IB_MGMT_METHOD_SET:
1432                 switch (smp->attr_id) {
1433                 case IB_SMP_ATTR_GUID_INFO:
1434                         ret = recv_subn_set_guidinfo(smp, ibdev);
1435                         goto bail;
1436                 case IB_SMP_ATTR_PORT_INFO:
1437                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1438                         goto bail;
1439                 case IB_SMP_ATTR_PKEY_TABLE:
1440                         ret = recv_subn_set_pkeytable(smp, ibdev);
1441                         goto bail;
1442                 case IB_SMP_ATTR_SM_INFO:
1443                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1444                                 ret = IB_MAD_RESULT_SUCCESS |
1445                                         IB_MAD_RESULT_CONSUMED;
1446                                 goto bail;
1447                         }
1448                         if (dev->port_cap_flags & IB_PORT_SM) {
1449                                 ret = IB_MAD_RESULT_SUCCESS;
1450                                 goto bail;
1451                         }
1452                         /* FALLTHROUGH */
1453                 default:
1454                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1455                         ret = reply(smp);
1456                         goto bail;
1457                 }
1458
1459         case IB_MGMT_METHOD_GET_RESP:
1460                 /*
1461                  * The ib_mad module will call us to process responses
1462                  * before checking for other consumers.
1463                  * Just tell the caller to process it normally.
1464                  */
1465                 ret = IB_MAD_RESULT_SUCCESS;
1466                 goto bail;
1467         default:
1468                 smp->status |= IB_SMP_UNSUP_METHOD;
1469                 ret = reply(smp);
1470         }
1471
1472 bail:
1473         return ret;
1474 }
1475
1476 static int process_perf(struct ib_device *ibdev, u8 port_num,
1477                         struct ib_mad *in_mad,
1478                         struct ib_mad *out_mad)
1479 {
1480         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1481         int ret;
1482
1483         *out_mad = *in_mad;
1484         if (pmp->class_version != 1) {
1485                 pmp->status |= IB_SMP_UNSUP_VERSION;
1486                 ret = reply((struct ib_smp *) pmp);
1487                 goto bail;
1488         }
1489
1490         switch (pmp->method) {
1491         case IB_MGMT_METHOD_GET:
1492                 switch (pmp->attr_id) {
1493                 case IB_PMA_CLASS_PORT_INFO:
1494                         ret = recv_pma_get_classportinfo(pmp);
1495                         goto bail;
1496                 case IB_PMA_PORT_SAMPLES_CONTROL:
1497                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1498                                                               port_num);
1499                         goto bail;
1500                 case IB_PMA_PORT_SAMPLES_RESULT:
1501                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1502                         goto bail;
1503                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1504                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1505                                                                  ibdev);
1506                         goto bail;
1507                 case IB_PMA_PORT_COUNTERS:
1508                         ret = recv_pma_get_portcounters(pmp, ibdev,
1509                                                         port_num);
1510                         goto bail;
1511                 case IB_PMA_PORT_COUNTERS_EXT:
1512                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1513                                                             port_num);
1514                         goto bail;
1515                 default:
1516                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1517                         ret = reply((struct ib_smp *) pmp);
1518                         goto bail;
1519                 }
1520
1521         case IB_MGMT_METHOD_SET:
1522                 switch (pmp->attr_id) {
1523                 case IB_PMA_PORT_SAMPLES_CONTROL:
1524                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1525                                                               port_num);
1526                         goto bail;
1527                 case IB_PMA_PORT_COUNTERS:
1528                         ret = recv_pma_set_portcounters(pmp, ibdev,
1529                                                         port_num);
1530                         goto bail;
1531                 case IB_PMA_PORT_COUNTERS_EXT:
1532                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1533                                                             port_num);
1534                         goto bail;
1535                 default:
1536                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1537                         ret = reply((struct ib_smp *) pmp);
1538                         goto bail;
1539                 }
1540
1541         case IB_MGMT_METHOD_GET_RESP:
1542                 /*
1543                  * The ib_mad module will call us to process responses
1544                  * before checking for other consumers.
1545                  * Just tell the caller to process it normally.
1546                  */
1547                 ret = IB_MAD_RESULT_SUCCESS;
1548                 goto bail;
1549         default:
1550                 pmp->status |= IB_SMP_UNSUP_METHOD;
1551                 ret = reply((struct ib_smp *) pmp);
1552         }
1553
1554 bail:
1555         return ret;
1556 }
1557
1558 /**
1559  * ipath_process_mad - process an incoming MAD packet
1560  * @ibdev: the infiniband device this packet came in on
1561  * @mad_flags: MAD flags
1562  * @port_num: the port number this packet came in on
1563  * @in_wc: the work completion entry for this packet
1564  * @in_grh: the global route header for this packet
1565  * @in_mad: the incoming MAD
1566  * @out_mad: any outgoing MAD reply
1567  *
1568  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1569  * interested in processing.
1570  *
1571  * Note that the verbs framework has already done the MAD sanity checks,
1572  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1573  * MADs.
1574  *
1575  * This is called by the ib_mad module.
1576  */
1577 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1578                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1579                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1580 {
1581         int ret;
1582
1583         switch (in_mad->mad_hdr.mgmt_class) {
1584         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1585         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1586                 ret = process_subn(ibdev, mad_flags, port_num,
1587                                    in_mad, out_mad);
1588                 goto bail;
1589         case IB_MGMT_CLASS_PERF_MGMT:
1590                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1591                 goto bail;
1592         default:
1593                 ret = IB_MAD_RESULT_SUCCESS;
1594         }
1595
1596 bail:
1597         return ret;
1598 }