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         char clientrereg = 0;
 406         u16 lid, smlid;
 407         u8 lwe;
 408         u8 lse;
 409         u8 state;
 410         u16 lstate;
 411         u32 mtu;
 412         int ret, ore;
 413
 414         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
 415                 goto err;
 416
 417         dev = to_idev(ibdev);
 418         dd = dev->dd;
 419         event.device = ibdev;
 420         event.element.port_num = port;
 421
 422         dev->mkey = pip->mkey;
 423         dev->gid_prefix = pip->gid_prefix;
 424         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
 425
 426         lid = be16_to_cpu(pip->lid);
 427         if (dd->ipath_lid != lid ||
 428             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
 429                 /* Must be a valid unicast LID address. */
 430                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
 431                         goto err;
 432                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
 433                 event.event = IB_EVENT_LID_CHANGE;
 434                 ib_dispatch_event(&event);
 435         }
 436
 437         smlid = be16_to_cpu(pip->sm_lid);
 438         if (smlid != dev->sm_lid) {
 439                 /* Must be a valid unicast LID address. */
 440                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
 441                         goto err;
 442                 dev->sm_lid = smlid;
 443                 event.event = IB_EVENT_SM_CHANGE;
 444                 ib_dispatch_event(&event);
 445         }
 446
 447         /* Only 4x supported but allow 1x or 4x to be set (see 14.2.6.6). */
 448         lwe = pip->link_width_enabled;
 449         if ((lwe >= 4 && lwe <= 8) || (lwe >= 0xC && lwe <= 0xFE))
 450                 goto err;
 451         if (lwe == 0xFF)
 452                 dev->link_width_enabled = 3;    /* 1x or 4x */
 453         else if (lwe)
 454                 dev->link_width_enabled = lwe;
 455
 456         /* Only 2.5 Gbs supported. */
 457         lse = pip->linkspeedactive_enabled & 0xF;
 458         if (lse >= 2 && lse <= 0xE)
 459                 goto err;
 460
 461         /* Set link down default state. */
 462         switch (pip->portphysstate_linkdown & 0xF) {
 463         case 0: /* NOP */
 464                 break;
 465         case 1: /* SLEEP */
 466                 if (set_linkdowndefaultstate(dd, 1))
 467                         goto err;
 468                 break;
 469         case 2: /* POLL */
 470                 if (set_linkdowndefaultstate(dd, 0))
 471                         goto err;
 472                 break;
 473         default:
 474                 goto err;
 475         }
 476
 477         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
 478         dev->vl_high_limit = pip->vl_high_limit;
 479
 480         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
 481         case IB_MTU_256:
 482                 mtu = 256;
 483                 break;
 484         case IB_MTU_512:
 485                 mtu = 512;
 486                 break;
 487         case IB_MTU_1024:
 488                 mtu = 1024;
 489                 break;
 490         case IB_MTU_2048:
 491                 mtu = 2048;
 492                 break;
 493         case IB_MTU_4096:
 494                 mtu = 4096;
 495                 break;
 496         default:
 497                 /* XXX We have already partially updated our state! */
 498                 goto err;
 499         }
 500         ipath_set_mtu(dd, mtu);
 501
 502         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
 503
 504         /* We only support VL0 */
 505         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
 506                 goto err;
 507
 508         if (pip->mkey_violations == 0)
 509                 dev->mkey_violations = 0;
 510
 511         /*
 512          * Hardware counter can't be reset so snapshot and subtract
 513          * later.
 514          */
 515         if (pip->pkey_violations == 0)
 516                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
 517
 518         if (pip->qkey_violations == 0)
 519                 dev->qkey_violations = 0;
 520
 521         ore = pip->localphyerrors_overrunerrors;
 522         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
 523                 goto err;
 524
 525         if (set_overrunthreshold(dd, (ore & 0xF)))
 526                 goto err;
 527
 528         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 529
 530         if (pip->clientrereg_resv_subnetto & 0x80) {
 531                 clientrereg = 1;
 532                 event.event = IB_EVENT_CLIENT_REREGISTER;
 533                 ib_dispatch_event(&event);
 534         }
 535
 536         /*
 537          * Do the port state change now that the other link parameters
 538          * have been set.
 539          * Changing the port physical state only makes sense if the link
 540          * is down or is being set to down.
 541          */
 542         state = pip->linkspeed_portstate & 0xF;
 543         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 544         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
 545                 goto err;
 546
 547         /*
 548          * Only state changes of DOWN, ARM, and ACTIVE are valid
 549          * and must be in the correct state to take effect (see 7.2.6).
 550          */
 551         switch (state) {
 552         case IB_PORT_NOP:
 553                 if (lstate == 0)
 554                         break;
 555                 /* FALLTHROUGH */
 556         case IB_PORT_DOWN:
 557                 if (lstate == 0)
 558                         if (get_linkdowndefaultstate(dd))
 559                                 lstate = IPATH_IB_LINKDOWN_SLEEP;
 560                         else
 561                                 lstate = IPATH_IB_LINKDOWN;
 562                 else if (lstate == 1)
 563                         lstate = IPATH_IB_LINKDOWN_SLEEP;
 564                 else if (lstate == 2)
 565                         lstate = IPATH_IB_LINKDOWN;
 566                 else if (lstate == 3)
 567                         lstate = IPATH_IB_LINKDOWN_DISABLE;
 568                 else
 569                         goto err;
 570                 ipath_set_linkstate(dd, lstate);
 571                 break;
 572         case IB_PORT_ARMED:
 573                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
 574                 break;
 575         case IB_PORT_ACTIVE:
 576                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
 577                 break;
 578         default:
 579                 /* XXX We have already partially updated our state! */
 580                 goto err;
 581         }
 582
 583         ret = recv_subn_get_portinfo(smp, ibdev, port);
 584
 585         if (clientrereg)
 586                 pip->clientrereg_resv_subnetto |= 0x80;
 587
 588         goto done;
 589
 590 err:
 591         smp->status |= IB_SMP_INVALID_FIELD;
 592         ret = recv_subn_get_portinfo(smp, ibdev, port);
 593
 594 done:
 595         return ret;
 596 }
 597
 598 /**
 599  * rm_pkey - decrecment the reference count for the given PKEY
 600  * @dd: the infinipath device
 601  * @key: the PKEY index
 602  *
 603  * Return true if this was the last reference and the hardware table entry
 604  * needs to be changed.
 605  */
 606 static int rm_pkey(struct ipath_devdata *dd, u16 key)
 607 {
 608         int i;
 609         int ret;
 610
 611         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 612                 if (dd->ipath_pkeys[i] != key)
 613                         continue;
 614                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
 615                         dd->ipath_pkeys[i] = 0;
 616                         ret = 1;
 617                         goto bail;
 618                 }
 619                 break;
 620         }
 621
 622         ret = 0;
 623
 624 bail:
 625         return ret;
 626 }
 627
 628 /**
 629  * add_pkey - add the given PKEY to the hardware table
 630  * @dd: the infinipath device
 631  * @key: the PKEY
 632  *
 633  * Return an error code if unable to add the entry, zero if no change,
 634  * or 1 if the hardware PKEY register needs to be updated.
 635  */
 636 static int add_pkey(struct ipath_devdata *dd, u16 key)
 637 {
 638         int i;
 639         u16 lkey = key & 0x7FFF;
 640         int any = 0;
 641         int ret;
 642
 643         if (lkey == 0x7FFF) {
 644                 ret = 0;
 645                 goto bail;
 646         }
 647
 648         /* Look for an empty slot or a matching PKEY. */
 649         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 650                 if (!dd->ipath_pkeys[i]) {
 651                         any++;
 652                         continue;
 653                 }
 654                 /* If it matches exactly, try to increment the ref count */
 655                 if (dd->ipath_pkeys[i] == key) {
 656                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
 657                                 ret = 0;
 658                                 goto bail;
 659                         }
 660                         /* Lost the race. Look for an empty slot below. */
 661                         atomic_dec(&dd->ipath_pkeyrefs[i]);
 662                         any++;
 663                 }
 664                 /*
 665                  * It makes no sense to have both the limited and unlimited
 666                  * PKEY set at the same time since the unlimited one will
 667                  * disable the limited one.
 668                  */
 669                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
 670                         ret = -EEXIST;
 671                         goto bail;
 672                 }
 673         }
 674         if (!any) {
 675                 ret = -EBUSY;
 676                 goto bail;
 677         }
 678         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 679                 if (!dd->ipath_pkeys[i] &&
 680                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
 681                         /* for ipathstats, etc. */
 682                         ipath_stats.sps_pkeys[i] = lkey;
 683                         dd->ipath_pkeys[i] = key;
 684                         ret = 1;
 685                         goto bail;
 686                 }
 687         }
 688         ret = -EBUSY;
 689
 690 bail:
 691         return ret;
 692 }
 693
 694 /**
 695  * set_pkeys - set the PKEY table for port 0
 696  * @dd: the infinipath device
 697  * @pkeys: the PKEY table
 698  */
 699 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 700 {
 701         struct ipath_portdata *pd;
 702         int i;
 703         int changed = 0;
 704
 705         pd = dd->ipath_pd[0];
 706
 707         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
 708                 u16 key = pkeys[i];
 709                 u16 okey = pd->port_pkeys[i];
 710
 711                 if (key == okey)
 712                         continue;
 713                 /*
 714                  * The value of this PKEY table entry is changing.
 715                  * Remove the old entry in the hardware's array of PKEYs.
 716                  */
 717                 if (okey & 0x7FFF)
 718                         changed |= rm_pkey(dd, okey);
 719                 if (key & 0x7FFF) {
 720                         int ret = add_pkey(dd, key);
 721
 722                         if (ret < 0)
 723                                 key = 0;
 724                         else
 725                                 changed |= ret;
 726                 }
 727                 pd->port_pkeys[i] = key;
 728         }
 729         if (changed) {
 730                 u64 pkey;
 731
 732                 pkey = (u64) dd->ipath_pkeys[0] |
 733                         ((u64) dd->ipath_pkeys[1] << 16) |
 734                         ((u64) dd->ipath_pkeys[2] << 32) |
 735                         ((u64) dd->ipath_pkeys[3] << 48);
 736                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
 737                            (unsigned long long) pkey);
 738                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
 739                                  pkey);
 740         }
 741         return 0;
 742 }
 743
 744 static int recv_subn_set_pkeytable(struct ib_smp *smp,
 745                                    struct ib_device *ibdev)
 746 {
 747         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 748         __be16 *p = (__be16 *) smp->data;
 749         u16 *q = (u16 *) smp->data;
 750         struct ipath_ibdev *dev = to_idev(ibdev);
 751         unsigned i, n = ipath_get_npkeys(dev->dd);
 752
 753         for (i = 0; i < n; i++)
 754                 q[i] = be16_to_cpu(p[i]);
 755
 756         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
 757                 smp->status |= IB_SMP_INVALID_FIELD;
 758
 759         return recv_subn_get_pkeytable(smp, ibdev);
 760 }
 761
 762 #define IB_PMA_CLASS_PORT_INFO          __constant_htons(0x0001)
 763 #define IB_PMA_PORT_SAMPLES_CONTROL     __constant_htons(0x0010)
 764 #define IB_PMA_PORT_SAMPLES_RESULT      __constant_htons(0x0011)
 765 #define IB_PMA_PORT_COUNTERS            __constant_htons(0x0012)
 766 #define IB_PMA_PORT_COUNTERS_EXT        __constant_htons(0x001D)
 767 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  __constant_htons(0x001E)
 768
 769 struct ib_perf {
 770         u8 base_version;
 771         u8 mgmt_class;
 772         u8 class_version;
 773         u8 method;
 774         __be16 status;
 775         __be16 unused;
 776         __be64 tid;
 777         __be16 attr_id;
 778         __be16 resv;
 779         __be32 attr_mod;
 780         u8 reserved[40];
 781         u8 data[192];
 782 } __attribute__ ((packed));
 783
 784 struct ib_pma_classportinfo {
 785         u8 base_version;
 786         u8 class_version;
 787         __be16 cap_mask;
 788         u8 reserved[3];
 789         u8 resp_time_value;     /* only lower 5 bits */
 790         union ib_gid redirect_gid;
 791         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
 792         __be16 redirect_lid;
 793         __be16 redirect_pkey;
 794         __be32 redirect_qp;     /* only lower 24 bits */
 795         __be32 redirect_qkey;
 796         union ib_gid trap_gid;
 797         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
 798         __be16 trap_lid;
 799         __be16 trap_pkey;
 800         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
 801         __be32 trap_qkey;
 802 } __attribute__ ((packed));
 803
 804 struct ib_pma_portsamplescontrol {
 805         u8 opcode;
 806         u8 port_select;
 807         u8 tick;
 808         u8 counter_width;       /* only lower 3 bits */
 809         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
 810         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
 811         u8 sample_mechanisms;
 812         u8 sample_status;       /* only lower 2 bits */
 813         __be64 option_mask;
 814         __be64 vendor_mask;
 815         __be32 sample_start;
 816         __be32 sample_interval;
 817         __be16 tag;
 818         __be16 counter_select[15];
 819 } __attribute__ ((packed));
 820
 821 struct ib_pma_portsamplesresult {
 822         __be16 tag;
 823         __be16 sample_status;   /* only lower 2 bits */
 824         __be32 counter[15];
 825 } __attribute__ ((packed));
 826
 827 struct ib_pma_portsamplesresult_ext {
 828         __be16 tag;
 829         __be16 sample_status;   /* only lower 2 bits */
 830         __be32 extended_width;  /* only upper 2 bits */
 831         __be64 counter[15];
 832 } __attribute__ ((packed));
 833
 834 struct ib_pma_portcounters {
 835         u8 reserved;
 836         u8 port_select;
 837         __be16 counter_select;
 838         __be16 symbol_error_counter;
 839         u8 link_error_recovery_counter;
 840         u8 link_downed_counter;
 841         __be16 port_rcv_errors;
 842         __be16 port_rcv_remphys_errors;
 843         __be16 port_rcv_switch_relay_errors;
 844         __be16 port_xmit_discards;
 845         u8 port_xmit_constraint_errors;
 846         u8 port_rcv_constraint_errors;
 847         u8 reserved1;
 848         u8 lli_ebor_errors;     /* 4, 4, bits */
 849         __be16 reserved2;
 850         __be16 vl15_dropped;
 851         __be32 port_xmit_data;
 852         __be32 port_rcv_data;
 853         __be32 port_xmit_packets;
 854         __be32 port_rcv_packets;
 855 } __attribute__ ((packed));
 856
 857 #define IB_PMA_SEL_SYMBOL_ERROR                 __constant_htons(0x0001)
 858 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          __constant_htons(0x0002)
 859 #define IB_PMA_SEL_LINK_DOWNED                  __constant_htons(0x0004)
 860 #define IB_PMA_SEL_PORT_RCV_ERRORS              __constant_htons(0x0008)
 861 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      __constant_htons(0x0010)
 862 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           __constant_htons(0x0040)
 863 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  __constant_htons(0x0200)
 864 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    __constant_htons(0x0400)
 865 #define IB_PMA_SEL_PORT_VL15_DROPPED            __constant_htons(0x0800)
 866 #define IB_PMA_SEL_PORT_XMIT_DATA               __constant_htons(0x1000)
 867 #define IB_PMA_SEL_PORT_RCV_DATA                __constant_htons(0x2000)
 868 #define IB_PMA_SEL_PORT_XMIT_PACKETS            __constant_htons(0x4000)
 869 #define IB_PMA_SEL_PORT_RCV_PACKETS             __constant_htons(0x8000)
 870
 871 struct ib_pma_portcounters_ext {
 872         u8 reserved;
 873         u8 port_select;
 874         __be16 counter_select;
 875         __be32 reserved1;
 876         __be64 port_xmit_data;
 877         __be64 port_rcv_data;
 878         __be64 port_xmit_packets;
 879         __be64 port_rcv_packets;
 880         __be64 port_unicast_xmit_packets;
 881         __be64 port_unicast_rcv_packets;
 882         __be64 port_multicast_xmit_packets;
 883         __be64 port_multicast_rcv_packets;
 884 } __attribute__ ((packed));
 885
 886 #define IB_PMA_SELX_PORT_XMIT_DATA              __constant_htons(0x0001)
 887 #define IB_PMA_SELX_PORT_RCV_DATA               __constant_htons(0x0002)
 888 #define IB_PMA_SELX_PORT_XMIT_PACKETS           __constant_htons(0x0004)
 889 #define IB_PMA_SELX_PORT_RCV_PACKETS            __constant_htons(0x0008)
 890 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       __constant_htons(0x0010)
 891 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        __constant_htons(0x0020)
 892 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     __constant_htons(0x0040)
 893 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      __constant_htons(0x0080)
 894
 895 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
 896 {
 897         struct ib_pma_classportinfo *p =
 898                 (struct ib_pma_classportinfo *)pmp->data;
 899
 900         memset(pmp->data, 0, sizeof(pmp->data));
 901
 902         if (pmp->attr_mod != 0)
 903                 pmp->status |= IB_SMP_INVALID_FIELD;
 904
 905         /* Indicate AllPortSelect is valid (only one port anyway) */
 906         p->cap_mask = __constant_cpu_to_be16(1 << 8);
 907         p->base_version = 1;
 908         p->class_version = 1;
 909         /*
 910          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
 911          * sec.
 912          */
 913         p->resp_time_value = 18;
 914
 915         return reply((struct ib_smp *) pmp);
 916 }
 917
 918 /*
 919  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
 920  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
 921  * We support 5 counters which only count the mandatory quantities.
 922  */
 923 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
 924 #define COUNTER_MASK0_9 \
 925         __constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
 926                                COUNTER_MASK(1, 1) | \
 927                                COUNTER_MASK(1, 2) | \
 928                                COUNTER_MASK(1, 3) | \
 929                                COUNTER_MASK(1, 4))
 930
 931 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 932                                            struct ib_device *ibdev, u8 port)
 933 {
 934         struct ib_pma_portsamplescontrol *p =
 935                 (struct ib_pma_portsamplescontrol *)pmp->data;
 936         struct ipath_ibdev *dev = to_idev(ibdev);
 937         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 938         unsigned long flags;
 939         u8 port_select = p->port_select;
 940
 941         memset(pmp->data, 0, sizeof(pmp->data));
 942
 943         p->port_select = port_select;
 944         if (pmp->attr_mod != 0 ||
 945             (port_select != port && port_select != 0xFF))
 946                 pmp->status |= IB_SMP_INVALID_FIELD;
 947         /*
 948          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 949          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
 950          * intervals are counted in ticks.  Since we use Linux timers, that
 951          * count in jiffies, we can't sample for less than 1000 ticks if HZ
 952          * == 1000 (4000 ticks if HZ is 250).
 953          */
 954         /* XXX This is WRONG. */
 955         p->tick = 250;          /* 1 usec. */
 956         p->counter_width = 4;   /* 32 bit counters */
 957         p->counter_mask0_9 = COUNTER_MASK0_9;
 958         spin_lock_irqsave(&dev->pending_lock, flags);
 959         if (crp->cr_psstat)
 960                 p->sample_status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 961         else
 962                 p->sample_status = dev->pma_sample_status;
 963         p->sample_start = cpu_to_be32(dev->pma_sample_start);
 964         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
 965         p->tag = cpu_to_be16(dev->pma_tag);
 966         p->counter_select[0] = dev->pma_counter_select[0];
 967         p->counter_select[1] = dev->pma_counter_select[1];
 968         p->counter_select[2] = dev->pma_counter_select[2];
 969         p->counter_select[3] = dev->pma_counter_select[3];
 970         p->counter_select[4] = dev->pma_counter_select[4];
 971         spin_unlock_irqrestore(&dev->pending_lock, flags);
 972
 973         return reply((struct ib_smp *) pmp);
 974 }
 975
 976 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
 977                                            struct ib_device *ibdev, u8 port)
 978 {
 979         struct ib_pma_portsamplescontrol *p =
 980                 (struct ib_pma_portsamplescontrol *)pmp->data;
 981         struct ipath_ibdev *dev = to_idev(ibdev);
 982         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 983         unsigned long flags;
 984         u8 status;
 985         int ret;
 986
 987         if (pmp->attr_mod != 0 ||
 988             (p->port_select != port && p->port_select != 0xFF)) {
 989                 pmp->status |= IB_SMP_INVALID_FIELD;
 990                 ret = reply((struct ib_smp *) pmp);
 991                 goto bail;
 992         }
 993
 994         spin_lock_irqsave(&dev->pending_lock, flags);
 995         if (crp->cr_psstat)
 996                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 997         else
 998                 status = dev->pma_sample_status;
 999         if (status == IB_PMA_SAMPLE_STATUS_DONE) {
1000                 dev->pma_sample_start = be32_to_cpu(p->sample_start);
1001                 dev->pma_sample_interval = be32_to_cpu(p->sample_interval);
1002                 dev->pma_tag = be16_to_cpu(p->tag);
1003                 dev->pma_counter_select[0] = p->counter_select[0];
1004                 dev->pma_counter_select[1] = p->counter_select[1];
1005                 dev->pma_counter_select[2] = p->counter_select[2];
1006                 dev->pma_counter_select[3] = p->counter_select[3];
1007                 dev->pma_counter_select[4] = p->counter_select[4];
1008                 if (crp->cr_psstat) {
1009                         ipath_write_creg(dev->dd, crp->cr_psinterval,
1010                                          dev->pma_sample_interval);
1011                         ipath_write_creg(dev->dd, crp->cr_psstart,
1012                                          dev->pma_sample_start);
1013                 } else
1014                         dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
1015         }
1016         spin_unlock_irqrestore(&dev->pending_lock, flags);
1017
1018         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1019
1020 bail:
1021         return ret;
1022 }
1023
1024 static u64 get_counter(struct ipath_ibdev *dev,
1025                        struct ipath_cregs const *crp,
1026                        __be16 sel)
1027 {
1028         u64 ret;
1029
1030         switch (sel) {
1031         case IB_PMA_PORT_XMIT_DATA:
1032                 ret = (crp->cr_psxmitdatacount) ?
1033                         ipath_read_creg32(dev->dd, crp->cr_psxmitdatacount) :
1034                         dev->ipath_sword;
1035                 break;
1036         case IB_PMA_PORT_RCV_DATA:
1037                 ret = (crp->cr_psrcvdatacount) ?
1038                         ipath_read_creg32(dev->dd, crp->cr_psrcvdatacount) :
1039                         dev->ipath_rword;
1040                 break;
1041         case IB_PMA_PORT_XMIT_PKTS:
1042                 ret = (crp->cr_psxmitpktscount) ?
1043                         ipath_read_creg32(dev->dd, crp->cr_psxmitpktscount) :
1044                         dev->ipath_spkts;
1045                 break;
1046         case IB_PMA_PORT_RCV_PKTS:
1047                 ret = (crp->cr_psrcvpktscount) ?
1048                         ipath_read_creg32(dev->dd, crp->cr_psrcvpktscount) :
1049                         dev->ipath_rpkts;
1050                 break;
1051         case IB_PMA_PORT_XMIT_WAIT:
1052                 ret = (crp->cr_psxmitwaitcount) ?
1053                         ipath_read_creg32(dev->dd, crp->cr_psxmitwaitcount) :
1054                         dev->ipath_xmit_wait;
1055                 break;
1056         default:
1057                 ret = 0;
1058         }
1059
1060         return ret;
1061 }
1062
1063 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1064                                           struct ib_device *ibdev)
1065 {
1066         struct ib_pma_portsamplesresult *p =
1067                 (struct ib_pma_portsamplesresult *)pmp->data;
1068         struct ipath_ibdev *dev = to_idev(ibdev);
1069         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1070         u8 status;
1071         int i;
1072
1073         memset(pmp->data, 0, sizeof(pmp->data));
1074         p->tag = cpu_to_be16(dev->pma_tag);
1075         if (crp->cr_psstat)
1076                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1077         else
1078                 status = dev->pma_sample_status;
1079         p->sample_status = cpu_to_be16(status);
1080         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1081                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1082                     cpu_to_be32(
1083                         get_counter(dev, crp, dev->pma_counter_select[i]));
1084
1085         return reply((struct ib_smp *) pmp);
1086 }
1087
1088 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1089                                               struct ib_device *ibdev)
1090 {
1091         struct ib_pma_portsamplesresult_ext *p =
1092                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1093         struct ipath_ibdev *dev = to_idev(ibdev);
1094         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1095         u8 status;
1096         int i;
1097
1098         memset(pmp->data, 0, sizeof(pmp->data));
1099         p->tag = cpu_to_be16(dev->pma_tag);
1100         if (crp->cr_psstat)
1101                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1102         else
1103                 status = dev->pma_sample_status;
1104         p->sample_status = cpu_to_be16(status);
1105         /* 64 bits */
1106         p->extended_width = __constant_cpu_to_be32(0x80000000);
1107         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1108                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1109                     cpu_to_be64(
1110                         get_counter(dev, crp, dev->pma_counter_select[i]));
1111
1112         return reply((struct ib_smp *) pmp);
1113 }
1114
1115 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1116                                      struct ib_device *ibdev, u8 port)
1117 {
1118         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1119                 pmp->data;
1120         struct ipath_ibdev *dev = to_idev(ibdev);
1121         struct ipath_verbs_counters cntrs;
1122         u8 port_select = p->port_select;
1123
1124         ipath_get_counters(dev->dd, &cntrs);
1125
1126         /* Adjust counters for any resets done. */
1127         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1128         cntrs.link_error_recovery_counter -=
1129                 dev->z_link_error_recovery_counter;
1130         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1131         cntrs.port_rcv_errors += dev->rcv_errors;
1132         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1133         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1134         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1135         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1136         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1137         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1138         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1139         cntrs.local_link_integrity_errors -=
1140                 dev->z_local_link_integrity_errors;
1141         cntrs.excessive_buffer_overrun_errors -=
1142                 dev->z_excessive_buffer_overrun_errors;
1143         cntrs.vl15_dropped -= dev->z_vl15_dropped;
1144         cntrs.vl15_dropped += dev->n_vl15_dropped;
1145
1146         memset(pmp->data, 0, sizeof(pmp->data));
1147
1148         p->port_select = port_select;
1149         if (pmp->attr_mod != 0 ||
1150             (port_select != port && port_select != 0xFF))
1151                 pmp->status |= IB_SMP_INVALID_FIELD;
1152
1153         if (cntrs.symbol_error_counter > 0xFFFFUL)
1154                 p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
1155         else
1156                 p->symbol_error_counter =
1157                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1158         if (cntrs.link_error_recovery_counter > 0xFFUL)
1159                 p->link_error_recovery_counter = 0xFF;
1160         else
1161                 p->link_error_recovery_counter =
1162                         (u8)cntrs.link_error_recovery_counter;
1163         if (cntrs.link_downed_counter > 0xFFUL)
1164                 p->link_downed_counter = 0xFF;
1165         else
1166                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1167         if (cntrs.port_rcv_errors > 0xFFFFUL)
1168                 p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
1169         else
1170                 p->port_rcv_errors =
1171                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1172         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1173                 p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
1174         else
1175                 p->port_rcv_remphys_errors =
1176                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1177         if (cntrs.port_xmit_discards > 0xFFFFUL)
1178                 p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
1179         else
1180                 p->port_xmit_discards =
1181                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1182         if (cntrs.local_link_integrity_errors > 0xFUL)
1183                 cntrs.local_link_integrity_errors = 0xFUL;
1184         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1185                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1186         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1187                 cntrs.excessive_buffer_overrun_errors;
1188         if (cntrs.vl15_dropped > 0xFFFFUL)
1189                 p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
1190         else
1191                 p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1192         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1193                 p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
1194         else
1195                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1196         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1197                 p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
1198         else
1199                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1200         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1201                 p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1202         else
1203                 p->port_xmit_packets =
1204                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1205         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1206                 p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
1207         else
1208                 p->port_rcv_packets =
1209                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1210
1211         return reply((struct ib_smp *) pmp);
1212 }
1213
1214 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1215                                          struct ib_device *ibdev, u8 port)
1216 {
1217         struct ib_pma_portcounters_ext *p =
1218                 (struct ib_pma_portcounters_ext *)pmp->data;
1219         struct ipath_ibdev *dev = to_idev(ibdev);
1220         u64 swords, rwords, spkts, rpkts, xwait;
1221         u8 port_select = p->port_select;
1222
1223         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1224                                 &rpkts, &xwait);
1225
1226         /* Adjust counters for any resets done. */
1227         swords -= dev->z_port_xmit_data;
1228         rwords -= dev->z_port_rcv_data;
1229         spkts -= dev->z_port_xmit_packets;
1230         rpkts -= dev->z_port_rcv_packets;
1231
1232         memset(pmp->data, 0, sizeof(pmp->data));
1233
1234         p->port_select = port_select;
1235         if (pmp->attr_mod != 0 ||
1236             (port_select != port && port_select != 0xFF))
1237                 pmp->status |= IB_SMP_INVALID_FIELD;
1238
1239         p->port_xmit_data = cpu_to_be64(swords);
1240         p->port_rcv_data = cpu_to_be64(rwords);
1241         p->port_xmit_packets = cpu_to_be64(spkts);
1242         p->port_rcv_packets = cpu_to_be64(rpkts);
1243         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1244         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1245         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1246         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1247
1248         return reply((struct ib_smp *) pmp);
1249 }
1250
1251 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1252                                      struct ib_device *ibdev, u8 port)
1253 {
1254         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1255                 pmp->data;
1256         struct ipath_ibdev *dev = to_idev(ibdev);
1257         struct ipath_verbs_counters cntrs;
1258
1259         /*
1260          * Since the HW doesn't support clearing counters, we save the
1261          * current count and subtract it from future responses.
1262          */
1263         ipath_get_counters(dev->dd, &cntrs);
1264
1265         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1266                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1267
1268         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1269                 dev->z_link_error_recovery_counter =
1270                         cntrs.link_error_recovery_counter;
1271
1272         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1273                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1274
1275         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1276                 dev->z_port_rcv_errors =
1277                         cntrs.port_rcv_errors + dev->rcv_errors;
1278
1279         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1280                 dev->z_port_rcv_remphys_errors =
1281                         cntrs.port_rcv_remphys_errors;
1282
1283         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1284                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1285
1286         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1287                 dev->z_local_link_integrity_errors =
1288                         cntrs.local_link_integrity_errors;
1289
1290         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1291                 dev->z_excessive_buffer_overrun_errors =
1292                         cntrs.excessive_buffer_overrun_errors;
1293
1294         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1295                 dev->n_vl15_dropped = 0;
1296                 dev->z_vl15_dropped = cntrs.vl15_dropped;
1297         }
1298
1299         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1300                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1301
1302         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1303                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1304
1305         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1306                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1307
1308         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1309                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1310
1311         return recv_pma_get_portcounters(pmp, ibdev, port);
1312 }
1313
1314 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1315                                          struct ib_device *ibdev, u8 port)
1316 {
1317         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1318                 pmp->data;
1319         struct ipath_ibdev *dev = to_idev(ibdev);
1320         u64 swords, rwords, spkts, rpkts, xwait;
1321
1322         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1323                                 &rpkts, &xwait);
1324
1325         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1326                 dev->z_port_xmit_data = swords;
1327
1328         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1329                 dev->z_port_rcv_data = rwords;
1330
1331         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1332                 dev->z_port_xmit_packets = spkts;
1333
1334         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1335                 dev->z_port_rcv_packets = rpkts;
1336
1337         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1338                 dev->n_unicast_xmit = 0;
1339
1340         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1341                 dev->n_unicast_rcv = 0;
1342
1343         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1344                 dev->n_multicast_xmit = 0;
1345
1346         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1347                 dev->n_multicast_rcv = 0;
1348
1349         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1350 }
1351
1352 static int process_subn(struct ib_device *ibdev, int mad_flags,
1353                         u8 port_num, struct ib_mad *in_mad,
1354                         struct ib_mad *out_mad)
1355 {
1356         struct ib_smp *smp = (struct ib_smp *)out_mad;
1357         struct ipath_ibdev *dev = to_idev(ibdev);
1358         int ret;
1359
1360         *out_mad = *in_mad;
1361         if (smp->class_version != 1) {
1362                 smp->status |= IB_SMP_UNSUP_VERSION;
1363                 ret = reply(smp);
1364                 goto bail;
1365         }
1366
1367         /* Is the mkey in the process of expiring? */
1368         if (dev->mkey_lease_timeout && jiffies >= dev->mkey_lease_timeout) {
1369                 /* Clear timeout and mkey protection field. */
1370                 dev->mkey_lease_timeout = 0;
1371                 dev->mkeyprot = 0;
1372         }
1373
1374         /*
1375          * M_Key checking depends on
1376          * Portinfo:M_Key_protect_bits
1377          */
1378         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1379             dev->mkey != smp->mkey &&
1380             (smp->method == IB_MGMT_METHOD_SET ||
1381              (smp->method == IB_MGMT_METHOD_GET &&
1382               dev->mkeyprot >= 2))) {
1383                 if (dev->mkey_violations != 0xFFFF)
1384                         ++dev->mkey_violations;
1385                 if (dev->mkey_lease_timeout ||
1386                     dev->mkey_lease_period == 0) {
1387                         ret = IB_MAD_RESULT_SUCCESS |
1388                                 IB_MAD_RESULT_CONSUMED;
1389                         goto bail;
1390                 }
1391                 dev->mkey_lease_timeout = jiffies +
1392                         dev->mkey_lease_period * HZ;
1393                 /* Future: Generate a trap notice. */
1394                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1395                 goto bail;
1396         } else if (dev->mkey_lease_timeout)
1397                 dev->mkey_lease_timeout = 0;
1398
1399         switch (smp->method) {
1400         case IB_MGMT_METHOD_GET:
1401                 switch (smp->attr_id) {
1402                 case IB_SMP_ATTR_NODE_DESC:
1403                         ret = recv_subn_get_nodedescription(smp, ibdev);
1404                         goto bail;
1405                 case IB_SMP_ATTR_NODE_INFO:
1406                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1407                         goto bail;
1408                 case IB_SMP_ATTR_GUID_INFO:
1409                         ret = recv_subn_get_guidinfo(smp, ibdev);
1410                         goto bail;
1411                 case IB_SMP_ATTR_PORT_INFO:
1412                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1413                         goto bail;
1414                 case IB_SMP_ATTR_PKEY_TABLE:
1415                         ret = recv_subn_get_pkeytable(smp, ibdev);
1416                         goto bail;
1417                 case IB_SMP_ATTR_SM_INFO:
1418                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1419                                 ret = IB_MAD_RESULT_SUCCESS |
1420                                         IB_MAD_RESULT_CONSUMED;
1421                                 goto bail;
1422                         }
1423                         if (dev->port_cap_flags & IB_PORT_SM) {
1424                                 ret = IB_MAD_RESULT_SUCCESS;
1425                                 goto bail;
1426                         }
1427                         /* FALLTHROUGH */
1428                 default:
1429                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1430                         ret = reply(smp);
1431                         goto bail;
1432                 }
1433
1434         case IB_MGMT_METHOD_SET:
1435                 switch (smp->attr_id) {
1436                 case IB_SMP_ATTR_GUID_INFO:
1437                         ret = recv_subn_set_guidinfo(smp, ibdev);
1438                         goto bail;
1439                 case IB_SMP_ATTR_PORT_INFO:
1440                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1441                         goto bail;
1442                 case IB_SMP_ATTR_PKEY_TABLE:
1443                         ret = recv_subn_set_pkeytable(smp, ibdev);
1444                         goto bail;
1445                 case IB_SMP_ATTR_SM_INFO:
1446                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1447                                 ret = IB_MAD_RESULT_SUCCESS |
1448                                         IB_MAD_RESULT_CONSUMED;
1449                                 goto bail;
1450                         }
1451                         if (dev->port_cap_flags & IB_PORT_SM) {
1452                                 ret = IB_MAD_RESULT_SUCCESS;
1453                                 goto bail;
1454                         }
1455                         /* FALLTHROUGH */
1456                 default:
1457                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1458                         ret = reply(smp);
1459                         goto bail;
1460                 }
1461
1462         case IB_MGMT_METHOD_GET_RESP:
1463                 /*
1464                  * The ib_mad module will call us to process responses
1465                  * before checking for other consumers.
1466                  * Just tell the caller to process it normally.
1467                  */
1468                 ret = IB_MAD_RESULT_SUCCESS;
1469                 goto bail;
1470         default:
1471                 smp->status |= IB_SMP_UNSUP_METHOD;
1472                 ret = reply(smp);
1473         }
1474
1475 bail:
1476         return ret;
1477 }
1478
1479 static int process_perf(struct ib_device *ibdev, u8 port_num,
1480                         struct ib_mad *in_mad,
1481                         struct ib_mad *out_mad)
1482 {
1483         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1484         int ret;
1485
1486         *out_mad = *in_mad;
1487         if (pmp->class_version != 1) {
1488                 pmp->status |= IB_SMP_UNSUP_VERSION;
1489                 ret = reply((struct ib_smp *) pmp);
1490                 goto bail;
1491         }
1492
1493         switch (pmp->method) {
1494         case IB_MGMT_METHOD_GET:
1495                 switch (pmp->attr_id) {
1496                 case IB_PMA_CLASS_PORT_INFO:
1497                         ret = recv_pma_get_classportinfo(pmp);
1498                         goto bail;
1499                 case IB_PMA_PORT_SAMPLES_CONTROL:
1500                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1501                                                               port_num);
1502                         goto bail;
1503                 case IB_PMA_PORT_SAMPLES_RESULT:
1504                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1505                         goto bail;
1506                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1507                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1508                                                                  ibdev);
1509                         goto bail;
1510                 case IB_PMA_PORT_COUNTERS:
1511                         ret = recv_pma_get_portcounters(pmp, ibdev,
1512                                                         port_num);
1513                         goto bail;
1514                 case IB_PMA_PORT_COUNTERS_EXT:
1515                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1516                                                             port_num);
1517                         goto bail;
1518                 default:
1519                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1520                         ret = reply((struct ib_smp *) pmp);
1521                         goto bail;
1522                 }
1523
1524         case IB_MGMT_METHOD_SET:
1525                 switch (pmp->attr_id) {
1526                 case IB_PMA_PORT_SAMPLES_CONTROL:
1527                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1528                                                               port_num);
1529                         goto bail;
1530                 case IB_PMA_PORT_COUNTERS:
1531                         ret = recv_pma_set_portcounters(pmp, ibdev,
1532                                                         port_num);
1533                         goto bail;
1534                 case IB_PMA_PORT_COUNTERS_EXT:
1535                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1536                                                             port_num);
1537                         goto bail;
1538                 default:
1539                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1540                         ret = reply((struct ib_smp *) pmp);
1541                         goto bail;
1542                 }
1543
1544         case IB_MGMT_METHOD_GET_RESP:
1545                 /*
1546                  * The ib_mad module will call us to process responses
1547                  * before checking for other consumers.
1548                  * Just tell the caller to process it normally.
1549                  */
1550                 ret = IB_MAD_RESULT_SUCCESS;
1551                 goto bail;
1552         default:
1553                 pmp->status |= IB_SMP_UNSUP_METHOD;
1554                 ret = reply((struct ib_smp *) pmp);
1555         }
1556
1557 bail:
1558         return ret;
1559 }
1560
1561 /**
1562  * ipath_process_mad - process an incoming MAD packet
1563  * @ibdev: the infiniband device this packet came in on
1564  * @mad_flags: MAD flags
1565  * @port_num: the port number this packet came in on
1566  * @in_wc: the work completion entry for this packet
1567  * @in_grh: the global route header for this packet
1568  * @in_mad: the incoming MAD
1569  * @out_mad: any outgoing MAD reply
1570  *
1571  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1572  * interested in processing.
1573  *
1574  * Note that the verbs framework has already done the MAD sanity checks,
1575  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1576  * MADs.
1577  *
1578  * This is called by the ib_mad module.
1579  */
1580 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1581                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1582                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1583 {
1584         int ret;
1585
1586         switch (in_mad->mad_hdr.mgmt_class) {
1587         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1588         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1589                 ret = process_subn(ibdev, mad_flags, port_num,
1590                                    in_mad, out_mad);
1591                 goto bail;
1592         case IB_MGMT_CLASS_PERF_MGMT:
1593                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1594                 goto bail;
1595         default:
1596                 ret = IB_MAD_RESULT_SUCCESS;
1597         }
1598
1599 bail:
1600         return ret;
1601 }