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