git.oblomov.eu Git - linux-2.6/blob - drivers/infiniband/hw/ipath/ipath_mad.c

   1 /*
   2  * Copyright (c) 2006, 2007, 2008 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    cpu_to_be16(0x0004)
  41 #define IB_SMP_UNSUP_METHOD     cpu_to_be16(0x0008)
  42 #define IB_SMP_UNSUP_METH_ATTR  cpu_to_be16(0x000C)
  43 #define IB_SMP_INVALID_FIELD    cpu_to_be16(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] = IPATH_SRC_OUI_1;
 115         nip->vendor_id[1] = IPATH_SRC_OUI_2;
 116         nip->vendor_id[2] = IPATH_SRC_OUI_3;
 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 static void set_link_width_enabled(struct ipath_devdata *dd, u32 w)
 150 {
 151         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LWID_ENB, w);
 152 }
 153
 154 static void set_link_speed_enabled(struct ipath_devdata *dd, u32 s)
 155 {
 156         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_SPD_ENB, s);
 157 }
 158
 159 static int get_overrunthreshold(struct ipath_devdata *dd)
 160 {
 161         return (dd->ipath_ibcctrl >>
 162                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 163                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 164 }
 165
 166 /**
 167  * set_overrunthreshold - set the overrun threshold
 168  * @dd: the infinipath device
 169  * @n: the new threshold
 170  *
 171  * Note that this will only take effect when the link state changes.
 172  */
 173 static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
 174 {
 175         unsigned v;
 176
 177         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 178                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 179         if (v != n) {
 180                 dd->ipath_ibcctrl &=
 181                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
 182                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
 183                 dd->ipath_ibcctrl |=
 184                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
 185                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 186                                  dd->ipath_ibcctrl);
 187         }
 188         return 0;
 189 }
 190
 191 static int get_phyerrthreshold(struct ipath_devdata *dd)
 192 {
 193         return (dd->ipath_ibcctrl >>
 194                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 195                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 196 }
 197
 198 /**
 199  * set_phyerrthreshold - set the physical error threshold
 200  * @dd: the infinipath device
 201  * @n: the new threshold
 202  *
 203  * Note that this will only take effect when the link state changes.
 204  */
 205 static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
 206 {
 207         unsigned v;
 208
 209         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 210                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 211         if (v != n) {
 212                 dd->ipath_ibcctrl &=
 213                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
 214                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
 215                 dd->ipath_ibcctrl |=
 216                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
 217                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 218                                  dd->ipath_ibcctrl);
 219         }
 220         return 0;
 221 }
 222
 223 /**
 224  * get_linkdowndefaultstate - get the default linkdown state
 225  * @dd: the infinipath device
 226  *
 227  * Returns zero if the default is POLL, 1 if the default is SLEEP.
 228  */
 229 static int get_linkdowndefaultstate(struct ipath_devdata *dd)
 230 {
 231         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
 232 }
 233
 234 static int recv_subn_get_portinfo(struct ib_smp *smp,
 235                                   struct ib_device *ibdev, u8 port)
 236 {
 237         struct ipath_ibdev *dev;
 238         struct ipath_devdata *dd;
 239         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 240         u16 lid;
 241         u8 ibcstat;
 242         u8 mtu;
 243         int ret;
 244
 245         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) {
 246                 smp->status |= IB_SMP_INVALID_FIELD;
 247                 ret = reply(smp);
 248                 goto bail;
 249         }
 250
 251         dev = to_idev(ibdev);
 252         dd = dev->dd;
 253
 254         /* Clear all fields.  Only set the non-zero fields. */
 255         memset(smp->data, 0, sizeof(smp->data));
 256
 257         /* Only return the mkey if the protection field allows it. */
 258         if (smp->method == IB_MGMT_METHOD_SET || dev->mkey == smp->mkey ||
 259             dev->mkeyprot == 0)
 260                 pip->mkey = dev->mkey;
 261         pip->gid_prefix = dev->gid_prefix;
 262         lid = dd->ipath_lid;
 263         pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
 264         pip->sm_lid = cpu_to_be16(dev->sm_lid);
 265         pip->cap_mask = cpu_to_be32(dev->port_cap_flags);
 266         /* pip->diag_code; */
 267         pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period);
 268         pip->local_port_num = port;
 269         pip->link_width_enabled = dd->ipath_link_width_enabled;
 270         pip->link_width_supported = dd->ipath_link_width_supported;
 271         pip->link_width_active = dd->ipath_link_width_active;
 272         pip->linkspeed_portstate = dd->ipath_link_speed_supported << 4;
 273         ibcstat = dd->ipath_lastibcstat;
 274         /* map LinkState to IB portinfo values.  */
 275         pip->linkspeed_portstate |= ipath_ib_linkstate(dd, ibcstat) + 1;
 276
 277         pip->portphysstate_linkdown =
 278                 (ipath_cvt_physportstate[ibcstat & dd->ibcs_lts_mask] << 4) |
 279                 (get_linkdowndefaultstate(dd) ? 1 : 2);
 280         pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dd->ipath_lmc;
 281         pip->linkspeedactive_enabled = (dd->ipath_link_speed_active << 4) |
 282                 dd->ipath_link_speed_enabled;
 283         switch (dd->ipath_ibmtu) {
 284         case 4096:
 285                 mtu = IB_MTU_4096;
 286                 break;
 287         case 2048:
 288                 mtu = IB_MTU_2048;
 289                 break;
 290         case 1024:
 291                 mtu = IB_MTU_1024;
 292                 break;
 293         case 512:
 294                 mtu = IB_MTU_512;
 295                 break;
 296         case 256:
 297                 mtu = IB_MTU_256;
 298                 break;
 299         default:                /* oops, something is wrong */
 300                 mtu = IB_MTU_2048;
 301                 break;
 302         }
 303         pip->neighbormtu_mastersmsl = (mtu << 4) | dev->sm_sl;
 304         pip->vlcap_inittype = 0x10;     /* VLCap = VL0, InitType = 0 */
 305         pip->vl_high_limit = dev->vl_high_limit;
 306         /* pip->vl_arb_high_cap; // only one VL */
 307         /* pip->vl_arb_low_cap; // only one VL */
 308         /* InitTypeReply = 0 */
 309         /* our mtu cap depends on whether 4K MTU enabled or not */
 310         pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
 311         /* HCAs ignore VLStallCount and HOQLife */
 312         /* pip->vlstallcnt_hoqlife; */
 313         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
 314         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
 315         /* P_KeyViolations are counted by hardware. */
 316         pip->pkey_violations =
 317                 cpu_to_be16((ipath_get_cr_errpkey(dd) -
 318                              dev->z_pkey_violations) & 0xFFFF);
 319         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
 320         /* Only the hardware GUID is supported for now */
 321         pip->guid_cap = 1;
 322         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
 323         /* 32.768 usec. response time (guessing) */
 324         pip->resv_resptimevalue = 3;
 325         pip->localphyerrors_overrunerrors =
 326                 (get_phyerrthreshold(dd) << 4) |
 327                 get_overrunthreshold(dd);
 328         /* pip->max_credit_hint; */
 329         if (dev->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
 330                 u32 v;
 331
 332                 v = dd->ipath_f_get_ib_cfg(dd, IPATH_IB_CFG_LINKLATENCY);
 333                 pip->link_roundtrip_latency[0] = v >> 16;
 334                 pip->link_roundtrip_latency[1] = v >> 8;
 335                 pip->link_roundtrip_latency[2] = v;
 336         }
 337
 338         ret = reply(smp);
 339
 340 bail:
 341         return ret;
 342 }
 343
 344 /**
 345  * get_pkeys - return the PKEY table for port 0
 346  * @dd: the infinipath device
 347  * @pkeys: the pkey table is placed here
 348  */
 349 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
 350 {
 351         /* always a kernel port, no locking needed */
 352         struct ipath_portdata *pd = dd->ipath_pd[0];
 353
 354         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
 355
 356         return 0;
 357 }
 358
 359 static int recv_subn_get_pkeytable(struct ib_smp *smp,
 360                                    struct ib_device *ibdev)
 361 {
 362         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 363         u16 *p = (u16 *) smp->data;
 364         __be16 *q = (__be16 *) smp->data;
 365
 366         /* 64 blocks of 32 16-bit P_Key entries */
 367
 368         memset(smp->data, 0, sizeof(smp->data));
 369         if (startpx == 0) {
 370                 struct ipath_ibdev *dev = to_idev(ibdev);
 371                 unsigned i, n = ipath_get_npkeys(dev->dd);
 372
 373                 get_pkeys(dev->dd, p);
 374
 375                 for (i = 0; i < n; i++)
 376                         q[i] = cpu_to_be16(p[i]);
 377         } else
 378                 smp->status |= IB_SMP_INVALID_FIELD;
 379
 380         return reply(smp);
 381 }
 382
 383 static int recv_subn_set_guidinfo(struct ib_smp *smp,
 384                                   struct ib_device *ibdev)
 385 {
 386         /* The only GUID we support is the first read-only entry. */
 387         return recv_subn_get_guidinfo(smp, ibdev);
 388 }
 389
 390 /**
 391  * set_linkdowndefaultstate - set the default linkdown state
 392  * @dd: the infinipath device
 393  * @sleep: the new state
 394  *
 395  * Note that this will only take effect when the link state changes.
 396  */
 397 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
 398 {
 399         if (sleep)
 400                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 401         else
 402                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 403         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 404                          dd->ipath_ibcctrl);
 405         return 0;
 406 }
 407
 408 /**
 409  * recv_subn_set_portinfo - set port information
 410  * @smp: the incoming SM packet
 411  * @ibdev: the infiniband device
 412  * @port: the port on the device
 413  *
 414  * Set Portinfo (see ch. 14.2.5.6).
 415  */
 416 static int recv_subn_set_portinfo(struct ib_smp *smp,
 417                                   struct ib_device *ibdev, u8 port)
 418 {
 419         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 420         struct ib_event event;
 421         struct ipath_ibdev *dev;
 422         struct ipath_devdata *dd;
 423         char clientrereg = 0;
 424         u16 lid, smlid;
 425         u8 lwe;
 426         u8 lse;
 427         u8 state;
 428         u16 lstate;
 429         u32 mtu;
 430         int ret, ore;
 431
 432         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
 433                 goto err;
 434
 435         dev = to_idev(ibdev);
 436         dd = dev->dd;
 437         event.device = ibdev;
 438         event.element.port_num = port;
 439
 440         dev->mkey = pip->mkey;
 441         dev->gid_prefix = pip->gid_prefix;
 442         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
 443
 444         lid = be16_to_cpu(pip->lid);
 445         if (dd->ipath_lid != lid ||
 446             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
 447                 /* Must be a valid unicast LID address. */
 448                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
 449                         goto err;
 450                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
 451                 event.event = IB_EVENT_LID_CHANGE;
 452                 ib_dispatch_event(&event);
 453         }
 454
 455         smlid = be16_to_cpu(pip->sm_lid);
 456         if (smlid != dev->sm_lid) {
 457                 /* Must be a valid unicast LID address. */
 458                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
 459                         goto err;
 460                 dev->sm_lid = smlid;
 461                 event.event = IB_EVENT_SM_CHANGE;
 462                 ib_dispatch_event(&event);
 463         }
 464
 465         /* Allow 1x or 4x to be set (see 14.2.6.6). */
 466         lwe = pip->link_width_enabled;
 467         if (lwe) {
 468                 if (lwe == 0xFF)
 469                         lwe = dd->ipath_link_width_supported;
 470                 else if (lwe >= 16 || (lwe & ~dd->ipath_link_width_supported))
 471                         goto err;
 472                 set_link_width_enabled(dd, lwe);
 473         }
 474
 475         /* Allow 2.5 or 5.0 Gbs. */
 476         lse = pip->linkspeedactive_enabled & 0xF;
 477         if (lse) {
 478                 if (lse == 15)
 479                         lse = dd->ipath_link_speed_supported;
 480                 else if (lse >= 8 || (lse & ~dd->ipath_link_speed_supported))
 481                         goto err;
 482                 set_link_speed_enabled(dd, lse);
 483         }
 484
 485         /* Set link down default state. */
 486         switch (pip->portphysstate_linkdown & 0xF) {
 487         case 0: /* NOP */
 488                 break;
 489         case 1: /* SLEEP */
 490                 if (set_linkdowndefaultstate(dd, 1))
 491                         goto err;
 492                 break;
 493         case 2: /* POLL */
 494                 if (set_linkdowndefaultstate(dd, 0))
 495                         goto err;
 496                 break;
 497         default:
 498                 goto err;
 499         }
 500
 501         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
 502         dev->vl_high_limit = pip->vl_high_limit;
 503
 504         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
 505         case IB_MTU_256:
 506                 mtu = 256;
 507                 break;
 508         case IB_MTU_512:
 509                 mtu = 512;
 510                 break;
 511         case IB_MTU_1024:
 512                 mtu = 1024;
 513                 break;
 514         case IB_MTU_2048:
 515                 mtu = 2048;
 516                 break;
 517         case IB_MTU_4096:
 518                 if (!ipath_mtu4096)
 519                         goto err;
 520                 mtu = 4096;
 521                 break;
 522         default:
 523                 /* XXX We have already partially updated our state! */
 524                 goto err;
 525         }
 526         ipath_set_mtu(dd, mtu);
 527
 528         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
 529
 530         /* We only support VL0 */
 531         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
 532                 goto err;
 533
 534         if (pip->mkey_violations == 0)
 535                 dev->mkey_violations = 0;
 536
 537         /*
 538          * Hardware counter can't be reset so snapshot and subtract
 539          * later.
 540          */
 541         if (pip->pkey_violations == 0)
 542                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
 543
 544         if (pip->qkey_violations == 0)
 545                 dev->qkey_violations = 0;
 546
 547         ore = pip->localphyerrors_overrunerrors;
 548         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
 549                 goto err;
 550
 551         if (set_overrunthreshold(dd, (ore & 0xF)))
 552                 goto err;
 553
 554         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 555
 556         if (pip->clientrereg_resv_subnetto & 0x80) {
 557                 clientrereg = 1;
 558                 event.event = IB_EVENT_CLIENT_REREGISTER;
 559                 ib_dispatch_event(&event);
 560         }
 561
 562         /*
 563          * Do the port state change now that the other link parameters
 564          * have been set.
 565          * Changing the port physical state only makes sense if the link
 566          * is down or is being set to down.
 567          */
 568         state = pip->linkspeed_portstate & 0xF;
 569         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 570         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
 571                 goto err;
 572
 573         /*
 574          * Only state changes of DOWN, ARM, and ACTIVE are valid
 575          * and must be in the correct state to take effect (see 7.2.6).
 576          */
 577         switch (state) {
 578         case IB_PORT_NOP:
 579                 if (lstate == 0)
 580                         break;
 581                 /* FALLTHROUGH */
 582         case IB_PORT_DOWN:
 583                 if (lstate == 0)
 584                         lstate = IPATH_IB_LINKDOWN_ONLY;
 585                 else if (lstate == 1)
 586                         lstate = IPATH_IB_LINKDOWN_SLEEP;
 587                 else if (lstate == 2)
 588                         lstate = IPATH_IB_LINKDOWN;
 589                 else if (lstate == 3)
 590                         lstate = IPATH_IB_LINKDOWN_DISABLE;
 591                 else
 592                         goto err;
 593                 ipath_set_linkstate(dd, lstate);
 594                 if (lstate == IPATH_IB_LINKDOWN_DISABLE) {
 595                         ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
 596                         goto done;
 597                 }
 598                 ipath_wait_linkstate(dd, IPATH_LINKINIT | IPATH_LINKARMED |
 599                                 IPATH_LINKACTIVE, 1000);
 600                 break;
 601         case IB_PORT_ARMED:
 602                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
 603                 break;
 604         case IB_PORT_ACTIVE:
 605                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
 606                 break;
 607         default:
 608                 /* XXX We have already partially updated our state! */
 609                 goto err;
 610         }
 611
 612         ret = recv_subn_get_portinfo(smp, ibdev, port);
 613
 614         if (clientrereg)
 615                 pip->clientrereg_resv_subnetto |= 0x80;
 616
 617         goto done;
 618
 619 err:
 620         smp->status |= IB_SMP_INVALID_FIELD;
 621         ret = recv_subn_get_portinfo(smp, ibdev, port);
 622
 623 done:
 624         return ret;
 625 }
 626
 627 /**
 628  * rm_pkey - decrecment the reference count for the given PKEY
 629  * @dd: the infinipath device
 630  * @key: the PKEY index
 631  *
 632  * Return true if this was the last reference and the hardware table entry
 633  * needs to be changed.
 634  */
 635 static int rm_pkey(struct ipath_devdata *dd, u16 key)
 636 {
 637         int i;
 638         int ret;
 639
 640         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 641                 if (dd->ipath_pkeys[i] != key)
 642                         continue;
 643                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
 644                         dd->ipath_pkeys[i] = 0;
 645                         ret = 1;
 646                         goto bail;
 647                 }
 648                 break;
 649         }
 650
 651         ret = 0;
 652
 653 bail:
 654         return ret;
 655 }
 656
 657 /**
 658  * add_pkey - add the given PKEY to the hardware table
 659  * @dd: the infinipath device
 660  * @key: the PKEY
 661  *
 662  * Return an error code if unable to add the entry, zero if no change,
 663  * or 1 if the hardware PKEY register needs to be updated.
 664  */
 665 static int add_pkey(struct ipath_devdata *dd, u16 key)
 666 {
 667         int i;
 668         u16 lkey = key & 0x7FFF;
 669         int any = 0;
 670         int ret;
 671
 672         if (lkey == 0x7FFF) {
 673                 ret = 0;
 674                 goto bail;
 675         }
 676
 677         /* Look for an empty slot or a matching PKEY. */
 678         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 679                 if (!dd->ipath_pkeys[i]) {
 680                         any++;
 681                         continue;
 682                 }
 683                 /* If it matches exactly, try to increment the ref count */
 684                 if (dd->ipath_pkeys[i] == key) {
 685                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
 686                                 ret = 0;
 687                                 goto bail;
 688                         }
 689                         /* Lost the race. Look for an empty slot below. */
 690                         atomic_dec(&dd->ipath_pkeyrefs[i]);
 691                         any++;
 692                 }
 693                 /*
 694                  * It makes no sense to have both the limited and unlimited
 695                  * PKEY set at the same time since the unlimited one will
 696                  * disable the limited one.
 697                  */
 698                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
 699                         ret = -EEXIST;
 700                         goto bail;
 701                 }
 702         }
 703         if (!any) {
 704                 ret = -EBUSY;
 705                 goto bail;
 706         }
 707         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 708                 if (!dd->ipath_pkeys[i] &&
 709                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
 710                         /* for ipathstats, etc. */
 711                         ipath_stats.sps_pkeys[i] = lkey;
 712                         dd->ipath_pkeys[i] = key;
 713                         ret = 1;
 714                         goto bail;
 715                 }
 716         }
 717         ret = -EBUSY;
 718
 719 bail:
 720         return ret;
 721 }
 722
 723 /**
 724  * set_pkeys - set the PKEY table for port 0
 725  * @dd: the infinipath device
 726  * @pkeys: the PKEY table
 727  */
 728 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 729 {
 730         struct ipath_portdata *pd;
 731         int i;
 732         int changed = 0;
 733
 734         /* always a kernel port, no locking needed */
 735         pd = dd->ipath_pd[0];
 736
 737         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
 738                 u16 key = pkeys[i];
 739                 u16 okey = pd->port_pkeys[i];
 740
 741                 if (key == okey)
 742                         continue;
 743                 /*
 744                  * The value of this PKEY table entry is changing.
 745                  * Remove the old entry in the hardware's array of PKEYs.
 746                  */
 747                 if (okey & 0x7FFF)
 748                         changed |= rm_pkey(dd, okey);
 749                 if (key & 0x7FFF) {
 750                         int ret = add_pkey(dd, key);
 751
 752                         if (ret < 0)
 753                                 key = 0;
 754                         else
 755                                 changed |= ret;
 756                 }
 757                 pd->port_pkeys[i] = key;
 758         }
 759         if (changed) {
 760                 u64 pkey;
 761
 762                 pkey = (u64) dd->ipath_pkeys[0] |
 763                         ((u64) dd->ipath_pkeys[1] << 16) |
 764                         ((u64) dd->ipath_pkeys[2] << 32) |
 765                         ((u64) dd->ipath_pkeys[3] << 48);
 766                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
 767                            (unsigned long long) pkey);
 768                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
 769                                  pkey);
 770         }
 771         return 0;
 772 }
 773
 774 static int recv_subn_set_pkeytable(struct ib_smp *smp,
 775                                    struct ib_device *ibdev)
 776 {
 777         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 778         __be16 *p = (__be16 *) smp->data;
 779         u16 *q = (u16 *) smp->data;
 780         struct ipath_ibdev *dev = to_idev(ibdev);
 781         unsigned i, n = ipath_get_npkeys(dev->dd);
 782
 783         for (i = 0; i < n; i++)
 784                 q[i] = be16_to_cpu(p[i]);
 785
 786         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
 787                 smp->status |= IB_SMP_INVALID_FIELD;
 788
 789         return recv_subn_get_pkeytable(smp, ibdev);
 790 }
 791
 792 #define IB_PMA_CLASS_PORT_INFO          cpu_to_be16(0x0001)
 793 #define IB_PMA_PORT_SAMPLES_CONTROL     cpu_to_be16(0x0010)
 794 #define IB_PMA_PORT_SAMPLES_RESULT      cpu_to_be16(0x0011)
 795 #define IB_PMA_PORT_COUNTERS            cpu_to_be16(0x0012)
 796 #define IB_PMA_PORT_COUNTERS_EXT        cpu_to_be16(0x001D)
 797 #define IB_PMA_PORT_SAMPLES_RESULT_EXT  cpu_to_be16(0x001E)
 798
 799 struct ib_perf {
 800         u8 base_version;
 801         u8 mgmt_class;
 802         u8 class_version;
 803         u8 method;
 804         __be16 status;
 805         __be16 unused;
 806         __be64 tid;
 807         __be16 attr_id;
 808         __be16 resv;
 809         __be32 attr_mod;
 810         u8 reserved[40];
 811         u8 data[192];
 812 } __attribute__ ((packed));
 813
 814 struct ib_pma_classportinfo {
 815         u8 base_version;
 816         u8 class_version;
 817         __be16 cap_mask;
 818         u8 reserved[3];
 819         u8 resp_time_value;     /* only lower 5 bits */
 820         union ib_gid redirect_gid;
 821         __be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
 822         __be16 redirect_lid;
 823         __be16 redirect_pkey;
 824         __be32 redirect_qp;     /* only lower 24 bits */
 825         __be32 redirect_qkey;
 826         union ib_gid trap_gid;
 827         __be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
 828         __be16 trap_lid;
 829         __be16 trap_pkey;
 830         __be32 trap_hl_qp;      /* 8, 24 bits respectively */
 831         __be32 trap_qkey;
 832 } __attribute__ ((packed));
 833
 834 struct ib_pma_portsamplescontrol {
 835         u8 opcode;
 836         u8 port_select;
 837         u8 tick;
 838         u8 counter_width;       /* only lower 3 bits */
 839         __be32 counter_mask0_9; /* 2, 10 * 3, bits */
 840         __be16 counter_mask10_14;       /* 1, 5 * 3, bits */
 841         u8 sample_mechanisms;
 842         u8 sample_status;       /* only lower 2 bits */
 843         __be64 option_mask;
 844         __be64 vendor_mask;
 845         __be32 sample_start;
 846         __be32 sample_interval;
 847         __be16 tag;
 848         __be16 counter_select[15];
 849 } __attribute__ ((packed));
 850
 851 struct ib_pma_portsamplesresult {
 852         __be16 tag;
 853         __be16 sample_status;   /* only lower 2 bits */
 854         __be32 counter[15];
 855 } __attribute__ ((packed));
 856
 857 struct ib_pma_portsamplesresult_ext {
 858         __be16 tag;
 859         __be16 sample_status;   /* only lower 2 bits */
 860         __be32 extended_width;  /* only upper 2 bits */
 861         __be64 counter[15];
 862 } __attribute__ ((packed));
 863
 864 struct ib_pma_portcounters {
 865         u8 reserved;
 866         u8 port_select;
 867         __be16 counter_select;
 868         __be16 symbol_error_counter;
 869         u8 link_error_recovery_counter;
 870         u8 link_downed_counter;
 871         __be16 port_rcv_errors;
 872         __be16 port_rcv_remphys_errors;
 873         __be16 port_rcv_switch_relay_errors;
 874         __be16 port_xmit_discards;
 875         u8 port_xmit_constraint_errors;
 876         u8 port_rcv_constraint_errors;
 877         u8 reserved1;
 878         u8 lli_ebor_errors;     /* 4, 4, bits */
 879         __be16 reserved2;
 880         __be16 vl15_dropped;
 881         __be32 port_xmit_data;
 882         __be32 port_rcv_data;
 883         __be32 port_xmit_packets;
 884         __be32 port_rcv_packets;
 885 } __attribute__ ((packed));
 886
 887 #define IB_PMA_SEL_SYMBOL_ERROR                 cpu_to_be16(0x0001)
 888 #define IB_PMA_SEL_LINK_ERROR_RECOVERY          cpu_to_be16(0x0002)
 889 #define IB_PMA_SEL_LINK_DOWNED                  cpu_to_be16(0x0004)
 890 #define IB_PMA_SEL_PORT_RCV_ERRORS              cpu_to_be16(0x0008)
 891 #define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      cpu_to_be16(0x0010)
 892 #define IB_PMA_SEL_PORT_XMIT_DISCARDS           cpu_to_be16(0x0040)
 893 #define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  cpu_to_be16(0x0200)
 894 #define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    cpu_to_be16(0x0400)
 895 #define IB_PMA_SEL_PORT_VL15_DROPPED            cpu_to_be16(0x0800)
 896 #define IB_PMA_SEL_PORT_XMIT_DATA               cpu_to_be16(0x1000)
 897 #define IB_PMA_SEL_PORT_RCV_DATA                cpu_to_be16(0x2000)
 898 #define IB_PMA_SEL_PORT_XMIT_PACKETS            cpu_to_be16(0x4000)
 899 #define IB_PMA_SEL_PORT_RCV_PACKETS             cpu_to_be16(0x8000)
 900
 901 struct ib_pma_portcounters_ext {
 902         u8 reserved;
 903         u8 port_select;
 904         __be16 counter_select;
 905         __be32 reserved1;
 906         __be64 port_xmit_data;
 907         __be64 port_rcv_data;
 908         __be64 port_xmit_packets;
 909         __be64 port_rcv_packets;
 910         __be64 port_unicast_xmit_packets;
 911         __be64 port_unicast_rcv_packets;
 912         __be64 port_multicast_xmit_packets;
 913         __be64 port_multicast_rcv_packets;
 914 } __attribute__ ((packed));
 915
 916 #define IB_PMA_SELX_PORT_XMIT_DATA              cpu_to_be16(0x0001)
 917 #define IB_PMA_SELX_PORT_RCV_DATA               cpu_to_be16(0x0002)
 918 #define IB_PMA_SELX_PORT_XMIT_PACKETS           cpu_to_be16(0x0004)
 919 #define IB_PMA_SELX_PORT_RCV_PACKETS            cpu_to_be16(0x0008)
 920 #define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       cpu_to_be16(0x0010)
 921 #define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        cpu_to_be16(0x0020)
 922 #define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     cpu_to_be16(0x0040)
 923 #define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      cpu_to_be16(0x0080)
 924
 925 static int recv_pma_get_classportinfo(struct ib_perf *pmp)
 926 {
 927         struct ib_pma_classportinfo *p =
 928                 (struct ib_pma_classportinfo *)pmp->data;
 929
 930         memset(pmp->data, 0, sizeof(pmp->data));
 931
 932         if (pmp->attr_mod != 0)
 933                 pmp->status |= IB_SMP_INVALID_FIELD;
 934
 935         /* Indicate AllPortSelect is valid (only one port anyway) */
 936         p->cap_mask = cpu_to_be16(1 << 8);
 937         p->base_version = 1;
 938         p->class_version = 1;
 939         /*
 940          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
 941          * sec.
 942          */
 943         p->resp_time_value = 18;
 944
 945         return reply((struct ib_smp *) pmp);
 946 }
 947
 948 /*
 949  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
 950  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
 951  * We support 5 counters which only count the mandatory quantities.
 952  */
 953 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
 954 #define COUNTER_MASK0_9 cpu_to_be32(COUNTER_MASK(1, 0) | \
 955                                     COUNTER_MASK(1, 1) | \
 956                                     COUNTER_MASK(1, 2) | \
 957                                     COUNTER_MASK(1, 3) | \
 958                                     COUNTER_MASK(1, 4))
 959
 960 static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 961                                            struct ib_device *ibdev, u8 port)
 962 {
 963         struct ib_pma_portsamplescontrol *p =
 964                 (struct ib_pma_portsamplescontrol *)pmp->data;
 965         struct ipath_ibdev *dev = to_idev(ibdev);
 966         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 967         unsigned long flags;
 968         u8 port_select = p->port_select;
 969
 970         memset(pmp->data, 0, sizeof(pmp->data));
 971
 972         p->port_select = port_select;
 973         if (pmp->attr_mod != 0 ||
 974             (port_select != port && port_select != 0xFF))
 975                 pmp->status |= IB_SMP_INVALID_FIELD;
 976         /*
 977          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 978          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
 979          * intervals are counted in ticks.  Since we use Linux timers, that
 980          * count in jiffies, we can't sample for less than 1000 ticks if HZ
 981          * == 1000 (4000 ticks if HZ is 250).  link_speed_active returns 2 for
 982          * DDR, 1 for SDR, set the tick to 1 for DDR, 0 for SDR on chips that
 983          * have hardware support for delaying packets.
 984          */
 985         if (crp->cr_psstat)
 986                 p->tick = dev->dd->ipath_link_speed_active - 1;
 987         else
 988                 p->tick = 250;          /* 1 usec. */
 989         p->counter_width = 4;   /* 32 bit counters */
 990         p->counter_mask0_9 = COUNTER_MASK0_9;
 991         spin_lock_irqsave(&dev->pending_lock, flags);
 992         if (crp->cr_psstat)
 993                 p->sample_status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 994         else
 995                 p->sample_status = dev->pma_sample_status;
 996         p->sample_start = cpu_to_be32(dev->pma_sample_start);
 997         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
 998         p->tag = cpu_to_be16(dev->pma_tag);
 999         p->counter_select[0] = dev->pma_counter_select[0];
1000         p->counter_select[1] = dev->pma_counter_select[1];
1001         p->counter_select[2] = dev->pma_counter_select[2];
1002         p->counter_select[3] = dev->pma_counter_select[3];
1003         p->counter_select[4] = dev->pma_counter_select[4];
1004         spin_unlock_irqrestore(&dev->pending_lock, flags);
1005
1006         return reply((struct ib_smp *) pmp);
1007 }
1008
1009 static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
1010                                            struct ib_device *ibdev, u8 port)
1011 {
1012         struct ib_pma_portsamplescontrol *p =
1013                 (struct ib_pma_portsamplescontrol *)pmp->data;
1014         struct ipath_ibdev *dev = to_idev(ibdev);
1015         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1016         unsigned long flags;
1017         u8 status;
1018         int ret;
1019
1020         if (pmp->attr_mod != 0 ||
1021             (p->port_select != port && p->port_select != 0xFF)) {
1022                 pmp->status |= IB_SMP_INVALID_FIELD;
1023                 ret = reply((struct ib_smp *) pmp);
1024                 goto bail;
1025         }
1026
1027         spin_lock_irqsave(&dev->pending_lock, flags);
1028         if (crp->cr_psstat)
1029                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1030         else
1031                 status = dev->pma_sample_status;
1032         if (status == IB_PMA_SAMPLE_STATUS_DONE) {
1033                 dev->pma_sample_start = be32_to_cpu(p->sample_start);
1034                 dev->pma_sample_interval = be32_to_cpu(p->sample_interval);
1035                 dev->pma_tag = be16_to_cpu(p->tag);
1036                 dev->pma_counter_select[0] = p->counter_select[0];
1037                 dev->pma_counter_select[1] = p->counter_select[1];
1038                 dev->pma_counter_select[2] = p->counter_select[2];
1039                 dev->pma_counter_select[3] = p->counter_select[3];
1040                 dev->pma_counter_select[4] = p->counter_select[4];
1041                 if (crp->cr_psstat) {
1042                         ipath_write_creg(dev->dd, crp->cr_psinterval,
1043                                          dev->pma_sample_interval);
1044                         ipath_write_creg(dev->dd, crp->cr_psstart,
1045                                          dev->pma_sample_start);
1046                 } else
1047                         dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
1048         }
1049         spin_unlock_irqrestore(&dev->pending_lock, flags);
1050
1051         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
1052
1053 bail:
1054         return ret;
1055 }
1056
1057 static u64 get_counter(struct ipath_ibdev *dev,
1058                        struct ipath_cregs const *crp,
1059                        __be16 sel)
1060 {
1061         u64 ret;
1062
1063         switch (sel) {
1064         case IB_PMA_PORT_XMIT_DATA:
1065                 ret = (crp->cr_psxmitdatacount) ?
1066                         ipath_read_creg32(dev->dd, crp->cr_psxmitdatacount) :
1067                         dev->ipath_sword;
1068                 break;
1069         case IB_PMA_PORT_RCV_DATA:
1070                 ret = (crp->cr_psrcvdatacount) ?
1071                         ipath_read_creg32(dev->dd, crp->cr_psrcvdatacount) :
1072                         dev->ipath_rword;
1073                 break;
1074         case IB_PMA_PORT_XMIT_PKTS:
1075                 ret = (crp->cr_psxmitpktscount) ?
1076                         ipath_read_creg32(dev->dd, crp->cr_psxmitpktscount) :
1077                         dev->ipath_spkts;
1078                 break;
1079         case IB_PMA_PORT_RCV_PKTS:
1080                 ret = (crp->cr_psrcvpktscount) ?
1081                         ipath_read_creg32(dev->dd, crp->cr_psrcvpktscount) :
1082                         dev->ipath_rpkts;
1083                 break;
1084         case IB_PMA_PORT_XMIT_WAIT:
1085                 ret = (crp->cr_psxmitwaitcount) ?
1086                         ipath_read_creg32(dev->dd, crp->cr_psxmitwaitcount) :
1087                         dev->ipath_xmit_wait;
1088                 break;
1089         default:
1090                 ret = 0;
1091         }
1092
1093         return ret;
1094 }
1095
1096 static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
1097                                           struct ib_device *ibdev)
1098 {
1099         struct ib_pma_portsamplesresult *p =
1100                 (struct ib_pma_portsamplesresult *)pmp->data;
1101         struct ipath_ibdev *dev = to_idev(ibdev);
1102         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1103         u8 status;
1104         int i;
1105
1106         memset(pmp->data, 0, sizeof(pmp->data));
1107         p->tag = cpu_to_be16(dev->pma_tag);
1108         if (crp->cr_psstat)
1109                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1110         else
1111                 status = dev->pma_sample_status;
1112         p->sample_status = cpu_to_be16(status);
1113         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1114                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1115                     cpu_to_be32(
1116                         get_counter(dev, crp, dev->pma_counter_select[i]));
1117
1118         return reply((struct ib_smp *) pmp);
1119 }
1120
1121 static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
1122                                               struct ib_device *ibdev)
1123 {
1124         struct ib_pma_portsamplesresult_ext *p =
1125                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
1126         struct ipath_ibdev *dev = to_idev(ibdev);
1127         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
1128         u8 status;
1129         int i;
1130
1131         memset(pmp->data, 0, sizeof(pmp->data));
1132         p->tag = cpu_to_be16(dev->pma_tag);
1133         if (crp->cr_psstat)
1134                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1135         else
1136                 status = dev->pma_sample_status;
1137         p->sample_status = cpu_to_be16(status);
1138         /* 64 bits */
1139         p->extended_width = cpu_to_be32(0x80000000);
1140         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1141                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1142                     cpu_to_be64(
1143                         get_counter(dev, crp, dev->pma_counter_select[i]));
1144
1145         return reply((struct ib_smp *) pmp);
1146 }
1147
1148 static int recv_pma_get_portcounters(struct ib_perf *pmp,
1149                                      struct ib_device *ibdev, u8 port)
1150 {
1151         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1152                 pmp->data;
1153         struct ipath_ibdev *dev = to_idev(ibdev);
1154         struct ipath_verbs_counters cntrs;
1155         u8 port_select = p->port_select;
1156
1157         ipath_get_counters(dev->dd, &cntrs);
1158
1159         /* Adjust counters for any resets done. */
1160         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1161         cntrs.link_error_recovery_counter -=
1162                 dev->z_link_error_recovery_counter;
1163         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1164         cntrs.port_rcv_errors += dev->rcv_errors;
1165         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1166         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1167         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1168         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1169         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1170         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1171         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1172         cntrs.local_link_integrity_errors -=
1173                 dev->z_local_link_integrity_errors;
1174         cntrs.excessive_buffer_overrun_errors -=
1175                 dev->z_excessive_buffer_overrun_errors;
1176         cntrs.vl15_dropped -= dev->z_vl15_dropped;
1177         cntrs.vl15_dropped += dev->n_vl15_dropped;
1178
1179         memset(pmp->data, 0, sizeof(pmp->data));
1180
1181         p->port_select = port_select;
1182         if (pmp->attr_mod != 0 ||
1183             (port_select != port && port_select != 0xFF))
1184                 pmp->status |= IB_SMP_INVALID_FIELD;
1185
1186         if (cntrs.symbol_error_counter > 0xFFFFUL)
1187                 p->symbol_error_counter = cpu_to_be16(0xFFFF);
1188         else
1189                 p->symbol_error_counter =
1190                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1191         if (cntrs.link_error_recovery_counter > 0xFFUL)
1192                 p->link_error_recovery_counter = 0xFF;
1193         else
1194                 p->link_error_recovery_counter =
1195                         (u8)cntrs.link_error_recovery_counter;
1196         if (cntrs.link_downed_counter > 0xFFUL)
1197                 p->link_downed_counter = 0xFF;
1198         else
1199                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1200         if (cntrs.port_rcv_errors > 0xFFFFUL)
1201                 p->port_rcv_errors = cpu_to_be16(0xFFFF);
1202         else
1203                 p->port_rcv_errors =
1204                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1205         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1206                 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
1207         else
1208                 p->port_rcv_remphys_errors =
1209                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1210         if (cntrs.port_xmit_discards > 0xFFFFUL)
1211                 p->port_xmit_discards = cpu_to_be16(0xFFFF);
1212         else
1213                 p->port_xmit_discards =
1214                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1215         if (cntrs.local_link_integrity_errors > 0xFUL)
1216                 cntrs.local_link_integrity_errors = 0xFUL;
1217         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1218                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1219         p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
1220                 cntrs.excessive_buffer_overrun_errors;
1221         if (cntrs.vl15_dropped > 0xFFFFUL)
1222                 p->vl15_dropped = cpu_to_be16(0xFFFF);
1223         else
1224                 p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1225         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1226                 p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
1227         else
1228                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1229         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1230                 p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
1231         else
1232                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1233         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1234                 p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
1235         else
1236                 p->port_xmit_packets =
1237                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1238         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1239                 p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
1240         else
1241                 p->port_rcv_packets =
1242                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1243
1244         return reply((struct ib_smp *) pmp);
1245 }
1246
1247 static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
1248                                          struct ib_device *ibdev, u8 port)
1249 {
1250         struct ib_pma_portcounters_ext *p =
1251                 (struct ib_pma_portcounters_ext *)pmp->data;
1252         struct ipath_ibdev *dev = to_idev(ibdev);
1253         u64 swords, rwords, spkts, rpkts, xwait;
1254         u8 port_select = p->port_select;
1255
1256         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1257                                 &rpkts, &xwait);
1258
1259         /* Adjust counters for any resets done. */
1260         swords -= dev->z_port_xmit_data;
1261         rwords -= dev->z_port_rcv_data;
1262         spkts -= dev->z_port_xmit_packets;
1263         rpkts -= dev->z_port_rcv_packets;
1264
1265         memset(pmp->data, 0, sizeof(pmp->data));
1266
1267         p->port_select = port_select;
1268         if (pmp->attr_mod != 0 ||
1269             (port_select != port && port_select != 0xFF))
1270                 pmp->status |= IB_SMP_INVALID_FIELD;
1271
1272         p->port_xmit_data = cpu_to_be64(swords);
1273         p->port_rcv_data = cpu_to_be64(rwords);
1274         p->port_xmit_packets = cpu_to_be64(spkts);
1275         p->port_rcv_packets = cpu_to_be64(rpkts);
1276         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1277         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1278         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1279         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1280
1281         return reply((struct ib_smp *) pmp);
1282 }
1283
1284 static int recv_pma_set_portcounters(struct ib_perf *pmp,
1285                                      struct ib_device *ibdev, u8 port)
1286 {
1287         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1288                 pmp->data;
1289         struct ipath_ibdev *dev = to_idev(ibdev);
1290         struct ipath_verbs_counters cntrs;
1291
1292         /*
1293          * Since the HW doesn't support clearing counters, we save the
1294          * current count and subtract it from future responses.
1295          */
1296         ipath_get_counters(dev->dd, &cntrs);
1297
1298         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1299                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1300
1301         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1302                 dev->z_link_error_recovery_counter =
1303                         cntrs.link_error_recovery_counter;
1304
1305         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1306                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1307
1308         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1309                 dev->z_port_rcv_errors =
1310                         cntrs.port_rcv_errors + dev->rcv_errors;
1311
1312         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1313                 dev->z_port_rcv_remphys_errors =
1314                         cntrs.port_rcv_remphys_errors;
1315
1316         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1317                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1318
1319         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1320                 dev->z_local_link_integrity_errors =
1321                         cntrs.local_link_integrity_errors;
1322
1323         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1324                 dev->z_excessive_buffer_overrun_errors =
1325                         cntrs.excessive_buffer_overrun_errors;
1326
1327         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1328                 dev->n_vl15_dropped = 0;
1329                 dev->z_vl15_dropped = cntrs.vl15_dropped;
1330         }
1331
1332         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1333                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1334
1335         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1336                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1337
1338         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1339                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1340
1341         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1342                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1343
1344         return recv_pma_get_portcounters(pmp, ibdev, port);
1345 }
1346
1347 static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
1348                                          struct ib_device *ibdev, u8 port)
1349 {
1350         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1351                 pmp->data;
1352         struct ipath_ibdev *dev = to_idev(ibdev);
1353         u64 swords, rwords, spkts, rpkts, xwait;
1354
1355         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1356                                 &rpkts, &xwait);
1357
1358         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1359                 dev->z_port_xmit_data = swords;
1360
1361         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1362                 dev->z_port_rcv_data = rwords;
1363
1364         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1365                 dev->z_port_xmit_packets = spkts;
1366
1367         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1368                 dev->z_port_rcv_packets = rpkts;
1369
1370         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1371                 dev->n_unicast_xmit = 0;
1372
1373         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1374                 dev->n_unicast_rcv = 0;
1375
1376         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1377                 dev->n_multicast_xmit = 0;
1378
1379         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1380                 dev->n_multicast_rcv = 0;
1381
1382         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1383 }
1384
1385 static int process_subn(struct ib_device *ibdev, int mad_flags,
1386                         u8 port_num, struct ib_mad *in_mad,
1387                         struct ib_mad *out_mad)
1388 {
1389         struct ib_smp *smp = (struct ib_smp *)out_mad;
1390         struct ipath_ibdev *dev = to_idev(ibdev);
1391         int ret;
1392
1393         *out_mad = *in_mad;
1394         if (smp->class_version != 1) {
1395                 smp->status |= IB_SMP_UNSUP_VERSION;
1396                 ret = reply(smp);
1397                 goto bail;
1398         }
1399
1400         /* Is the mkey in the process of expiring? */
1401         if (dev->mkey_lease_timeout &&
1402             time_after_eq(jiffies, dev->mkey_lease_timeout)) {
1403                 /* Clear timeout and mkey protection field. */
1404                 dev->mkey_lease_timeout = 0;
1405                 dev->mkeyprot = 0;
1406         }
1407
1408         /*
1409          * M_Key checking depends on
1410          * Portinfo:M_Key_protect_bits
1411          */
1412         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1413             dev->mkey != smp->mkey &&
1414             (smp->method == IB_MGMT_METHOD_SET ||
1415              (smp->method == IB_MGMT_METHOD_GET &&
1416               dev->mkeyprot >= 2))) {
1417                 if (dev->mkey_violations != 0xFFFF)
1418                         ++dev->mkey_violations;
1419                 if (dev->mkey_lease_timeout ||
1420                     dev->mkey_lease_period == 0) {
1421                         ret = IB_MAD_RESULT_SUCCESS |
1422                                 IB_MAD_RESULT_CONSUMED;
1423                         goto bail;
1424                 }
1425                 dev->mkey_lease_timeout = jiffies +
1426                         dev->mkey_lease_period * HZ;
1427                 /* Future: Generate a trap notice. */
1428                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1429                 goto bail;
1430         } else if (dev->mkey_lease_timeout)
1431                 dev->mkey_lease_timeout = 0;
1432
1433         switch (smp->method) {
1434         case IB_MGMT_METHOD_GET:
1435                 switch (smp->attr_id) {
1436                 case IB_SMP_ATTR_NODE_DESC:
1437                         ret = recv_subn_get_nodedescription(smp, ibdev);
1438                         goto bail;
1439                 case IB_SMP_ATTR_NODE_INFO:
1440                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1441                         goto bail;
1442                 case IB_SMP_ATTR_GUID_INFO:
1443                         ret = recv_subn_get_guidinfo(smp, ibdev);
1444                         goto bail;
1445                 case IB_SMP_ATTR_PORT_INFO:
1446                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1447                         goto bail;
1448                 case IB_SMP_ATTR_PKEY_TABLE:
1449                         ret = recv_subn_get_pkeytable(smp, ibdev);
1450                         goto bail;
1451                 case IB_SMP_ATTR_SM_INFO:
1452                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1453                                 ret = IB_MAD_RESULT_SUCCESS |
1454                                         IB_MAD_RESULT_CONSUMED;
1455                                 goto bail;
1456                         }
1457                         if (dev->port_cap_flags & IB_PORT_SM) {
1458                                 ret = IB_MAD_RESULT_SUCCESS;
1459                                 goto bail;
1460                         }
1461                         /* FALLTHROUGH */
1462                 default:
1463                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1464                         ret = reply(smp);
1465                         goto bail;
1466                 }
1467
1468         case IB_MGMT_METHOD_SET:
1469                 switch (smp->attr_id) {
1470                 case IB_SMP_ATTR_GUID_INFO:
1471                         ret = recv_subn_set_guidinfo(smp, ibdev);
1472                         goto bail;
1473                 case IB_SMP_ATTR_PORT_INFO:
1474                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1475                         goto bail;
1476                 case IB_SMP_ATTR_PKEY_TABLE:
1477                         ret = recv_subn_set_pkeytable(smp, ibdev);
1478                         goto bail;
1479                 case IB_SMP_ATTR_SM_INFO:
1480                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1481                                 ret = IB_MAD_RESULT_SUCCESS |
1482                                         IB_MAD_RESULT_CONSUMED;
1483                                 goto bail;
1484                         }
1485                         if (dev->port_cap_flags & IB_PORT_SM) {
1486                                 ret = IB_MAD_RESULT_SUCCESS;
1487                                 goto bail;
1488                         }
1489                         /* FALLTHROUGH */
1490                 default:
1491                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1492                         ret = reply(smp);
1493                         goto bail;
1494                 }
1495
1496         case IB_MGMT_METHOD_TRAP:
1497         case IB_MGMT_METHOD_REPORT:
1498         case IB_MGMT_METHOD_REPORT_RESP:
1499         case IB_MGMT_METHOD_TRAP_REPRESS:
1500         case IB_MGMT_METHOD_GET_RESP:
1501                 /*
1502                  * The ib_mad module will call us to process responses
1503                  * before checking for other consumers.
1504                  * Just tell the caller to process it normally.
1505                  */
1506                 ret = IB_MAD_RESULT_SUCCESS;
1507                 goto bail;
1508         default:
1509                 smp->status |= IB_SMP_UNSUP_METHOD;
1510                 ret = reply(smp);
1511         }
1512
1513 bail:
1514         return ret;
1515 }
1516
1517 static int process_perf(struct ib_device *ibdev, u8 port_num,
1518                         struct ib_mad *in_mad,
1519                         struct ib_mad *out_mad)
1520 {
1521         struct ib_perf *pmp = (struct ib_perf *)out_mad;
1522         int ret;
1523
1524         *out_mad = *in_mad;
1525         if (pmp->class_version != 1) {
1526                 pmp->status |= IB_SMP_UNSUP_VERSION;
1527                 ret = reply((struct ib_smp *) pmp);
1528                 goto bail;
1529         }
1530
1531         switch (pmp->method) {
1532         case IB_MGMT_METHOD_GET:
1533                 switch (pmp->attr_id) {
1534                 case IB_PMA_CLASS_PORT_INFO:
1535                         ret = recv_pma_get_classportinfo(pmp);
1536                         goto bail;
1537                 case IB_PMA_PORT_SAMPLES_CONTROL:
1538                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1539                                                               port_num);
1540                         goto bail;
1541                 case IB_PMA_PORT_SAMPLES_RESULT:
1542                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1543                         goto bail;
1544                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1545                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1546                                                                  ibdev);
1547                         goto bail;
1548                 case IB_PMA_PORT_COUNTERS:
1549                         ret = recv_pma_get_portcounters(pmp, ibdev,
1550                                                         port_num);
1551                         goto bail;
1552                 case IB_PMA_PORT_COUNTERS_EXT:
1553                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1554                                                             port_num);
1555                         goto bail;
1556                 default:
1557                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1558                         ret = reply((struct ib_smp *) pmp);
1559                         goto bail;
1560                 }
1561
1562         case IB_MGMT_METHOD_SET:
1563                 switch (pmp->attr_id) {
1564                 case IB_PMA_PORT_SAMPLES_CONTROL:
1565                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1566                                                               port_num);
1567                         goto bail;
1568                 case IB_PMA_PORT_COUNTERS:
1569                         ret = recv_pma_set_portcounters(pmp, ibdev,
1570                                                         port_num);
1571                         goto bail;
1572                 case IB_PMA_PORT_COUNTERS_EXT:
1573                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1574                                                             port_num);
1575                         goto bail;
1576                 default:
1577                         pmp->status |= IB_SMP_UNSUP_METH_ATTR;
1578                         ret = reply((struct ib_smp *) pmp);
1579                         goto bail;
1580                 }
1581
1582         case IB_MGMT_METHOD_GET_RESP:
1583                 /*
1584                  * The ib_mad module will call us to process responses
1585                  * before checking for other consumers.
1586                  * Just tell the caller to process it normally.
1587                  */
1588                 ret = IB_MAD_RESULT_SUCCESS;
1589                 goto bail;
1590         default:
1591                 pmp->status |= IB_SMP_UNSUP_METHOD;
1592                 ret = reply((struct ib_smp *) pmp);
1593         }
1594
1595 bail:
1596         return ret;
1597 }
1598
1599 /**
1600  * ipath_process_mad - process an incoming MAD packet
1601  * @ibdev: the infiniband device this packet came in on
1602  * @mad_flags: MAD flags
1603  * @port_num: the port number this packet came in on
1604  * @in_wc: the work completion entry for this packet
1605  * @in_grh: the global route header for this packet
1606  * @in_mad: the incoming MAD
1607  * @out_mad: any outgoing MAD reply
1608  *
1609  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1610  * interested in processing.
1611  *
1612  * Note that the verbs framework has already done the MAD sanity checks,
1613  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1614  * MADs.
1615  *
1616  * This is called by the ib_mad module.
1617  */
1618 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1619                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1620                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1621 {
1622         int ret;
1623
1624         switch (in_mad->mad_hdr.mgmt_class) {
1625         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1626         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1627                 ret = process_subn(ibdev, mad_flags, port_num,
1628                                    in_mad, out_mad);
1629                 goto bail;
1630         case IB_MGMT_CLASS_PERF_MGMT:
1631                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1632                 goto bail;
1633         default:
1634                 ret = IB_MAD_RESULT_SUCCESS;
1635         }
1636
1637 bail:
1638         return ret;
1639 }