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 #include <rdma/ib_pma.h>
  36
  37 #include "ipath_kernel.h"
  38 #include "ipath_verbs.h"
  39 #include "ipath_common.h"
  40
  41 #define IB_SMP_UNSUP_VERSION    cpu_to_be16(0x0004)
  42 #define IB_SMP_UNSUP_METHOD     cpu_to_be16(0x0008)
  43 #define IB_SMP_UNSUP_METH_ATTR  cpu_to_be16(0x000C)
  44 #define IB_SMP_INVALID_FIELD    cpu_to_be16(0x001C)
  45
  46 static int reply(struct ib_smp *smp)
  47 {
  48         /*
  49          * The verbs framework will handle the directed/LID route
  50          * packet changes.
  51          */
  52         smp->method = IB_MGMT_METHOD_GET_RESP;
  53         if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
  54                 smp->status |= IB_SMP_DIRECTION;
  55         return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
  56 }
  57
  58 static int recv_subn_get_nodedescription(struct ib_smp *smp,
  59                                          struct ib_device *ibdev)
  60 {
  61         if (smp->attr_mod)
  62                 smp->status |= IB_SMP_INVALID_FIELD;
  63
  64         memcpy(smp->data, ibdev->node_desc, sizeof(smp->data));
  65
  66         return reply(smp);
  67 }
  68
  69 struct nodeinfo {
  70         u8 base_version;
  71         u8 class_version;
  72         u8 node_type;
  73         u8 num_ports;
  74         __be64 sys_guid;
  75         __be64 node_guid;
  76         __be64 port_guid;
  77         __be16 partition_cap;
  78         __be16 device_id;
  79         __be32 revision;
  80         u8 local_port_num;
  81         u8 vendor_id[3];
  82 } __attribute__ ((packed));
  83
  84 static int recv_subn_get_nodeinfo(struct ib_smp *smp,
  85                                   struct ib_device *ibdev, u8 port)
  86 {
  87         struct nodeinfo *nip = (struct nodeinfo *)&smp->data;
  88         struct ipath_devdata *dd = to_idev(ibdev)->dd;
  89         u32 vendor, majrev, minrev;
  90
  91         /* GUID 0 is illegal */
  92         if (smp->attr_mod || (dd->ipath_guid == 0))
  93                 smp->status |= IB_SMP_INVALID_FIELD;
  94
  95         nip->base_version = 1;
  96         nip->class_version = 1;
  97         nip->node_type = 1;     /* channel adapter */
  98         /*
  99          * XXX The num_ports value will need a layer function to get
 100          * the value if we ever have more than one IB port on a chip.
 101          * We will also need to get the GUID for the port.
 102          */
 103         nip->num_ports = ibdev->phys_port_cnt;
 104         /* This is already in network order */
 105         nip->sys_guid = to_idev(ibdev)->sys_image_guid;
 106         nip->node_guid = dd->ipath_guid;
 107         nip->port_guid = dd->ipath_guid;
 108         nip->partition_cap = cpu_to_be16(ipath_get_npkeys(dd));
 109         nip->device_id = cpu_to_be16(dd->ipath_deviceid);
 110         majrev = dd->ipath_majrev;
 111         minrev = dd->ipath_minrev;
 112         nip->revision = cpu_to_be32((majrev << 16) | minrev);
 113         nip->local_port_num = port;
 114         vendor = dd->ipath_vendorid;
 115         nip->vendor_id[0] = IPATH_SRC_OUI_1;
 116         nip->vendor_id[1] = IPATH_SRC_OUI_2;
 117         nip->vendor_id[2] = IPATH_SRC_OUI_3;
 118
 119         return reply(smp);
 120 }
 121
 122 static int recv_subn_get_guidinfo(struct ib_smp *smp,
 123                                   struct ib_device *ibdev)
 124 {
 125         u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
 126         __be64 *p = (__be64 *) smp->data;
 127
 128         /* 32 blocks of 8 64-bit GUIDs per block */
 129
 130         memset(smp->data, 0, sizeof(smp->data));
 131
 132         /*
 133          * We only support one GUID for now.  If this changes, the
 134          * portinfo.guid_cap field needs to be updated too.
 135          */
 136         if (startgx == 0) {
 137                 __be64 g = to_idev(ibdev)->dd->ipath_guid;
 138                 if (g == 0)
 139                         /* GUID 0 is illegal */
 140                         smp->status |= IB_SMP_INVALID_FIELD;
 141                 else
 142                         /* The first is a copy of the read-only HW GUID. */
 143                         *p = g;
 144         } else
 145                 smp->status |= IB_SMP_INVALID_FIELD;
 146
 147         return reply(smp);
 148 }
 149
 150 static void set_link_width_enabled(struct ipath_devdata *dd, u32 w)
 151 {
 152         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LWID_ENB, w);
 153 }
 154
 155 static void set_link_speed_enabled(struct ipath_devdata *dd, u32 s)
 156 {
 157         (void) dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_SPD_ENB, s);
 158 }
 159
 160 static int get_overrunthreshold(struct ipath_devdata *dd)
 161 {
 162         return (dd->ipath_ibcctrl >>
 163                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 164                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 165 }
 166
 167 /**
 168  * set_overrunthreshold - set the overrun threshold
 169  * @dd: the infinipath device
 170  * @n: the new threshold
 171  *
 172  * Note that this will only take effect when the link state changes.
 173  */
 174 static int set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
 175 {
 176         unsigned v;
 177
 178         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
 179                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
 180         if (v != n) {
 181                 dd->ipath_ibcctrl &=
 182                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
 183                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
 184                 dd->ipath_ibcctrl |=
 185                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
 186                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 187                                  dd->ipath_ibcctrl);
 188         }
 189         return 0;
 190 }
 191
 192 static int get_phyerrthreshold(struct ipath_devdata *dd)
 193 {
 194         return (dd->ipath_ibcctrl >>
 195                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 196                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 197 }
 198
 199 /**
 200  * set_phyerrthreshold - set the physical error threshold
 201  * @dd: the infinipath device
 202  * @n: the new threshold
 203  *
 204  * Note that this will only take effect when the link state changes.
 205  */
 206 static int set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
 207 {
 208         unsigned v;
 209
 210         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
 211                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
 212         if (v != n) {
 213                 dd->ipath_ibcctrl &=
 214                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
 215                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
 216                 dd->ipath_ibcctrl |=
 217                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
 218                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 219                                  dd->ipath_ibcctrl);
 220         }
 221         return 0;
 222 }
 223
 224 /**
 225  * get_linkdowndefaultstate - get the default linkdown state
 226  * @dd: the infinipath device
 227  *
 228  * Returns zero if the default is POLL, 1 if the default is SLEEP.
 229  */
 230 static int get_linkdowndefaultstate(struct ipath_devdata *dd)
 231 {
 232         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
 233 }
 234
 235 static int recv_subn_get_portinfo(struct ib_smp *smp,
 236                                   struct ib_device *ibdev, u8 port)
 237 {
 238         struct ipath_ibdev *dev;
 239         struct ipath_devdata *dd;
 240         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 241         u16 lid;
 242         u8 ibcstat;
 243         u8 mtu;
 244         int ret;
 245
 246         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt) {
 247                 smp->status |= IB_SMP_INVALID_FIELD;
 248                 ret = reply(smp);
 249                 goto bail;
 250         }
 251
 252         dev = to_idev(ibdev);
 253         dd = dev->dd;
 254
 255         /* Clear all fields.  Only set the non-zero fields. */
 256         memset(smp->data, 0, sizeof(smp->data));
 257
 258         /* Only return the mkey if the protection field allows it. */
 259         if (smp->method == IB_MGMT_METHOD_SET || dev->mkey == smp->mkey ||
 260             dev->mkeyprot == 0)
 261                 pip->mkey = dev->mkey;
 262         pip->gid_prefix = dev->gid_prefix;
 263         lid = dd->ipath_lid;
 264         pip->lid = lid ? cpu_to_be16(lid) : IB_LID_PERMISSIVE;
 265         pip->sm_lid = cpu_to_be16(dev->sm_lid);
 266         pip->cap_mask = cpu_to_be32(dev->port_cap_flags);
 267         /* pip->diag_code; */
 268         pip->mkey_lease_period = cpu_to_be16(dev->mkey_lease_period);
 269         pip->local_port_num = port;
 270         pip->link_width_enabled = dd->ipath_link_width_enabled;
 271         pip->link_width_supported = dd->ipath_link_width_supported;
 272         pip->link_width_active = dd->ipath_link_width_active;
 273         pip->linkspeed_portstate = dd->ipath_link_speed_supported << 4;
 274         ibcstat = dd->ipath_lastibcstat;
 275         /* map LinkState to IB portinfo values.  */
 276         pip->linkspeed_portstate |= ipath_ib_linkstate(dd, ibcstat) + 1;
 277
 278         pip->portphysstate_linkdown =
 279                 (ipath_cvt_physportstate[ibcstat & dd->ibcs_lts_mask] << 4) |
 280                 (get_linkdowndefaultstate(dd) ? 1 : 2);
 281         pip->mkeyprot_resv_lmc = (dev->mkeyprot << 6) | dd->ipath_lmc;
 282         pip->linkspeedactive_enabled = (dd->ipath_link_speed_active << 4) |
 283                 dd->ipath_link_speed_enabled;
 284         switch (dd->ipath_ibmtu) {
 285         case 4096:
 286                 mtu = IB_MTU_4096;
 287                 break;
 288         case 2048:
 289                 mtu = IB_MTU_2048;
 290                 break;
 291         case 1024:
 292                 mtu = IB_MTU_1024;
 293                 break;
 294         case 512:
 295                 mtu = IB_MTU_512;
 296                 break;
 297         case 256:
 298                 mtu = IB_MTU_256;
 299                 break;
 300         default:                /* oops, something is wrong */
 301                 mtu = IB_MTU_2048;
 302                 break;
 303         }
 304         pip->neighbormtu_mastersmsl = (mtu << 4) | dev->sm_sl;
 305         pip->vlcap_inittype = 0x10;     /* VLCap = VL0, InitType = 0 */
 306         pip->vl_high_limit = dev->vl_high_limit;
 307         /* pip->vl_arb_high_cap; // only one VL */
 308         /* pip->vl_arb_low_cap; // only one VL */
 309         /* InitTypeReply = 0 */
 310         /* our mtu cap depends on whether 4K MTU enabled or not */
 311         pip->inittypereply_mtucap = ipath_mtu4096 ? IB_MTU_4096 : IB_MTU_2048;
 312         /* HCAs ignore VLStallCount and HOQLife */
 313         /* pip->vlstallcnt_hoqlife; */
 314         pip->operationalvl_pei_peo_fpi_fpo = 0x10;      /* OVLs = 1 */
 315         pip->mkey_violations = cpu_to_be16(dev->mkey_violations);
 316         /* P_KeyViolations are counted by hardware. */
 317         pip->pkey_violations =
 318                 cpu_to_be16((ipath_get_cr_errpkey(dd) -
 319                              dev->z_pkey_violations) & 0xFFFF);
 320         pip->qkey_violations = cpu_to_be16(dev->qkey_violations);
 321         /* Only the hardware GUID is supported for now */
 322         pip->guid_cap = 1;
 323         pip->clientrereg_resv_subnetto = dev->subnet_timeout;
 324         /* 32.768 usec. response time (guessing) */
 325         pip->resv_resptimevalue = 3;
 326         pip->localphyerrors_overrunerrors =
 327                 (get_phyerrthreshold(dd) << 4) |
 328                 get_overrunthreshold(dd);
 329         /* pip->max_credit_hint; */
 330         if (dev->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
 331                 u32 v;
 332
 333                 v = dd->ipath_f_get_ib_cfg(dd, IPATH_IB_CFG_LINKLATENCY);
 334                 pip->link_roundtrip_latency[0] = v >> 16;
 335                 pip->link_roundtrip_latency[1] = v >> 8;
 336                 pip->link_roundtrip_latency[2] = v;
 337         }
 338
 339         ret = reply(smp);
 340
 341 bail:
 342         return ret;
 343 }
 344
 345 /**
 346  * get_pkeys - return the PKEY table for port 0
 347  * @dd: the infinipath device
 348  * @pkeys: the pkey table is placed here
 349  */
 350 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
 351 {
 352         /* always a kernel port, no locking needed */
 353         struct ipath_portdata *pd = dd->ipath_pd[0];
 354
 355         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
 356
 357         return 0;
 358 }
 359
 360 static int recv_subn_get_pkeytable(struct ib_smp *smp,
 361                                    struct ib_device *ibdev)
 362 {
 363         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 364         u16 *p = (u16 *) smp->data;
 365         __be16 *q = (__be16 *) smp->data;
 366
 367         /* 64 blocks of 32 16-bit P_Key entries */
 368
 369         memset(smp->data, 0, sizeof(smp->data));
 370         if (startpx == 0) {
 371                 struct ipath_ibdev *dev = to_idev(ibdev);
 372                 unsigned i, n = ipath_get_npkeys(dev->dd);
 373
 374                 get_pkeys(dev->dd, p);
 375
 376                 for (i = 0; i < n; i++)
 377                         q[i] = cpu_to_be16(p[i]);
 378         } else
 379                 smp->status |= IB_SMP_INVALID_FIELD;
 380
 381         return reply(smp);
 382 }
 383
 384 static int recv_subn_set_guidinfo(struct ib_smp *smp,
 385                                   struct ib_device *ibdev)
 386 {
 387         /* The only GUID we support is the first read-only entry. */
 388         return recv_subn_get_guidinfo(smp, ibdev);
 389 }
 390
 391 /**
 392  * set_linkdowndefaultstate - set the default linkdown state
 393  * @dd: the infinipath device
 394  * @sleep: the new state
 395  *
 396  * Note that this will only take effect when the link state changes.
 397  */
 398 static int set_linkdowndefaultstate(struct ipath_devdata *dd, int sleep)
 399 {
 400         if (sleep)
 401                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 402         else
 403                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
 404         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 405                          dd->ipath_ibcctrl);
 406         return 0;
 407 }
 408
 409 /**
 410  * recv_subn_set_portinfo - set port information
 411  * @smp: the incoming SM packet
 412  * @ibdev: the infiniband device
 413  * @port: the port on the device
 414  *
 415  * Set Portinfo (see ch. 14.2.5.6).
 416  */
 417 static int recv_subn_set_portinfo(struct ib_smp *smp,
 418                                   struct ib_device *ibdev, u8 port)
 419 {
 420         struct ib_port_info *pip = (struct ib_port_info *)smp->data;
 421         struct ib_event event;
 422         struct ipath_ibdev *dev;
 423         struct ipath_devdata *dd;
 424         char clientrereg = 0;
 425         u16 lid, smlid;
 426         u8 lwe;
 427         u8 lse;
 428         u8 state;
 429         u16 lstate;
 430         u32 mtu;
 431         int ret, ore;
 432
 433         if (be32_to_cpu(smp->attr_mod) > ibdev->phys_port_cnt)
 434                 goto err;
 435
 436         dev = to_idev(ibdev);
 437         dd = dev->dd;
 438         event.device = ibdev;
 439         event.element.port_num = port;
 440
 441         dev->mkey = pip->mkey;
 442         dev->gid_prefix = pip->gid_prefix;
 443         dev->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
 444
 445         lid = be16_to_cpu(pip->lid);
 446         if (dd->ipath_lid != lid ||
 447             dd->ipath_lmc != (pip->mkeyprot_resv_lmc & 7)) {
 448                 /* Must be a valid unicast LID address. */
 449                 if (lid == 0 || lid >= IPATH_MULTICAST_LID_BASE)
 450                         goto err;
 451                 ipath_set_lid(dd, lid, pip->mkeyprot_resv_lmc & 7);
 452                 event.event = IB_EVENT_LID_CHANGE;
 453                 ib_dispatch_event(&event);
 454         }
 455
 456         smlid = be16_to_cpu(pip->sm_lid);
 457         if (smlid != dev->sm_lid) {
 458                 /* Must be a valid unicast LID address. */
 459                 if (smlid == 0 || smlid >= IPATH_MULTICAST_LID_BASE)
 460                         goto err;
 461                 dev->sm_lid = smlid;
 462                 event.event = IB_EVENT_SM_CHANGE;
 463                 ib_dispatch_event(&event);
 464         }
 465
 466         /* Allow 1x or 4x to be set (see 14.2.6.6). */
 467         lwe = pip->link_width_enabled;
 468         if (lwe) {
 469                 if (lwe == 0xFF)
 470                         lwe = dd->ipath_link_width_supported;
 471                 else if (lwe >= 16 || (lwe & ~dd->ipath_link_width_supported))
 472                         goto err;
 473                 set_link_width_enabled(dd, lwe);
 474         }
 475
 476         /* Allow 2.5 or 5.0 Gbs. */
 477         lse = pip->linkspeedactive_enabled & 0xF;
 478         if (lse) {
 479                 if (lse == 15)
 480                         lse = dd->ipath_link_speed_supported;
 481                 else if (lse >= 8 || (lse & ~dd->ipath_link_speed_supported))
 482                         goto err;
 483                 set_link_speed_enabled(dd, lse);
 484         }
 485
 486         /* Set link down default state. */
 487         switch (pip->portphysstate_linkdown & 0xF) {
 488         case 0: /* NOP */
 489                 break;
 490         case 1: /* SLEEP */
 491                 if (set_linkdowndefaultstate(dd, 1))
 492                         goto err;
 493                 break;
 494         case 2: /* POLL */
 495                 if (set_linkdowndefaultstate(dd, 0))
 496                         goto err;
 497                 break;
 498         default:
 499                 goto err;
 500         }
 501
 502         dev->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
 503         dev->vl_high_limit = pip->vl_high_limit;
 504
 505         switch ((pip->neighbormtu_mastersmsl >> 4) & 0xF) {
 506         case IB_MTU_256:
 507                 mtu = 256;
 508                 break;
 509         case IB_MTU_512:
 510                 mtu = 512;
 511                 break;
 512         case IB_MTU_1024:
 513                 mtu = 1024;
 514                 break;
 515         case IB_MTU_2048:
 516                 mtu = 2048;
 517                 break;
 518         case IB_MTU_4096:
 519                 if (!ipath_mtu4096)
 520                         goto err;
 521                 mtu = 4096;
 522                 break;
 523         default:
 524                 /* XXX We have already partially updated our state! */
 525                 goto err;
 526         }
 527         ipath_set_mtu(dd, mtu);
 528
 529         dev->sm_sl = pip->neighbormtu_mastersmsl & 0xF;
 530
 531         /* We only support VL0 */
 532         if (((pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF) > 1)
 533                 goto err;
 534
 535         if (pip->mkey_violations == 0)
 536                 dev->mkey_violations = 0;
 537
 538         /*
 539          * Hardware counter can't be reset so snapshot and subtract
 540          * later.
 541          */
 542         if (pip->pkey_violations == 0)
 543                 dev->z_pkey_violations = ipath_get_cr_errpkey(dd);
 544
 545         if (pip->qkey_violations == 0)
 546                 dev->qkey_violations = 0;
 547
 548         ore = pip->localphyerrors_overrunerrors;
 549         if (set_phyerrthreshold(dd, (ore >> 4) & 0xF))
 550                 goto err;
 551
 552         if (set_overrunthreshold(dd, (ore & 0xF)))
 553                 goto err;
 554
 555         dev->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
 556
 557         if (pip->clientrereg_resv_subnetto & 0x80) {
 558                 clientrereg = 1;
 559                 event.event = IB_EVENT_CLIENT_REREGISTER;
 560                 ib_dispatch_event(&event);
 561         }
 562
 563         /*
 564          * Do the port state change now that the other link parameters
 565          * have been set.
 566          * Changing the port physical state only makes sense if the link
 567          * is down or is being set to down.
 568          */
 569         state = pip->linkspeed_portstate & 0xF;
 570         lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
 571         if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
 572                 goto err;
 573
 574         /*
 575          * Only state changes of DOWN, ARM, and ACTIVE are valid
 576          * and must be in the correct state to take effect (see 7.2.6).
 577          */
 578         switch (state) {
 579         case IB_PORT_NOP:
 580                 if (lstate == 0)
 581                         break;
 582                 /* FALLTHROUGH */
 583         case IB_PORT_DOWN:
 584                 if (lstate == 0)
 585                         lstate = IPATH_IB_LINKDOWN_ONLY;
 586                 else if (lstate == 1)
 587                         lstate = IPATH_IB_LINKDOWN_SLEEP;
 588                 else if (lstate == 2)
 589                         lstate = IPATH_IB_LINKDOWN;
 590                 else if (lstate == 3)
 591                         lstate = IPATH_IB_LINKDOWN_DISABLE;
 592                 else
 593                         goto err;
 594                 ipath_set_linkstate(dd, lstate);
 595                 if (lstate == IPATH_IB_LINKDOWN_DISABLE) {
 596                         ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
 597                         goto done;
 598                 }
 599                 ipath_wait_linkstate(dd, IPATH_LINKINIT | IPATH_LINKARMED |
 600                                 IPATH_LINKACTIVE, 1000);
 601                 break;
 602         case IB_PORT_ARMED:
 603                 ipath_set_linkstate(dd, IPATH_IB_LINKARM);
 604                 break;
 605         case IB_PORT_ACTIVE:
 606                 ipath_set_linkstate(dd, IPATH_IB_LINKACTIVE);
 607                 break;
 608         default:
 609                 /* XXX We have already partially updated our state! */
 610                 goto err;
 611         }
 612
 613         ret = recv_subn_get_portinfo(smp, ibdev, port);
 614
 615         if (clientrereg)
 616                 pip->clientrereg_resv_subnetto |= 0x80;
 617
 618         goto done;
 619
 620 err:
 621         smp->status |= IB_SMP_INVALID_FIELD;
 622         ret = recv_subn_get_portinfo(smp, ibdev, port);
 623
 624 done:
 625         return ret;
 626 }
 627
 628 /**
 629  * rm_pkey - decrecment the reference count for the given PKEY
 630  * @dd: the infinipath device
 631  * @key: the PKEY index
 632  *
 633  * Return true if this was the last reference and the hardware table entry
 634  * needs to be changed.
 635  */
 636 static int rm_pkey(struct ipath_devdata *dd, u16 key)
 637 {
 638         int i;
 639         int ret;
 640
 641         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 642                 if (dd->ipath_pkeys[i] != key)
 643                         continue;
 644                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
 645                         dd->ipath_pkeys[i] = 0;
 646                         ret = 1;
 647                         goto bail;
 648                 }
 649                 break;
 650         }
 651
 652         ret = 0;
 653
 654 bail:
 655         return ret;
 656 }
 657
 658 /**
 659  * add_pkey - add the given PKEY to the hardware table
 660  * @dd: the infinipath device
 661  * @key: the PKEY
 662  *
 663  * Return an error code if unable to add the entry, zero if no change,
 664  * or 1 if the hardware PKEY register needs to be updated.
 665  */
 666 static int add_pkey(struct ipath_devdata *dd, u16 key)
 667 {
 668         int i;
 669         u16 lkey = key & 0x7FFF;
 670         int any = 0;
 671         int ret;
 672
 673         if (lkey == 0x7FFF) {
 674                 ret = 0;
 675                 goto bail;
 676         }
 677
 678         /* Look for an empty slot or a matching PKEY. */
 679         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 680                 if (!dd->ipath_pkeys[i]) {
 681                         any++;
 682                         continue;
 683                 }
 684                 /* If it matches exactly, try to increment the ref count */
 685                 if (dd->ipath_pkeys[i] == key) {
 686                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
 687                                 ret = 0;
 688                                 goto bail;
 689                         }
 690                         /* Lost the race. Look for an empty slot below. */
 691                         atomic_dec(&dd->ipath_pkeyrefs[i]);
 692                         any++;
 693                 }
 694                 /*
 695                  * It makes no sense to have both the limited and unlimited
 696                  * PKEY set at the same time since the unlimited one will
 697                  * disable the limited one.
 698                  */
 699                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
 700                         ret = -EEXIST;
 701                         goto bail;
 702                 }
 703         }
 704         if (!any) {
 705                 ret = -EBUSY;
 706                 goto bail;
 707         }
 708         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
 709                 if (!dd->ipath_pkeys[i] &&
 710                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
 711                         /* for ipathstats, etc. */
 712                         ipath_stats.sps_pkeys[i] = lkey;
 713                         dd->ipath_pkeys[i] = key;
 714                         ret = 1;
 715                         goto bail;
 716                 }
 717         }
 718         ret = -EBUSY;
 719
 720 bail:
 721         return ret;
 722 }
 723
 724 /**
 725  * set_pkeys - set the PKEY table for port 0
 726  * @dd: the infinipath device
 727  * @pkeys: the PKEY table
 728  */
 729 static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 730 {
 731         struct ipath_portdata *pd;
 732         int i;
 733         int changed = 0;
 734
 735         /* always a kernel port, no locking needed */
 736         pd = dd->ipath_pd[0];
 737
 738         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
 739                 u16 key = pkeys[i];
 740                 u16 okey = pd->port_pkeys[i];
 741
 742                 if (key == okey)
 743                         continue;
 744                 /*
 745                  * The value of this PKEY table entry is changing.
 746                  * Remove the old entry in the hardware's array of PKEYs.
 747                  */
 748                 if (okey & 0x7FFF)
 749                         changed |= rm_pkey(dd, okey);
 750                 if (key & 0x7FFF) {
 751                         int ret = add_pkey(dd, key);
 752
 753                         if (ret < 0)
 754                                 key = 0;
 755                         else
 756                                 changed |= ret;
 757                 }
 758                 pd->port_pkeys[i] = key;
 759         }
 760         if (changed) {
 761                 u64 pkey;
 762
 763                 pkey = (u64) dd->ipath_pkeys[0] |
 764                         ((u64) dd->ipath_pkeys[1] << 16) |
 765                         ((u64) dd->ipath_pkeys[2] << 32) |
 766                         ((u64) dd->ipath_pkeys[3] << 48);
 767                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
 768                            (unsigned long long) pkey);
 769                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
 770                                  pkey);
 771         }
 772         return 0;
 773 }
 774
 775 static int recv_subn_set_pkeytable(struct ib_smp *smp,
 776                                    struct ib_device *ibdev)
 777 {
 778         u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
 779         __be16 *p = (__be16 *) smp->data;
 780         u16 *q = (u16 *) smp->data;
 781         struct ipath_ibdev *dev = to_idev(ibdev);
 782         unsigned i, n = ipath_get_npkeys(dev->dd);
 783
 784         for (i = 0; i < n; i++)
 785                 q[i] = be16_to_cpu(p[i]);
 786
 787         if (startpx != 0 || set_pkeys(dev->dd, q) != 0)
 788                 smp->status |= IB_SMP_INVALID_FIELD;
 789
 790         return recv_subn_get_pkeytable(smp, ibdev);
 791 }
 792
 793 static int recv_pma_get_classportinfo(struct ib_pma_mad *pmp)
 794 {
 795         struct ib_class_port_info *p =
 796                 (struct ib_class_port_info *)pmp->data;
 797
 798         memset(pmp->data, 0, sizeof(pmp->data));
 799
 800         if (pmp->mad_hdr.attr_mod != 0)
 801                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 802
 803         /* Indicate AllPortSelect is valid (only one port anyway) */
 804         p->capability_mask = cpu_to_be16(1 << 8);
 805         p->base_version = 1;
 806         p->class_version = 1;
 807         /*
 808          * Expected response time is 4.096 usec. * 2^18 == 1.073741824
 809          * sec.
 810          */
 811         p->resp_time_value = 18;
 812
 813         return reply((struct ib_smp *) pmp);
 814 }
 815
 816 /*
 817  * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
 818  * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
 819  * We support 5 counters which only count the mandatory quantities.
 820  */
 821 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
 822 #define COUNTER_MASK0_9 cpu_to_be32(COUNTER_MASK(1, 0) | \
 823                                     COUNTER_MASK(1, 1) | \
 824                                     COUNTER_MASK(1, 2) | \
 825                                     COUNTER_MASK(1, 3) | \
 826                                     COUNTER_MASK(1, 4))
 827
 828 static int recv_pma_get_portsamplescontrol(struct ib_pma_mad *pmp,
 829                                            struct ib_device *ibdev, u8 port)
 830 {
 831         struct ib_pma_portsamplescontrol *p =
 832                 (struct ib_pma_portsamplescontrol *)pmp->data;
 833         struct ipath_ibdev *dev = to_idev(ibdev);
 834         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 835         unsigned long flags;
 836         u8 port_select = p->port_select;
 837
 838         memset(pmp->data, 0, sizeof(pmp->data));
 839
 840         p->port_select = port_select;
 841         if (pmp->mad_hdr.attr_mod != 0 ||
 842             (port_select != port && port_select != 0xFF))
 843                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 844         /*
 845          * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 846          * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
 847          * intervals are counted in ticks.  Since we use Linux timers, that
 848          * count in jiffies, we can't sample for less than 1000 ticks if HZ
 849          * == 1000 (4000 ticks if HZ is 250).  link_speed_active returns 2 for
 850          * DDR, 1 for SDR, set the tick to 1 for DDR, 0 for SDR on chips that
 851          * have hardware support for delaying packets.
 852          */
 853         if (crp->cr_psstat)
 854                 p->tick = dev->dd->ipath_link_speed_active - 1;
 855         else
 856                 p->tick = 250;          /* 1 usec. */
 857         p->counter_width = 4;   /* 32 bit counters */
 858         p->counter_mask0_9 = COUNTER_MASK0_9;
 859         spin_lock_irqsave(&dev->pending_lock, flags);
 860         if (crp->cr_psstat)
 861                 p->sample_status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 862         else
 863                 p->sample_status = dev->pma_sample_status;
 864         p->sample_start = cpu_to_be32(dev->pma_sample_start);
 865         p->sample_interval = cpu_to_be32(dev->pma_sample_interval);
 866         p->tag = cpu_to_be16(dev->pma_tag);
 867         p->counter_select[0] = dev->pma_counter_select[0];
 868         p->counter_select[1] = dev->pma_counter_select[1];
 869         p->counter_select[2] = dev->pma_counter_select[2];
 870         p->counter_select[3] = dev->pma_counter_select[3];
 871         p->counter_select[4] = dev->pma_counter_select[4];
 872         spin_unlock_irqrestore(&dev->pending_lock, flags);
 873
 874         return reply((struct ib_smp *) pmp);
 875 }
 876
 877 static int recv_pma_set_portsamplescontrol(struct ib_pma_mad *pmp,
 878                                            struct ib_device *ibdev, u8 port)
 879 {
 880         struct ib_pma_portsamplescontrol *p =
 881                 (struct ib_pma_portsamplescontrol *)pmp->data;
 882         struct ipath_ibdev *dev = to_idev(ibdev);
 883         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 884         unsigned long flags;
 885         u8 status;
 886         int ret;
 887
 888         if (pmp->mad_hdr.attr_mod != 0 ||
 889             (p->port_select != port && p->port_select != 0xFF)) {
 890                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 891                 ret = reply((struct ib_smp *) pmp);
 892                 goto bail;
 893         }
 894
 895         spin_lock_irqsave(&dev->pending_lock, flags);
 896         if (crp->cr_psstat)
 897                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 898         else
 899                 status = dev->pma_sample_status;
 900         if (status == IB_PMA_SAMPLE_STATUS_DONE) {
 901                 dev->pma_sample_start = be32_to_cpu(p->sample_start);
 902                 dev->pma_sample_interval = be32_to_cpu(p->sample_interval);
 903                 dev->pma_tag = be16_to_cpu(p->tag);
 904                 dev->pma_counter_select[0] = p->counter_select[0];
 905                 dev->pma_counter_select[1] = p->counter_select[1];
 906                 dev->pma_counter_select[2] = p->counter_select[2];
 907                 dev->pma_counter_select[3] = p->counter_select[3];
 908                 dev->pma_counter_select[4] = p->counter_select[4];
 909                 if (crp->cr_psstat) {
 910                         ipath_write_creg(dev->dd, crp->cr_psinterval,
 911                                          dev->pma_sample_interval);
 912                         ipath_write_creg(dev->dd, crp->cr_psstart,
 913                                          dev->pma_sample_start);
 914                 } else
 915                         dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
 916         }
 917         spin_unlock_irqrestore(&dev->pending_lock, flags);
 918
 919         ret = recv_pma_get_portsamplescontrol(pmp, ibdev, port);
 920
 921 bail:
 922         return ret;
 923 }
 924
 925 static u64 get_counter(struct ipath_ibdev *dev,
 926                        struct ipath_cregs const *crp,
 927                        __be16 sel)
 928 {
 929         u64 ret;
 930
 931         switch (sel) {
 932         case IB_PMA_PORT_XMIT_DATA:
 933                 ret = (crp->cr_psxmitdatacount) ?
 934                         ipath_read_creg32(dev->dd, crp->cr_psxmitdatacount) :
 935                         dev->ipath_sword;
 936                 break;
 937         case IB_PMA_PORT_RCV_DATA:
 938                 ret = (crp->cr_psrcvdatacount) ?
 939                         ipath_read_creg32(dev->dd, crp->cr_psrcvdatacount) :
 940                         dev->ipath_rword;
 941                 break;
 942         case IB_PMA_PORT_XMIT_PKTS:
 943                 ret = (crp->cr_psxmitpktscount) ?
 944                         ipath_read_creg32(dev->dd, crp->cr_psxmitpktscount) :
 945                         dev->ipath_spkts;
 946                 break;
 947         case IB_PMA_PORT_RCV_PKTS:
 948                 ret = (crp->cr_psrcvpktscount) ?
 949                         ipath_read_creg32(dev->dd, crp->cr_psrcvpktscount) :
 950                         dev->ipath_rpkts;
 951                 break;
 952         case IB_PMA_PORT_XMIT_WAIT:
 953                 ret = (crp->cr_psxmitwaitcount) ?
 954                         ipath_read_creg32(dev->dd, crp->cr_psxmitwaitcount) :
 955                         dev->ipath_xmit_wait;
 956                 break;
 957         default:
 958                 ret = 0;
 959         }
 960
 961         return ret;
 962 }
 963
 964 static int recv_pma_get_portsamplesresult(struct ib_pma_mad *pmp,
 965                                           struct ib_device *ibdev)
 966 {
 967         struct ib_pma_portsamplesresult *p =
 968                 (struct ib_pma_portsamplesresult *)pmp->data;
 969         struct ipath_ibdev *dev = to_idev(ibdev);
 970         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 971         u8 status;
 972         int i;
 973
 974         memset(pmp->data, 0, sizeof(pmp->data));
 975         p->tag = cpu_to_be16(dev->pma_tag);
 976         if (crp->cr_psstat)
 977                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
 978         else
 979                 status = dev->pma_sample_status;
 980         p->sample_status = cpu_to_be16(status);
 981         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
 982                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
 983                     cpu_to_be32(
 984                         get_counter(dev, crp, dev->pma_counter_select[i]));
 985
 986         return reply((struct ib_smp *) pmp);
 987 }
 988
 989 static int recv_pma_get_portsamplesresult_ext(struct ib_pma_mad *pmp,
 990                                               struct ib_device *ibdev)
 991 {
 992         struct ib_pma_portsamplesresult_ext *p =
 993                 (struct ib_pma_portsamplesresult_ext *)pmp->data;
 994         struct ipath_ibdev *dev = to_idev(ibdev);
 995         struct ipath_cregs const *crp = dev->dd->ipath_cregs;
 996         u8 status;
 997         int i;
 998
 999         memset(pmp->data, 0, sizeof(pmp->data));
1000         p->tag = cpu_to_be16(dev->pma_tag);
1001         if (crp->cr_psstat)
1002                 status = ipath_read_creg32(dev->dd, crp->cr_psstat);
1003         else
1004                 status = dev->pma_sample_status;
1005         p->sample_status = cpu_to_be16(status);
1006         /* 64 bits */
1007         p->extended_width = cpu_to_be32(0x80000000);
1008         for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
1009                 p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
1010                     cpu_to_be64(
1011                         get_counter(dev, crp, dev->pma_counter_select[i]));
1012
1013         return reply((struct ib_smp *) pmp);
1014 }
1015
1016 static int recv_pma_get_portcounters(struct ib_pma_mad *pmp,
1017                                      struct ib_device *ibdev, u8 port)
1018 {
1019         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1020                 pmp->data;
1021         struct ipath_ibdev *dev = to_idev(ibdev);
1022         struct ipath_verbs_counters cntrs;
1023         u8 port_select = p->port_select;
1024
1025         ipath_get_counters(dev->dd, &cntrs);
1026
1027         /* Adjust counters for any resets done. */
1028         cntrs.symbol_error_counter -= dev->z_symbol_error_counter;
1029         cntrs.link_error_recovery_counter -=
1030                 dev->z_link_error_recovery_counter;
1031         cntrs.link_downed_counter -= dev->z_link_downed_counter;
1032         cntrs.port_rcv_errors += dev->rcv_errors;
1033         cntrs.port_rcv_errors -= dev->z_port_rcv_errors;
1034         cntrs.port_rcv_remphys_errors -= dev->z_port_rcv_remphys_errors;
1035         cntrs.port_xmit_discards -= dev->z_port_xmit_discards;
1036         cntrs.port_xmit_data -= dev->z_port_xmit_data;
1037         cntrs.port_rcv_data -= dev->z_port_rcv_data;
1038         cntrs.port_xmit_packets -= dev->z_port_xmit_packets;
1039         cntrs.port_rcv_packets -= dev->z_port_rcv_packets;
1040         cntrs.local_link_integrity_errors -=
1041                 dev->z_local_link_integrity_errors;
1042         cntrs.excessive_buffer_overrun_errors -=
1043                 dev->z_excessive_buffer_overrun_errors;
1044         cntrs.vl15_dropped -= dev->z_vl15_dropped;
1045         cntrs.vl15_dropped += dev->n_vl15_dropped;
1046
1047         memset(pmp->data, 0, sizeof(pmp->data));
1048
1049         p->port_select = port_select;
1050         if (pmp->mad_hdr.attr_mod != 0 ||
1051             (port_select != port && port_select != 0xFF))
1052                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1053
1054         if (cntrs.symbol_error_counter > 0xFFFFUL)
1055                 p->symbol_error_counter = cpu_to_be16(0xFFFF);
1056         else
1057                 p->symbol_error_counter =
1058                         cpu_to_be16((u16)cntrs.symbol_error_counter);
1059         if (cntrs.link_error_recovery_counter > 0xFFUL)
1060                 p->link_error_recovery_counter = 0xFF;
1061         else
1062                 p->link_error_recovery_counter =
1063                         (u8)cntrs.link_error_recovery_counter;
1064         if (cntrs.link_downed_counter > 0xFFUL)
1065                 p->link_downed_counter = 0xFF;
1066         else
1067                 p->link_downed_counter = (u8)cntrs.link_downed_counter;
1068         if (cntrs.port_rcv_errors > 0xFFFFUL)
1069                 p->port_rcv_errors = cpu_to_be16(0xFFFF);
1070         else
1071                 p->port_rcv_errors =
1072                         cpu_to_be16((u16) cntrs.port_rcv_errors);
1073         if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
1074                 p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
1075         else
1076                 p->port_rcv_remphys_errors =
1077                         cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
1078         if (cntrs.port_xmit_discards > 0xFFFFUL)
1079                 p->port_xmit_discards = cpu_to_be16(0xFFFF);
1080         else
1081                 p->port_xmit_discards =
1082                         cpu_to_be16((u16)cntrs.port_xmit_discards);
1083         if (cntrs.local_link_integrity_errors > 0xFUL)
1084                 cntrs.local_link_integrity_errors = 0xFUL;
1085         if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
1086                 cntrs.excessive_buffer_overrun_errors = 0xFUL;
1087         p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
1088                 cntrs.excessive_buffer_overrun_errors;
1089         if (cntrs.vl15_dropped > 0xFFFFUL)
1090                 p->vl15_dropped = cpu_to_be16(0xFFFF);
1091         else
1092                 p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
1093         if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
1094                 p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
1095         else
1096                 p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
1097         if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
1098                 p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
1099         else
1100                 p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
1101         if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
1102                 p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
1103         else
1104                 p->port_xmit_packets =
1105                         cpu_to_be32((u32)cntrs.port_xmit_packets);
1106         if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
1107                 p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
1108         else
1109                 p->port_rcv_packets =
1110                         cpu_to_be32((u32) cntrs.port_rcv_packets);
1111
1112         return reply((struct ib_smp *) pmp);
1113 }
1114
1115 static int recv_pma_get_portcounters_ext(struct ib_pma_mad *pmp,
1116                                          struct ib_device *ibdev, u8 port)
1117 {
1118         struct ib_pma_portcounters_ext *p =
1119                 (struct ib_pma_portcounters_ext *)pmp->data;
1120         struct ipath_ibdev *dev = to_idev(ibdev);
1121         u64 swords, rwords, spkts, rpkts, xwait;
1122         u8 port_select = p->port_select;
1123
1124         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1125                                 &rpkts, &xwait);
1126
1127         /* Adjust counters for any resets done. */
1128         swords -= dev->z_port_xmit_data;
1129         rwords -= dev->z_port_rcv_data;
1130         spkts -= dev->z_port_xmit_packets;
1131         rpkts -= dev->z_port_rcv_packets;
1132
1133         memset(pmp->data, 0, sizeof(pmp->data));
1134
1135         p->port_select = port_select;
1136         if (pmp->mad_hdr.attr_mod != 0 ||
1137             (port_select != port && port_select != 0xFF))
1138                 pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
1139
1140         p->port_xmit_data = cpu_to_be64(swords);
1141         p->port_rcv_data = cpu_to_be64(rwords);
1142         p->port_xmit_packets = cpu_to_be64(spkts);
1143         p->port_rcv_packets = cpu_to_be64(rpkts);
1144         p->port_unicast_xmit_packets = cpu_to_be64(dev->n_unicast_xmit);
1145         p->port_unicast_rcv_packets = cpu_to_be64(dev->n_unicast_rcv);
1146         p->port_multicast_xmit_packets = cpu_to_be64(dev->n_multicast_xmit);
1147         p->port_multicast_rcv_packets = cpu_to_be64(dev->n_multicast_rcv);
1148
1149         return reply((struct ib_smp *) pmp);
1150 }
1151
1152 static int recv_pma_set_portcounters(struct ib_pma_mad *pmp,
1153                                      struct ib_device *ibdev, u8 port)
1154 {
1155         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1156                 pmp->data;
1157         struct ipath_ibdev *dev = to_idev(ibdev);
1158         struct ipath_verbs_counters cntrs;
1159
1160         /*
1161          * Since the HW doesn't support clearing counters, we save the
1162          * current count and subtract it from future responses.
1163          */
1164         ipath_get_counters(dev->dd, &cntrs);
1165
1166         if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
1167                 dev->z_symbol_error_counter = cntrs.symbol_error_counter;
1168
1169         if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
1170                 dev->z_link_error_recovery_counter =
1171                         cntrs.link_error_recovery_counter;
1172
1173         if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
1174                 dev->z_link_downed_counter = cntrs.link_downed_counter;
1175
1176         if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
1177                 dev->z_port_rcv_errors =
1178                         cntrs.port_rcv_errors + dev->rcv_errors;
1179
1180         if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
1181                 dev->z_port_rcv_remphys_errors =
1182                         cntrs.port_rcv_remphys_errors;
1183
1184         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
1185                 dev->z_port_xmit_discards = cntrs.port_xmit_discards;
1186
1187         if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
1188                 dev->z_local_link_integrity_errors =
1189                         cntrs.local_link_integrity_errors;
1190
1191         if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
1192                 dev->z_excessive_buffer_overrun_errors =
1193                         cntrs.excessive_buffer_overrun_errors;
1194
1195         if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
1196                 dev->n_vl15_dropped = 0;
1197                 dev->z_vl15_dropped = cntrs.vl15_dropped;
1198         }
1199
1200         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
1201                 dev->z_port_xmit_data = cntrs.port_xmit_data;
1202
1203         if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
1204                 dev->z_port_rcv_data = cntrs.port_rcv_data;
1205
1206         if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
1207                 dev->z_port_xmit_packets = cntrs.port_xmit_packets;
1208
1209         if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
1210                 dev->z_port_rcv_packets = cntrs.port_rcv_packets;
1211
1212         return recv_pma_get_portcounters(pmp, ibdev, port);
1213 }
1214
1215 static int recv_pma_set_portcounters_ext(struct ib_pma_mad *pmp,
1216                                          struct ib_device *ibdev, u8 port)
1217 {
1218         struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
1219                 pmp->data;
1220         struct ipath_ibdev *dev = to_idev(ibdev);
1221         u64 swords, rwords, spkts, rpkts, xwait;
1222
1223         ipath_snapshot_counters(dev->dd, &swords, &rwords, &spkts,
1224                                 &rpkts, &xwait);
1225
1226         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
1227                 dev->z_port_xmit_data = swords;
1228
1229         if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
1230                 dev->z_port_rcv_data = rwords;
1231
1232         if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
1233                 dev->z_port_xmit_packets = spkts;
1234
1235         if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
1236                 dev->z_port_rcv_packets = rpkts;
1237
1238         if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
1239                 dev->n_unicast_xmit = 0;
1240
1241         if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
1242                 dev->n_unicast_rcv = 0;
1243
1244         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
1245                 dev->n_multicast_xmit = 0;
1246
1247         if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
1248                 dev->n_multicast_rcv = 0;
1249
1250         return recv_pma_get_portcounters_ext(pmp, ibdev, port);
1251 }
1252
1253 static int process_subn(struct ib_device *ibdev, int mad_flags,
1254                         u8 port_num, struct ib_mad *in_mad,
1255                         struct ib_mad *out_mad)
1256 {
1257         struct ib_smp *smp = (struct ib_smp *)out_mad;
1258         struct ipath_ibdev *dev = to_idev(ibdev);
1259         int ret;
1260
1261         *out_mad = *in_mad;
1262         if (smp->class_version != 1) {
1263                 smp->status |= IB_SMP_UNSUP_VERSION;
1264                 ret = reply(smp);
1265                 goto bail;
1266         }
1267
1268         /* Is the mkey in the process of expiring? */
1269         if (dev->mkey_lease_timeout &&
1270             time_after_eq(jiffies, dev->mkey_lease_timeout)) {
1271                 /* Clear timeout and mkey protection field. */
1272                 dev->mkey_lease_timeout = 0;
1273                 dev->mkeyprot = 0;
1274         }
1275
1276         /*
1277          * M_Key checking depends on
1278          * Portinfo:M_Key_protect_bits
1279          */
1280         if ((mad_flags & IB_MAD_IGNORE_MKEY) == 0 && dev->mkey != 0 &&
1281             dev->mkey != smp->mkey &&
1282             (smp->method == IB_MGMT_METHOD_SET ||
1283              (smp->method == IB_MGMT_METHOD_GET &&
1284               dev->mkeyprot >= 2))) {
1285                 if (dev->mkey_violations != 0xFFFF)
1286                         ++dev->mkey_violations;
1287                 if (dev->mkey_lease_timeout ||
1288                     dev->mkey_lease_period == 0) {
1289                         ret = IB_MAD_RESULT_SUCCESS |
1290                                 IB_MAD_RESULT_CONSUMED;
1291                         goto bail;
1292                 }
1293                 dev->mkey_lease_timeout = jiffies +
1294                         dev->mkey_lease_period * HZ;
1295                 /* Future: Generate a trap notice. */
1296                 ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
1297                 goto bail;
1298         } else if (dev->mkey_lease_timeout)
1299                 dev->mkey_lease_timeout = 0;
1300
1301         switch (smp->method) {
1302         case IB_MGMT_METHOD_GET:
1303                 switch (smp->attr_id) {
1304                 case IB_SMP_ATTR_NODE_DESC:
1305                         ret = recv_subn_get_nodedescription(smp, ibdev);
1306                         goto bail;
1307                 case IB_SMP_ATTR_NODE_INFO:
1308                         ret = recv_subn_get_nodeinfo(smp, ibdev, port_num);
1309                         goto bail;
1310                 case IB_SMP_ATTR_GUID_INFO:
1311                         ret = recv_subn_get_guidinfo(smp, ibdev);
1312                         goto bail;
1313                 case IB_SMP_ATTR_PORT_INFO:
1314                         ret = recv_subn_get_portinfo(smp, ibdev, port_num);
1315                         goto bail;
1316                 case IB_SMP_ATTR_PKEY_TABLE:
1317                         ret = recv_subn_get_pkeytable(smp, ibdev);
1318                         goto bail;
1319                 case IB_SMP_ATTR_SM_INFO:
1320                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1321                                 ret = IB_MAD_RESULT_SUCCESS |
1322                                         IB_MAD_RESULT_CONSUMED;
1323                                 goto bail;
1324                         }
1325                         if (dev->port_cap_flags & IB_PORT_SM) {
1326                                 ret = IB_MAD_RESULT_SUCCESS;
1327                                 goto bail;
1328                         }
1329                         /* FALLTHROUGH */
1330                 default:
1331                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1332                         ret = reply(smp);
1333                         goto bail;
1334                 }
1335
1336         case IB_MGMT_METHOD_SET:
1337                 switch (smp->attr_id) {
1338                 case IB_SMP_ATTR_GUID_INFO:
1339                         ret = recv_subn_set_guidinfo(smp, ibdev);
1340                         goto bail;
1341                 case IB_SMP_ATTR_PORT_INFO:
1342                         ret = recv_subn_set_portinfo(smp, ibdev, port_num);
1343                         goto bail;
1344                 case IB_SMP_ATTR_PKEY_TABLE:
1345                         ret = recv_subn_set_pkeytable(smp, ibdev);
1346                         goto bail;
1347                 case IB_SMP_ATTR_SM_INFO:
1348                         if (dev->port_cap_flags & IB_PORT_SM_DISABLED) {
1349                                 ret = IB_MAD_RESULT_SUCCESS |
1350                                         IB_MAD_RESULT_CONSUMED;
1351                                 goto bail;
1352                         }
1353                         if (dev->port_cap_flags & IB_PORT_SM) {
1354                                 ret = IB_MAD_RESULT_SUCCESS;
1355                                 goto bail;
1356                         }
1357                         /* FALLTHROUGH */
1358                 default:
1359                         smp->status |= IB_SMP_UNSUP_METH_ATTR;
1360                         ret = reply(smp);
1361                         goto bail;
1362                 }
1363
1364         case IB_MGMT_METHOD_TRAP:
1365         case IB_MGMT_METHOD_REPORT:
1366         case IB_MGMT_METHOD_REPORT_RESP:
1367         case IB_MGMT_METHOD_TRAP_REPRESS:
1368         case IB_MGMT_METHOD_GET_RESP:
1369                 /*
1370                  * The ib_mad module will call us to process responses
1371                  * before checking for other consumers.
1372                  * Just tell the caller to process it normally.
1373                  */
1374                 ret = IB_MAD_RESULT_SUCCESS;
1375                 goto bail;
1376         default:
1377                 smp->status |= IB_SMP_UNSUP_METHOD;
1378                 ret = reply(smp);
1379         }
1380
1381 bail:
1382         return ret;
1383 }
1384
1385 static int process_perf(struct ib_device *ibdev, u8 port_num,
1386                         struct ib_mad *in_mad,
1387                         struct ib_mad *out_mad)
1388 {
1389         struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
1390         int ret;
1391
1392         *out_mad = *in_mad;
1393         if (pmp->mad_hdr.class_version != 1) {
1394                 pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
1395                 ret = reply((struct ib_smp *) pmp);
1396                 goto bail;
1397         }
1398
1399         switch (pmp->mad_hdr.method) {
1400         case IB_MGMT_METHOD_GET:
1401                 switch (pmp->mad_hdr.attr_id) {
1402                 case IB_PMA_CLASS_PORT_INFO:
1403                         ret = recv_pma_get_classportinfo(pmp);
1404                         goto bail;
1405                 case IB_PMA_PORT_SAMPLES_CONTROL:
1406                         ret = recv_pma_get_portsamplescontrol(pmp, ibdev,
1407                                                               port_num);
1408                         goto bail;
1409                 case IB_PMA_PORT_SAMPLES_RESULT:
1410                         ret = recv_pma_get_portsamplesresult(pmp, ibdev);
1411                         goto bail;
1412                 case IB_PMA_PORT_SAMPLES_RESULT_EXT:
1413                         ret = recv_pma_get_portsamplesresult_ext(pmp,
1414                                                                  ibdev);
1415                         goto bail;
1416                 case IB_PMA_PORT_COUNTERS:
1417                         ret = recv_pma_get_portcounters(pmp, ibdev,
1418                                                         port_num);
1419                         goto bail;
1420                 case IB_PMA_PORT_COUNTERS_EXT:
1421                         ret = recv_pma_get_portcounters_ext(pmp, ibdev,
1422                                                             port_num);
1423                         goto bail;
1424                 default:
1425                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
1426                         ret = reply((struct ib_smp *) pmp);
1427                         goto bail;
1428                 }
1429
1430         case IB_MGMT_METHOD_SET:
1431                 switch (pmp->mad_hdr.attr_id) {
1432                 case IB_PMA_PORT_SAMPLES_CONTROL:
1433                         ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
1434                                                               port_num);
1435                         goto bail;
1436                 case IB_PMA_PORT_COUNTERS:
1437                         ret = recv_pma_set_portcounters(pmp, ibdev,
1438                                                         port_num);
1439                         goto bail;
1440                 case IB_PMA_PORT_COUNTERS_EXT:
1441                         ret = recv_pma_set_portcounters_ext(pmp, ibdev,
1442                                                             port_num);
1443                         goto bail;
1444                 default:
1445                         pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
1446                         ret = reply((struct ib_smp *) pmp);
1447                         goto bail;
1448                 }
1449
1450         case IB_MGMT_METHOD_GET_RESP:
1451                 /*
1452                  * The ib_mad module will call us to process responses
1453                  * before checking for other consumers.
1454                  * Just tell the caller to process it normally.
1455                  */
1456                 ret = IB_MAD_RESULT_SUCCESS;
1457                 goto bail;
1458         default:
1459                 pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
1460                 ret = reply((struct ib_smp *) pmp);
1461         }
1462
1463 bail:
1464         return ret;
1465 }
1466
1467 /**
1468  * ipath_process_mad - process an incoming MAD packet
1469  * @ibdev: the infiniband device this packet came in on
1470  * @mad_flags: MAD flags
1471  * @port_num: the port number this packet came in on
1472  * @in_wc: the work completion entry for this packet
1473  * @in_grh: the global route header for this packet
1474  * @in_mad: the incoming MAD
1475  * @out_mad: any outgoing MAD reply
1476  *
1477  * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
1478  * interested in processing.
1479  *
1480  * Note that the verbs framework has already done the MAD sanity checks,
1481  * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
1482  * MADs.
1483  *
1484  * This is called by the ib_mad module.
1485  */
1486 int ipath_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1487                       struct ib_wc *in_wc, struct ib_grh *in_grh,
1488                       struct ib_mad *in_mad, struct ib_mad *out_mad)
1489 {
1490         int ret;
1491
1492         switch (in_mad->mad_hdr.mgmt_class) {
1493         case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
1494         case IB_MGMT_CLASS_SUBN_LID_ROUTED:
1495                 ret = process_subn(ibdev, mad_flags, port_num,
1496                                    in_mad, out_mad);
1497                 goto bail;
1498         case IB_MGMT_CLASS_PERF_MGMT:
1499                 ret = process_perf(ibdev, port_num, in_mad, out_mad);
1500                 goto bail;
1501         default:
1502                 ret = IB_MAD_RESULT_SUCCESS;
1503         }
1504
1505 bail:
1506         return ret;
1507 }