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

   1 /*
   2  * Copyright (c) 2006 QLogic, Inc. All rights reserved.
   3  * Copyright (c) 2003, 2004, 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 /*
  35  * These are the routines used by layered drivers, currently just the
  36  * layered ethernet driver and verbs layer.
  37  */
  38
  39 #include <linux/io.h>
  40 #include <linux/pci.h>
  41 #include <asm/byteorder.h>
  42
  43 #include "ipath_kernel.h"
  44 #include "ipath_layer.h"
  45 #include "ipath_common.h"
  46
  47 /* Acquire before ipath_devs_lock. */
  48 static DEFINE_MUTEX(ipath_layer_mutex);
  49
  50 static int ipath_verbs_registered;
  51
  52 u16 ipath_layer_rcv_opcode;
  53
  54 static int (*layer_intr)(void *, u32);
  55 static int (*layer_rcv)(void *, void *, struct sk_buff *);
  56 static int (*layer_rcv_lid)(void *, void *);
  57 static int (*verbs_piobufavail)(void *);
  58 static void (*verbs_rcv)(void *, void *, void *, u32);
  59
  60 static void *(*layer_add_one)(int, struct ipath_devdata *);
  61 static void (*layer_remove_one)(void *);
  62 static void *(*verbs_add_one)(int, struct ipath_devdata *);
  63 static void (*verbs_remove_one)(void *);
  64 static void (*verbs_timer_cb)(void *);
  65
  66 int __ipath_layer_intr(struct ipath_devdata *dd, u32 arg)
  67 {
  68         int ret = -ENODEV;
  69
  70         if (dd->ipath_layer.l_arg && layer_intr)
  71                 ret = layer_intr(dd->ipath_layer.l_arg, arg);
  72
  73         return ret;
  74 }
  75
  76 int ipath_layer_intr(struct ipath_devdata *dd, u32 arg)
  77 {
  78         int ret;
  79
  80         mutex_lock(&ipath_layer_mutex);
  81
  82         ret = __ipath_layer_intr(dd, arg);
  83
  84         mutex_unlock(&ipath_layer_mutex);
  85
  86         return ret;
  87 }
  88
  89 int __ipath_layer_rcv(struct ipath_devdata *dd, void *hdr,
  90                       struct sk_buff *skb)
  91 {
  92         int ret = -ENODEV;
  93
  94         if (dd->ipath_layer.l_arg && layer_rcv)
  95                 ret = layer_rcv(dd->ipath_layer.l_arg, hdr, skb);
  96
  97         return ret;
  98 }
  99
 100 int __ipath_layer_rcv_lid(struct ipath_devdata *dd, void *hdr)
 101 {
 102         int ret = -ENODEV;
 103
 104         if (dd->ipath_layer.l_arg && layer_rcv_lid)
 105                 ret = layer_rcv_lid(dd->ipath_layer.l_arg, hdr);
 106
 107         return ret;
 108 }
 109
 110 int __ipath_verbs_piobufavail(struct ipath_devdata *dd)
 111 {
 112         int ret = -ENODEV;
 113
 114         if (dd->verbs_layer.l_arg && verbs_piobufavail)
 115                 ret = verbs_piobufavail(dd->verbs_layer.l_arg);
 116
 117         return ret;
 118 }
 119
 120 int __ipath_verbs_rcv(struct ipath_devdata *dd, void *rc, void *ebuf,
 121                       u32 tlen)
 122 {
 123         int ret = -ENODEV;
 124
 125         if (dd->verbs_layer.l_arg && verbs_rcv) {
 126                 verbs_rcv(dd->verbs_layer.l_arg, rc, ebuf, tlen);
 127                 ret = 0;
 128         }
 129
 130         return ret;
 131 }
 132
 133 int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 newstate)
 134 {
 135         u32 lstate;
 136         int ret;
 137
 138         switch (newstate) {
 139         case IPATH_IB_LINKDOWN:
 140                 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL <<
 141                                     INFINIPATH_IBCC_LINKINITCMD_SHIFT);
 142                 /* don't wait */
 143                 ret = 0;
 144                 goto bail;
 145
 146         case IPATH_IB_LINKDOWN_SLEEP:
 147                 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_SLEEP <<
 148                                     INFINIPATH_IBCC_LINKINITCMD_SHIFT);
 149                 /* don't wait */
 150                 ret = 0;
 151                 goto bail;
 152
 153         case IPATH_IB_LINKDOWN_DISABLE:
 154                 ipath_set_ib_lstate(dd,
 155                                     INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
 156                                     INFINIPATH_IBCC_LINKINITCMD_SHIFT);
 157                 /* don't wait */
 158                 ret = 0;
 159                 goto bail;
 160
 161         case IPATH_IB_LINKINIT:
 162                 if (dd->ipath_flags & IPATH_LINKINIT) {
 163                         ret = 0;
 164                         goto bail;
 165                 }
 166                 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_INIT <<
 167                                     INFINIPATH_IBCC_LINKCMD_SHIFT);
 168                 lstate = IPATH_LINKINIT;
 169                 break;
 170
 171         case IPATH_IB_LINKARM:
 172                 if (dd->ipath_flags & IPATH_LINKARMED) {
 173                         ret = 0;
 174                         goto bail;
 175                 }
 176                 if (!(dd->ipath_flags &
 177                       (IPATH_LINKINIT | IPATH_LINKACTIVE))) {
 178                         ret = -EINVAL;
 179                         goto bail;
 180                 }
 181                 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED <<
 182                                     INFINIPATH_IBCC_LINKCMD_SHIFT);
 183                 /*
 184                  * Since the port can transition to ACTIVE by receiving
 185                  * a non VL 15 packet, wait for either state.
 186                  */
 187                 lstate = IPATH_LINKARMED | IPATH_LINKACTIVE;
 188                 break;
 189
 190         case IPATH_IB_LINKACTIVE:
 191                 if (dd->ipath_flags & IPATH_LINKACTIVE) {
 192                         ret = 0;
 193                         goto bail;
 194                 }
 195                 if (!(dd->ipath_flags & IPATH_LINKARMED)) {
 196                         ret = -EINVAL;
 197                         goto bail;
 198                 }
 199                 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE <<
 200                                     INFINIPATH_IBCC_LINKCMD_SHIFT);
 201                 lstate = IPATH_LINKACTIVE;
 202                 break;
 203
 204         default:
 205                 ipath_dbg("Invalid linkstate 0x%x requested\n", newstate);
 206                 ret = -EINVAL;
 207                 goto bail;
 208         }
 209         ret = ipath_wait_linkstate(dd, lstate, 2000);
 210
 211 bail:
 212         return ret;
 213 }
 214
 215 EXPORT_SYMBOL_GPL(ipath_layer_set_linkstate);
 216
 217 /**
 218  * ipath_layer_set_mtu - set the MTU
 219  * @dd: the infinipath device
 220  * @arg: the new MTU
 221  *
 222  * we can handle "any" incoming size, the issue here is whether we
 223  * need to restrict our outgoing size.   For now, we don't do any
 224  * sanity checking on this, and we don't deal with what happens to
 225  * programs that are already running when the size changes.
 226  * NOTE: changing the MTU will usually cause the IBC to go back to
 227  * link initialize (IPATH_IBSTATE_INIT) state...
 228  */
 229 int ipath_layer_set_mtu(struct ipath_devdata *dd, u16 arg)
 230 {
 231         u32 piosize;
 232         int changed = 0;
 233         int ret;
 234
 235         /*
 236          * mtu is IB data payload max.  It's the largest power of 2 less
 237          * than piosize (or even larger, since it only really controls the
 238          * largest we can receive; we can send the max of the mtu and
 239          * piosize).  We check that it's one of the valid IB sizes.
 240          */
 241         if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
 242             arg != 4096) {
 243                 ipath_dbg("Trying to set invalid mtu %u, failing\n", arg);
 244                 ret = -EINVAL;
 245                 goto bail;
 246         }
 247         if (dd->ipath_ibmtu == arg) {
 248                 ret = 0;        /* same as current */
 249                 goto bail;
 250         }
 251
 252         piosize = dd->ipath_ibmaxlen;
 253         dd->ipath_ibmtu = arg;
 254
 255         if (arg >= (piosize - IPATH_PIO_MAXIBHDR)) {
 256                 /* Only if it's not the initial value (or reset to it) */
 257                 if (piosize != dd->ipath_init_ibmaxlen) {
 258                         dd->ipath_ibmaxlen = piosize;
 259                         changed = 1;
 260                 }
 261         } else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) {
 262                 piosize = arg + IPATH_PIO_MAXIBHDR;
 263                 ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x "
 264                            "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize,
 265                            arg);
 266                 dd->ipath_ibmaxlen = piosize;
 267                 changed = 1;
 268         }
 269
 270         if (changed) {
 271                 /*
 272                  * set the IBC maxpktlength to the size of our pio
 273                  * buffers in words
 274                  */
 275                 u64 ibc = dd->ipath_ibcctrl;
 276                 ibc &= ~(INFINIPATH_IBCC_MAXPKTLEN_MASK <<
 277                          INFINIPATH_IBCC_MAXPKTLEN_SHIFT);
 278
 279                 piosize = piosize - 2 * sizeof(u32);    /* ignore pbc */
 280                 dd->ipath_ibmaxlen = piosize;
 281                 piosize /= sizeof(u32); /* in words */
 282                 /*
 283                  * for ICRC, which we only send in diag test pkt mode, and
 284                  * we don't need to worry about that for mtu
 285                  */
 286                 piosize += 1;
 287
 288                 ibc |= piosize << INFINIPATH_IBCC_MAXPKTLEN_SHIFT;
 289                 dd->ipath_ibcctrl = ibc;
 290                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
 291                                  dd->ipath_ibcctrl);
 292                 dd->ipath_f_tidtemplate(dd);
 293         }
 294
 295         ret = 0;
 296
 297 bail:
 298         return ret;
 299 }
 300
 301 EXPORT_SYMBOL_GPL(ipath_layer_set_mtu);
 302
 303 int ipath_set_lid(struct ipath_devdata *dd, u32 arg, u8 lmc)
 304 {
 305         dd->ipath_lid = arg;
 306         dd->ipath_lmc = lmc;
 307
 308         mutex_lock(&ipath_layer_mutex);
 309
 310         if (dd->ipath_layer.l_arg && layer_intr)
 311                 layer_intr(dd->ipath_layer.l_arg, IPATH_LAYER_INT_LID);
 312
 313         mutex_unlock(&ipath_layer_mutex);
 314
 315         return 0;
 316 }
 317
 318 EXPORT_SYMBOL_GPL(ipath_set_lid);
 319
 320 int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid)
 321 {
 322         /* XXX - need to inform anyone who cares this just happened. */
 323         dd->ipath_guid = guid;
 324         return 0;
 325 }
 326
 327 EXPORT_SYMBOL_GPL(ipath_layer_set_guid);
 328
 329 __be64 ipath_layer_get_guid(struct ipath_devdata *dd)
 330 {
 331         return dd->ipath_guid;
 332 }
 333
 334 EXPORT_SYMBOL_GPL(ipath_layer_get_guid);
 335
 336 u32 ipath_layer_get_nguid(struct ipath_devdata *dd)
 337 {
 338         return dd->ipath_nguid;
 339 }
 340
 341 EXPORT_SYMBOL_GPL(ipath_layer_get_nguid);
 342
 343 u32 ipath_layer_get_majrev(struct ipath_devdata *dd)
 344 {
 345         return dd->ipath_majrev;
 346 }
 347
 348 EXPORT_SYMBOL_GPL(ipath_layer_get_majrev);
 349
 350 u32 ipath_layer_get_minrev(struct ipath_devdata *dd)
 351 {
 352         return dd->ipath_minrev;
 353 }
 354
 355 EXPORT_SYMBOL_GPL(ipath_layer_get_minrev);
 356
 357 u32 ipath_layer_get_pcirev(struct ipath_devdata *dd)
 358 {
 359         return dd->ipath_pcirev;
 360 }
 361
 362 EXPORT_SYMBOL_GPL(ipath_layer_get_pcirev);
 363
 364 u32 ipath_layer_get_flags(struct ipath_devdata *dd)
 365 {
 366         return dd->ipath_flags;
 367 }
 368
 369 EXPORT_SYMBOL_GPL(ipath_layer_get_flags);
 370
 371 struct device *ipath_layer_get_device(struct ipath_devdata *dd)
 372 {
 373         return &dd->pcidev->dev;
 374 }
 375
 376 EXPORT_SYMBOL_GPL(ipath_layer_get_device);
 377
 378 u16 ipath_layer_get_deviceid(struct ipath_devdata *dd)
 379 {
 380         return dd->ipath_deviceid;
 381 }
 382
 383 EXPORT_SYMBOL_GPL(ipath_layer_get_deviceid);
 384
 385 u32 ipath_layer_get_vendorid(struct ipath_devdata *dd)
 386 {
 387         return dd->ipath_vendorid;
 388 }
 389
 390 EXPORT_SYMBOL_GPL(ipath_layer_get_vendorid);
 391
 392 u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd)
 393 {
 394         return dd->ipath_lastibcstat;
 395 }
 396
 397 EXPORT_SYMBOL_GPL(ipath_layer_get_lastibcstat);
 398
 399 u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd)
 400 {
 401         return dd->ipath_ibmtu;
 402 }
 403
 404 EXPORT_SYMBOL_GPL(ipath_layer_get_ibmtu);
 405
 406 void ipath_layer_add(struct ipath_devdata *dd)
 407 {
 408         mutex_lock(&ipath_layer_mutex);
 409
 410         if (layer_add_one)
 411                 dd->ipath_layer.l_arg =
 412                         layer_add_one(dd->ipath_unit, dd);
 413
 414         if (verbs_add_one)
 415                 dd->verbs_layer.l_arg =
 416                         verbs_add_one(dd->ipath_unit, dd);
 417
 418         mutex_unlock(&ipath_layer_mutex);
 419 }
 420
 421 void ipath_layer_remove(struct ipath_devdata *dd)
 422 {
 423         mutex_lock(&ipath_layer_mutex);
 424
 425         if (dd->ipath_layer.l_arg && layer_remove_one) {
 426                 layer_remove_one(dd->ipath_layer.l_arg);
 427                 dd->ipath_layer.l_arg = NULL;
 428         }
 429
 430         if (dd->verbs_layer.l_arg && verbs_remove_one) {
 431                 verbs_remove_one(dd->verbs_layer.l_arg);
 432                 dd->verbs_layer.l_arg = NULL;
 433         }
 434
 435         mutex_unlock(&ipath_layer_mutex);
 436 }
 437
 438 int ipath_layer_register(void *(*l_add)(int, struct ipath_devdata *),
 439                          void (*l_remove)(void *),
 440                          int (*l_intr)(void *, u32),
 441                          int (*l_rcv)(void *, void *, struct sk_buff *),
 442                          u16 l_rcv_opcode,
 443                          int (*l_rcv_lid)(void *, void *))
 444 {
 445         struct ipath_devdata *dd, *tmp;
 446         unsigned long flags;
 447
 448         mutex_lock(&ipath_layer_mutex);
 449
 450         layer_add_one = l_add;
 451         layer_remove_one = l_remove;
 452         layer_intr = l_intr;
 453         layer_rcv = l_rcv;
 454         layer_rcv_lid = l_rcv_lid;
 455         ipath_layer_rcv_opcode = l_rcv_opcode;
 456
 457         spin_lock_irqsave(&ipath_devs_lock, flags);
 458
 459         list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
 460                 if (!(dd->ipath_flags & IPATH_INITTED))
 461                         continue;
 462
 463                 if (dd->ipath_layer.l_arg)
 464                         continue;
 465
 466                 if (!(*dd->ipath_statusp & IPATH_STATUS_SMA))
 467                         *dd->ipath_statusp |= IPATH_STATUS_OIB_SMA;
 468
 469                 spin_unlock_irqrestore(&ipath_devs_lock, flags);
 470                 dd->ipath_layer.l_arg = l_add(dd->ipath_unit, dd);
 471                 spin_lock_irqsave(&ipath_devs_lock, flags);
 472         }
 473
 474         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 475         mutex_unlock(&ipath_layer_mutex);
 476
 477         return 0;
 478 }
 479
 480 EXPORT_SYMBOL_GPL(ipath_layer_register);
 481
 482 void ipath_layer_unregister(void)
 483 {
 484         struct ipath_devdata *dd, *tmp;
 485         unsigned long flags;
 486
 487         mutex_lock(&ipath_layer_mutex);
 488         spin_lock_irqsave(&ipath_devs_lock, flags);
 489
 490         list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
 491                 if (dd->ipath_layer.l_arg && layer_remove_one) {
 492                         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 493                         layer_remove_one(dd->ipath_layer.l_arg);
 494                         spin_lock_irqsave(&ipath_devs_lock, flags);
 495                         dd->ipath_layer.l_arg = NULL;
 496                 }
 497         }
 498
 499         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 500
 501         layer_add_one = NULL;
 502         layer_remove_one = NULL;
 503         layer_intr = NULL;
 504         layer_rcv = NULL;
 505         layer_rcv_lid = NULL;
 506
 507         mutex_unlock(&ipath_layer_mutex);
 508 }
 509
 510 EXPORT_SYMBOL_GPL(ipath_layer_unregister);
 511
 512 static void __ipath_verbs_timer(unsigned long arg)
 513 {
 514         struct ipath_devdata *dd = (struct ipath_devdata *) arg;
 515
 516         /*
 517          * If port 0 receive packet interrupts are not available, or
 518          * can be missed, poll the receive queue
 519          */
 520         if (dd->ipath_flags & IPATH_POLL_RX_INTR)
 521                 ipath_kreceive(dd);
 522
 523         /* Handle verbs layer timeouts. */
 524         if (dd->verbs_layer.l_arg && verbs_timer_cb)
 525                 verbs_timer_cb(dd->verbs_layer.l_arg);
 526
 527         mod_timer(&dd->verbs_layer.l_timer, jiffies + 1);
 528 }
 529
 530 /**
 531  * ipath_verbs_register - verbs layer registration
 532  * @l_piobufavail: callback for when PIO buffers become available
 533  * @l_rcv: callback for receiving a packet
 534  * @l_timer_cb: timer callback
 535  * @ipath_devdata: device data structure is put here
 536  */
 537 int ipath_verbs_register(void *(*l_add)(int, struct ipath_devdata *),
 538                          void (*l_remove)(void *arg),
 539                          int (*l_piobufavail) (void *arg),
 540                          void (*l_rcv) (void *arg, void *rhdr,
 541                                         void *data, u32 tlen),
 542                          void (*l_timer_cb) (void *arg))
 543 {
 544         struct ipath_devdata *dd, *tmp;
 545         unsigned long flags;
 546
 547         mutex_lock(&ipath_layer_mutex);
 548
 549         verbs_add_one = l_add;
 550         verbs_remove_one = l_remove;
 551         verbs_piobufavail = l_piobufavail;
 552         verbs_rcv = l_rcv;
 553         verbs_timer_cb = l_timer_cb;
 554
 555         spin_lock_irqsave(&ipath_devs_lock, flags);
 556
 557         list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
 558                 if (!(dd->ipath_flags & IPATH_INITTED))
 559                         continue;
 560
 561                 if (dd->verbs_layer.l_arg)
 562                         continue;
 563
 564                 spin_unlock_irqrestore(&ipath_devs_lock, flags);
 565                 dd->verbs_layer.l_arg = l_add(dd->ipath_unit, dd);
 566                 spin_lock_irqsave(&ipath_devs_lock, flags);
 567         }
 568
 569         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 570         mutex_unlock(&ipath_layer_mutex);
 571
 572         ipath_verbs_registered = 1;
 573
 574         return 0;
 575 }
 576
 577 EXPORT_SYMBOL_GPL(ipath_verbs_register);
 578
 579 void ipath_verbs_unregister(void)
 580 {
 581         struct ipath_devdata *dd, *tmp;
 582         unsigned long flags;
 583
 584         mutex_lock(&ipath_layer_mutex);
 585         spin_lock_irqsave(&ipath_devs_lock, flags);
 586
 587         list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
 588                 *dd->ipath_statusp &= ~IPATH_STATUS_OIB_SMA;
 589
 590                 if (dd->verbs_layer.l_arg && verbs_remove_one) {
 591                         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 592                         verbs_remove_one(dd->verbs_layer.l_arg);
 593                         spin_lock_irqsave(&ipath_devs_lock, flags);
 594                         dd->verbs_layer.l_arg = NULL;
 595                 }
 596         }
 597
 598         spin_unlock_irqrestore(&ipath_devs_lock, flags);
 599
 600         verbs_add_one = NULL;
 601         verbs_remove_one = NULL;
 602         verbs_piobufavail = NULL;
 603         verbs_rcv = NULL;
 604         verbs_timer_cb = NULL;
 605
 606         ipath_verbs_registered = 0;
 607
 608         mutex_unlock(&ipath_layer_mutex);
 609 }
 610
 611 EXPORT_SYMBOL_GPL(ipath_verbs_unregister);
 612
 613 int ipath_layer_open(struct ipath_devdata *dd, u32 * pktmax)
 614 {
 615         int ret;
 616         u32 intval = 0;
 617
 618         mutex_lock(&ipath_layer_mutex);
 619
 620         if (!dd->ipath_layer.l_arg) {
 621                 ret = -EINVAL;
 622                 goto bail;
 623         }
 624
 625         ret = ipath_setrcvhdrsize(dd, IPATH_HEADER_QUEUE_WORDS);
 626
 627         if (ret < 0)
 628                 goto bail;
 629
 630         *pktmax = dd->ipath_ibmaxlen;
 631
 632         if (*dd->ipath_statusp & IPATH_STATUS_IB_READY)
 633                 intval |= IPATH_LAYER_INT_IF_UP;
 634         if (dd->ipath_lid)
 635                 intval |= IPATH_LAYER_INT_LID;
 636         if (dd->ipath_mlid)
 637                 intval |= IPATH_LAYER_INT_BCAST;
 638         /*
 639          * do this on open, in case low level is already up and
 640          * just layered driver was reloaded, etc.
 641          */
 642         if (intval)
 643                 layer_intr(dd->ipath_layer.l_arg, intval);
 644
 645         ret = 0;
 646 bail:
 647         mutex_unlock(&ipath_layer_mutex);
 648
 649         return ret;
 650 }
 651
 652 EXPORT_SYMBOL_GPL(ipath_layer_open);
 653
 654 u16 ipath_layer_get_lid(struct ipath_devdata *dd)
 655 {
 656         return dd->ipath_lid;
 657 }
 658
 659 EXPORT_SYMBOL_GPL(ipath_layer_get_lid);
 660
 661 /**
 662  * ipath_layer_get_mac - get the MAC address
 663  * @dd: the infinipath device
 664  * @mac: the MAC is put here
 665  *
 666  * This is the EUID-64 OUI octets (top 3), then
 667  * skip the next 2 (which should both be zero or 0xff).
 668  * The returned MAC is in network order
 669  * mac points to at least 6 bytes of buffer
 670  * We assume that by the time the LID is set, that the GUID is as valid
 671  * as it's ever going to be, rather than adding yet another status bit.
 672  */
 673
 674 int ipath_layer_get_mac(struct ipath_devdata *dd, u8 * mac)
 675 {
 676         u8 *guid;
 677
 678         guid = (u8 *) &dd->ipath_guid;
 679
 680         mac[0] = guid[0];
 681         mac[1] = guid[1];
 682         mac[2] = guid[2];
 683         mac[3] = guid[5];
 684         mac[4] = guid[6];
 685         mac[5] = guid[7];
 686         if ((guid[3] || guid[4]) && !(guid[3] == 0xff && guid[4] == 0xff))
 687                 ipath_dbg("Warning, guid bytes 3 and 4 not 0 or 0xffff: "
 688                           "%x %x\n", guid[3], guid[4]);
 689         return 0;
 690 }
 691
 692 EXPORT_SYMBOL_GPL(ipath_layer_get_mac);
 693
 694 u16 ipath_layer_get_bcast(struct ipath_devdata *dd)
 695 {
 696         return dd->ipath_mlid;
 697 }
 698
 699 EXPORT_SYMBOL_GPL(ipath_layer_get_bcast);
 700
 701 u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd)
 702 {
 703         return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey);
 704 }
 705
 706 EXPORT_SYMBOL_GPL(ipath_layer_get_cr_errpkey);
 707
 708 static void update_sge(struct ipath_sge_state *ss, u32 length)
 709 {
 710         struct ipath_sge *sge = &ss->sge;
 711
 712         sge->vaddr += length;
 713         sge->length -= length;
 714         sge->sge_length -= length;
 715         if (sge->sge_length == 0) {
 716                 if (--ss->num_sge)
 717                         *sge = *ss->sg_list++;
 718         } else if (sge->length == 0 && sge->mr != NULL) {
 719                 if (++sge->n >= IPATH_SEGSZ) {
 720                         if (++sge->m >= sge->mr->mapsz)
 721                                 return;
 722                         sge->n = 0;
 723                 }
 724                 sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
 725                 sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
 726         }
 727 }
 728
 729 #ifdef __LITTLE_ENDIAN
 730 static inline u32 get_upper_bits(u32 data, u32 shift)
 731 {
 732         return data >> shift;
 733 }
 734
 735 static inline u32 set_upper_bits(u32 data, u32 shift)
 736 {
 737         return data << shift;
 738 }
 739
 740 static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
 741 {
 742         data <<= ((sizeof(u32) - n) * BITS_PER_BYTE);
 743         data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
 744         return data;
 745 }
 746 #else
 747 static inline u32 get_upper_bits(u32 data, u32 shift)
 748 {
 749         return data << shift;
 750 }
 751
 752 static inline u32 set_upper_bits(u32 data, u32 shift)
 753 {
 754         return data >> shift;
 755 }
 756
 757 static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
 758 {
 759         data >>= ((sizeof(u32) - n) * BITS_PER_BYTE);
 760         data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
 761         return data;
 762 }
 763 #endif
 764
 765 static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss,
 766                     u32 length)
 767 {
 768         u32 extra = 0;
 769         u32 data = 0;
 770         u32 last;
 771
 772         while (1) {
 773                 u32 len = ss->sge.length;
 774                 u32 off;
 775
 776                 BUG_ON(len == 0);
 777                 if (len > length)
 778                         len = length;
 779                 if (len > ss->sge.sge_length)
 780                         len = ss->sge.sge_length;
 781                 /* If the source address is not aligned, try to align it. */
 782                 off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1);
 783                 if (off) {
 784                         u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr &
 785                                             ~(sizeof(u32) - 1));
 786                         u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE);
 787                         u32 y;
 788
 789                         y = sizeof(u32) - off;
 790                         if (len > y)
 791                                 len = y;
 792                         if (len + extra >= sizeof(u32)) {
 793                                 data |= set_upper_bits(v, extra *
 794                                                        BITS_PER_BYTE);
 795                                 len = sizeof(u32) - extra;
 796                                 if (len == length) {
 797                                         last = data;
 798                                         break;
 799                                 }
 800                                 __raw_writel(data, piobuf);
 801                                 piobuf++;
 802                                 extra = 0;
 803                                 data = 0;
 804                         } else {
 805                                 /* Clear unused upper bytes */
 806                                 data |= clear_upper_bytes(v, len, extra);
 807                                 if (len == length) {
 808                                         last = data;
 809                                         break;
 810                                 }
 811                                 extra += len;
 812                         }
 813                 } else if (extra) {
 814                         /* Source address is aligned. */
 815                         u32 *addr = (u32 *) ss->sge.vaddr;
 816                         int shift = extra * BITS_PER_BYTE;
 817                         int ushift = 32 - shift;
 818                         u32 l = len;
 819
 820                         while (l >= sizeof(u32)) {
 821                                 u32 v = *addr;
 822
 823                                 data |= set_upper_bits(v, shift);
 824                                 __raw_writel(data, piobuf);
 825                                 data = get_upper_bits(v, ushift);
 826                                 piobuf++;
 827                                 addr++;
 828                                 l -= sizeof(u32);
 829                         }
 830                         /*
 831                          * We still have 'extra' number of bytes leftover.
 832                          */
 833                         if (l) {
 834                                 u32 v = *addr;
 835
 836                                 if (l + extra >= sizeof(u32)) {
 837                                         data |= set_upper_bits(v, shift);
 838                                         len -= l + extra - sizeof(u32);
 839                                         if (len == length) {
 840                                                 last = data;
 841                                                 break;
 842                                         }
 843                                         __raw_writel(data, piobuf);
 844                                         piobuf++;
 845                                         extra = 0;
 846                                         data = 0;
 847                                 } else {
 848                                         /* Clear unused upper bytes */
 849                                         data |= clear_upper_bytes(v, l,
 850                                                                   extra);
 851                                         if (len == length) {
 852                                                 last = data;
 853                                                 break;
 854                                         }
 855                                         extra += l;
 856                                 }
 857                         } else if (len == length) {
 858                                 last = data;
 859                                 break;
 860                         }
 861                 } else if (len == length) {
 862                         u32 w;
 863
 864                         /*
 865                          * Need to round up for the last dword in the
 866                          * packet.
 867                          */
 868                         w = (len + 3) >> 2;
 869                         __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1);
 870                         piobuf += w - 1;
 871                         last = ((u32 *) ss->sge.vaddr)[w - 1];
 872                         break;
 873                 } else {
 874                         u32 w = len >> 2;
 875
 876                         __iowrite32_copy(piobuf, ss->sge.vaddr, w);
 877                         piobuf += w;
 878
 879                         extra = len & (sizeof(u32) - 1);
 880                         if (extra) {
 881                                 u32 v = ((u32 *) ss->sge.vaddr)[w];
 882
 883                                 /* Clear unused upper bytes */
 884                                 data = clear_upper_bytes(v, extra, 0);
 885                         }
 886                 }
 887                 update_sge(ss, len);
 888                 length -= len;
 889         }
 890         /* Update address before sending packet. */
 891         update_sge(ss, length);
 892         /* must flush early everything before trigger word */
 893         ipath_flush_wc();
 894         __raw_writel(last, piobuf);
 895         /* be sure trigger word is written */
 896         ipath_flush_wc();
 897 }
 898
 899 /**
 900  * ipath_verbs_send - send a packet from the verbs layer
 901  * @dd: the infinipath device
 902  * @hdrwords: the number of words in the header
 903  * @hdr: the packet header
 904  * @len: the length of the packet in bytes
 905  * @ss: the SGE to send
 906  *
 907  * This is like ipath_sma_send_pkt() in that we need to be able to send
 908  * packets after the chip is initialized (MADs) but also like
 909  * ipath_layer_send_hdr() since its used by the verbs layer.
 910  */
 911 int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
 912                      u32 *hdr, u32 len, struct ipath_sge_state *ss)
 913 {
 914         u32 __iomem *piobuf;
 915         u32 plen;
 916         int ret;
 917
 918         /* +1 is for the qword padding of pbc */
 919         plen = hdrwords + ((len + 3) >> 2) + 1;
 920         if (unlikely((plen << 2) > dd->ipath_ibmaxlen)) {
 921                 ipath_dbg("packet len 0x%x too long, failing\n", plen);
 922                 ret = -EINVAL;
 923                 goto bail;
 924         }
 925
 926         /* Get a PIO buffer to use. */
 927         piobuf = ipath_getpiobuf(dd, NULL);
 928         if (unlikely(piobuf == NULL)) {
 929                 ret = -EBUSY;
 930                 goto bail;
 931         }
 932
 933         /*
 934          * Write len to control qword, no flags.
 935          * We have to flush after the PBC for correctness on some cpus
 936          * or WC buffer can be written out of order.
 937          */
 938         writeq(plen, piobuf);
 939         ipath_flush_wc();
 940         piobuf += 2;
 941         if (len == 0) {
 942                 /*
 943                  * If there is just the header portion, must flush before
 944                  * writing last word of header for correctness, and after
 945                  * the last header word (trigger word).
 946                  */
 947                 __iowrite32_copy(piobuf, hdr, hdrwords - 1);
 948                 ipath_flush_wc();
 949                 __raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1);
 950                 ipath_flush_wc();
 951                 ret = 0;
 952                 goto bail;
 953         }
 954
 955         __iowrite32_copy(piobuf, hdr, hdrwords);
 956         piobuf += hdrwords;
 957
 958         /* The common case is aligned and contained in one segment. */
 959         if (likely(ss->num_sge == 1 && len <= ss->sge.length &&
 960                    !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) {
 961                 u32 w;
 962                 u32 *addr = (u32 *) ss->sge.vaddr;
 963
 964                 /* Update address before sending packet. */
 965                 update_sge(ss, len);
 966                 /* Need to round up for the last dword in the packet. */
 967                 w = (len + 3) >> 2;
 968                 __iowrite32_copy(piobuf, addr, w - 1);
 969                 /* must flush early everything before trigger word */
 970                 ipath_flush_wc();
 971                 __raw_writel(addr[w - 1], piobuf + w - 1);
 972                 /* be sure trigger word is written */
 973                 ipath_flush_wc();
 974                 ret = 0;
 975                 goto bail;
 976         }
 977         copy_io(piobuf, ss, len);
 978         ret = 0;
 979
 980 bail:
 981         return ret;
 982 }
 983
 984 EXPORT_SYMBOL_GPL(ipath_verbs_send);
 985
 986 int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords,
 987                                   u64 *rwords, u64 *spkts, u64 *rpkts,
 988                                   u64 *xmit_wait)
 989 {
 990         int ret;
 991
 992         if (!(dd->ipath_flags & IPATH_INITTED)) {
 993                 /* no hardware, freeze, etc. */
 994                 ipath_dbg("unit %u not usable\n", dd->ipath_unit);
 995                 ret = -EINVAL;
 996                 goto bail;
 997         }
 998         *swords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
 999         *rwords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
1000         *spkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
1001         *rpkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
1002         *xmit_wait = ipath_snap_cntr(dd, dd->ipath_cregs->cr_sendstallcnt);
1003
1004         ret = 0;
1005
1006 bail:
1007         return ret;
1008 }
1009
1010 EXPORT_SYMBOL_GPL(ipath_layer_snapshot_counters);
1011
1012 /**
1013  * ipath_layer_get_counters - get various chip counters
1014  * @dd: the infinipath device
1015  * @cntrs: counters are placed here
1016  *
1017  * Return the counters needed by recv_pma_get_portcounters().
1018  */
1019 int ipath_layer_get_counters(struct ipath_devdata *dd,
1020                               struct ipath_layer_counters *cntrs)
1021 {
1022         int ret;
1023
1024         if (!(dd->ipath_flags & IPATH_INITTED)) {
1025                 /* no hardware, freeze, etc. */
1026                 ipath_dbg("unit %u not usable\n", dd->ipath_unit);
1027                 ret = -EINVAL;
1028                 goto bail;
1029         }
1030         cntrs->symbol_error_counter =
1031                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_ibsymbolerrcnt);
1032         cntrs->link_error_recovery_counter =
1033                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkerrrecovcnt);
1034         /*
1035          * The link downed counter counts when the other side downs the
1036          * connection.  We add in the number of times we downed the link
1037          * due to local link integrity errors to compensate.
1038          */
1039         cntrs->link_downed_counter =
1040                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkdowncnt);
1041         cntrs->port_rcv_errors =
1042                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rxdroppktcnt) +
1043                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvovflcnt) +
1044                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_portovflcnt) +
1045                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_err_rlencnt) +
1046                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_invalidrlencnt) +
1047                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) +
1048                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) +
1049                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) +
1050                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt);
1051         cntrs->port_rcv_remphys_errors =
1052                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt);
1053         cntrs->port_xmit_discards =
1054                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_unsupvlcnt);
1055         cntrs->port_xmit_data =
1056                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
1057         cntrs->port_rcv_data =
1058                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
1059         cntrs->port_xmit_packets =
1060                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
1061         cntrs->port_rcv_packets =
1062                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
1063         cntrs->local_link_integrity_errors = dd->ipath_lli_errors;
1064         cntrs->excessive_buffer_overrun_errors = 0; /* XXX */
1065
1066         ret = 0;
1067
1068 bail:
1069         return ret;
1070 }
1071
1072 EXPORT_SYMBOL_GPL(ipath_layer_get_counters);
1073
1074 int ipath_layer_want_buffer(struct ipath_devdata *dd)
1075 {
1076         set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
1077         ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
1078                          dd->ipath_sendctrl);
1079
1080         return 0;
1081 }
1082
1083 EXPORT_SYMBOL_GPL(ipath_layer_want_buffer);
1084
1085 int ipath_layer_send_hdr(struct ipath_devdata *dd, struct ether_header *hdr)
1086 {
1087         int ret = 0;
1088         u32 __iomem *piobuf;
1089         u32 plen, *uhdr;
1090         size_t count;
1091         __be16 vlsllnh;
1092
1093         if (!(dd->ipath_flags & IPATH_RCVHDRSZ_SET)) {
1094                 ipath_dbg("send while not open\n");
1095                 ret = -EINVAL;
1096         } else
1097                 if ((dd->ipath_flags & (IPATH_LINKUNK | IPATH_LINKDOWN)) ||
1098                     dd->ipath_lid == 0) {
1099                         /*
1100                          * lid check is for when sma hasn't yet configured
1101                          */
1102                         ret = -ENETDOWN;
1103                         ipath_cdbg(VERBOSE, "send while not ready, "
1104                                    "mylid=%u, flags=0x%x\n",
1105                                    dd->ipath_lid, dd->ipath_flags);
1106                 }
1107
1108         vlsllnh = *((__be16 *) hdr);
1109         if (vlsllnh != htons(IPATH_LRH_BTH)) {
1110                 ipath_dbg("Warning: lrh[0] wrong (%x, not %x); "
1111                           "not sending\n", be16_to_cpu(vlsllnh),
1112                           IPATH_LRH_BTH);
1113                 ret = -EINVAL;
1114         }
1115         if (ret)
1116                 goto done;
1117
1118         /* Get a PIO buffer to use. */
1119         piobuf = ipath_getpiobuf(dd, NULL);
1120         if (piobuf == NULL) {
1121                 ret = -EBUSY;
1122                 goto done;
1123         }
1124
1125         plen = (sizeof(*hdr) >> 2); /* actual length */
1126         ipath_cdbg(EPKT, "0x%x+1w pio %p\n", plen, piobuf);
1127
1128         writeq(plen+1, piobuf); /* len (+1 for pad) to pbc, no flags */
1129         ipath_flush_wc();
1130         piobuf += 2;
1131         uhdr = (u32 *)hdr;
1132         count = plen-1; /* amount we can copy before trigger word */
1133         __iowrite32_copy(piobuf, uhdr, count);
1134         ipath_flush_wc();
1135         __raw_writel(uhdr[count], piobuf + count);
1136         ipath_flush_wc(); /* ensure it's sent, now */
1137
1138         ipath_stats.sps_ether_spkts++;  /* ether packet sent */
1139
1140 done:
1141         return ret;
1142 }
1143
1144 EXPORT_SYMBOL_GPL(ipath_layer_send_hdr);
1145
1146 int ipath_layer_set_piointbufavail_int(struct ipath_devdata *dd)
1147 {
1148         set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
1149
1150         ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
1151                          dd->ipath_sendctrl);
1152         return 0;
1153 }
1154
1155 EXPORT_SYMBOL_GPL(ipath_layer_set_piointbufavail_int);
1156
1157 int ipath_layer_enable_timer(struct ipath_devdata *dd)
1158 {
1159         /*
1160          * HT-400 has a design flaw where the chip and kernel idea
1161          * of the tail register don't always agree, and therefore we won't
1162          * get an interrupt on the next packet received.
1163          * If the board supports per packet receive interrupts, use it.
1164          * Otherwise, the timer function periodically checks for packets
1165          * to cover this case.
1166          * Either way, the timer is needed for verbs layer related
1167          * processing.
1168          */
1169         if (dd->ipath_flags & IPATH_GPIO_INTR) {
1170                 ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect,
1171                                  0x2074076542310ULL);
1172                 /* Enable GPIO bit 2 interrupt */
1173                 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
1174                                  (u64) (1 << 2));
1175         }
1176
1177         init_timer(&dd->verbs_layer.l_timer);
1178         dd->verbs_layer.l_timer.function = __ipath_verbs_timer;
1179         dd->verbs_layer.l_timer.data = (unsigned long)dd;
1180         dd->verbs_layer.l_timer.expires = jiffies + 1;
1181         add_timer(&dd->verbs_layer.l_timer);
1182
1183         return 0;
1184 }
1185
1186 EXPORT_SYMBOL_GPL(ipath_layer_enable_timer);
1187
1188 int ipath_layer_disable_timer(struct ipath_devdata *dd)
1189 {
1190         /* Disable GPIO bit 2 interrupt */
1191         if (dd->ipath_flags & IPATH_GPIO_INTR)
1192                 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0);
1193
1194         del_timer_sync(&dd->verbs_layer.l_timer);
1195
1196         return 0;
1197 }
1198
1199 EXPORT_SYMBOL_GPL(ipath_layer_disable_timer);
1200
1201 /**
1202  * ipath_layer_set_verbs_flags - set the verbs layer flags
1203  * @dd: the infinipath device
1204  * @flags: the flags to set
1205  */
1206 int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags)
1207 {
1208         struct ipath_devdata *ss;
1209         unsigned long lflags;
1210
1211         spin_lock_irqsave(&ipath_devs_lock, lflags);
1212
1213         list_for_each_entry(ss, &ipath_dev_list, ipath_list) {
1214                 if (!(ss->ipath_flags & IPATH_INITTED))
1215                         continue;
1216                 if ((flags & IPATH_VERBS_KERNEL_SMA) &&
1217                     !(*ss->ipath_statusp & IPATH_STATUS_SMA))
1218                         *ss->ipath_statusp |= IPATH_STATUS_OIB_SMA;
1219                 else
1220                         *ss->ipath_statusp &= ~IPATH_STATUS_OIB_SMA;
1221         }
1222
1223         spin_unlock_irqrestore(&ipath_devs_lock, lflags);
1224
1225         return 0;
1226 }
1227
1228 EXPORT_SYMBOL_GPL(ipath_layer_set_verbs_flags);
1229
1230 /**
1231  * ipath_layer_get_npkeys - return the size of the PKEY table for port 0
1232  * @dd: the infinipath device
1233  */
1234 unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd)
1235 {
1236         return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys);
1237 }
1238
1239 EXPORT_SYMBOL_GPL(ipath_layer_get_npkeys);
1240
1241 /**
1242  * ipath_layer_get_pkey - return the indexed PKEY from the port 0 PKEY table
1243  * @dd: the infinipath device
1244  * @index: the PKEY index
1245  */
1246 unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index)
1247 {
1248         unsigned ret;
1249
1250         if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys))
1251                 ret = 0;
1252         else
1253                 ret = dd->ipath_pd[0]->port_pkeys[index];
1254
1255         return ret;
1256 }
1257
1258 EXPORT_SYMBOL_GPL(ipath_layer_get_pkey);
1259
1260 /**
1261  * ipath_layer_get_pkeys - return the PKEY table for port 0
1262  * @dd: the infinipath device
1263  * @pkeys: the pkey table is placed here
1264  */
1265 int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
1266 {
1267         struct ipath_portdata *pd = dd->ipath_pd[0];
1268
1269         memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
1270
1271         return 0;
1272 }
1273
1274 EXPORT_SYMBOL_GPL(ipath_layer_get_pkeys);
1275
1276 /**
1277  * rm_pkey - decrecment the reference count for the given PKEY
1278  * @dd: the infinipath device
1279  * @key: the PKEY index
1280  *
1281  * Return true if this was the last reference and the hardware table entry
1282  * needs to be changed.
1283  */
1284 static int rm_pkey(struct ipath_devdata *dd, u16 key)
1285 {
1286         int i;
1287         int ret;
1288
1289         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1290                 if (dd->ipath_pkeys[i] != key)
1291                         continue;
1292                 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
1293                         dd->ipath_pkeys[i] = 0;
1294                         ret = 1;
1295                         goto bail;
1296                 }
1297                 break;
1298         }
1299
1300         ret = 0;
1301
1302 bail:
1303         return ret;
1304 }
1305
1306 /**
1307  * add_pkey - add the given PKEY to the hardware table
1308  * @dd: the infinipath device
1309  * @key: the PKEY
1310  *
1311  * Return an error code if unable to add the entry, zero if no change,
1312  * or 1 if the hardware PKEY register needs to be updated.
1313  */
1314 static int add_pkey(struct ipath_devdata *dd, u16 key)
1315 {
1316         int i;
1317         u16 lkey = key & 0x7FFF;
1318         int any = 0;
1319         int ret;
1320
1321         if (lkey == 0x7FFF) {
1322                 ret = 0;
1323                 goto bail;
1324         }
1325
1326         /* Look for an empty slot or a matching PKEY. */
1327         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1328                 if (!dd->ipath_pkeys[i]) {
1329                         any++;
1330                         continue;
1331                 }
1332                 /* If it matches exactly, try to increment the ref count */
1333                 if (dd->ipath_pkeys[i] == key) {
1334                         if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
1335                                 ret = 0;
1336                                 goto bail;
1337                         }
1338                         /* Lost the race. Look for an empty slot below. */
1339                         atomic_dec(&dd->ipath_pkeyrefs[i]);
1340                         any++;
1341                 }
1342                 /*
1343                  * It makes no sense to have both the limited and unlimited
1344                  * PKEY set at the same time since the unlimited one will
1345                  * disable the limited one.
1346                  */
1347                 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
1348                         ret = -EEXIST;
1349                         goto bail;
1350                 }
1351         }
1352         if (!any) {
1353                 ret = -EBUSY;
1354                 goto bail;
1355         }
1356         for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1357                 if (!dd->ipath_pkeys[i] &&
1358                     atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
1359                         /* for ipathstats, etc. */
1360                         ipath_stats.sps_pkeys[i] = lkey;
1361                         dd->ipath_pkeys[i] = key;
1362                         ret = 1;
1363                         goto bail;
1364                 }
1365         }
1366         ret = -EBUSY;
1367
1368 bail:
1369         return ret;
1370 }
1371
1372 /**
1373  * ipath_layer_set_pkeys - set the PKEY table for port 0
1374  * @dd: the infinipath device
1375  * @pkeys: the PKEY table
1376  */
1377 int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 * pkeys)
1378 {
1379         struct ipath_portdata *pd;
1380         int i;
1381         int changed = 0;
1382
1383         pd = dd->ipath_pd[0];
1384
1385         for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
1386                 u16 key = pkeys[i];
1387                 u16 okey = pd->port_pkeys[i];
1388
1389                 if (key == okey)
1390                         continue;
1391                 /*
1392                  * The value of this PKEY table entry is changing.
1393                  * Remove the old entry in the hardware's array of PKEYs.
1394                  */
1395                 if (okey & 0x7FFF)
1396                         changed |= rm_pkey(dd, okey);
1397                 if (key & 0x7FFF) {
1398                         int ret = add_pkey(dd, key);
1399
1400                         if (ret < 0)
1401                                 key = 0;
1402                         else
1403                                 changed |= ret;
1404                 }
1405                 pd->port_pkeys[i] = key;
1406         }
1407         if (changed) {
1408                 u64 pkey;
1409
1410                 pkey = (u64) dd->ipath_pkeys[0] |
1411                         ((u64) dd->ipath_pkeys[1] << 16) |
1412                         ((u64) dd->ipath_pkeys[2] << 32) |
1413                         ((u64) dd->ipath_pkeys[3] << 48);
1414                 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
1415                            (unsigned long long) pkey);
1416                 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
1417                                  pkey);
1418         }
1419         return 0;
1420 }
1421
1422 EXPORT_SYMBOL_GPL(ipath_layer_set_pkeys);
1423
1424 /**
1425  * ipath_layer_get_linkdowndefaultstate - get the default linkdown state
1426  * @dd: the infinipath device
1427  *
1428  * Returns zero if the default is POLL, 1 if the default is SLEEP.
1429  */
1430 int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd)
1431 {
1432         return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
1433 }
1434
1435 EXPORT_SYMBOL_GPL(ipath_layer_get_linkdowndefaultstate);
1436
1437 /**
1438  * ipath_layer_set_linkdowndefaultstate - set the default linkdown state
1439  * @dd: the infinipath device
1440  * @sleep: the new state
1441  *
1442  * Note that this will only take effect when the link state changes.
1443  */
1444 int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd,
1445                                          int sleep)
1446 {
1447         if (sleep)
1448                 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
1449         else
1450                 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
1451         ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1452                          dd->ipath_ibcctrl);
1453         return 0;
1454 }
1455
1456 EXPORT_SYMBOL_GPL(ipath_layer_set_linkdowndefaultstate);
1457
1458 int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd)
1459 {
1460         return (dd->ipath_ibcctrl >>
1461                 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
1462                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
1463 }
1464
1465 EXPORT_SYMBOL_GPL(ipath_layer_get_phyerrthreshold);
1466
1467 /**
1468  * ipath_layer_set_phyerrthreshold - set the physical error threshold
1469  * @dd: the infinipath device
1470  * @n: the new threshold
1471  *
1472  * Note that this will only take effect when the link state changes.
1473  */
1474 int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
1475 {
1476         unsigned v;
1477
1478         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
1479                 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
1480         if (v != n) {
1481                 dd->ipath_ibcctrl &=
1482                         ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
1483                           INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
1484                 dd->ipath_ibcctrl |=
1485                         (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
1486                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1487                                  dd->ipath_ibcctrl);
1488         }
1489         return 0;
1490 }
1491
1492 EXPORT_SYMBOL_GPL(ipath_layer_set_phyerrthreshold);
1493
1494 int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd)
1495 {
1496         return (dd->ipath_ibcctrl >>
1497                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
1498                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
1499 }
1500
1501 EXPORT_SYMBOL_GPL(ipath_layer_get_overrunthreshold);
1502
1503 /**
1504  * ipath_layer_set_overrunthreshold - set the overrun threshold
1505  * @dd: the infinipath device
1506  * @n: the new threshold
1507  *
1508  * Note that this will only take effect when the link state changes.
1509  */
1510 int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
1511 {
1512         unsigned v;
1513
1514         v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
1515                 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
1516         if (v != n) {
1517                 dd->ipath_ibcctrl &=
1518                         ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
1519                           INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
1520                 dd->ipath_ibcctrl |=
1521                         (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
1522                 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1523                                  dd->ipath_ibcctrl);
1524         }
1525         return 0;
1526 }
1527
1528 EXPORT_SYMBOL_GPL(ipath_layer_set_overrunthreshold);
1529
1530 int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name,
1531                               size_t namelen)
1532 {
1533         return dd->ipath_f_get_boardname(dd, name, namelen);
1534 }
1535 EXPORT_SYMBOL_GPL(ipath_layer_get_boardname);
1536
1537 u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd)
1538 {
1539         return dd->ipath_rcvhdrentsize;
1540 }
1541 EXPORT_SYMBOL_GPL(ipath_layer_get_rcvhdrentsize);