drivers/misc/sgi-xp/xpc_main.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (c) 2004-2009 Silicon Graphics, Inc.  All Rights Reserved.
   7  */
   8
   9 /*
  10  * Cross Partition Communication (XPC) support - standard version.
  11  *
  12  *      XPC provides a message passing capability that crosses partition
  13  *      boundaries. This module is made up of two parts:
  14  *
  15  *          partition   This part detects the presence/absence of other
  16  *                      partitions. It provides a heartbeat and monitors
  17  *                      the heartbeats of other partitions.
  18  *
  19  *          channel     This part manages the channels and sends/receives
  20  *                      messages across them to/from other partitions.
  21  *
  22  *      There are a couple of additional functions residing in XP, which
  23  *      provide an interface to XPC for its users.
  24  *
  25  *
  26  *      Caveats:
  27  *
  28  *        . Currently on sn2, we have no way to determine which nasid an IRQ
  29  *          came from. Thus, xpc_send_IRQ_sn2() does a remote amo write
  30  *          followed by an IPI. The amo indicates where data is to be pulled
  31  *          from, so after the IPI arrives, the remote partition checks the amo
  32  *          word. The IPI can actually arrive before the amo however, so other
  33  *          code must periodically check for this case. Also, remote amo
  34  *          operations do not reliably time out. Thus we do a remote PIO read
  35  *          solely to know whether the remote partition is down and whether we
  36  *          should stop sending IPIs to it. This remote PIO read operation is
  37  *          set up in a special nofault region so SAL knows to ignore (and
  38  *          cleanup) any errors due to the remote amo write, PIO read, and/or
  39  *          PIO write operations.
  40  *
  41  *          If/when new hardware solves this IPI problem, we should abandon
  42  *          the current approach.
  43  *
  44  */
  45
  46 #include <linux/module.h>
  47 #include <linux/slab.h>
  48 #include <linux/sysctl.h>
  49 #include <linux/device.h>
  50 #include <linux/delay.h>
  51 #include <linux/reboot.h>
  52 #include <linux/kdebug.h>
  53 #include <linux/kthread.h>
  54 #include "xpc.h"
  55
  56 #ifdef CONFIG_X86_64
  57 #include <asm/traps.h>
  58 #endif
  59
  60 /* define two XPC debug device structures to be used with dev_dbg() et al */
  61
  62 static struct device_driver xpc_dbg_name = {
  63         .name = "xpc"
  64 };
  65
  66 static struct device xpc_part_dbg_subname = {
  67         .init_name = "",        /* set to "part" at xpc_init() time */
  68         .driver = &xpc_dbg_name
  69 };
  70
  71 static struct device xpc_chan_dbg_subname = {
  72         .init_name = "",        /* set to "chan" at xpc_init() time */
  73         .driver = &xpc_dbg_name
  74 };
  75
  76 struct device *xpc_part = &xpc_part_dbg_subname;
  77 struct device *xpc_chan = &xpc_chan_dbg_subname;
  78
  79 static int xpc_kdebug_ignore;
  80
  81 /* systune related variables for /proc/sys directories */
  82
  83 static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
  84 static int xpc_hb_min_interval = 1;
  85 static int xpc_hb_max_interval = 10;
  86
  87 static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL;
  88 static int xpc_hb_check_min_interval = 10;
  89 static int xpc_hb_check_max_interval = 120;
  90
  91 int xpc_disengage_timelimit = XPC_DISENGAGE_DEFAULT_TIMELIMIT;
  92 static int xpc_disengage_min_timelimit; /* = 0 */
  93 static int xpc_disengage_max_timelimit = 120;
  94
  95 static struct ctl_table xpc_sys_xpc_hb_dir[] = {
  96         {
  97          .procname = "hb_interval",
  98          .data = &xpc_hb_interval,
  99          .maxlen = sizeof(int),
 100          .mode = 0644,
 101          .proc_handler = proc_dointvec_minmax,
 102          .extra1 = &xpc_hb_min_interval,
 103          .extra2 = &xpc_hb_max_interval},
 104         {
 105          .procname = "hb_check_interval",
 106          .data = &xpc_hb_check_interval,
 107          .maxlen = sizeof(int),
 108          .mode = 0644,
 109          .proc_handler = proc_dointvec_minmax,
 110          .extra1 = &xpc_hb_check_min_interval,
 111          .extra2 = &xpc_hb_check_max_interval},
 112         {}
 113 };
 114 static struct ctl_table xpc_sys_xpc_dir[] = {
 115         {
 116          .procname = "hb",
 117          .mode = 0555,
 118          .child = xpc_sys_xpc_hb_dir},
 119         {
 120          .procname = "disengage_timelimit",
 121          .data = &xpc_disengage_timelimit,
 122          .maxlen = sizeof(int),
 123          .mode = 0644,
 124          .proc_handler = proc_dointvec_minmax,
 125          .extra1 = &xpc_disengage_min_timelimit,
 126          .extra2 = &xpc_disengage_max_timelimit},
 127         {}
 128 };
 129 static struct ctl_table xpc_sys_dir[] = {
 130         {
 131          .procname = "xpc",
 132          .mode = 0555,
 133          .child = xpc_sys_xpc_dir},
 134         {}
 135 };
 136 static struct ctl_table_header *xpc_sysctl;
 137
 138 /* non-zero if any remote partition disengage was timed out */
 139 int xpc_disengage_timedout;
 140
 141 /* #of activate IRQs received and not yet processed */
 142 int xpc_activate_IRQ_rcvd;
 143 DEFINE_SPINLOCK(xpc_activate_IRQ_rcvd_lock);
 144
 145 /* IRQ handler notifies this wait queue on receipt of an IRQ */
 146 DECLARE_WAIT_QUEUE_HEAD(xpc_activate_IRQ_wq);
 147
 148 static unsigned long xpc_hb_check_timeout;
 149 static struct timer_list xpc_hb_timer;
 150
 151 /* notification that the xpc_hb_checker thread has exited */
 152 static DECLARE_COMPLETION(xpc_hb_checker_exited);
 153
 154 /* notification that the xpc_discovery thread has exited */
 155 static DECLARE_COMPLETION(xpc_discovery_exited);
 156
 157 static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *);
 158
 159 static int xpc_system_reboot(struct notifier_block *, unsigned long, void *);
 160 static struct notifier_block xpc_reboot_notifier = {
 161         .notifier_call = xpc_system_reboot,
 162 };
 163
 164 static int xpc_system_die(struct notifier_block *, unsigned long, void *);
 165 static struct notifier_block xpc_die_notifier = {
 166         .notifier_call = xpc_system_die,
 167 };
 168
 169 struct xpc_arch_operations xpc_arch_ops;
 170
 171 /*
 172  * Timer function to enforce the timelimit on the partition disengage.
 173  */
 174 static void
 175 xpc_timeout_partition_disengage(struct timer_list *t)
 176 {
 177         struct xpc_partition *part = from_timer(part, t, disengage_timer);
 178
 179         DBUG_ON(time_is_after_jiffies(part->disengage_timeout));
 180
 181         (void)xpc_partition_disengaged(part);
 182
 183         DBUG_ON(part->disengage_timeout != 0);
 184         DBUG_ON(xpc_arch_ops.partition_engaged(XPC_PARTID(part)));
 185 }
 186
 187 /*
 188  * Timer to produce the heartbeat.  The timer structures function is
 189  * already set when this is initially called.  A tunable is used to
 190  * specify when the next timeout should occur.
 191  */
 192 static void
 193 xpc_hb_beater(struct timer_list *unused)
 194 {
 195         xpc_arch_ops.increment_heartbeat();
 196
 197         if (time_is_before_eq_jiffies(xpc_hb_check_timeout))
 198                 wake_up_interruptible(&xpc_activate_IRQ_wq);
 199
 200         xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ);
 201         add_timer(&xpc_hb_timer);
 202 }
 203
 204 static void
 205 xpc_start_hb_beater(void)
 206 {
 207         xpc_arch_ops.heartbeat_init();
 208         timer_setup(&xpc_hb_timer, xpc_hb_beater, 0);
 209         xpc_hb_beater(0);
 210 }
 211
 212 static void
 213 xpc_stop_hb_beater(void)
 214 {
 215         del_timer_sync(&xpc_hb_timer);
 216         xpc_arch_ops.heartbeat_exit();
 217 }
 218
 219 /*
 220  * At periodic intervals, scan through all active partitions and ensure
 221  * their heartbeat is still active.  If not, the partition is deactivated.
 222  */
 223 static void
 224 xpc_check_remote_hb(void)
 225 {
 226         struct xpc_partition *part;
 227         short partid;
 228         enum xp_retval ret;
 229
 230         for (partid = 0; partid < xp_max_npartitions; partid++) {
 231
 232                 if (xpc_exiting)
 233                         break;
 234
 235                 if (partid == xp_partition_id)
 236                         continue;
 237
 238                 part = &xpc_partitions[partid];
 239
 240                 if (part->act_state == XPC_P_AS_INACTIVE ||
 241                     part->act_state == XPC_P_AS_DEACTIVATING) {
 242                         continue;
 243                 }
 244
 245                 ret = xpc_arch_ops.get_remote_heartbeat(part);
 246                 if (ret != xpSuccess)
 247                         XPC_DEACTIVATE_PARTITION(part, ret);
 248         }
 249 }
 250
 251 /*
 252  * This thread is responsible for nearly all of the partition
 253  * activation/deactivation.
 254  */
 255 static int
 256 xpc_hb_checker(void *ignore)
 257 {
 258         int force_IRQ = 0;
 259
 260         /* this thread was marked active by xpc_hb_init() */
 261
 262         set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
 263
 264         /* set our heartbeating to other partitions into motion */
 265         xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
 266         xpc_start_hb_beater();
 267
 268         while (!xpc_exiting) {
 269
 270                 dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have "
 271                         "been received\n",
 272                         (int)(xpc_hb_check_timeout - jiffies),
 273                         xpc_activate_IRQ_rcvd);
 274
 275                 /* checking of remote heartbeats is skewed by IRQ handling */
 276                 if (time_is_before_eq_jiffies(xpc_hb_check_timeout)) {
 277                         xpc_hb_check_timeout = jiffies +
 278                             (xpc_hb_check_interval * HZ);
 279
 280                         dev_dbg(xpc_part, "checking remote heartbeats\n");
 281                         xpc_check_remote_hb();
 282                 }
 283
 284                 /* check for outstanding IRQs */
 285                 if (xpc_activate_IRQ_rcvd > 0 || force_IRQ != 0) {
 286                         force_IRQ = 0;
 287                         dev_dbg(xpc_part, "processing activate IRQs "
 288                                 "received\n");
 289                         xpc_arch_ops.process_activate_IRQ_rcvd();
 290                 }
 291
 292                 /* wait for IRQ or timeout */
 293                 (void)wait_event_interruptible(xpc_activate_IRQ_wq,
 294                                                (time_is_before_eq_jiffies(
 295                                                 xpc_hb_check_timeout) ||
 296                                                 xpc_activate_IRQ_rcvd > 0 ||
 297                                                 xpc_exiting));
 298         }
 299
 300         xpc_stop_hb_beater();
 301
 302         dev_dbg(xpc_part, "heartbeat checker is exiting\n");
 303
 304         /* mark this thread as having exited */
 305         complete(&xpc_hb_checker_exited);
 306         return 0;
 307 }
 308
 309 /*
 310  * This thread will attempt to discover other partitions to activate
 311  * based on info provided by SAL. This new thread is short lived and
 312  * will exit once discovery is complete.
 313  */
 314 static int
 315 xpc_initiate_discovery(void *ignore)
 316 {
 317         xpc_discovery();
 318
 319         dev_dbg(xpc_part, "discovery thread is exiting\n");
 320
 321         /* mark this thread as having exited */
 322         complete(&xpc_discovery_exited);
 323         return 0;
 324 }
 325
 326 /*
 327  * The first kthread assigned to a newly activated partition is the one
 328  * created by XPC HB with which it calls xpc_activating(). XPC hangs on to
 329  * that kthread until the partition is brought down, at which time that kthread
 330  * returns back to XPC HB. (The return of that kthread will signify to XPC HB
 331  * that XPC has dismantled all communication infrastructure for the associated
 332  * partition.) This kthread becomes the channel manager for that partition.
 333  *
 334  * Each active partition has a channel manager, who, besides connecting and
 335  * disconnecting channels, will ensure that each of the partition's connected
 336  * channels has the required number of assigned kthreads to get the work done.
 337  */
 338 static void
 339 xpc_channel_mgr(struct xpc_partition *part)
 340 {
 341         while (part->act_state != XPC_P_AS_DEACTIVATING ||
 342                atomic_read(&part->nchannels_active) > 0 ||
 343                !xpc_partition_disengaged(part)) {
 344
 345                 xpc_process_sent_chctl_flags(part);
 346
 347                 /*
 348                  * Wait until we've been requested to activate kthreads or
 349                  * all of the channel's message queues have been torn down or
 350                  * a signal is pending.
 351                  *
 352                  * The channel_mgr_requests is set to 1 after being awakened,
 353                  * This is done to prevent the channel mgr from making one pass
 354                  * through the loop for each request, since he will
 355                  * be servicing all the requests in one pass. The reason it's
 356                  * set to 1 instead of 0 is so that other kthreads will know
 357                  * that the channel mgr is running and won't bother trying to
 358                  * wake him up.
 359                  */
 360                 atomic_dec(&part->channel_mgr_requests);
 361                 (void)wait_event_interruptible(part->channel_mgr_wq,
 362                                 (atomic_read(&part->channel_mgr_requests) > 0 ||
 363                                  part->chctl.all_flags != 0 ||
 364                                  (part->act_state == XPC_P_AS_DEACTIVATING &&
 365                                  atomic_read(&part->nchannels_active) == 0 &&
 366                                  xpc_partition_disengaged(part))));
 367                 atomic_set(&part->channel_mgr_requests, 1);
 368         }
 369 }
 370
 371 /*
 372  * Guarantee that the kzalloc'd memory is cacheline aligned.
 373  */
 374 void *
 375 xpc_kzalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
 376 {
 377         /* see if kzalloc will give us cachline aligned memory by default */
 378         *base = kzalloc(size, flags);
 379         if (*base == NULL)
 380                 return NULL;
 381
 382         if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
 383                 return *base;
 384
 385         kfree(*base);
 386
 387         /* nope, we'll have to do it ourselves */
 388         *base = kzalloc(size + L1_CACHE_BYTES, flags);
 389         if (*base == NULL)
 390                 return NULL;
 391
 392         return (void *)L1_CACHE_ALIGN((u64)*base);
 393 }
 394
 395 /*
 396  * Setup the channel structures necessary to support XPartition Communication
 397  * between the specified remote partition and the local one.
 398  */
 399 static enum xp_retval
 400 xpc_setup_ch_structures(struct xpc_partition *part)
 401 {
 402         enum xp_retval ret;
 403         int ch_number;
 404         struct xpc_channel *ch;
 405         short partid = XPC_PARTID(part);
 406
 407         /*
 408          * Allocate all of the channel structures as a contiguous chunk of
 409          * memory.
 410          */
 411         DBUG_ON(part->channels != NULL);
 412         part->channels = kcalloc(XPC_MAX_NCHANNELS,
 413                                  sizeof(struct xpc_channel),
 414                                  GFP_KERNEL);
 415         if (part->channels == NULL) {
 416                 dev_err(xpc_chan, "can't get memory for channels\n");
 417                 return xpNoMemory;
 418         }
 419
 420         /* allocate the remote open and close args */
 421
 422         part->remote_openclose_args =
 423             xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE,
 424                                           GFP_KERNEL, &part->
 425                                           remote_openclose_args_base);
 426         if (part->remote_openclose_args == NULL) {
 427                 dev_err(xpc_chan, "can't get memory for remote connect args\n");
 428                 ret = xpNoMemory;
 429                 goto out_1;
 430         }
 431
 432         part->chctl.all_flags = 0;
 433         spin_lock_init(&part->chctl_lock);
 434
 435         atomic_set(&part->channel_mgr_requests, 1);
 436         init_waitqueue_head(&part->channel_mgr_wq);
 437
 438         part->nchannels = XPC_MAX_NCHANNELS;
 439
 440         atomic_set(&part->nchannels_active, 0);
 441         atomic_set(&part->nchannels_engaged, 0);
 442
 443         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
 444                 ch = &part->channels[ch_number];
 445
 446                 ch->partid = partid;
 447                 ch->number = ch_number;
 448                 ch->flags = XPC_C_DISCONNECTED;
 449
 450                 atomic_set(&ch->kthreads_assigned, 0);
 451                 atomic_set(&ch->kthreads_idle, 0);
 452                 atomic_set(&ch->kthreads_active, 0);
 453
 454                 atomic_set(&ch->references, 0);
 455                 atomic_set(&ch->n_to_notify, 0);
 456
 457                 spin_lock_init(&ch->lock);
 458                 init_completion(&ch->wdisconnect_wait);
 459
 460                 atomic_set(&ch->n_on_msg_allocate_wq, 0);
 461                 init_waitqueue_head(&ch->msg_allocate_wq);
 462                 init_waitqueue_head(&ch->idle_wq);
 463         }
 464
 465         ret = xpc_arch_ops.setup_ch_structures(part);
 466         if (ret != xpSuccess)
 467                 goto out_2;
 468
 469         /*
 470          * With the setting of the partition setup_state to XPC_P_SS_SETUP,
 471          * we're declaring that this partition is ready to go.
 472          */
 473         part->setup_state = XPC_P_SS_SETUP;
 474
 475         return xpSuccess;
 476
 477         /* setup of ch structures failed */
 478 out_2:
 479         kfree(part->remote_openclose_args_base);
 480         part->remote_openclose_args = NULL;
 481 out_1:
 482         kfree(part->channels);
 483         part->channels = NULL;
 484         return ret;
 485 }
 486
 487 /*
 488  * Teardown the channel structures necessary to support XPartition Communication
 489  * between the specified remote partition and the local one.
 490  */
 491 static void
 492 xpc_teardown_ch_structures(struct xpc_partition *part)
 493 {
 494         DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
 495         DBUG_ON(atomic_read(&part->nchannels_active) != 0);
 496
 497         /*
 498          * Make this partition inaccessible to local processes by marking it
 499          * as no longer setup. Then wait before proceeding with the teardown
 500          * until all existing references cease.
 501          */
 502         DBUG_ON(part->setup_state != XPC_P_SS_SETUP);
 503         part->setup_state = XPC_P_SS_WTEARDOWN;
 504
 505         wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
 506
 507         /* now we can begin tearing down the infrastructure */
 508
 509         xpc_arch_ops.teardown_ch_structures(part);
 510
 511         kfree(part->remote_openclose_args_base);
 512         part->remote_openclose_args = NULL;
 513         kfree(part->channels);
 514         part->channels = NULL;
 515
 516         part->setup_state = XPC_P_SS_TORNDOWN;
 517 }
 518
 519 /*
 520  * When XPC HB determines that a partition has come up, it will create a new
 521  * kthread and that kthread will call this function to attempt to set up the
 522  * basic infrastructure used for Cross Partition Communication with the newly
 523  * upped partition.
 524  *
 525  * The kthread that was created by XPC HB and which setup the XPC
 526  * infrastructure will remain assigned to the partition becoming the channel
 527  * manager for that partition until the partition is deactivating, at which
 528  * time the kthread will teardown the XPC infrastructure and then exit.
 529  */
 530 static int
 531 xpc_activating(void *__partid)
 532 {
 533         short partid = (u64)__partid;
 534         struct xpc_partition *part = &xpc_partitions[partid];
 535         unsigned long irq_flags;
 536
 537         DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
 538
 539         spin_lock_irqsave(&part->act_lock, irq_flags);
 540
 541         if (part->act_state == XPC_P_AS_DEACTIVATING) {
 542                 part->act_state = XPC_P_AS_INACTIVE;
 543                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 544                 part->remote_rp_pa = 0;
 545                 return 0;
 546         }
 547
 548         /* indicate the thread is activating */
 549         DBUG_ON(part->act_state != XPC_P_AS_ACTIVATION_REQ);
 550         part->act_state = XPC_P_AS_ACTIVATING;
 551
 552         XPC_SET_REASON(part, 0, 0);
 553         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 554
 555         dev_dbg(xpc_part, "activating partition %d\n", partid);
 556
 557         xpc_arch_ops.allow_hb(partid);
 558
 559         if (xpc_setup_ch_structures(part) == xpSuccess) {
 560                 (void)xpc_part_ref(part);       /* this will always succeed */
 561
 562                 if (xpc_arch_ops.make_first_contact(part) == xpSuccess) {
 563                         xpc_mark_partition_active(part);
 564                         xpc_channel_mgr(part);
 565                         /* won't return until partition is deactivating */
 566                 }
 567
 568                 xpc_part_deref(part);
 569                 xpc_teardown_ch_structures(part);
 570         }
 571
 572         xpc_arch_ops.disallow_hb(partid);
 573         xpc_mark_partition_inactive(part);
 574
 575         if (part->reason == xpReactivating) {
 576                 /* interrupting ourselves results in activating partition */
 577                 xpc_arch_ops.request_partition_reactivation(part);
 578         }
 579
 580         return 0;
 581 }
 582
 583 void
 584 xpc_activate_partition(struct xpc_partition *part)
 585 {
 586         short partid = XPC_PARTID(part);
 587         unsigned long irq_flags;
 588         struct task_struct *kthread;
 589
 590         spin_lock_irqsave(&part->act_lock, irq_flags);
 591
 592         DBUG_ON(part->act_state != XPC_P_AS_INACTIVE);
 593
 594         part->act_state = XPC_P_AS_ACTIVATION_REQ;
 595         XPC_SET_REASON(part, xpCloneKThread, __LINE__);
 596
 597         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 598
 599         kthread = kthread_run(xpc_activating, (void *)((u64)partid), "xpc%02d",
 600                               partid);
 601         if (IS_ERR(kthread)) {
 602                 spin_lock_irqsave(&part->act_lock, irq_flags);
 603                 part->act_state = XPC_P_AS_INACTIVE;
 604                 XPC_SET_REASON(part, xpCloneKThreadFailed, __LINE__);
 605                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 606         }
 607 }
 608
 609 void
 610 xpc_activate_kthreads(struct xpc_channel *ch, int needed)
 611 {
 612         int idle = atomic_read(&ch->kthreads_idle);
 613         int assigned = atomic_read(&ch->kthreads_assigned);
 614         int wakeup;
 615
 616         DBUG_ON(needed <= 0);
 617
 618         if (idle > 0) {
 619                 wakeup = (needed > idle) ? idle : needed;
 620                 needed -= wakeup;
 621
 622                 dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, "
 623                         "channel=%d\n", wakeup, ch->partid, ch->number);
 624
 625                 /* only wakeup the requested number of kthreads */
 626                 wake_up_nr(&ch->idle_wq, wakeup);
 627         }
 628
 629         if (needed <= 0)
 630                 return;
 631
 632         if (needed + assigned > ch->kthreads_assigned_limit) {
 633                 needed = ch->kthreads_assigned_limit - assigned;
 634                 if (needed <= 0)
 635                         return;
 636         }
 637
 638         dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n",
 639                 needed, ch->partid, ch->number);
 640
 641         xpc_create_kthreads(ch, needed, 0);
 642 }
 643
 644 /*
 645  * This function is where XPC's kthreads wait for messages to deliver.
 646  */
 647 static void
 648 xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
 649 {
 650         int (*n_of_deliverable_payloads) (struct xpc_channel *) =
 651                 xpc_arch_ops.n_of_deliverable_payloads;
 652
 653         do {
 654                 /* deliver messages to their intended recipients */
 655
 656                 while (n_of_deliverable_payloads(ch) > 0 &&
 657                        !(ch->flags & XPC_C_DISCONNECTING)) {
 658                         xpc_deliver_payload(ch);
 659                 }
 660
 661                 if (atomic_inc_return(&ch->kthreads_idle) >
 662                     ch->kthreads_idle_limit) {
 663                         /* too many idle kthreads on this channel */
 664                         atomic_dec(&ch->kthreads_idle);
 665                         break;
 666                 }
 667
 668                 dev_dbg(xpc_chan, "idle kthread calling "
 669                         "wait_event_interruptible_exclusive()\n");
 670
 671                 (void)wait_event_interruptible_exclusive(ch->idle_wq,
 672                                 (n_of_deliverable_payloads(ch) > 0 ||
 673                                  (ch->flags & XPC_C_DISCONNECTING)));
 674
 675                 atomic_dec(&ch->kthreads_idle);
 676
 677         } while (!(ch->flags & XPC_C_DISCONNECTING));
 678 }
 679
 680 static int
 681 xpc_kthread_start(void *args)
 682 {
 683         short partid = XPC_UNPACK_ARG1(args);
 684         u16 ch_number = XPC_UNPACK_ARG2(args);
 685         struct xpc_partition *part = &xpc_partitions[partid];
 686         struct xpc_channel *ch;
 687         int n_needed;
 688         unsigned long irq_flags;
 689         int (*n_of_deliverable_payloads) (struct xpc_channel *) =
 690                 xpc_arch_ops.n_of_deliverable_payloads;
 691
 692         dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n",
 693                 partid, ch_number);
 694
 695         ch = &part->channels[ch_number];
 696
 697         if (!(ch->flags & XPC_C_DISCONNECTING)) {
 698
 699                 /* let registerer know that connection has been established */
 700
 701                 spin_lock_irqsave(&ch->lock, irq_flags);
 702                 if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) {
 703                         ch->flags |= XPC_C_CONNECTEDCALLOUT;
 704                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 705
 706                         xpc_connected_callout(ch);
 707
 708                         spin_lock_irqsave(&ch->lock, irq_flags);
 709                         ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE;
 710                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 711
 712                         /*
 713                          * It is possible that while the callout was being
 714                          * made that the remote partition sent some messages.
 715                          * If that is the case, we may need to activate
 716                          * additional kthreads to help deliver them. We only
 717                          * need one less than total #of messages to deliver.
 718                          */
 719                         n_needed = n_of_deliverable_payloads(ch) - 1;
 720                         if (n_needed > 0 && !(ch->flags & XPC_C_DISCONNECTING))
 721                                 xpc_activate_kthreads(ch, n_needed);
 722
 723                 } else {
 724                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 725                 }
 726
 727                 xpc_kthread_waitmsgs(part, ch);
 728         }
 729
 730         /* let registerer know that connection is disconnecting */
 731
 732         spin_lock_irqsave(&ch->lock, irq_flags);
 733         if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
 734             !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) {
 735                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT;
 736                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 737
 738                 xpc_disconnect_callout(ch, xpDisconnecting);
 739
 740                 spin_lock_irqsave(&ch->lock, irq_flags);
 741                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE;
 742         }
 743         spin_unlock_irqrestore(&ch->lock, irq_flags);
 744
 745         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 746             atomic_dec_return(&part->nchannels_engaged) == 0) {
 747                 xpc_arch_ops.indicate_partition_disengaged(part);
 748         }
 749
 750         xpc_msgqueue_deref(ch);
 751
 752         dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n",
 753                 partid, ch_number);
 754
 755         xpc_part_deref(part);
 756         return 0;
 757 }
 758
 759 /*
 760  * For each partition that XPC has established communications with, there is
 761  * a minimum of one kernel thread assigned to perform any operation that
 762  * may potentially sleep or block (basically the callouts to the asynchronous
 763  * functions registered via xpc_connect()).
 764  *
 765  * Additional kthreads are created and destroyed by XPC as the workload
 766  * demands.
 767  *
 768  * A kthread is assigned to one of the active channels that exists for a given
 769  * partition.
 770  */
 771 void
 772 xpc_create_kthreads(struct xpc_channel *ch, int needed,
 773                     int ignore_disconnecting)
 774 {
 775         unsigned long irq_flags;
 776         u64 args = XPC_PACK_ARGS(ch->partid, ch->number);
 777         struct xpc_partition *part = &xpc_partitions[ch->partid];
 778         struct task_struct *kthread;
 779         void (*indicate_partition_disengaged) (struct xpc_partition *) =
 780                 xpc_arch_ops.indicate_partition_disengaged;
 781
 782         while (needed-- > 0) {
 783
 784                 /*
 785                  * The following is done on behalf of the newly created
 786                  * kthread. That kthread is responsible for doing the
 787                  * counterpart to the following before it exits.
 788                  */
 789                 if (ignore_disconnecting) {
 790                         if (!atomic_inc_not_zero(&ch->kthreads_assigned)) {
 791                                 /* kthreads assigned had gone to zero */
 792                                 BUG_ON(!(ch->flags &
 793                                          XPC_C_DISCONNECTINGCALLOUT_MADE));
 794                                 break;
 795                         }
 796
 797                 } else if (ch->flags & XPC_C_DISCONNECTING) {
 798                         break;
 799
 800                 } else if (atomic_inc_return(&ch->kthreads_assigned) == 1 &&
 801                            atomic_inc_return(&part->nchannels_engaged) == 1) {
 802                         xpc_arch_ops.indicate_partition_engaged(part);
 803                 }
 804                 (void)xpc_part_ref(part);
 805                 xpc_msgqueue_ref(ch);
 806
 807                 kthread = kthread_run(xpc_kthread_start, (void *)args,
 808                                       "xpc%02dc%d", ch->partid, ch->number);
 809                 if (IS_ERR(kthread)) {
 810                         /* the fork failed */
 811
 812                         /*
 813                          * NOTE: if (ignore_disconnecting &&
 814                          * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true,
 815                          * then we'll deadlock if all other kthreads assigned
 816                          * to this channel are blocked in the channel's
 817                          * registerer, because the only thing that will unblock
 818                          * them is the xpDisconnecting callout that this
 819                          * failed kthread_run() would have made.
 820                          */
 821
 822                         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 823                             atomic_dec_return(&part->nchannels_engaged) == 0) {
 824                                 indicate_partition_disengaged(part);
 825                         }
 826                         xpc_msgqueue_deref(ch);
 827                         xpc_part_deref(part);
 828
 829                         if (atomic_read(&ch->kthreads_assigned) <
 830                             ch->kthreads_idle_limit) {
 831                                 /*
 832                                  * Flag this as an error only if we have an
 833                                  * insufficient #of kthreads for the channel
 834                                  * to function.
 835                                  */
 836                                 spin_lock_irqsave(&ch->lock, irq_flags);
 837                                 XPC_DISCONNECT_CHANNEL(ch, xpLackOfResources,
 838                                                        &irq_flags);
 839                                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 840                         }
 841                         break;
 842                 }
 843         }
 844 }
 845
 846 void
 847 xpc_disconnect_wait(int ch_number)
 848 {
 849         unsigned long irq_flags;
 850         short partid;
 851         struct xpc_partition *part;
 852         struct xpc_channel *ch;
 853         int wakeup_channel_mgr;
 854
 855         /* now wait for all callouts to the caller's function to cease */
 856         for (partid = 0; partid < xp_max_npartitions; partid++) {
 857                 part = &xpc_partitions[partid];
 858
 859                 if (!xpc_part_ref(part))
 860                         continue;
 861
 862                 ch = &part->channels[ch_number];
 863
 864                 if (!(ch->flags & XPC_C_WDISCONNECT)) {
 865                         xpc_part_deref(part);
 866                         continue;
 867                 }
 868
 869                 wait_for_completion(&ch->wdisconnect_wait);
 870
 871                 spin_lock_irqsave(&ch->lock, irq_flags);
 872                 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
 873                 wakeup_channel_mgr = 0;
 874
 875                 if (ch->delayed_chctl_flags) {
 876                         if (part->act_state != XPC_P_AS_DEACTIVATING) {
 877                                 spin_lock(&part->chctl_lock);
 878                                 part->chctl.flags[ch->number] |=
 879                                     ch->delayed_chctl_flags;
 880                                 spin_unlock(&part->chctl_lock);
 881                                 wakeup_channel_mgr = 1;
 882                         }
 883                         ch->delayed_chctl_flags = 0;
 884                 }
 885
 886                 ch->flags &= ~XPC_C_WDISCONNECT;
 887                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 888
 889                 if (wakeup_channel_mgr)
 890                         xpc_wakeup_channel_mgr(part);
 891
 892                 xpc_part_deref(part);
 893         }
 894 }
 895
 896 static int
 897 xpc_setup_partitions(void)
 898 {
 899         short partid;
 900         struct xpc_partition *part;
 901
 902         xpc_partitions = kcalloc(xp_max_npartitions,
 903                                  sizeof(struct xpc_partition),
 904                                  GFP_KERNEL);
 905         if (xpc_partitions == NULL) {
 906                 dev_err(xpc_part, "can't get memory for partition structure\n");
 907                 return -ENOMEM;
 908         }
 909
 910         /*
 911          * The first few fields of each entry of xpc_partitions[] need to
 912          * be initialized now so that calls to xpc_connect() and
 913          * xpc_disconnect() can be made prior to the activation of any remote
 914          * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE
 915          * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING
 916          * PARTITION HAS BEEN ACTIVATED.
 917          */
 918         for (partid = 0; partid < xp_max_npartitions; partid++) {
 919                 part = &xpc_partitions[partid];
 920
 921                 DBUG_ON((u64)part != L1_CACHE_ALIGN((u64)part));
 922
 923                 part->activate_IRQ_rcvd = 0;
 924                 spin_lock_init(&part->act_lock);
 925                 part->act_state = XPC_P_AS_INACTIVE;
 926                 XPC_SET_REASON(part, 0, 0);
 927
 928                 timer_setup(&part->disengage_timer,
 929                             xpc_timeout_partition_disengage, 0);
 930
 931                 part->setup_state = XPC_P_SS_UNSET;
 932                 init_waitqueue_head(&part->teardown_wq);
 933                 atomic_set(&part->references, 0);
 934         }
 935
 936         return xpc_arch_ops.setup_partitions();
 937 }
 938
 939 static void
 940 xpc_teardown_partitions(void)
 941 {
 942         xpc_arch_ops.teardown_partitions();
 943         kfree(xpc_partitions);
 944 }
 945
 946 static void
 947 xpc_do_exit(enum xp_retval reason)
 948 {
 949         short partid;
 950         int active_part_count, printed_waiting_msg = 0;
 951         struct xpc_partition *part;
 952         unsigned long printmsg_time, disengage_timeout = 0;
 953
 954         /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
 955         DBUG_ON(xpc_exiting == 1);
 956
 957         /*
 958          * Let the heartbeat checker thread and the discovery thread
 959          * (if one is running) know that they should exit. Also wake up
 960          * the heartbeat checker thread in case it's sleeping.
 961          */
 962         xpc_exiting = 1;
 963         wake_up_interruptible(&xpc_activate_IRQ_wq);
 964
 965         /* wait for the discovery thread to exit */
 966         wait_for_completion(&xpc_discovery_exited);
 967
 968         /* wait for the heartbeat checker thread to exit */
 969         wait_for_completion(&xpc_hb_checker_exited);
 970
 971         /* sleep for a 1/3 of a second or so */
 972         (void)msleep_interruptible(300);
 973
 974         /* wait for all partitions to become inactive */
 975
 976         printmsg_time = jiffies + (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
 977         xpc_disengage_timedout = 0;
 978
 979         do {
 980                 active_part_count = 0;
 981
 982                 for (partid = 0; partid < xp_max_npartitions; partid++) {
 983                         part = &xpc_partitions[partid];
 984
 985                         if (xpc_partition_disengaged(part) &&
 986                             part->act_state == XPC_P_AS_INACTIVE) {
 987                                 continue;
 988                         }
 989
 990                         active_part_count++;
 991
 992                         XPC_DEACTIVATE_PARTITION(part, reason);
 993
 994                         if (part->disengage_timeout > disengage_timeout)
 995                                 disengage_timeout = part->disengage_timeout;
 996                 }
 997
 998                 if (xpc_arch_ops.any_partition_engaged()) {
 999                         if (time_is_before_jiffies(printmsg_time)) {
1000                                 dev_info(xpc_part, "waiting for remote "
1001                                          "partitions to deactivate, timeout in "
1002                                          "%ld seconds\n", (disengage_timeout -
1003                                          jiffies) / HZ);
1004                                 printmsg_time = jiffies +
1005                                     (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
1006                                 printed_waiting_msg = 1;
1007                         }
1008
1009                 } else if (active_part_count > 0) {
1010                         if (printed_waiting_msg) {
1011                                 dev_info(xpc_part, "waiting for local partition"
1012                                          " to deactivate\n");
1013                                 printed_waiting_msg = 0;
1014                         }
1015
1016                 } else {
1017                         if (!xpc_disengage_timedout) {
1018                                 dev_info(xpc_part, "all partitions have "
1019                                          "deactivated\n");
1020                         }
1021                         break;
1022                 }
1023
1024                 /* sleep for a 1/3 of a second or so */
1025                 (void)msleep_interruptible(300);
1026
1027         } while (1);
1028
1029         DBUG_ON(xpc_arch_ops.any_partition_engaged());
1030
1031         xpc_teardown_rsvd_page();
1032
1033         if (reason == xpUnloading) {
1034                 (void)unregister_die_notifier(&xpc_die_notifier);
1035                 (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1036         }
1037
1038         /* clear the interface to XPC's functions */
1039         xpc_clear_interface();
1040
1041         if (xpc_sysctl)
1042                 unregister_sysctl_table(xpc_sysctl);
1043
1044         xpc_teardown_partitions();
1045
1046         if (is_uv())
1047                 xpc_exit_uv();
1048 }
1049
1050 /*
1051  * This function is called when the system is being rebooted.
1052  */
1053 static int
1054 xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
1055 {
1056         enum xp_retval reason;
1057
1058         switch (event) {
1059         case SYS_RESTART:
1060                 reason = xpSystemReboot;
1061                 break;
1062         case SYS_HALT:
1063                 reason = xpSystemHalt;
1064                 break;
1065         case SYS_POWER_OFF:
1066                 reason = xpSystemPoweroff;
1067                 break;
1068         default:
1069                 reason = xpSystemGoingDown;
1070         }
1071
1072         xpc_do_exit(reason);
1073         return NOTIFY_DONE;
1074 }
1075
1076 /* Used to only allow one cpu to complete disconnect */
1077 static unsigned int xpc_die_disconnecting;
1078
1079 /*
1080  * Notify other partitions to deactivate from us by first disengaging from all
1081  * references to our memory.
1082  */
1083 static void
1084 xpc_die_deactivate(void)
1085 {
1086         struct xpc_partition *part;
1087         short partid;
1088         int any_engaged;
1089         long keep_waiting;
1090         long wait_to_print;
1091
1092         if (cmpxchg(&xpc_die_disconnecting, 0, 1))
1093                 return;
1094
1095         /* keep xpc_hb_checker thread from doing anything (just in case) */
1096         xpc_exiting = 1;
1097
1098         xpc_arch_ops.disallow_all_hbs();   /*indicate we're deactivated */
1099
1100         for (partid = 0; partid < xp_max_npartitions; partid++) {
1101                 part = &xpc_partitions[partid];
1102
1103                 if (xpc_arch_ops.partition_engaged(partid) ||
1104                     part->act_state != XPC_P_AS_INACTIVE) {
1105                         xpc_arch_ops.request_partition_deactivation(part);
1106                         xpc_arch_ops.indicate_partition_disengaged(part);
1107                 }
1108         }
1109
1110         /*
1111          * Though we requested that all other partitions deactivate from us,
1112          * we only wait until they've all disengaged or we've reached the
1113          * defined timelimit.
1114          *
1115          * Given that one iteration through the following while-loop takes
1116          * approximately 200 microseconds, calculate the #of loops to take
1117          * before bailing and the #of loops before printing a waiting message.
1118          */
1119         keep_waiting = xpc_disengage_timelimit * 1000 * 5;
1120         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL * 1000 * 5;
1121
1122         while (1) {
1123                 any_engaged = xpc_arch_ops.any_partition_engaged();
1124                 if (!any_engaged) {
1125                         dev_info(xpc_part, "all partitions have deactivated\n");
1126                         break;
1127                 }
1128
1129                 if (!keep_waiting--) {
1130                         for (partid = 0; partid < xp_max_npartitions;
1131                              partid++) {
1132                                 if (xpc_arch_ops.partition_engaged(partid)) {
1133                                         dev_info(xpc_part, "deactivate from "
1134                                                  "remote partition %d timed "
1135                                                  "out\n", partid);
1136                                 }
1137                         }
1138                         break;
1139                 }
1140
1141                 if (!wait_to_print--) {
1142                         dev_info(xpc_part, "waiting for remote partitions to "
1143                                  "deactivate, timeout in %ld seconds\n",
1144                                  keep_waiting / (1000 * 5));
1145                         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL *
1146                             1000 * 5;
1147                 }
1148
1149                 udelay(200);
1150         }
1151 }
1152
1153 /*
1154  * This function is called when the system is being restarted or halted due
1155  * to some sort of system failure. If this is the case we need to notify the
1156  * other partitions to disengage from all references to our memory.
1157  * This function can also be called when our heartbeater could be offlined
1158  * for a time. In this case we need to notify other partitions to not worry
1159  * about the lack of a heartbeat.
1160  */
1161 static int
1162 xpc_system_die(struct notifier_block *nb, unsigned long event, void *_die_args)
1163 {
1164 #ifdef CONFIG_IA64              /* !!! temporary kludge */
1165         switch (event) {
1166         case DIE_MACHINE_RESTART:
1167         case DIE_MACHINE_HALT:
1168                 xpc_die_deactivate();
1169                 break;
1170
1171         case DIE_KDEBUG_ENTER:
1172                 /* Should lack of heartbeat be ignored by other partitions? */
1173                 if (!xpc_kdebug_ignore)
1174                         break;
1175
1176                 /* fall through */
1177         case DIE_MCA_MONARCH_ENTER:
1178         case DIE_INIT_MONARCH_ENTER:
1179                 xpc_arch_ops.offline_heartbeat();
1180                 break;
1181
1182         case DIE_KDEBUG_LEAVE:
1183                 /* Is lack of heartbeat being ignored by other partitions? */
1184                 if (!xpc_kdebug_ignore)
1185                         break;
1186
1187                 /* fall through */
1188         case DIE_MCA_MONARCH_LEAVE:
1189         case DIE_INIT_MONARCH_LEAVE:
1190                 xpc_arch_ops.online_heartbeat();
1191                 break;
1192         }
1193 #else
1194         struct die_args *die_args = _die_args;
1195
1196         switch (event) {
1197         case DIE_TRAP:
1198                 if (die_args->trapnr == X86_TRAP_DF)
1199                         xpc_die_deactivate();
1200
1201                 if (((die_args->trapnr == X86_TRAP_MF) ||
1202                      (die_args->trapnr == X86_TRAP_XF)) &&
1203                     !user_mode(die_args->regs))
1204                         xpc_die_deactivate();
1205
1206                 break;
1207         case DIE_INT3:
1208         case DIE_DEBUG:
1209                 break;
1210         case DIE_OOPS:
1211         case DIE_GPF:
1212         default:
1213                 xpc_die_deactivate();
1214         }
1215 #endif
1216
1217         return NOTIFY_DONE;
1218 }
1219
1220 static int __init
1221 xpc_init(void)
1222 {
1223         int ret;
1224         struct task_struct *kthread;
1225
1226         dev_set_name(xpc_part, "part");
1227         dev_set_name(xpc_chan, "chan");
1228
1229         if (is_uv()) {
1230                 ret = xpc_init_uv();
1231
1232         } else {
1233                 ret = -ENODEV;
1234         }
1235
1236         if (ret != 0)
1237                 return ret;
1238
1239         ret = xpc_setup_partitions();
1240         if (ret != 0) {
1241                 dev_err(xpc_part, "can't get memory for partition structure\n");
1242                 goto out_1;
1243         }
1244
1245         xpc_sysctl = register_sysctl_table(xpc_sys_dir);
1246
1247         /*
1248          * Fill the partition reserved page with the information needed by
1249          * other partitions to discover we are alive and establish initial
1250          * communications.
1251          */
1252         ret = xpc_setup_rsvd_page();
1253         if (ret != 0) {
1254                 dev_err(xpc_part, "can't setup our reserved page\n");
1255                 goto out_2;
1256         }
1257
1258         /* add ourselves to the reboot_notifier_list */
1259         ret = register_reboot_notifier(&xpc_reboot_notifier);
1260         if (ret != 0)
1261                 dev_warn(xpc_part, "can't register reboot notifier\n");
1262
1263         /* add ourselves to the die_notifier list */
1264         ret = register_die_notifier(&xpc_die_notifier);
1265         if (ret != 0)
1266                 dev_warn(xpc_part, "can't register die notifier\n");
1267
1268         /*
1269          * The real work-horse behind xpc.  This processes incoming
1270          * interrupts and monitors remote heartbeats.
1271          */
1272         kthread = kthread_run(xpc_hb_checker, NULL, XPC_HB_CHECK_THREAD_NAME);
1273         if (IS_ERR(kthread)) {
1274                 dev_err(xpc_part, "failed while forking hb check thread\n");
1275                 ret = -EBUSY;
1276                 goto out_3;
1277         }
1278
1279         /*
1280          * Startup a thread that will attempt to discover other partitions to
1281          * activate based on info provided by SAL. This new thread is short
1282          * lived and will exit once discovery is complete.
1283          */
1284         kthread = kthread_run(xpc_initiate_discovery, NULL,
1285                               XPC_DISCOVERY_THREAD_NAME);
1286         if (IS_ERR(kthread)) {
1287                 dev_err(xpc_part, "failed while forking discovery thread\n");
1288
1289                 /* mark this new thread as a non-starter */
1290                 complete(&xpc_discovery_exited);
1291
1292                 xpc_do_exit(xpUnloading);
1293                 return -EBUSY;
1294         }
1295
1296         /* set the interface to point at XPC's functions */
1297         xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect,
1298                           xpc_initiate_send, xpc_initiate_send_notify,
1299                           xpc_initiate_received, xpc_initiate_partid_to_nasids);
1300
1301         return 0;
1302
1303         /* initialization was not successful */
1304 out_3:
1305         xpc_teardown_rsvd_page();
1306
1307         (void)unregister_die_notifier(&xpc_die_notifier);
1308         (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1309 out_2:
1310         if (xpc_sysctl)
1311                 unregister_sysctl_table(xpc_sysctl);
1312
1313         xpc_teardown_partitions();
1314 out_1:
1315         if (is_uv())
1316                 xpc_exit_uv();
1317         return ret;
1318 }
1319
1320 module_init(xpc_init);
1321
1322 static void __exit
1323 xpc_exit(void)
1324 {
1325         xpc_do_exit(xpUnloading);
1326 }
1327
1328 module_exit(xpc_exit);
1329
1330 MODULE_AUTHOR("Silicon Graphics, Inc.");
1331 MODULE_DESCRIPTION("Cross Partition Communication (XPC) support");
1332 MODULE_LICENSE("GPL");
1333
1334 module_param(xpc_hb_interval, int, 0);
1335 MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between "
1336                  "heartbeat increments.");
1337
1338 module_param(xpc_hb_check_interval, int, 0);
1339 MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between "
1340                  "heartbeat checks.");
1341
1342 module_param(xpc_disengage_timelimit, int, 0);
1343 MODULE_PARM_DESC(xpc_disengage_timelimit, "Number of seconds to wait "
1344                  "for disengage to complete.");
1345
1346 module_param(xpc_kdebug_ignore, int, 0);
1347 MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by "
1348                  "other partitions when dropping into kdebug.");