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  * (C) Copyright 2020 Hewlett Packard Enterprise Development LP
   7  * Copyright (c) 2004-2009 Silicon Graphics, Inc.  All Rights Reserved.
   8  */
   9
  10 /*
  11  * Cross Partition Communication (XPC) support - standard version.
  12  *
  13  *      XPC provides a message passing capability that crosses partition
  14  *      boundaries. This module is made up of two parts:
  15  *
  16  *          partition   This part detects the presence/absence of other
  17  *                      partitions. It provides a heartbeat and monitors
  18  *                      the heartbeats of other partitions.
  19  *
  20  *          channel     This part manages the channels and sends/receives
  21  *                      messages across them to/from other partitions.
  22  *
  23  *      There are a couple of additional functions residing in XP, which
  24  *      provide an interface to XPC for its users.
  25  *
  26  *
  27  *      Caveats:
  28  *
  29  *        . Currently on sn2, we have no way to determine which nasid an IRQ
  30  *          came from. Thus, xpc_send_IRQ_sn2() does a remote amo write
  31  *          followed by an IPI. The amo indicates where data is to be pulled
  32  *          from, so after the IPI arrives, the remote partition checks the amo
  33  *          word. The IPI can actually arrive before the amo however, so other
  34  *          code must periodically check for this case. Also, remote amo
  35  *          operations do not reliably time out. Thus we do a remote PIO read
  36  *          solely to know whether the remote partition is down and whether we
  37  *          should stop sending IPIs to it. This remote PIO read operation is
  38  *          set up in a special nofault region so SAL knows to ignore (and
  39  *          cleanup) any errors due to the remote amo write, PIO read, and/or
  40  *          PIO write operations.
  41  *
  42  *          If/when new hardware solves this IPI problem, we should abandon
  43  *          the current approach.
  44  *
  45  */
  46
  47 #include <linux/module.h>
  48 #include <linux/slab.h>
  49 #include <linux/sysctl.h>
  50 #include <linux/device.h>
  51 #include <linux/delay.h>
  52 #include <linux/reboot.h>
  53 #include <linux/kdebug.h>
  54 #include <linux/kthread.h>
  55 #include "xpc.h"
  56
  57 #ifdef CONFIG_X86_64
  58 #include <asm/traps.h>
  59 #endif
  60
  61 /* define two XPC debug device structures to be used with dev_dbg() et al */
  62
  63 static struct device_driver xpc_dbg_name = {
  64         .name = "xpc"
  65 };
  66
  67 static struct device xpc_part_dbg_subname = {
  68         .init_name = "",        /* set to "part" at xpc_init() time */
  69         .driver = &xpc_dbg_name
  70 };
  71
  72 static struct device xpc_chan_dbg_subname = {
  73         .init_name = "",        /* set to "chan" at xpc_init() time */
  74         .driver = &xpc_dbg_name
  75 };
  76
  77 struct device *xpc_part = &xpc_part_dbg_subname;
  78 struct device *xpc_chan = &xpc_chan_dbg_subname;
  79
  80 static int xpc_kdebug_ignore;
  81
  82 /* systune related variables for /proc/sys directories */
  83
  84 static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
  85 static int xpc_hb_min_interval = 1;
  86 static int xpc_hb_max_interval = 10;
  87
  88 static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL;
  89 static int xpc_hb_check_min_interval = 10;
  90 static int xpc_hb_check_max_interval = 120;
  91
  92 int xpc_disengage_timelimit = XPC_DISENGAGE_DEFAULT_TIMELIMIT;
  93 static int xpc_disengage_min_timelimit; /* = 0 */
  94 static int xpc_disengage_max_timelimit = 120;
  95
  96 static struct ctl_table xpc_sys_xpc_hb[] = {
  97         {
  98          .procname = "hb_interval",
  99          .data = &xpc_hb_interval,
 100          .maxlen = sizeof(int),
 101          .mode = 0644,
 102          .proc_handler = proc_dointvec_minmax,
 103          .extra1 = &xpc_hb_min_interval,
 104          .extra2 = &xpc_hb_max_interval},
 105         {
 106          .procname = "hb_check_interval",
 107          .data = &xpc_hb_check_interval,
 108          .maxlen = sizeof(int),
 109          .mode = 0644,
 110          .proc_handler = proc_dointvec_minmax,
 111          .extra1 = &xpc_hb_check_min_interval,
 112          .extra2 = &xpc_hb_check_max_interval},
 113         {}
 114 };
 115 static struct ctl_table xpc_sys_xpc[] = {
 116         {
 117          .procname = "disengage_timelimit",
 118          .data = &xpc_disengage_timelimit,
 119          .maxlen = sizeof(int),
 120          .mode = 0644,
 121          .proc_handler = proc_dointvec_minmax,
 122          .extra1 = &xpc_disengage_min_timelimit,
 123          .extra2 = &xpc_disengage_max_timelimit},
 124         {}
 125 };
 126
 127 static struct ctl_table_header *xpc_sysctl;
 128 static struct ctl_table_header *xpc_sysctl_hb;
 129
 130 /* non-zero if any remote partition disengage was timed out */
 131 int xpc_disengage_timedout;
 132
 133 /* #of activate IRQs received and not yet processed */
 134 int xpc_activate_IRQ_rcvd;
 135 DEFINE_SPINLOCK(xpc_activate_IRQ_rcvd_lock);
 136
 137 /* IRQ handler notifies this wait queue on receipt of an IRQ */
 138 DECLARE_WAIT_QUEUE_HEAD(xpc_activate_IRQ_wq);
 139
 140 static unsigned long xpc_hb_check_timeout;
 141 static struct timer_list xpc_hb_timer;
 142
 143 /* notification that the xpc_hb_checker thread has exited */
 144 static DECLARE_COMPLETION(xpc_hb_checker_exited);
 145
 146 /* notification that the xpc_discovery thread has exited */
 147 static DECLARE_COMPLETION(xpc_discovery_exited);
 148
 149 static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *);
 150
 151 static int xpc_system_reboot(struct notifier_block *, unsigned long, void *);
 152 static struct notifier_block xpc_reboot_notifier = {
 153         .notifier_call = xpc_system_reboot,
 154 };
 155
 156 static int xpc_system_die(struct notifier_block *, unsigned long, void *);
 157 static struct notifier_block xpc_die_notifier = {
 158         .notifier_call = xpc_system_die,
 159 };
 160
 161 struct xpc_arch_operations xpc_arch_ops;
 162
 163 /*
 164  * Timer function to enforce the timelimit on the partition disengage.
 165  */
 166 static void
 167 xpc_timeout_partition_disengage(struct timer_list *t)
 168 {
 169         struct xpc_partition *part = from_timer(part, t, disengage_timer);
 170
 171         DBUG_ON(time_is_after_jiffies(part->disengage_timeout));
 172
 173         xpc_partition_disengaged_from_timer(part);
 174
 175         DBUG_ON(part->disengage_timeout != 0);
 176         DBUG_ON(xpc_arch_ops.partition_engaged(XPC_PARTID(part)));
 177 }
 178
 179 /*
 180  * Timer to produce the heartbeat.  The timer structures function is
 181  * already set when this is initially called.  A tunable is used to
 182  * specify when the next timeout should occur.
 183  */
 184 static void
 185 xpc_hb_beater(struct timer_list *unused)
 186 {
 187         xpc_arch_ops.increment_heartbeat();
 188
 189         if (time_is_before_eq_jiffies(xpc_hb_check_timeout))
 190                 wake_up_interruptible(&xpc_activate_IRQ_wq);
 191
 192         xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ);
 193         add_timer(&xpc_hb_timer);
 194 }
 195
 196 static void
 197 xpc_start_hb_beater(void)
 198 {
 199         xpc_arch_ops.heartbeat_init();
 200         timer_setup(&xpc_hb_timer, xpc_hb_beater, 0);
 201         xpc_hb_beater(NULL);
 202 }
 203
 204 static void
 205 xpc_stop_hb_beater(void)
 206 {
 207         del_timer_sync(&xpc_hb_timer);
 208         xpc_arch_ops.heartbeat_exit();
 209 }
 210
 211 /*
 212  * At periodic intervals, scan through all active partitions and ensure
 213  * their heartbeat is still active.  If not, the partition is deactivated.
 214  */
 215 static void
 216 xpc_check_remote_hb(void)
 217 {
 218         struct xpc_partition *part;
 219         short partid;
 220         enum xp_retval ret;
 221
 222         for (partid = 0; partid < xp_max_npartitions; partid++) {
 223
 224                 if (xpc_exiting)
 225                         break;
 226
 227                 if (partid == xp_partition_id)
 228                         continue;
 229
 230                 part = &xpc_partitions[partid];
 231
 232                 if (part->act_state == XPC_P_AS_INACTIVE ||
 233                     part->act_state == XPC_P_AS_DEACTIVATING) {
 234                         continue;
 235                 }
 236
 237                 ret = xpc_arch_ops.get_remote_heartbeat(part);
 238                 if (ret != xpSuccess)
 239                         XPC_DEACTIVATE_PARTITION(part, ret);
 240         }
 241 }
 242
 243 /*
 244  * This thread is responsible for nearly all of the partition
 245  * activation/deactivation.
 246  */
 247 static int
 248 xpc_hb_checker(void *ignore)
 249 {
 250         int force_IRQ = 0;
 251
 252         /* this thread was marked active by xpc_hb_init() */
 253
 254         set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
 255
 256         /* set our heartbeating to other partitions into motion */
 257         xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
 258         xpc_start_hb_beater();
 259
 260         while (!xpc_exiting) {
 261
 262                 dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have "
 263                         "been received\n",
 264                         (int)(xpc_hb_check_timeout - jiffies),
 265                         xpc_activate_IRQ_rcvd);
 266
 267                 /* checking of remote heartbeats is skewed by IRQ handling */
 268                 if (time_is_before_eq_jiffies(xpc_hb_check_timeout)) {
 269                         xpc_hb_check_timeout = jiffies +
 270                             (xpc_hb_check_interval * HZ);
 271
 272                         dev_dbg(xpc_part, "checking remote heartbeats\n");
 273                         xpc_check_remote_hb();
 274                 }
 275
 276                 /* check for outstanding IRQs */
 277                 if (xpc_activate_IRQ_rcvd > 0 || force_IRQ != 0) {
 278                         force_IRQ = 0;
 279                         dev_dbg(xpc_part, "processing activate IRQs "
 280                                 "received\n");
 281                         xpc_arch_ops.process_activate_IRQ_rcvd();
 282                 }
 283
 284                 /* wait for IRQ or timeout */
 285                 (void)wait_event_interruptible(xpc_activate_IRQ_wq,
 286                                                (time_is_before_eq_jiffies(
 287                                                 xpc_hb_check_timeout) ||
 288                                                 xpc_activate_IRQ_rcvd > 0 ||
 289                                                 xpc_exiting));
 290         }
 291
 292         xpc_stop_hb_beater();
 293
 294         dev_dbg(xpc_part, "heartbeat checker is exiting\n");
 295
 296         /* mark this thread as having exited */
 297         complete(&xpc_hb_checker_exited);
 298         return 0;
 299 }
 300
 301 /*
 302  * This thread will attempt to discover other partitions to activate
 303  * based on info provided by SAL. This new thread is short lived and
 304  * will exit once discovery is complete.
 305  */
 306 static int
 307 xpc_initiate_discovery(void *ignore)
 308 {
 309         xpc_discovery();
 310
 311         dev_dbg(xpc_part, "discovery thread is exiting\n");
 312
 313         /* mark this thread as having exited */
 314         complete(&xpc_discovery_exited);
 315         return 0;
 316 }
 317
 318 /*
 319  * The first kthread assigned to a newly activated partition is the one
 320  * created by XPC HB with which it calls xpc_activating(). XPC hangs on to
 321  * that kthread until the partition is brought down, at which time that kthread
 322  * returns back to XPC HB. (The return of that kthread will signify to XPC HB
 323  * that XPC has dismantled all communication infrastructure for the associated
 324  * partition.) This kthread becomes the channel manager for that partition.
 325  *
 326  * Each active partition has a channel manager, who, besides connecting and
 327  * disconnecting channels, will ensure that each of the partition's connected
 328  * channels has the required number of assigned kthreads to get the work done.
 329  */
 330 static void
 331 xpc_channel_mgr(struct xpc_partition *part)
 332 {
 333         while (part->act_state != XPC_P_AS_DEACTIVATING ||
 334                atomic_read(&part->nchannels_active) > 0 ||
 335                !xpc_partition_disengaged(part)) {
 336
 337                 xpc_process_sent_chctl_flags(part);
 338
 339                 /*
 340                  * Wait until we've been requested to activate kthreads or
 341                  * all of the channel's message queues have been torn down or
 342                  * a signal is pending.
 343                  *
 344                  * The channel_mgr_requests is set to 1 after being awakened,
 345                  * This is done to prevent the channel mgr from making one pass
 346                  * through the loop for each request, since he will
 347                  * be servicing all the requests in one pass. The reason it's
 348                  * set to 1 instead of 0 is so that other kthreads will know
 349                  * that the channel mgr is running and won't bother trying to
 350                  * wake him up.
 351                  */
 352                 atomic_dec(&part->channel_mgr_requests);
 353                 (void)wait_event_interruptible(part->channel_mgr_wq,
 354                                 (atomic_read(&part->channel_mgr_requests) > 0 ||
 355                                  part->chctl.all_flags != 0 ||
 356                                  (part->act_state == XPC_P_AS_DEACTIVATING &&
 357                                  atomic_read(&part->nchannels_active) == 0 &&
 358                                  xpc_partition_disengaged(part))));
 359                 atomic_set(&part->channel_mgr_requests, 1);
 360         }
 361 }
 362
 363 /*
 364  * Guarantee that the kzalloc'd memory is cacheline aligned.
 365  */
 366 void *
 367 xpc_kzalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
 368 {
 369         /* see if kzalloc will give us cachline aligned memory by default */
 370         *base = kzalloc(size, flags);
 371         if (*base == NULL)
 372                 return NULL;
 373
 374         if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
 375                 return *base;
 376
 377         kfree(*base);
 378
 379         /* nope, we'll have to do it ourselves */
 380         *base = kzalloc(size + L1_CACHE_BYTES, flags);
 381         if (*base == NULL)
 382                 return NULL;
 383
 384         return (void *)L1_CACHE_ALIGN((u64)*base);
 385 }
 386
 387 /*
 388  * Setup the channel structures necessary to support XPartition Communication
 389  * between the specified remote partition and the local one.
 390  */
 391 static enum xp_retval
 392 xpc_setup_ch_structures(struct xpc_partition *part)
 393 {
 394         enum xp_retval ret;
 395         int ch_number;
 396         struct xpc_channel *ch;
 397         short partid = XPC_PARTID(part);
 398
 399         /*
 400          * Allocate all of the channel structures as a contiguous chunk of
 401          * memory.
 402          */
 403         DBUG_ON(part->channels != NULL);
 404         part->channels = kcalloc(XPC_MAX_NCHANNELS,
 405                                  sizeof(struct xpc_channel),
 406                                  GFP_KERNEL);
 407         if (part->channels == NULL) {
 408                 dev_err(xpc_chan, "can't get memory for channels\n");
 409                 return xpNoMemory;
 410         }
 411
 412         /* allocate the remote open and close args */
 413
 414         part->remote_openclose_args =
 415             xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE,
 416                                           GFP_KERNEL, &part->
 417                                           remote_openclose_args_base);
 418         if (part->remote_openclose_args == NULL) {
 419                 dev_err(xpc_chan, "can't get memory for remote connect args\n");
 420                 ret = xpNoMemory;
 421                 goto out_1;
 422         }
 423
 424         part->chctl.all_flags = 0;
 425         spin_lock_init(&part->chctl_lock);
 426
 427         atomic_set(&part->channel_mgr_requests, 1);
 428         init_waitqueue_head(&part->channel_mgr_wq);
 429
 430         part->nchannels = XPC_MAX_NCHANNELS;
 431
 432         atomic_set(&part->nchannels_active, 0);
 433         atomic_set(&part->nchannels_engaged, 0);
 434
 435         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
 436                 ch = &part->channels[ch_number];
 437
 438                 ch->partid = partid;
 439                 ch->number = ch_number;
 440                 ch->flags = XPC_C_DISCONNECTED;
 441
 442                 atomic_set(&ch->kthreads_assigned, 0);
 443                 atomic_set(&ch->kthreads_idle, 0);
 444                 atomic_set(&ch->kthreads_active, 0);
 445
 446                 atomic_set(&ch->references, 0);
 447                 atomic_set(&ch->n_to_notify, 0);
 448
 449                 spin_lock_init(&ch->lock);
 450                 init_completion(&ch->wdisconnect_wait);
 451
 452                 atomic_set(&ch->n_on_msg_allocate_wq, 0);
 453                 init_waitqueue_head(&ch->msg_allocate_wq);
 454                 init_waitqueue_head(&ch->idle_wq);
 455         }
 456
 457         ret = xpc_arch_ops.setup_ch_structures(part);
 458         if (ret != xpSuccess)
 459                 goto out_2;
 460
 461         /*
 462          * With the setting of the partition setup_state to XPC_P_SS_SETUP,
 463          * we're declaring that this partition is ready to go.
 464          */
 465         part->setup_state = XPC_P_SS_SETUP;
 466
 467         return xpSuccess;
 468
 469         /* setup of ch structures failed */
 470 out_2:
 471         kfree(part->remote_openclose_args_base);
 472         part->remote_openclose_args = NULL;
 473 out_1:
 474         kfree(part->channels);
 475         part->channels = NULL;
 476         return ret;
 477 }
 478
 479 /*
 480  * Teardown the channel structures necessary to support XPartition Communication
 481  * between the specified remote partition and the local one.
 482  */
 483 static void
 484 xpc_teardown_ch_structures(struct xpc_partition *part)
 485 {
 486         DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
 487         DBUG_ON(atomic_read(&part->nchannels_active) != 0);
 488
 489         /*
 490          * Make this partition inaccessible to local processes by marking it
 491          * as no longer setup. Then wait before proceeding with the teardown
 492          * until all existing references cease.
 493          */
 494         DBUG_ON(part->setup_state != XPC_P_SS_SETUP);
 495         part->setup_state = XPC_P_SS_WTEARDOWN;
 496
 497         wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
 498
 499         /* now we can begin tearing down the infrastructure */
 500
 501         xpc_arch_ops.teardown_ch_structures(part);
 502
 503         kfree(part->remote_openclose_args_base);
 504         part->remote_openclose_args = NULL;
 505         kfree(part->channels);
 506         part->channels = NULL;
 507
 508         part->setup_state = XPC_P_SS_TORNDOWN;
 509 }
 510
 511 /*
 512  * When XPC HB determines that a partition has come up, it will create a new
 513  * kthread and that kthread will call this function to attempt to set up the
 514  * basic infrastructure used for Cross Partition Communication with the newly
 515  * upped partition.
 516  *
 517  * The kthread that was created by XPC HB and which setup the XPC
 518  * infrastructure will remain assigned to the partition becoming the channel
 519  * manager for that partition until the partition is deactivating, at which
 520  * time the kthread will teardown the XPC infrastructure and then exit.
 521  */
 522 static int
 523 xpc_activating(void *__partid)
 524 {
 525         short partid = (u64)__partid;
 526         struct xpc_partition *part = &xpc_partitions[partid];
 527         unsigned long irq_flags;
 528
 529         DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
 530
 531         spin_lock_irqsave(&part->act_lock, irq_flags);
 532
 533         if (part->act_state == XPC_P_AS_DEACTIVATING) {
 534                 part->act_state = XPC_P_AS_INACTIVE;
 535                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 536                 part->remote_rp_pa = 0;
 537                 return 0;
 538         }
 539
 540         /* indicate the thread is activating */
 541         DBUG_ON(part->act_state != XPC_P_AS_ACTIVATION_REQ);
 542         part->act_state = XPC_P_AS_ACTIVATING;
 543
 544         XPC_SET_REASON(part, 0, 0);
 545         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 546
 547         dev_dbg(xpc_part, "activating partition %d\n", partid);
 548
 549         xpc_arch_ops.allow_hb(partid);
 550
 551         if (xpc_setup_ch_structures(part) == xpSuccess) {
 552                 (void)xpc_part_ref(part);       /* this will always succeed */
 553
 554                 if (xpc_arch_ops.make_first_contact(part) == xpSuccess) {
 555                         xpc_mark_partition_active(part);
 556                         xpc_channel_mgr(part);
 557                         /* won't return until partition is deactivating */
 558                 }
 559
 560                 xpc_part_deref(part);
 561                 xpc_teardown_ch_structures(part);
 562         }
 563
 564         xpc_arch_ops.disallow_hb(partid);
 565         xpc_mark_partition_inactive(part);
 566
 567         if (part->reason == xpReactivating) {
 568                 /* interrupting ourselves results in activating partition */
 569                 xpc_arch_ops.request_partition_reactivation(part);
 570         }
 571
 572         return 0;
 573 }
 574
 575 void
 576 xpc_activate_partition(struct xpc_partition *part)
 577 {
 578         short partid = XPC_PARTID(part);
 579         unsigned long irq_flags;
 580         struct task_struct *kthread;
 581
 582         spin_lock_irqsave(&part->act_lock, irq_flags);
 583
 584         DBUG_ON(part->act_state != XPC_P_AS_INACTIVE);
 585
 586         part->act_state = XPC_P_AS_ACTIVATION_REQ;
 587         XPC_SET_REASON(part, xpCloneKThread, __LINE__);
 588
 589         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 590
 591         kthread = kthread_run(xpc_activating, (void *)((u64)partid), "xpc%02d",
 592                               partid);
 593         if (IS_ERR(kthread)) {
 594                 spin_lock_irqsave(&part->act_lock, irq_flags);
 595                 part->act_state = XPC_P_AS_INACTIVE;
 596                 XPC_SET_REASON(part, xpCloneKThreadFailed, __LINE__);
 597                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 598         }
 599 }
 600
 601 void
 602 xpc_activate_kthreads(struct xpc_channel *ch, int needed)
 603 {
 604         int idle = atomic_read(&ch->kthreads_idle);
 605         int assigned = atomic_read(&ch->kthreads_assigned);
 606         int wakeup;
 607
 608         DBUG_ON(needed <= 0);
 609
 610         if (idle > 0) {
 611                 wakeup = (needed > idle) ? idle : needed;
 612                 needed -= wakeup;
 613
 614                 dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, "
 615                         "channel=%d\n", wakeup, ch->partid, ch->number);
 616
 617                 /* only wakeup the requested number of kthreads */
 618                 wake_up_nr(&ch->idle_wq, wakeup);
 619         }
 620
 621         if (needed <= 0)
 622                 return;
 623
 624         if (needed + assigned > ch->kthreads_assigned_limit) {
 625                 needed = ch->kthreads_assigned_limit - assigned;
 626                 if (needed <= 0)
 627                         return;
 628         }
 629
 630         dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n",
 631                 needed, ch->partid, ch->number);
 632
 633         xpc_create_kthreads(ch, needed, 0);
 634 }
 635
 636 /*
 637  * This function is where XPC's kthreads wait for messages to deliver.
 638  */
 639 static void
 640 xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
 641 {
 642         int (*n_of_deliverable_payloads) (struct xpc_channel *) =
 643                 xpc_arch_ops.n_of_deliverable_payloads;
 644
 645         do {
 646                 /* deliver messages to their intended recipients */
 647
 648                 while (n_of_deliverable_payloads(ch) > 0 &&
 649                        !(ch->flags & XPC_C_DISCONNECTING)) {
 650                         xpc_deliver_payload(ch);
 651                 }
 652
 653                 if (atomic_inc_return(&ch->kthreads_idle) >
 654                     ch->kthreads_idle_limit) {
 655                         /* too many idle kthreads on this channel */
 656                         atomic_dec(&ch->kthreads_idle);
 657                         break;
 658                 }
 659
 660                 dev_dbg(xpc_chan, "idle kthread calling "
 661                         "wait_event_interruptible_exclusive()\n");
 662
 663                 (void)wait_event_interruptible_exclusive(ch->idle_wq,
 664                                 (n_of_deliverable_payloads(ch) > 0 ||
 665                                  (ch->flags & XPC_C_DISCONNECTING)));
 666
 667                 atomic_dec(&ch->kthreads_idle);
 668
 669         } while (!(ch->flags & XPC_C_DISCONNECTING));
 670 }
 671
 672 static int
 673 xpc_kthread_start(void *args)
 674 {
 675         short partid = XPC_UNPACK_ARG1(args);
 676         u16 ch_number = XPC_UNPACK_ARG2(args);
 677         struct xpc_partition *part = &xpc_partitions[partid];
 678         struct xpc_channel *ch;
 679         int n_needed;
 680         unsigned long irq_flags;
 681         int (*n_of_deliverable_payloads) (struct xpc_channel *) =
 682                 xpc_arch_ops.n_of_deliverable_payloads;
 683
 684         dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n",
 685                 partid, ch_number);
 686
 687         ch = &part->channels[ch_number];
 688
 689         if (!(ch->flags & XPC_C_DISCONNECTING)) {
 690
 691                 /* let registerer know that connection has been established */
 692
 693                 spin_lock_irqsave(&ch->lock, irq_flags);
 694                 if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) {
 695                         ch->flags |= XPC_C_CONNECTEDCALLOUT;
 696                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 697
 698                         xpc_connected_callout(ch);
 699
 700                         spin_lock_irqsave(&ch->lock, irq_flags);
 701                         ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE;
 702                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 703
 704                         /*
 705                          * It is possible that while the callout was being
 706                          * made that the remote partition sent some messages.
 707                          * If that is the case, we may need to activate
 708                          * additional kthreads to help deliver them. We only
 709                          * need one less than total #of messages to deliver.
 710                          */
 711                         n_needed = n_of_deliverable_payloads(ch) - 1;
 712                         if (n_needed > 0 && !(ch->flags & XPC_C_DISCONNECTING))
 713                                 xpc_activate_kthreads(ch, n_needed);
 714
 715                 } else {
 716                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 717                 }
 718
 719                 xpc_kthread_waitmsgs(part, ch);
 720         }
 721
 722         /* let registerer know that connection is disconnecting */
 723
 724         spin_lock_irqsave(&ch->lock, irq_flags);
 725         if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
 726             !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) {
 727                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT;
 728                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 729
 730                 xpc_disconnect_callout(ch, xpDisconnecting);
 731
 732                 spin_lock_irqsave(&ch->lock, irq_flags);
 733                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE;
 734         }
 735         spin_unlock_irqrestore(&ch->lock, irq_flags);
 736
 737         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 738             atomic_dec_return(&part->nchannels_engaged) == 0) {
 739                 xpc_arch_ops.indicate_partition_disengaged(part);
 740         }
 741
 742         xpc_msgqueue_deref(ch);
 743
 744         dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n",
 745                 partid, ch_number);
 746
 747         xpc_part_deref(part);
 748         return 0;
 749 }
 750
 751 /*
 752  * For each partition that XPC has established communications with, there is
 753  * a minimum of one kernel thread assigned to perform any operation that
 754  * may potentially sleep or block (basically the callouts to the asynchronous
 755  * functions registered via xpc_connect()).
 756  *
 757  * Additional kthreads are created and destroyed by XPC as the workload
 758  * demands.
 759  *
 760  * A kthread is assigned to one of the active channels that exists for a given
 761  * partition.
 762  */
 763 void
 764 xpc_create_kthreads(struct xpc_channel *ch, int needed,
 765                     int ignore_disconnecting)
 766 {
 767         unsigned long irq_flags;
 768         u64 args = XPC_PACK_ARGS(ch->partid, ch->number);
 769         struct xpc_partition *part = &xpc_partitions[ch->partid];
 770         struct task_struct *kthread;
 771         void (*indicate_partition_disengaged) (struct xpc_partition *) =
 772                 xpc_arch_ops.indicate_partition_disengaged;
 773
 774         while (needed-- > 0) {
 775
 776                 /*
 777                  * The following is done on behalf of the newly created
 778                  * kthread. That kthread is responsible for doing the
 779                  * counterpart to the following before it exits.
 780                  */
 781                 if (ignore_disconnecting) {
 782                         if (!atomic_inc_not_zero(&ch->kthreads_assigned)) {
 783                                 /* kthreads assigned had gone to zero */
 784                                 BUG_ON(!(ch->flags &
 785                                          XPC_C_DISCONNECTINGCALLOUT_MADE));
 786                                 break;
 787                         }
 788
 789                 } else if (ch->flags & XPC_C_DISCONNECTING) {
 790                         break;
 791
 792                 } else if (atomic_inc_return(&ch->kthreads_assigned) == 1 &&
 793                            atomic_inc_return(&part->nchannels_engaged) == 1) {
 794                         xpc_arch_ops.indicate_partition_engaged(part);
 795                 }
 796                 (void)xpc_part_ref(part);
 797                 xpc_msgqueue_ref(ch);
 798
 799                 kthread = kthread_run(xpc_kthread_start, (void *)args,
 800                                       "xpc%02dc%d", ch->partid, ch->number);
 801                 if (IS_ERR(kthread)) {
 802                         /* the fork failed */
 803
 804                         /*
 805                          * NOTE: if (ignore_disconnecting &&
 806                          * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true,
 807                          * then we'll deadlock if all other kthreads assigned
 808                          * to this channel are blocked in the channel's
 809                          * registerer, because the only thing that will unblock
 810                          * them is the xpDisconnecting callout that this
 811                          * failed kthread_run() would have made.
 812                          */
 813
 814                         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 815                             atomic_dec_return(&part->nchannels_engaged) == 0) {
 816                                 indicate_partition_disengaged(part);
 817                         }
 818                         xpc_msgqueue_deref(ch);
 819                         xpc_part_deref(part);
 820
 821                         if (atomic_read(&ch->kthreads_assigned) <
 822                             ch->kthreads_idle_limit) {
 823                                 /*
 824                                  * Flag this as an error only if we have an
 825                                  * insufficient #of kthreads for the channel
 826                                  * to function.
 827                                  */
 828                                 spin_lock_irqsave(&ch->lock, irq_flags);
 829                                 XPC_DISCONNECT_CHANNEL(ch, xpLackOfResources,
 830                                                        &irq_flags);
 831                                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 832                         }
 833                         break;
 834                 }
 835         }
 836 }
 837
 838 void
 839 xpc_disconnect_wait(int ch_number)
 840 {
 841         unsigned long irq_flags;
 842         short partid;
 843         struct xpc_partition *part;
 844         struct xpc_channel *ch;
 845         int wakeup_channel_mgr;
 846
 847         /* now wait for all callouts to the caller's function to cease */
 848         for (partid = 0; partid < xp_max_npartitions; partid++) {
 849                 part = &xpc_partitions[partid];
 850
 851                 if (!xpc_part_ref(part))
 852                         continue;
 853
 854                 ch = &part->channels[ch_number];
 855
 856                 if (!(ch->flags & XPC_C_WDISCONNECT)) {
 857                         xpc_part_deref(part);
 858                         continue;
 859                 }
 860
 861                 wait_for_completion(&ch->wdisconnect_wait);
 862
 863                 spin_lock_irqsave(&ch->lock, irq_flags);
 864                 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
 865                 wakeup_channel_mgr = 0;
 866
 867                 if (ch->delayed_chctl_flags) {
 868                         if (part->act_state != XPC_P_AS_DEACTIVATING) {
 869                                 spin_lock(&part->chctl_lock);
 870                                 part->chctl.flags[ch->number] |=
 871                                     ch->delayed_chctl_flags;
 872                                 spin_unlock(&part->chctl_lock);
 873                                 wakeup_channel_mgr = 1;
 874                         }
 875                         ch->delayed_chctl_flags = 0;
 876                 }
 877
 878                 ch->flags &= ~XPC_C_WDISCONNECT;
 879                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 880
 881                 if (wakeup_channel_mgr)
 882                         xpc_wakeup_channel_mgr(part);
 883
 884                 xpc_part_deref(part);
 885         }
 886 }
 887
 888 static int
 889 xpc_setup_partitions(void)
 890 {
 891         short partid;
 892         struct xpc_partition *part;
 893
 894         xpc_partitions = kcalloc(xp_max_npartitions,
 895                                  sizeof(struct xpc_partition),
 896                                  GFP_KERNEL);
 897         if (xpc_partitions == NULL) {
 898                 dev_err(xpc_part, "can't get memory for partition structure\n");
 899                 return -ENOMEM;
 900         }
 901
 902         /*
 903          * The first few fields of each entry of xpc_partitions[] need to
 904          * be initialized now so that calls to xpc_connect() and
 905          * xpc_disconnect() can be made prior to the activation of any remote
 906          * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE
 907          * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING
 908          * PARTITION HAS BEEN ACTIVATED.
 909          */
 910         for (partid = 0; partid < xp_max_npartitions; partid++) {
 911                 part = &xpc_partitions[partid];
 912
 913                 DBUG_ON((u64)part != L1_CACHE_ALIGN((u64)part));
 914
 915                 part->activate_IRQ_rcvd = 0;
 916                 spin_lock_init(&part->act_lock);
 917                 part->act_state = XPC_P_AS_INACTIVE;
 918                 XPC_SET_REASON(part, 0, 0);
 919
 920                 timer_setup(&part->disengage_timer,
 921                             xpc_timeout_partition_disengage, 0);
 922
 923                 part->setup_state = XPC_P_SS_UNSET;
 924                 init_waitqueue_head(&part->teardown_wq);
 925                 atomic_set(&part->references, 0);
 926         }
 927
 928         return xpc_arch_ops.setup_partitions();
 929 }
 930
 931 static void
 932 xpc_teardown_partitions(void)
 933 {
 934         xpc_arch_ops.teardown_partitions();
 935         kfree(xpc_partitions);
 936 }
 937
 938 static void
 939 xpc_do_exit(enum xp_retval reason)
 940 {
 941         short partid;
 942         int active_part_count, printed_waiting_msg = 0;
 943         struct xpc_partition *part;
 944         unsigned long printmsg_time, disengage_timeout = 0;
 945
 946         /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
 947         DBUG_ON(xpc_exiting == 1);
 948
 949         /*
 950          * Let the heartbeat checker thread and the discovery thread
 951          * (if one is running) know that they should exit. Also wake up
 952          * the heartbeat checker thread in case it's sleeping.
 953          */
 954         xpc_exiting = 1;
 955         wake_up_interruptible(&xpc_activate_IRQ_wq);
 956
 957         /* wait for the discovery thread to exit */
 958         wait_for_completion(&xpc_discovery_exited);
 959
 960         /* wait for the heartbeat checker thread to exit */
 961         wait_for_completion(&xpc_hb_checker_exited);
 962
 963         /* sleep for a 1/3 of a second or so */
 964         (void)msleep_interruptible(300);
 965
 966         /* wait for all partitions to become inactive */
 967
 968         printmsg_time = jiffies + (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
 969         xpc_disengage_timedout = 0;
 970
 971         do {
 972                 active_part_count = 0;
 973
 974                 for (partid = 0; partid < xp_max_npartitions; partid++) {
 975                         part = &xpc_partitions[partid];
 976
 977                         if (xpc_partition_disengaged(part) &&
 978                             part->act_state == XPC_P_AS_INACTIVE) {
 979                                 continue;
 980                         }
 981
 982                         active_part_count++;
 983
 984                         XPC_DEACTIVATE_PARTITION(part, reason);
 985
 986                         if (part->disengage_timeout > disengage_timeout)
 987                                 disengage_timeout = part->disengage_timeout;
 988                 }
 989
 990                 if (xpc_arch_ops.any_partition_engaged()) {
 991                         if (time_is_before_jiffies(printmsg_time)) {
 992                                 dev_info(xpc_part, "waiting for remote "
 993                                          "partitions to deactivate, timeout in "
 994                                          "%ld seconds\n", (disengage_timeout -
 995                                          jiffies) / HZ);
 996                                 printmsg_time = jiffies +
 997                                     (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
 998                                 printed_waiting_msg = 1;
 999                         }
1000
1001                 } else if (active_part_count > 0) {
1002                         if (printed_waiting_msg) {
1003                                 dev_info(xpc_part, "waiting for local partition"
1004                                          " to deactivate\n");
1005                                 printed_waiting_msg = 0;
1006                         }
1007
1008                 } else {
1009                         if (!xpc_disengage_timedout) {
1010                                 dev_info(xpc_part, "all partitions have "
1011                                          "deactivated\n");
1012                         }
1013                         break;
1014                 }
1015
1016                 /* sleep for a 1/3 of a second or so */
1017                 (void)msleep_interruptible(300);
1018
1019         } while (1);
1020
1021         DBUG_ON(xpc_arch_ops.any_partition_engaged());
1022
1023         xpc_teardown_rsvd_page();
1024
1025         if (reason == xpUnloading) {
1026                 (void)unregister_die_notifier(&xpc_die_notifier);
1027                 (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1028         }
1029
1030         /* clear the interface to XPC's functions */
1031         xpc_clear_interface();
1032
1033         if (xpc_sysctl)
1034                 unregister_sysctl_table(xpc_sysctl);
1035         if (xpc_sysctl_hb)
1036                 unregister_sysctl_table(xpc_sysctl_hb);
1037
1038         xpc_teardown_partitions();
1039
1040         if (is_uv_system())
1041                 xpc_exit_uv();
1042 }
1043
1044 /*
1045  * This function is called when the system is being rebooted.
1046  */
1047 static int
1048 xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
1049 {
1050         enum xp_retval reason;
1051
1052         switch (event) {
1053         case SYS_RESTART:
1054                 reason = xpSystemReboot;
1055                 break;
1056         case SYS_HALT:
1057                 reason = xpSystemHalt;
1058                 break;
1059         case SYS_POWER_OFF:
1060                 reason = xpSystemPoweroff;
1061                 break;
1062         default:
1063                 reason = xpSystemGoingDown;
1064         }
1065
1066         xpc_do_exit(reason);
1067         return NOTIFY_DONE;
1068 }
1069
1070 /* Used to only allow one cpu to complete disconnect */
1071 static unsigned int xpc_die_disconnecting;
1072
1073 /*
1074  * Notify other partitions to deactivate from us by first disengaging from all
1075  * references to our memory.
1076  */
1077 static void
1078 xpc_die_deactivate(void)
1079 {
1080         struct xpc_partition *part;
1081         short partid;
1082         int any_engaged;
1083         long keep_waiting;
1084         long wait_to_print;
1085
1086         if (cmpxchg(&xpc_die_disconnecting, 0, 1))
1087                 return;
1088
1089         /* keep xpc_hb_checker thread from doing anything (just in case) */
1090         xpc_exiting = 1;
1091
1092         xpc_arch_ops.disallow_all_hbs();   /*indicate we're deactivated */
1093
1094         for (partid = 0; partid < xp_max_npartitions; partid++) {
1095                 part = &xpc_partitions[partid];
1096
1097                 if (xpc_arch_ops.partition_engaged(partid) ||
1098                     part->act_state != XPC_P_AS_INACTIVE) {
1099                         xpc_arch_ops.request_partition_deactivation(part);
1100                         xpc_arch_ops.indicate_partition_disengaged(part);
1101                 }
1102         }
1103
1104         /*
1105          * Though we requested that all other partitions deactivate from us,
1106          * we only wait until they've all disengaged or we've reached the
1107          * defined timelimit.
1108          *
1109          * Given that one iteration through the following while-loop takes
1110          * approximately 200 microseconds, calculate the #of loops to take
1111          * before bailing and the #of loops before printing a waiting message.
1112          */
1113         keep_waiting = xpc_disengage_timelimit * 1000 * 5;
1114         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL * 1000 * 5;
1115
1116         while (1) {
1117                 any_engaged = xpc_arch_ops.any_partition_engaged();
1118                 if (!any_engaged) {
1119                         dev_info(xpc_part, "all partitions have deactivated\n");
1120                         break;
1121                 }
1122
1123                 if (!keep_waiting--) {
1124                         for (partid = 0; partid < xp_max_npartitions;
1125                              partid++) {
1126                                 if (xpc_arch_ops.partition_engaged(partid)) {
1127                                         dev_info(xpc_part, "deactivate from "
1128                                                  "remote partition %d timed "
1129                                                  "out\n", partid);
1130                                 }
1131                         }
1132                         break;
1133                 }
1134
1135                 if (!wait_to_print--) {
1136                         dev_info(xpc_part, "waiting for remote partitions to "
1137                                  "deactivate, timeout in %ld seconds\n",
1138                                  keep_waiting / (1000 * 5));
1139                         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL *
1140                             1000 * 5;
1141                 }
1142
1143                 udelay(200);
1144         }
1145 }
1146
1147 /*
1148  * This function is called when the system is being restarted or halted due
1149  * to some sort of system failure. If this is the case we need to notify the
1150  * other partitions to disengage from all references to our memory.
1151  * This function can also be called when our heartbeater could be offlined
1152  * for a time. In this case we need to notify other partitions to not worry
1153  * about the lack of a heartbeat.
1154  */
1155 static int
1156 xpc_system_die(struct notifier_block *nb, unsigned long event, void *_die_args)
1157 {
1158 #ifdef CONFIG_IA64              /* !!! temporary kludge */
1159         switch (event) {
1160         case DIE_MACHINE_RESTART:
1161         case DIE_MACHINE_HALT:
1162                 xpc_die_deactivate();
1163                 break;
1164
1165         case DIE_KDEBUG_ENTER:
1166                 /* Should lack of heartbeat be ignored by other partitions? */
1167                 if (!xpc_kdebug_ignore)
1168                         break;
1169
1170                 fallthrough;
1171         case DIE_MCA_MONARCH_ENTER:
1172         case DIE_INIT_MONARCH_ENTER:
1173                 xpc_arch_ops.offline_heartbeat();
1174                 break;
1175
1176         case DIE_KDEBUG_LEAVE:
1177                 /* Is lack of heartbeat being ignored by other partitions? */
1178                 if (!xpc_kdebug_ignore)
1179                         break;
1180
1181                 fallthrough;
1182         case DIE_MCA_MONARCH_LEAVE:
1183         case DIE_INIT_MONARCH_LEAVE:
1184                 xpc_arch_ops.online_heartbeat();
1185                 break;
1186         }
1187 #else
1188         struct die_args *die_args = _die_args;
1189
1190         switch (event) {
1191         case DIE_TRAP:
1192                 if (die_args->trapnr == X86_TRAP_DF)
1193                         xpc_die_deactivate();
1194
1195                 if (((die_args->trapnr == X86_TRAP_MF) ||
1196                      (die_args->trapnr == X86_TRAP_XF)) &&
1197                     !user_mode(die_args->regs))
1198                         xpc_die_deactivate();
1199
1200                 break;
1201         case DIE_INT3:
1202         case DIE_DEBUG:
1203                 break;
1204         case DIE_OOPS:
1205         case DIE_GPF:
1206         default:
1207                 xpc_die_deactivate();
1208         }
1209 #endif
1210
1211         return NOTIFY_DONE;
1212 }
1213
1214 static int __init
1215 xpc_init(void)
1216 {
1217         int ret;
1218         struct task_struct *kthread;
1219
1220         dev_set_name(xpc_part, "part");
1221         dev_set_name(xpc_chan, "chan");
1222
1223         if (is_uv_system()) {
1224                 ret = xpc_init_uv();
1225
1226         } else {
1227                 ret = -ENODEV;
1228         }
1229
1230         if (ret != 0)
1231                 return ret;
1232
1233         ret = xpc_setup_partitions();
1234         if (ret != 0) {
1235                 dev_err(xpc_part, "can't get memory for partition structure\n");
1236                 goto out_1;
1237         }
1238
1239         xpc_sysctl = register_sysctl("xpc", xpc_sys_xpc);
1240         xpc_sysctl_hb = register_sysctl("xpc/hb", xpc_sys_xpc_hb);
1241
1242         /*
1243          * Fill the partition reserved page with the information needed by
1244          * other partitions to discover we are alive and establish initial
1245          * communications.
1246          */
1247         ret = xpc_setup_rsvd_page();
1248         if (ret != 0) {
1249                 dev_err(xpc_part, "can't setup our reserved page\n");
1250                 goto out_2;
1251         }
1252
1253         /* add ourselves to the reboot_notifier_list */
1254         ret = register_reboot_notifier(&xpc_reboot_notifier);
1255         if (ret != 0)
1256                 dev_warn(xpc_part, "can't register reboot notifier\n");
1257
1258         /* add ourselves to the die_notifier list */
1259         ret = register_die_notifier(&xpc_die_notifier);
1260         if (ret != 0)
1261                 dev_warn(xpc_part, "can't register die notifier\n");
1262
1263         /*
1264          * The real work-horse behind xpc.  This processes incoming
1265          * interrupts and monitors remote heartbeats.
1266          */
1267         kthread = kthread_run(xpc_hb_checker, NULL, XPC_HB_CHECK_THREAD_NAME);
1268         if (IS_ERR(kthread)) {
1269                 dev_err(xpc_part, "failed while forking hb check thread\n");
1270                 ret = -EBUSY;
1271                 goto out_3;
1272         }
1273
1274         /*
1275          * Startup a thread that will attempt to discover other partitions to
1276          * activate based on info provided by SAL. This new thread is short
1277          * lived and will exit once discovery is complete.
1278          */
1279         kthread = kthread_run(xpc_initiate_discovery, NULL,
1280                               XPC_DISCOVERY_THREAD_NAME);
1281         if (IS_ERR(kthread)) {
1282                 dev_err(xpc_part, "failed while forking discovery thread\n");
1283
1284                 /* mark this new thread as a non-starter */
1285                 complete(&xpc_discovery_exited);
1286
1287                 xpc_do_exit(xpUnloading);
1288                 return -EBUSY;
1289         }
1290
1291         /* set the interface to point at XPC's functions */
1292         xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect,
1293                           xpc_initiate_send, xpc_initiate_send_notify,
1294                           xpc_initiate_received, xpc_initiate_partid_to_nasids);
1295
1296         return 0;
1297
1298         /* initialization was not successful */
1299 out_3:
1300         xpc_teardown_rsvd_page();
1301
1302         (void)unregister_die_notifier(&xpc_die_notifier);
1303         (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1304 out_2:
1305         if (xpc_sysctl_hb)
1306                 unregister_sysctl_table(xpc_sysctl_hb);
1307         if (xpc_sysctl)
1308                 unregister_sysctl_table(xpc_sysctl);
1309
1310         xpc_teardown_partitions();
1311 out_1:
1312         if (is_uv_system())
1313                 xpc_exit_uv();
1314         return ret;
1315 }
1316
1317 module_init(xpc_init);
1318
1319 static void __exit
1320 xpc_exit(void)
1321 {
1322         xpc_do_exit(xpUnloading);
1323 }
1324
1325 module_exit(xpc_exit);
1326
1327 MODULE_AUTHOR("Silicon Graphics, Inc.");
1328 MODULE_DESCRIPTION("Cross Partition Communication (XPC) support");
1329 MODULE_LICENSE("GPL");
1330
1331 module_param(xpc_hb_interval, int, 0);
1332 MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between "
1333                  "heartbeat increments.");
1334
1335 module_param(xpc_hb_check_interval, int, 0);
1336 MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between "
1337                  "heartbeat checks.");
1338
1339 module_param(xpc_disengage_timelimit, int, 0);
1340 MODULE_PARM_DESC(xpc_disengage_timelimit, "Number of seconds to wait "
1341                  "for disengage to complete.");
1342
1343 module_param(xpc_kdebug_ignore, int, 0);
1344 MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by "
1345                  "other partitions when dropping into kdebug.");