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/sysctl.h>
  48 #include <linux/device.h>
  49 #include <linux/delay.h>
  50 #include <linux/reboot.h>
  51 #include <linux/kdebug.h>
  52 #include <linux/kthread.h>
  53 #include "xpc.h"
  54
  55 /* define two XPC debug device structures to be used with dev_dbg() et al */
  56
  57 struct device_driver xpc_dbg_name = {
  58         .name = "xpc"
  59 };
  60
  61 struct device xpc_part_dbg_subname = {
  62         .init_name = "",        /* set to "part" at xpc_init() time */
  63         .driver = &xpc_dbg_name
  64 };
  65
  66 struct device xpc_chan_dbg_subname = {
  67         .init_name = "",        /* set to "chan" at xpc_init() time */
  68         .driver = &xpc_dbg_name
  69 };
  70
  71 struct device *xpc_part = &xpc_part_dbg_subname;
  72 struct device *xpc_chan = &xpc_chan_dbg_subname;
  73
  74 static int xpc_kdebug_ignore;
  75
  76 /* systune related variables for /proc/sys directories */
  77
  78 static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
  79 static int xpc_hb_min_interval = 1;
  80 static int xpc_hb_max_interval = 10;
  81
  82 static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL;
  83 static int xpc_hb_check_min_interval = 10;
  84 static int xpc_hb_check_max_interval = 120;
  85
  86 int xpc_disengage_timelimit = XPC_DISENGAGE_DEFAULT_TIMELIMIT;
  87 static int xpc_disengage_min_timelimit; /* = 0 */
  88 static int xpc_disengage_max_timelimit = 120;
  89
  90 static ctl_table xpc_sys_xpc_hb_dir[] = {
  91         {
  92          .ctl_name = CTL_UNNUMBERED,
  93          .procname = "hb_interval",
  94          .data = &xpc_hb_interval,
  95          .maxlen = sizeof(int),
  96          .mode = 0644,
  97          .proc_handler = &proc_dointvec_minmax,
  98          .strategy = &sysctl_intvec,
  99          .extra1 = &xpc_hb_min_interval,
 100          .extra2 = &xpc_hb_max_interval},
 101         {
 102          .ctl_name = CTL_UNNUMBERED,
 103          .procname = "hb_check_interval",
 104          .data = &xpc_hb_check_interval,
 105          .maxlen = sizeof(int),
 106          .mode = 0644,
 107          .proc_handler = &proc_dointvec_minmax,
 108          .strategy = &sysctl_intvec,
 109          .extra1 = &xpc_hb_check_min_interval,
 110          .extra2 = &xpc_hb_check_max_interval},
 111         {}
 112 };
 113 static ctl_table xpc_sys_xpc_dir[] = {
 114         {
 115          .ctl_name = CTL_UNNUMBERED,
 116          .procname = "hb",
 117          .mode = 0555,
 118          .child = xpc_sys_xpc_hb_dir},
 119         {
 120          .ctl_name = CTL_UNNUMBERED,
 121          .procname = "disengage_timelimit",
 122          .data = &xpc_disengage_timelimit,
 123          .maxlen = sizeof(int),
 124          .mode = 0644,
 125          .proc_handler = &proc_dointvec_minmax,
 126          .strategy = &sysctl_intvec,
 127          .extra1 = &xpc_disengage_min_timelimit,
 128          .extra2 = &xpc_disengage_max_timelimit},
 129         {}
 130 };
 131 static ctl_table xpc_sys_dir[] = {
 132         {
 133          .ctl_name = CTL_UNNUMBERED,
 134          .procname = "xpc",
 135          .mode = 0555,
 136          .child = xpc_sys_xpc_dir},
 137         {}
 138 };
 139 static struct ctl_table_header *xpc_sysctl;
 140
 141 /* non-zero if any remote partition disengage was timed out */
 142 int xpc_disengage_timedout;
 143
 144 /* #of activate IRQs received and not yet processed */
 145 int xpc_activate_IRQ_rcvd;
 146 DEFINE_SPINLOCK(xpc_activate_IRQ_rcvd_lock);
 147
 148 /* IRQ handler notifies this wait queue on receipt of an IRQ */
 149 DECLARE_WAIT_QUEUE_HEAD(xpc_activate_IRQ_wq);
 150
 151 static unsigned long xpc_hb_check_timeout;
 152 static struct timer_list xpc_hb_timer;
 153
 154 /* notification that the xpc_hb_checker thread has exited */
 155 static DECLARE_COMPLETION(xpc_hb_checker_exited);
 156
 157 /* notification that the xpc_discovery thread has exited */
 158 static DECLARE_COMPLETION(xpc_discovery_exited);
 159
 160 static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *);
 161
 162 static int xpc_system_reboot(struct notifier_block *, unsigned long, void *);
 163 static struct notifier_block xpc_reboot_notifier = {
 164         .notifier_call = xpc_system_reboot,
 165 };
 166
 167 static int xpc_system_die(struct notifier_block *, unsigned long, void *);
 168 static struct notifier_block xpc_die_notifier = {
 169         .notifier_call = xpc_system_die,
 170 };
 171
 172 int (*xpc_setup_partitions_sn) (void);
 173 void (*xpc_teardown_partitions_sn) (void);
 174 enum xp_retval (*xpc_get_partition_rsvd_page_pa) (void *buf, u64 *cookie,
 175                                                   unsigned long *rp_pa,
 176                                                   size_t *len);
 177 int (*xpc_setup_rsvd_page_sn) (struct xpc_rsvd_page *rp);
 178
 179 void (*xpc_allow_hb) (short partid);
 180 void (*xpc_disallow_hb) (short partid);
 181 void (*xpc_disallow_all_hbs) (void);
 182 void (*xpc_heartbeat_init) (void);
 183 void (*xpc_heartbeat_exit) (void);
 184 void (*xpc_increment_heartbeat) (void);
 185 void (*xpc_offline_heartbeat) (void);
 186 void (*xpc_online_heartbeat) (void);
 187 enum xp_retval (*xpc_get_remote_heartbeat) (struct xpc_partition *part);
 188
 189 enum xp_retval (*xpc_make_first_contact) (struct xpc_partition *part);
 190 void (*xpc_notify_senders_of_disconnect) (struct xpc_channel *ch);
 191 u64 (*xpc_get_chctl_all_flags) (struct xpc_partition *part);
 192 enum xp_retval (*xpc_setup_msg_structures) (struct xpc_channel *ch);
 193 void (*xpc_teardown_msg_structures) (struct xpc_channel *ch);
 194 void (*xpc_process_msg_chctl_flags) (struct xpc_partition *part, int ch_number);
 195 int (*xpc_n_of_deliverable_payloads) (struct xpc_channel *ch);
 196 void *(*xpc_get_deliverable_payload) (struct xpc_channel *ch);
 197
 198 void (*xpc_request_partition_activation) (struct xpc_rsvd_page *remote_rp,
 199                                           unsigned long remote_rp_pa,
 200                                           int nasid);
 201 void (*xpc_request_partition_reactivation) (struct xpc_partition *part);
 202 void (*xpc_request_partition_deactivation) (struct xpc_partition *part);
 203 void (*xpc_cancel_partition_deactivation_request) (struct xpc_partition *part);
 204
 205 void (*xpc_process_activate_IRQ_rcvd) (void);
 206 enum xp_retval (*xpc_setup_ch_structures_sn) (struct xpc_partition *part);
 207 void (*xpc_teardown_ch_structures_sn) (struct xpc_partition *part);
 208
 209 void (*xpc_indicate_partition_engaged) (struct xpc_partition *part);
 210 int (*xpc_partition_engaged) (short partid);
 211 int (*xpc_any_partition_engaged) (void);
 212 void (*xpc_indicate_partition_disengaged) (struct xpc_partition *part);
 213 void (*xpc_assume_partition_disengaged) (short partid);
 214
 215 void (*xpc_send_chctl_closerequest) (struct xpc_channel *ch,
 216                                      unsigned long *irq_flags);
 217 void (*xpc_send_chctl_closereply) (struct xpc_channel *ch,
 218                                    unsigned long *irq_flags);
 219 void (*xpc_send_chctl_openrequest) (struct xpc_channel *ch,
 220                                     unsigned long *irq_flags);
 221 void (*xpc_send_chctl_openreply) (struct xpc_channel *ch,
 222                                   unsigned long *irq_flags);
 223 void (*xpc_send_chctl_opencomplete) (struct xpc_channel *ch,
 224                                   unsigned long *irq_flags);
 225
 226 enum xp_retval (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *ch,
 227                                                unsigned long msgqueue_pa);
 228
 229 enum xp_retval (*xpc_send_payload) (struct xpc_channel *ch, u32 flags,
 230                                     void *payload, u16 payload_size,
 231                                     u8 notify_type, xpc_notify_func func,
 232                                     void *key);
 233 void (*xpc_received_payload) (struct xpc_channel *ch, void *payload);
 234
 235 /*
 236  * Timer function to enforce the timelimit on the partition disengage.
 237  */
 238 static void
 239 xpc_timeout_partition_disengage(unsigned long data)
 240 {
 241         struct xpc_partition *part = (struct xpc_partition *)data;
 242
 243         DBUG_ON(time_is_after_jiffies(part->disengage_timeout));
 244
 245         (void)xpc_partition_disengaged(part);
 246
 247         DBUG_ON(part->disengage_timeout != 0);
 248         DBUG_ON(xpc_partition_engaged(XPC_PARTID(part)));
 249 }
 250
 251 /*
 252  * Timer to produce the heartbeat.  The timer structures function is
 253  * already set when this is initially called.  A tunable is used to
 254  * specify when the next timeout should occur.
 255  */
 256 static void
 257 xpc_hb_beater(unsigned long dummy)
 258 {
 259         xpc_increment_heartbeat();
 260
 261         if (time_is_before_eq_jiffies(xpc_hb_check_timeout))
 262                 wake_up_interruptible(&xpc_activate_IRQ_wq);
 263
 264         xpc_hb_timer.expires = jiffies + (xpc_hb_interval * HZ);
 265         add_timer(&xpc_hb_timer);
 266 }
 267
 268 static void
 269 xpc_start_hb_beater(void)
 270 {
 271         xpc_heartbeat_init();
 272         init_timer(&xpc_hb_timer);
 273         xpc_hb_timer.function = xpc_hb_beater;
 274         xpc_hb_beater(0);
 275 }
 276
 277 static void
 278 xpc_stop_hb_beater(void)
 279 {
 280         del_timer_sync(&xpc_hb_timer);
 281         xpc_heartbeat_exit();
 282 }
 283
 284 /*
 285  * At periodic intervals, scan through all active partitions and ensure
 286  * their heartbeat is still active.  If not, the partition is deactivated.
 287  */
 288 static void
 289 xpc_check_remote_hb(void)
 290 {
 291         struct xpc_partition *part;
 292         short partid;
 293         enum xp_retval ret;
 294
 295         for (partid = 0; partid < xp_max_npartitions; partid++) {
 296
 297                 if (xpc_exiting)
 298                         break;
 299
 300                 if (partid == xp_partition_id)
 301                         continue;
 302
 303                 part = &xpc_partitions[partid];
 304
 305                 if (part->act_state == XPC_P_AS_INACTIVE ||
 306                     part->act_state == XPC_P_AS_DEACTIVATING) {
 307                         continue;
 308                 }
 309
 310                 ret = xpc_get_remote_heartbeat(part);
 311                 if (ret != xpSuccess)
 312                         XPC_DEACTIVATE_PARTITION(part, ret);
 313         }
 314 }
 315
 316 /*
 317  * This thread is responsible for nearly all of the partition
 318  * activation/deactivation.
 319  */
 320 static int
 321 xpc_hb_checker(void *ignore)
 322 {
 323         int force_IRQ = 0;
 324
 325         /* this thread was marked active by xpc_hb_init() */
 326
 327         set_cpus_allowed_ptr(current, cpumask_of(XPC_HB_CHECK_CPU));
 328
 329         /* set our heartbeating to other partitions into motion */
 330         xpc_hb_check_timeout = jiffies + (xpc_hb_check_interval * HZ);
 331         xpc_start_hb_beater();
 332
 333         while (!xpc_exiting) {
 334
 335                 dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have "
 336                         "been received\n",
 337                         (int)(xpc_hb_check_timeout - jiffies),
 338                         xpc_activate_IRQ_rcvd);
 339
 340                 /* checking of remote heartbeats is skewed by IRQ handling */
 341                 if (time_is_before_eq_jiffies(xpc_hb_check_timeout)) {
 342                         xpc_hb_check_timeout = jiffies +
 343                             (xpc_hb_check_interval * HZ);
 344
 345                         dev_dbg(xpc_part, "checking remote heartbeats\n");
 346                         xpc_check_remote_hb();
 347
 348                         /*
 349                          * On sn2 we need to periodically recheck to ensure no
 350                          * IRQ/amo pairs have been missed.
 351                          */
 352                         if (is_shub())
 353                                 force_IRQ = 1;
 354                 }
 355
 356                 /* check for outstanding IRQs */
 357                 if (xpc_activate_IRQ_rcvd > 0 || force_IRQ != 0) {
 358                         force_IRQ = 0;
 359                         dev_dbg(xpc_part, "processing activate IRQs "
 360                                 "received\n");
 361                         xpc_process_activate_IRQ_rcvd();
 362                 }
 363
 364                 /* wait for IRQ or timeout */
 365                 (void)wait_event_interruptible(xpc_activate_IRQ_wq,
 366                                                (time_is_before_eq_jiffies(
 367                                                 xpc_hb_check_timeout) ||
 368                                                 xpc_activate_IRQ_rcvd > 0 ||
 369                                                 xpc_exiting));
 370         }
 371
 372         xpc_stop_hb_beater();
 373
 374         dev_dbg(xpc_part, "heartbeat checker is exiting\n");
 375
 376         /* mark this thread as having exited */
 377         complete(&xpc_hb_checker_exited);
 378         return 0;
 379 }
 380
 381 /*
 382  * This thread will attempt to discover other partitions to activate
 383  * based on info provided by SAL. This new thread is short lived and
 384  * will exit once discovery is complete.
 385  */
 386 static int
 387 xpc_initiate_discovery(void *ignore)
 388 {
 389         xpc_discovery();
 390
 391         dev_dbg(xpc_part, "discovery thread is exiting\n");
 392
 393         /* mark this thread as having exited */
 394         complete(&xpc_discovery_exited);
 395         return 0;
 396 }
 397
 398 /*
 399  * The first kthread assigned to a newly activated partition is the one
 400  * created by XPC HB with which it calls xpc_activating(). XPC hangs on to
 401  * that kthread until the partition is brought down, at which time that kthread
 402  * returns back to XPC HB. (The return of that kthread will signify to XPC HB
 403  * that XPC has dismantled all communication infrastructure for the associated
 404  * partition.) This kthread becomes the channel manager for that partition.
 405  *
 406  * Each active partition has a channel manager, who, besides connecting and
 407  * disconnecting channels, will ensure that each of the partition's connected
 408  * channels has the required number of assigned kthreads to get the work done.
 409  */
 410 static void
 411 xpc_channel_mgr(struct xpc_partition *part)
 412 {
 413         while (part->act_state != XPC_P_AS_DEACTIVATING ||
 414                atomic_read(&part->nchannels_active) > 0 ||
 415                !xpc_partition_disengaged(part)) {
 416
 417                 xpc_process_sent_chctl_flags(part);
 418
 419                 /*
 420                  * Wait until we've been requested to activate kthreads or
 421                  * all of the channel's message queues have been torn down or
 422                  * a signal is pending.
 423                  *
 424                  * The channel_mgr_requests is set to 1 after being awakened,
 425                  * This is done to prevent the channel mgr from making one pass
 426                  * through the loop for each request, since he will
 427                  * be servicing all the requests in one pass. The reason it's
 428                  * set to 1 instead of 0 is so that other kthreads will know
 429                  * that the channel mgr is running and won't bother trying to
 430                  * wake him up.
 431                  */
 432                 atomic_dec(&part->channel_mgr_requests);
 433                 (void)wait_event_interruptible(part->channel_mgr_wq,
 434                                 (atomic_read(&part->channel_mgr_requests) > 0 ||
 435                                  part->chctl.all_flags != 0 ||
 436                                  (part->act_state == XPC_P_AS_DEACTIVATING &&
 437                                  atomic_read(&part->nchannels_active) == 0 &&
 438                                  xpc_partition_disengaged(part))));
 439                 atomic_set(&part->channel_mgr_requests, 1);
 440         }
 441 }
 442
 443 /*
 444  * Guarantee that the kzalloc'd memory is cacheline aligned.
 445  */
 446 void *
 447 xpc_kzalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
 448 {
 449         /* see if kzalloc will give us cachline aligned memory by default */
 450         *base = kzalloc(size, flags);
 451         if (*base == NULL)
 452                 return NULL;
 453
 454         if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
 455                 return *base;
 456
 457         kfree(*base);
 458
 459         /* nope, we'll have to do it ourselves */
 460         *base = kzalloc(size + L1_CACHE_BYTES, flags);
 461         if (*base == NULL)
 462                 return NULL;
 463
 464         return (void *)L1_CACHE_ALIGN((u64)*base);
 465 }
 466
 467 /*
 468  * Setup the channel structures necessary to support XPartition Communication
 469  * between the specified remote partition and the local one.
 470  */
 471 static enum xp_retval
 472 xpc_setup_ch_structures(struct xpc_partition *part)
 473 {
 474         enum xp_retval ret;
 475         int ch_number;
 476         struct xpc_channel *ch;
 477         short partid = XPC_PARTID(part);
 478
 479         /*
 480          * Allocate all of the channel structures as a contiguous chunk of
 481          * memory.
 482          */
 483         DBUG_ON(part->channels != NULL);
 484         part->channels = kzalloc(sizeof(struct xpc_channel) * XPC_MAX_NCHANNELS,
 485                                  GFP_KERNEL);
 486         if (part->channels == NULL) {
 487                 dev_err(xpc_chan, "can't get memory for channels\n");
 488                 return xpNoMemory;
 489         }
 490
 491         /* allocate the remote open and close args */
 492
 493         part->remote_openclose_args =
 494             xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE,
 495                                           GFP_KERNEL, &part->
 496                                           remote_openclose_args_base);
 497         if (part->remote_openclose_args == NULL) {
 498                 dev_err(xpc_chan, "can't get memory for remote connect args\n");
 499                 ret = xpNoMemory;
 500                 goto out_1;
 501         }
 502
 503         part->chctl.all_flags = 0;
 504         spin_lock_init(&part->chctl_lock);
 505
 506         atomic_set(&part->channel_mgr_requests, 1);
 507         init_waitqueue_head(&part->channel_mgr_wq);
 508
 509         part->nchannels = XPC_MAX_NCHANNELS;
 510
 511         atomic_set(&part->nchannels_active, 0);
 512         atomic_set(&part->nchannels_engaged, 0);
 513
 514         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
 515                 ch = &part->channels[ch_number];
 516
 517                 ch->partid = partid;
 518                 ch->number = ch_number;
 519                 ch->flags = XPC_C_DISCONNECTED;
 520
 521                 atomic_set(&ch->kthreads_assigned, 0);
 522                 atomic_set(&ch->kthreads_idle, 0);
 523                 atomic_set(&ch->kthreads_active, 0);
 524
 525                 atomic_set(&ch->references, 0);
 526                 atomic_set(&ch->n_to_notify, 0);
 527
 528                 spin_lock_init(&ch->lock);
 529                 init_completion(&ch->wdisconnect_wait);
 530
 531                 atomic_set(&ch->n_on_msg_allocate_wq, 0);
 532                 init_waitqueue_head(&ch->msg_allocate_wq);
 533                 init_waitqueue_head(&ch->idle_wq);
 534         }
 535
 536         ret = xpc_setup_ch_structures_sn(part);
 537         if (ret != xpSuccess)
 538                 goto out_2;
 539
 540         /*
 541          * With the setting of the partition setup_state to XPC_P_SS_SETUP,
 542          * we're declaring that this partition is ready to go.
 543          */
 544         part->setup_state = XPC_P_SS_SETUP;
 545
 546         return xpSuccess;
 547
 548         /* setup of ch structures failed */
 549 out_2:
 550         kfree(part->remote_openclose_args_base);
 551         part->remote_openclose_args = NULL;
 552 out_1:
 553         kfree(part->channels);
 554         part->channels = NULL;
 555         return ret;
 556 }
 557
 558 /*
 559  * Teardown the channel structures necessary to support XPartition Communication
 560  * between the specified remote partition and the local one.
 561  */
 562 static void
 563 xpc_teardown_ch_structures(struct xpc_partition *part)
 564 {
 565         DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
 566         DBUG_ON(atomic_read(&part->nchannels_active) != 0);
 567
 568         /*
 569          * Make this partition inaccessible to local processes by marking it
 570          * as no longer setup. Then wait before proceeding with the teardown
 571          * until all existing references cease.
 572          */
 573         DBUG_ON(part->setup_state != XPC_P_SS_SETUP);
 574         part->setup_state = XPC_P_SS_WTEARDOWN;
 575
 576         wait_event(part->teardown_wq, (atomic_read(&part->references) == 0));
 577
 578         /* now we can begin tearing down the infrastructure */
 579
 580         xpc_teardown_ch_structures_sn(part);
 581
 582         kfree(part->remote_openclose_args_base);
 583         part->remote_openclose_args = NULL;
 584         kfree(part->channels);
 585         part->channels = NULL;
 586
 587         part->setup_state = XPC_P_SS_TORNDOWN;
 588 }
 589
 590 /*
 591  * When XPC HB determines that a partition has come up, it will create a new
 592  * kthread and that kthread will call this function to attempt to set up the
 593  * basic infrastructure used for Cross Partition Communication with the newly
 594  * upped partition.
 595  *
 596  * The kthread that was created by XPC HB and which setup the XPC
 597  * infrastructure will remain assigned to the partition becoming the channel
 598  * manager for that partition until the partition is deactivating, at which
 599  * time the kthread will teardown the XPC infrastructure and then exit.
 600  */
 601 static int
 602 xpc_activating(void *__partid)
 603 {
 604         short partid = (u64)__partid;
 605         struct xpc_partition *part = &xpc_partitions[partid];
 606         unsigned long irq_flags;
 607
 608         DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
 609
 610         spin_lock_irqsave(&part->act_lock, irq_flags);
 611
 612         if (part->act_state == XPC_P_AS_DEACTIVATING) {
 613                 part->act_state = XPC_P_AS_INACTIVE;
 614                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 615                 part->remote_rp_pa = 0;
 616                 return 0;
 617         }
 618
 619         /* indicate the thread is activating */
 620         DBUG_ON(part->act_state != XPC_P_AS_ACTIVATION_REQ);
 621         part->act_state = XPC_P_AS_ACTIVATING;
 622
 623         XPC_SET_REASON(part, 0, 0);
 624         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 625
 626         dev_dbg(xpc_part, "activating partition %d\n", partid);
 627
 628         xpc_allow_hb(partid);
 629
 630         if (xpc_setup_ch_structures(part) == xpSuccess) {
 631                 (void)xpc_part_ref(part);       /* this will always succeed */
 632
 633                 if (xpc_make_first_contact(part) == xpSuccess) {
 634                         xpc_mark_partition_active(part);
 635                         xpc_channel_mgr(part);
 636                         /* won't return until partition is deactivating */
 637                 }
 638
 639                 xpc_part_deref(part);
 640                 xpc_teardown_ch_structures(part);
 641         }
 642
 643         xpc_disallow_hb(partid);
 644         xpc_mark_partition_inactive(part);
 645
 646         if (part->reason == xpReactivating) {
 647                 /* interrupting ourselves results in activating partition */
 648                 xpc_request_partition_reactivation(part);
 649         }
 650
 651         return 0;
 652 }
 653
 654 void
 655 xpc_activate_partition(struct xpc_partition *part)
 656 {
 657         short partid = XPC_PARTID(part);
 658         unsigned long irq_flags;
 659         struct task_struct *kthread;
 660
 661         spin_lock_irqsave(&part->act_lock, irq_flags);
 662
 663         DBUG_ON(part->act_state != XPC_P_AS_INACTIVE);
 664
 665         part->act_state = XPC_P_AS_ACTIVATION_REQ;
 666         XPC_SET_REASON(part, xpCloneKThread, __LINE__);
 667
 668         spin_unlock_irqrestore(&part->act_lock, irq_flags);
 669
 670         kthread = kthread_run(xpc_activating, (void *)((u64)partid), "xpc%02d",
 671                               partid);
 672         if (IS_ERR(kthread)) {
 673                 spin_lock_irqsave(&part->act_lock, irq_flags);
 674                 part->act_state = XPC_P_AS_INACTIVE;
 675                 XPC_SET_REASON(part, xpCloneKThreadFailed, __LINE__);
 676                 spin_unlock_irqrestore(&part->act_lock, irq_flags);
 677         }
 678 }
 679
 680 void
 681 xpc_activate_kthreads(struct xpc_channel *ch, int needed)
 682 {
 683         int idle = atomic_read(&ch->kthreads_idle);
 684         int assigned = atomic_read(&ch->kthreads_assigned);
 685         int wakeup;
 686
 687         DBUG_ON(needed <= 0);
 688
 689         if (idle > 0) {
 690                 wakeup = (needed > idle) ? idle : needed;
 691                 needed -= wakeup;
 692
 693                 dev_dbg(xpc_chan, "wakeup %d idle kthreads, partid=%d, "
 694                         "channel=%d\n", wakeup, ch->partid, ch->number);
 695
 696                 /* only wakeup the requested number of kthreads */
 697                 wake_up_nr(&ch->idle_wq, wakeup);
 698         }
 699
 700         if (needed <= 0)
 701                 return;
 702
 703         if (needed + assigned > ch->kthreads_assigned_limit) {
 704                 needed = ch->kthreads_assigned_limit - assigned;
 705                 if (needed <= 0)
 706                         return;
 707         }
 708
 709         dev_dbg(xpc_chan, "create %d new kthreads, partid=%d, channel=%d\n",
 710                 needed, ch->partid, ch->number);
 711
 712         xpc_create_kthreads(ch, needed, 0);
 713 }
 714
 715 /*
 716  * This function is where XPC's kthreads wait for messages to deliver.
 717  */
 718 static void
 719 xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
 720 {
 721         do {
 722                 /* deliver messages to their intended recipients */
 723
 724                 while (xpc_n_of_deliverable_payloads(ch) > 0 &&
 725                        !(ch->flags & XPC_C_DISCONNECTING)) {
 726                         xpc_deliver_payload(ch);
 727                 }
 728
 729                 if (atomic_inc_return(&ch->kthreads_idle) >
 730                     ch->kthreads_idle_limit) {
 731                         /* too many idle kthreads on this channel */
 732                         atomic_dec(&ch->kthreads_idle);
 733                         break;
 734                 }
 735
 736                 dev_dbg(xpc_chan, "idle kthread calling "
 737                         "wait_event_interruptible_exclusive()\n");
 738
 739                 (void)wait_event_interruptible_exclusive(ch->idle_wq,
 740                                 (xpc_n_of_deliverable_payloads(ch) > 0 ||
 741                                  (ch->flags & XPC_C_DISCONNECTING)));
 742
 743                 atomic_dec(&ch->kthreads_idle);
 744
 745         } while (!(ch->flags & XPC_C_DISCONNECTING));
 746 }
 747
 748 static int
 749 xpc_kthread_start(void *args)
 750 {
 751         short partid = XPC_UNPACK_ARG1(args);
 752         u16 ch_number = XPC_UNPACK_ARG2(args);
 753         struct xpc_partition *part = &xpc_partitions[partid];
 754         struct xpc_channel *ch;
 755         int n_needed;
 756         unsigned long irq_flags;
 757
 758         dev_dbg(xpc_chan, "kthread starting, partid=%d, channel=%d\n",
 759                 partid, ch_number);
 760
 761         ch = &part->channels[ch_number];
 762
 763         if (!(ch->flags & XPC_C_DISCONNECTING)) {
 764
 765                 /* let registerer know that connection has been established */
 766
 767                 spin_lock_irqsave(&ch->lock, irq_flags);
 768                 if (!(ch->flags & XPC_C_CONNECTEDCALLOUT)) {
 769                         ch->flags |= XPC_C_CONNECTEDCALLOUT;
 770                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 771
 772                         xpc_connected_callout(ch);
 773
 774                         spin_lock_irqsave(&ch->lock, irq_flags);
 775                         ch->flags |= XPC_C_CONNECTEDCALLOUT_MADE;
 776                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 777
 778                         /*
 779                          * It is possible that while the callout was being
 780                          * made that the remote partition sent some messages.
 781                          * If that is the case, we may need to activate
 782                          * additional kthreads to help deliver them. We only
 783                          * need one less than total #of messages to deliver.
 784                          */
 785                         n_needed = xpc_n_of_deliverable_payloads(ch) - 1;
 786                         if (n_needed > 0 && !(ch->flags & XPC_C_DISCONNECTING))
 787                                 xpc_activate_kthreads(ch, n_needed);
 788
 789                 } else {
 790                         spin_unlock_irqrestore(&ch->lock, irq_flags);
 791                 }
 792
 793                 xpc_kthread_waitmsgs(part, ch);
 794         }
 795
 796         /* let registerer know that connection is disconnecting */
 797
 798         spin_lock_irqsave(&ch->lock, irq_flags);
 799         if ((ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) &&
 800             !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) {
 801                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT;
 802                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 803
 804                 xpc_disconnect_callout(ch, xpDisconnecting);
 805
 806                 spin_lock_irqsave(&ch->lock, irq_flags);
 807                 ch->flags |= XPC_C_DISCONNECTINGCALLOUT_MADE;
 808         }
 809         spin_unlock_irqrestore(&ch->lock, irq_flags);
 810
 811         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 812             atomic_dec_return(&part->nchannels_engaged) == 0) {
 813                 xpc_indicate_partition_disengaged(part);
 814         }
 815
 816         xpc_msgqueue_deref(ch);
 817
 818         dev_dbg(xpc_chan, "kthread exiting, partid=%d, channel=%d\n",
 819                 partid, ch_number);
 820
 821         xpc_part_deref(part);
 822         return 0;
 823 }
 824
 825 /*
 826  * For each partition that XPC has established communications with, there is
 827  * a minimum of one kernel thread assigned to perform any operation that
 828  * may potentially sleep or block (basically the callouts to the asynchronous
 829  * functions registered via xpc_connect()).
 830  *
 831  * Additional kthreads are created and destroyed by XPC as the workload
 832  * demands.
 833  *
 834  * A kthread is assigned to one of the active channels that exists for a given
 835  * partition.
 836  */
 837 void
 838 xpc_create_kthreads(struct xpc_channel *ch, int needed,
 839                     int ignore_disconnecting)
 840 {
 841         unsigned long irq_flags;
 842         u64 args = XPC_PACK_ARGS(ch->partid, ch->number);
 843         struct xpc_partition *part = &xpc_partitions[ch->partid];
 844         struct task_struct *kthread;
 845
 846         while (needed-- > 0) {
 847
 848                 /*
 849                  * The following is done on behalf of the newly created
 850                  * kthread. That kthread is responsible for doing the
 851                  * counterpart to the following before it exits.
 852                  */
 853                 if (ignore_disconnecting) {
 854                         if (!atomic_inc_not_zero(&ch->kthreads_assigned)) {
 855                                 /* kthreads assigned had gone to zero */
 856                                 BUG_ON(!(ch->flags &
 857                                          XPC_C_DISCONNECTINGCALLOUT_MADE));
 858                                 break;
 859                         }
 860
 861                 } else if (ch->flags & XPC_C_DISCONNECTING) {
 862                         break;
 863
 864                 } else if (atomic_inc_return(&ch->kthreads_assigned) == 1 &&
 865                            atomic_inc_return(&part->nchannels_engaged) == 1) {
 866                                 xpc_indicate_partition_engaged(part);
 867                 }
 868                 (void)xpc_part_ref(part);
 869                 xpc_msgqueue_ref(ch);
 870
 871                 kthread = kthread_run(xpc_kthread_start, (void *)args,
 872                                       "xpc%02dc%d", ch->partid, ch->number);
 873                 if (IS_ERR(kthread)) {
 874                         /* the fork failed */
 875
 876                         /*
 877                          * NOTE: if (ignore_disconnecting &&
 878                          * !(ch->flags & XPC_C_DISCONNECTINGCALLOUT)) is true,
 879                          * then we'll deadlock if all other kthreads assigned
 880                          * to this channel are blocked in the channel's
 881                          * registerer, because the only thing that will unblock
 882                          * them is the xpDisconnecting callout that this
 883                          * failed kthread_run() would have made.
 884                          */
 885
 886                         if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
 887                             atomic_dec_return(&part->nchannels_engaged) == 0) {
 888                                 xpc_indicate_partition_disengaged(part);
 889                         }
 890                         xpc_msgqueue_deref(ch);
 891                         xpc_part_deref(part);
 892
 893                         if (atomic_read(&ch->kthreads_assigned) <
 894                             ch->kthreads_idle_limit) {
 895                                 /*
 896                                  * Flag this as an error only if we have an
 897                                  * insufficient #of kthreads for the channel
 898                                  * to function.
 899                                  */
 900                                 spin_lock_irqsave(&ch->lock, irq_flags);
 901                                 XPC_DISCONNECT_CHANNEL(ch, xpLackOfResources,
 902                                                        &irq_flags);
 903                                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 904                         }
 905                         break;
 906                 }
 907         }
 908 }
 909
 910 void
 911 xpc_disconnect_wait(int ch_number)
 912 {
 913         unsigned long irq_flags;
 914         short partid;
 915         struct xpc_partition *part;
 916         struct xpc_channel *ch;
 917         int wakeup_channel_mgr;
 918
 919         /* now wait for all callouts to the caller's function to cease */
 920         for (partid = 0; partid < xp_max_npartitions; partid++) {
 921                 part = &xpc_partitions[partid];
 922
 923                 if (!xpc_part_ref(part))
 924                         continue;
 925
 926                 ch = &part->channels[ch_number];
 927
 928                 if (!(ch->flags & XPC_C_WDISCONNECT)) {
 929                         xpc_part_deref(part);
 930                         continue;
 931                 }
 932
 933                 wait_for_completion(&ch->wdisconnect_wait);
 934
 935                 spin_lock_irqsave(&ch->lock, irq_flags);
 936                 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTED));
 937                 wakeup_channel_mgr = 0;
 938
 939                 if (ch->delayed_chctl_flags) {
 940                         if (part->act_state != XPC_P_AS_DEACTIVATING) {
 941                                 spin_lock(&part->chctl_lock);
 942                                 part->chctl.flags[ch->number] |=
 943                                     ch->delayed_chctl_flags;
 944                                 spin_unlock(&part->chctl_lock);
 945                                 wakeup_channel_mgr = 1;
 946                         }
 947                         ch->delayed_chctl_flags = 0;
 948                 }
 949
 950                 ch->flags &= ~XPC_C_WDISCONNECT;
 951                 spin_unlock_irqrestore(&ch->lock, irq_flags);
 952
 953                 if (wakeup_channel_mgr)
 954                         xpc_wakeup_channel_mgr(part);
 955
 956                 xpc_part_deref(part);
 957         }
 958 }
 959
 960 static int
 961 xpc_setup_partitions(void)
 962 {
 963         short partid;
 964         struct xpc_partition *part;
 965
 966         xpc_partitions = kzalloc(sizeof(struct xpc_partition) *
 967                                  xp_max_npartitions, GFP_KERNEL);
 968         if (xpc_partitions == NULL) {
 969                 dev_err(xpc_part, "can't get memory for partition structure\n");
 970                 return -ENOMEM;
 971         }
 972
 973         /*
 974          * The first few fields of each entry of xpc_partitions[] need to
 975          * be initialized now so that calls to xpc_connect() and
 976          * xpc_disconnect() can be made prior to the activation of any remote
 977          * partition. NOTE THAT NONE OF THE OTHER FIELDS BELONGING TO THESE
 978          * ENTRIES ARE MEANINGFUL UNTIL AFTER AN ENTRY'S CORRESPONDING
 979          * PARTITION HAS BEEN ACTIVATED.
 980          */
 981         for (partid = 0; partid < xp_max_npartitions; partid++) {
 982                 part = &xpc_partitions[partid];
 983
 984                 DBUG_ON((u64)part != L1_CACHE_ALIGN((u64)part));
 985
 986                 part->activate_IRQ_rcvd = 0;
 987                 spin_lock_init(&part->act_lock);
 988                 part->act_state = XPC_P_AS_INACTIVE;
 989                 XPC_SET_REASON(part, 0, 0);
 990
 991                 init_timer(&part->disengage_timer);
 992                 part->disengage_timer.function =
 993                     xpc_timeout_partition_disengage;
 994                 part->disengage_timer.data = (unsigned long)part;
 995
 996                 part->setup_state = XPC_P_SS_UNSET;
 997                 init_waitqueue_head(&part->teardown_wq);
 998                 atomic_set(&part->references, 0);
 999         }
1000
1001         return xpc_setup_partitions_sn();
1002 }
1003
1004 static void
1005 xpc_teardown_partitions(void)
1006 {
1007         xpc_teardown_partitions_sn();
1008         kfree(xpc_partitions);
1009 }
1010
1011 static void
1012 xpc_do_exit(enum xp_retval reason)
1013 {
1014         short partid;
1015         int active_part_count, printed_waiting_msg = 0;
1016         struct xpc_partition *part;
1017         unsigned long printmsg_time, disengage_timeout = 0;
1018
1019         /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
1020         DBUG_ON(xpc_exiting == 1);
1021
1022         /*
1023          * Let the heartbeat checker thread and the discovery thread
1024          * (if one is running) know that they should exit. Also wake up
1025          * the heartbeat checker thread in case it's sleeping.
1026          */
1027         xpc_exiting = 1;
1028         wake_up_interruptible(&xpc_activate_IRQ_wq);
1029
1030         /* wait for the discovery thread to exit */
1031         wait_for_completion(&xpc_discovery_exited);
1032
1033         /* wait for the heartbeat checker thread to exit */
1034         wait_for_completion(&xpc_hb_checker_exited);
1035
1036         /* sleep for a 1/3 of a second or so */
1037         (void)msleep_interruptible(300);
1038
1039         /* wait for all partitions to become inactive */
1040
1041         printmsg_time = jiffies + (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
1042         xpc_disengage_timedout = 0;
1043
1044         do {
1045                 active_part_count = 0;
1046
1047                 for (partid = 0; partid < xp_max_npartitions; partid++) {
1048                         part = &xpc_partitions[partid];
1049
1050                         if (xpc_partition_disengaged(part) &&
1051                             part->act_state == XPC_P_AS_INACTIVE) {
1052                                 continue;
1053                         }
1054
1055                         active_part_count++;
1056
1057                         XPC_DEACTIVATE_PARTITION(part, reason);
1058
1059                         if (part->disengage_timeout > disengage_timeout)
1060                                 disengage_timeout = part->disengage_timeout;
1061                 }
1062
1063                 if (xpc_any_partition_engaged()) {
1064                         if (time_is_before_jiffies(printmsg_time)) {
1065                                 dev_info(xpc_part, "waiting for remote "
1066                                          "partitions to deactivate, timeout in "
1067                                          "%ld seconds\n", (disengage_timeout -
1068                                          jiffies) / HZ);
1069                                 printmsg_time = jiffies +
1070                                     (XPC_DEACTIVATE_PRINTMSG_INTERVAL * HZ);
1071                                 printed_waiting_msg = 1;
1072                         }
1073
1074                 } else if (active_part_count > 0) {
1075                         if (printed_waiting_msg) {
1076                                 dev_info(xpc_part, "waiting for local partition"
1077                                          " to deactivate\n");
1078                                 printed_waiting_msg = 0;
1079                         }
1080
1081                 } else {
1082                         if (!xpc_disengage_timedout) {
1083                                 dev_info(xpc_part, "all partitions have "
1084                                          "deactivated\n");
1085                         }
1086                         break;
1087                 }
1088
1089                 /* sleep for a 1/3 of a second or so */
1090                 (void)msleep_interruptible(300);
1091
1092         } while (1);
1093
1094         DBUG_ON(xpc_any_partition_engaged());
1095
1096         xpc_teardown_rsvd_page();
1097
1098         if (reason == xpUnloading) {
1099                 (void)unregister_die_notifier(&xpc_die_notifier);
1100                 (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1101         }
1102
1103         /* clear the interface to XPC's functions */
1104         xpc_clear_interface();
1105
1106         if (xpc_sysctl)
1107                 unregister_sysctl_table(xpc_sysctl);
1108
1109         xpc_teardown_partitions();
1110
1111         if (is_shub())
1112                 xpc_exit_sn2();
1113         else if (is_uv())
1114                 xpc_exit_uv();
1115 }
1116
1117 /*
1118  * This function is called when the system is being rebooted.
1119  */
1120 static int
1121 xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
1122 {
1123         enum xp_retval reason;
1124
1125         switch (event) {
1126         case SYS_RESTART:
1127                 reason = xpSystemReboot;
1128                 break;
1129         case SYS_HALT:
1130                 reason = xpSystemHalt;
1131                 break;
1132         case SYS_POWER_OFF:
1133                 reason = xpSystemPoweroff;
1134                 break;
1135         default:
1136                 reason = xpSystemGoingDown;
1137         }
1138
1139         xpc_do_exit(reason);
1140         return NOTIFY_DONE;
1141 }
1142
1143 /*
1144  * Notify other partitions to deactivate from us by first disengaging from all
1145  * references to our memory.
1146  */
1147 static void
1148 xpc_die_deactivate(void)
1149 {
1150         struct xpc_partition *part;
1151         short partid;
1152         int any_engaged;
1153         long keep_waiting;
1154         long wait_to_print;
1155
1156         /* keep xpc_hb_checker thread from doing anything (just in case) */
1157         xpc_exiting = 1;
1158
1159         xpc_disallow_all_hbs(); /*indicate we're deactivated */
1160
1161         for (partid = 0; partid < xp_max_npartitions; partid++) {
1162                 part = &xpc_partitions[partid];
1163
1164                 if (xpc_partition_engaged(partid) ||
1165                     part->act_state != XPC_P_AS_INACTIVE) {
1166                         xpc_request_partition_deactivation(part);
1167                         xpc_indicate_partition_disengaged(part);
1168                 }
1169         }
1170
1171         /*
1172          * Though we requested that all other partitions deactivate from us,
1173          * we only wait until they've all disengaged or we've reached the
1174          * defined timelimit.
1175          *
1176          * Given that one iteration through the following while-loop takes
1177          * approximately 200 microseconds, calculate the #of loops to take
1178          * before bailing and the #of loops before printing a waiting message.
1179          */
1180         keep_waiting = xpc_disengage_timelimit * 1000 * 5;
1181         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL * 1000 * 5;
1182
1183         while (1) {
1184                 any_engaged = xpc_any_partition_engaged();
1185                 if (!any_engaged) {
1186                         dev_info(xpc_part, "all partitions have deactivated\n");
1187                         break;
1188                 }
1189
1190                 if (!keep_waiting--) {
1191                         for (partid = 0; partid < xp_max_npartitions;
1192                              partid++) {
1193                                 if (xpc_partition_engaged(partid)) {
1194                                         dev_info(xpc_part, "deactivate from "
1195                                                  "remote partition %d timed "
1196                                                  "out\n", partid);
1197                                 }
1198                         }
1199                         break;
1200                 }
1201
1202                 if (!wait_to_print--) {
1203                         dev_info(xpc_part, "waiting for remote partitions to "
1204                                  "deactivate, timeout in %ld seconds\n",
1205                                  keep_waiting / (1000 * 5));
1206                         wait_to_print = XPC_DEACTIVATE_PRINTMSG_INTERVAL *
1207                             1000 * 5;
1208                 }
1209
1210                 udelay(200);
1211         }
1212 }
1213
1214 /*
1215  * This function is called when the system is being restarted or halted due
1216  * to some sort of system failure. If this is the case we need to notify the
1217  * other partitions to disengage from all references to our memory.
1218  * This function can also be called when our heartbeater could be offlined
1219  * for a time. In this case we need to notify other partitions to not worry
1220  * about the lack of a heartbeat.
1221  */
1222 static int
1223 xpc_system_die(struct notifier_block *nb, unsigned long event, void *unused)
1224 {
1225 #ifdef CONFIG_IA64              /* !!! temporary kludge */
1226         switch (event) {
1227         case DIE_MACHINE_RESTART:
1228         case DIE_MACHINE_HALT:
1229                 xpc_die_deactivate();
1230                 break;
1231
1232         case DIE_KDEBUG_ENTER:
1233                 /* Should lack of heartbeat be ignored by other partitions? */
1234                 if (!xpc_kdebug_ignore)
1235                         break;
1236
1237                 /* fall through */
1238         case DIE_MCA_MONARCH_ENTER:
1239         case DIE_INIT_MONARCH_ENTER:
1240                 xpc_offline_heartbeat();
1241                 break;
1242
1243         case DIE_KDEBUG_LEAVE:
1244                 /* Is lack of heartbeat being ignored by other partitions? */
1245                 if (!xpc_kdebug_ignore)
1246                         break;
1247
1248                 /* fall through */
1249         case DIE_MCA_MONARCH_LEAVE:
1250         case DIE_INIT_MONARCH_LEAVE:
1251                 xpc_online_heartbeat();
1252                 break;
1253         }
1254 #else
1255         xpc_die_deactivate();
1256 #endif
1257
1258         return NOTIFY_DONE;
1259 }
1260
1261 int __init
1262 xpc_init(void)
1263 {
1264         int ret;
1265         struct task_struct *kthread;
1266
1267         dev_set_name(xpc_part, "part");
1268         dev_set_name(xpc_chan, "chan");
1269
1270         if (is_shub()) {
1271                 /*
1272                  * The ia64-sn2 architecture supports at most 64 partitions.
1273                  * And the inability to unregister remote amos restricts us
1274                  * further to only support exactly 64 partitions on this
1275                  * architecture, no less.
1276                  */
1277                 if (xp_max_npartitions != 64) {
1278                         dev_err(xpc_part, "max #of partitions not set to 64\n");
1279                         ret = -EINVAL;
1280                 } else {
1281                         ret = xpc_init_sn2();
1282                 }
1283
1284         } else if (is_uv()) {
1285                 ret = xpc_init_uv();
1286
1287         } else {
1288                 ret = -ENODEV;
1289         }
1290
1291         if (ret != 0)
1292                 return ret;
1293
1294         ret = xpc_setup_partitions();
1295         if (ret != 0) {
1296                 dev_err(xpc_part, "can't get memory for partition structure\n");
1297                 goto out_1;
1298         }
1299
1300         xpc_sysctl = register_sysctl_table(xpc_sys_dir);
1301
1302         /*
1303          * Fill the partition reserved page with the information needed by
1304          * other partitions to discover we are alive and establish initial
1305          * communications.
1306          */
1307         ret = xpc_setup_rsvd_page();
1308         if (ret != 0) {
1309                 dev_err(xpc_part, "can't setup our reserved page\n");
1310                 goto out_2;
1311         }
1312
1313         /* add ourselves to the reboot_notifier_list */
1314         ret = register_reboot_notifier(&xpc_reboot_notifier);
1315         if (ret != 0)
1316                 dev_warn(xpc_part, "can't register reboot notifier\n");
1317
1318         /* add ourselves to the die_notifier list */
1319         ret = register_die_notifier(&xpc_die_notifier);
1320         if (ret != 0)
1321                 dev_warn(xpc_part, "can't register die notifier\n");
1322
1323         /*
1324          * The real work-horse behind xpc.  This processes incoming
1325          * interrupts and monitors remote heartbeats.
1326          */
1327         kthread = kthread_run(xpc_hb_checker, NULL, XPC_HB_CHECK_THREAD_NAME);
1328         if (IS_ERR(kthread)) {
1329                 dev_err(xpc_part, "failed while forking hb check thread\n");
1330                 ret = -EBUSY;
1331                 goto out_3;
1332         }
1333
1334         /*
1335          * Startup a thread that will attempt to discover other partitions to
1336          * activate based on info provided by SAL. This new thread is short
1337          * lived and will exit once discovery is complete.
1338          */
1339         kthread = kthread_run(xpc_initiate_discovery, NULL,
1340                               XPC_DISCOVERY_THREAD_NAME);
1341         if (IS_ERR(kthread)) {
1342                 dev_err(xpc_part, "failed while forking discovery thread\n");
1343
1344                 /* mark this new thread as a non-starter */
1345                 complete(&xpc_discovery_exited);
1346
1347                 xpc_do_exit(xpUnloading);
1348                 return -EBUSY;
1349         }
1350
1351         /* set the interface to point at XPC's functions */
1352         xpc_set_interface(xpc_initiate_connect, xpc_initiate_disconnect,
1353                           xpc_initiate_send, xpc_initiate_send_notify,
1354                           xpc_initiate_received, xpc_initiate_partid_to_nasids);
1355
1356         return 0;
1357
1358         /* initialization was not successful */
1359 out_3:
1360         xpc_teardown_rsvd_page();
1361
1362         (void)unregister_die_notifier(&xpc_die_notifier);
1363         (void)unregister_reboot_notifier(&xpc_reboot_notifier);
1364 out_2:
1365         if (xpc_sysctl)
1366                 unregister_sysctl_table(xpc_sysctl);
1367
1368         xpc_teardown_partitions();
1369 out_1:
1370         if (is_shub())
1371                 xpc_exit_sn2();
1372         else if (is_uv())
1373                 xpc_exit_uv();
1374         return ret;
1375 }
1376
1377 module_init(xpc_init);
1378
1379 void __exit
1380 xpc_exit(void)
1381 {
1382         xpc_do_exit(xpUnloading);
1383 }
1384
1385 module_exit(xpc_exit);
1386
1387 MODULE_AUTHOR("Silicon Graphics, Inc.");
1388 MODULE_DESCRIPTION("Cross Partition Communication (XPC) support");
1389 MODULE_LICENSE("GPL");
1390
1391 module_param(xpc_hb_interval, int, 0);
1392 MODULE_PARM_DESC(xpc_hb_interval, "Number of seconds between "
1393                  "heartbeat increments.");
1394
1395 module_param(xpc_hb_check_interval, int, 0);
1396 MODULE_PARM_DESC(xpc_hb_check_interval, "Number of seconds between "
1397                  "heartbeat checks.");
1398
1399 module_param(xpc_disengage_timelimit, int, 0);
1400 MODULE_PARM_DESC(xpc_disengage_timelimit, "Number of seconds to wait "
1401                  "for disengage to complete.");
1402
1403 module_param(xpc_kdebug_ignore, int, 0);
1404 MODULE_PARM_DESC(xpc_kdebug_ignore, "Should lack of heartbeat be ignored by "
1405                  "other partitions when dropping into kdebug.");