linux/drivers/misc/sgi-xp/xpc_sn2.c
<<
>>
Prefs
   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) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
   7 */
   8
   9/*
  10 * Cross Partition Communication (XPC) sn2-based functions.
  11 *
  12 *     Architecture specific implementation of common functions.
  13 *
  14 */
  15
  16#include <linux/delay.h>
  17#include <linux/slab.h>
  18#include <asm/uncached.h>
  19#include <asm/sn/mspec.h>
  20#include <asm/sn/sn_sal.h>
  21#include "xpc.h"
  22
  23/*
  24 * Define the number of u64s required to represent all the C-brick nasids
  25 * as a bitmap.  The cross-partition kernel modules deal only with
  26 * C-brick nasids, thus the need for bitmaps which don't account for
  27 * odd-numbered (non C-brick) nasids.
  28 */
  29#define XPC_MAX_PHYSNODES_SN2   (MAX_NUMALINK_NODES / 2)
  30#define XP_NASID_MASK_BYTES_SN2 ((XPC_MAX_PHYSNODES_SN2 + 7) / 8)
  31#define XP_NASID_MASK_WORDS_SN2 ((XPC_MAX_PHYSNODES_SN2 + 63) / 64)
  32
  33/*
  34 * Memory for XPC's amo variables is allocated by the MSPEC driver. These
  35 * pages are located in the lowest granule. The lowest granule uses 4k pages
  36 * for cached references and an alternate TLB handler to never provide a
  37 * cacheable mapping for the entire region. This will prevent speculative
  38 * reading of cached copies of our lines from being issued which will cause
  39 * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
  40 * amo variables (based on XP_MAX_NPARTITIONS_SN2) to identify the senders of
  41 * NOTIFY IRQs, 128 amo variables (based on XP_NASID_MASK_WORDS_SN2) to identify
  42 * the senders of ACTIVATE IRQs, 1 amo variable to identify which remote
  43 * partitions (i.e., XPCs) consider themselves currently engaged with the
  44 * local XPC and 1 amo variable to request partition deactivation.
  45 */
  46#define XPC_NOTIFY_IRQ_AMOS_SN2         0
  47#define XPC_ACTIVATE_IRQ_AMOS_SN2       (XPC_NOTIFY_IRQ_AMOS_SN2 + \
  48                                         XP_MAX_NPARTITIONS_SN2)
  49#define XPC_ENGAGED_PARTITIONS_AMO_SN2  (XPC_ACTIVATE_IRQ_AMOS_SN2 + \
  50                                         XP_NASID_MASK_WORDS_SN2)
  51#define XPC_DEACTIVATE_REQUEST_AMO_SN2  (XPC_ENGAGED_PARTITIONS_AMO_SN2 + 1)
  52
  53/*
  54 * Buffer used to store a local copy of portions of a remote partition's
  55 * reserved page (either its header and part_nasids mask, or its vars).
  56 */
  57static void *xpc_remote_copy_buffer_base_sn2;
  58static char *xpc_remote_copy_buffer_sn2;
  59
  60static struct xpc_vars_sn2 *xpc_vars_sn2;
  61static struct xpc_vars_part_sn2 *xpc_vars_part_sn2;
  62
  63static int
  64xpc_setup_partitions_sn2(void)
  65{
  66        /* nothing needs to be done */
  67        return 0;
  68}
  69
  70static void
  71xpc_teardown_partitions_sn2(void)
  72{
  73        /* nothing needs to be done */
  74}
  75
  76/* SH_IPI_ACCESS shub register value on startup */
  77static u64 xpc_sh1_IPI_access_sn2;
  78static u64 xpc_sh2_IPI_access0_sn2;
  79static u64 xpc_sh2_IPI_access1_sn2;
  80static u64 xpc_sh2_IPI_access2_sn2;
  81static u64 xpc_sh2_IPI_access3_sn2;
  82
  83/*
  84 * Change protections to allow IPI operations.
  85 */
  86static void
  87xpc_allow_IPI_ops_sn2(void)
  88{
  89        int node;
  90        int nasid;
  91
  92        /* !!! The following should get moved into SAL. */
  93        if (is_shub2()) {
  94                xpc_sh2_IPI_access0_sn2 =
  95                    (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
  96                xpc_sh2_IPI_access1_sn2 =
  97                    (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
  98                xpc_sh2_IPI_access2_sn2 =
  99                    (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
 100                xpc_sh2_IPI_access3_sn2 =
 101                    (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));
 102
 103                for_each_online_node(node) {
 104                        nasid = cnodeid_to_nasid(node);
 105                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 106                              -1UL);
 107                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 108                              -1UL);
 109                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 110                              -1UL);
 111                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 112                              -1UL);
 113                }
 114        } else {
 115                xpc_sh1_IPI_access_sn2 =
 116                    (u64)HUB_L((u64 *)LOCAL_MMR_ADDR(SH1_IPI_ACCESS));
 117
 118                for_each_online_node(node) {
 119                        nasid = cnodeid_to_nasid(node);
 120                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 121                              -1UL);
 122                }
 123        }
 124}
 125
 126/*
 127 * Restrict protections to disallow IPI operations.
 128 */
 129static void
 130xpc_disallow_IPI_ops_sn2(void)
 131{
 132        int node;
 133        int nasid;
 134
 135        /* !!! The following should get moved into SAL. */
 136        if (is_shub2()) {
 137                for_each_online_node(node) {
 138                        nasid = cnodeid_to_nasid(node);
 139                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
 140                              xpc_sh2_IPI_access0_sn2);
 141                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
 142                              xpc_sh2_IPI_access1_sn2);
 143                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
 144                              xpc_sh2_IPI_access2_sn2);
 145                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
 146                              xpc_sh2_IPI_access3_sn2);
 147                }
 148        } else {
 149                for_each_online_node(node) {
 150                        nasid = cnodeid_to_nasid(node);
 151                        HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
 152                              xpc_sh1_IPI_access_sn2);
 153                }
 154        }
 155}
 156
 157/*
 158 * The following set of functions are used for the sending and receiving of
 159 * IRQs (also known as IPIs). There are two flavors of IRQs, one that is
 160 * associated with partition activity (SGI_XPC_ACTIVATE) and the other that
 161 * is associated with channel activity (SGI_XPC_NOTIFY).
 162 */
 163
 164static u64
 165xpc_receive_IRQ_amo_sn2(struct amo *amo)
 166{
 167        return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_CLEAR);
 168}
 169
 170static enum xp_retval
 171xpc_send_IRQ_sn2(struct amo *amo, u64 flag, int nasid, int phys_cpuid,
 172                 int vector)
 173{
 174        int ret = 0;
 175        unsigned long irq_flags;
 176
 177        local_irq_save(irq_flags);
 178
 179        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR, flag);
 180        sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
 181
 182        /*
 183         * We must always use the nofault function regardless of whether we
 184         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
 185         * didn't, we'd never know that the other partition is down and would
 186         * keep sending IRQs and amos to it until the heartbeat times out.
 187         */
 188        ret = xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
 189                                                     xp_nofault_PIOR_target));
 190
 191        local_irq_restore(irq_flags);
 192
 193        return (ret == 0) ? xpSuccess : xpPioReadError;
 194}
 195
 196static struct amo *
 197xpc_init_IRQ_amo_sn2(int index)
 198{
 199        struct amo *amo = xpc_vars_sn2->amos_page + index;
 200
 201        (void)xpc_receive_IRQ_amo_sn2(amo);     /* clear amo variable */
 202        return amo;
 203}
 204
 205/*
 206 * Functions associated with SGI_XPC_ACTIVATE IRQ.
 207 */
 208
 209/*
 210 * Notify the heartbeat check thread that an activate IRQ has been received.
 211 */
 212static irqreturn_t
 213xpc_handle_activate_IRQ_sn2(int irq, void *dev_id)
 214{
 215        unsigned long irq_flags;
 216
 217        spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 218        xpc_activate_IRQ_rcvd++;
 219        spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 220
 221        wake_up_interruptible(&xpc_activate_IRQ_wq);
 222        return IRQ_HANDLED;
 223}
 224
 225/*
 226 * Flag the appropriate amo variable and send an IRQ to the specified node.
 227 */
 228static void
 229xpc_send_activate_IRQ_sn2(unsigned long amos_page_pa, int from_nasid,
 230                          int to_nasid, int to_phys_cpuid)
 231{
 232        struct amo *amos = (struct amo *)__va(amos_page_pa +
 233                                              (XPC_ACTIVATE_IRQ_AMOS_SN2 *
 234                                              sizeof(struct amo)));
 235
 236        (void)xpc_send_IRQ_sn2(&amos[BIT_WORD(from_nasid / 2)],
 237                               BIT_MASK(from_nasid / 2), to_nasid,
 238                               to_phys_cpuid, SGI_XPC_ACTIVATE);
 239}
 240
 241static void
 242xpc_send_local_activate_IRQ_sn2(int from_nasid)
 243{
 244        unsigned long irq_flags;
 245        struct amo *amos = (struct amo *)__va(xpc_vars_sn2->amos_page_pa +
 246                                              (XPC_ACTIVATE_IRQ_AMOS_SN2 *
 247                                              sizeof(struct amo)));
 248
 249        /* fake the sending and receipt of an activate IRQ from remote nasid */
 250        FETCHOP_STORE_OP(TO_AMO((u64)&amos[BIT_WORD(from_nasid / 2)].variable),
 251                         FETCHOP_OR, BIT_MASK(from_nasid / 2));
 252
 253        spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 254        xpc_activate_IRQ_rcvd++;
 255        spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 256
 257        wake_up_interruptible(&xpc_activate_IRQ_wq);
 258}
 259
 260/*
 261 * Functions associated with SGI_XPC_NOTIFY IRQ.
 262 */
 263
 264/*
 265 * Check to see if any chctl flags were sent from the specified partition.
 266 */
 267static void
 268xpc_check_for_sent_chctl_flags_sn2(struct xpc_partition *part)
 269{
 270        union xpc_channel_ctl_flags chctl;
 271        unsigned long irq_flags;
 272
 273        chctl.all_flags = xpc_receive_IRQ_amo_sn2(part->sn.sn2.
 274                                                  local_chctl_amo_va);
 275        if (chctl.all_flags == 0)
 276                return;
 277
 278        spin_lock_irqsave(&part->chctl_lock, irq_flags);
 279        part->chctl.all_flags |= chctl.all_flags;
 280        spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 281
 282        dev_dbg(xpc_chan, "received notify IRQ from partid=%d, chctl.all_flags="
 283                "0x%llx\n", XPC_PARTID(part), chctl.all_flags);
 284
 285        xpc_wakeup_channel_mgr(part);
 286}
 287
 288/*
 289 * Handle the receipt of a SGI_XPC_NOTIFY IRQ by seeing whether the specified
 290 * partition actually sent it. Since SGI_XPC_NOTIFY IRQs may be shared by more
 291 * than one partition, we use an amo structure per partition to indicate
 292 * whether a partition has sent an IRQ or not.  If it has, then wake up the
 293 * associated kthread to handle it.
 294 *
 295 * All SGI_XPC_NOTIFY IRQs received by XPC are the result of IRQs sent by XPC
 296 * running on other partitions.
 297 *
 298 * Noteworthy Arguments:
 299 *
 300 *      irq - Interrupt ReQuest number. NOT USED.
 301 *
 302 *      dev_id - partid of IRQ's potential sender.
 303 */
 304static irqreturn_t
 305xpc_handle_notify_IRQ_sn2(int irq, void *dev_id)
 306{
 307        short partid = (short)(u64)dev_id;
 308        struct xpc_partition *part = &xpc_partitions[partid];
 309
 310        DBUG_ON(partid < 0 || partid >= XP_MAX_NPARTITIONS_SN2);
 311
 312        if (xpc_part_ref(part)) {
 313                xpc_check_for_sent_chctl_flags_sn2(part);
 314
 315                xpc_part_deref(part);
 316        }
 317        return IRQ_HANDLED;
 318}
 319
 320/*
 321 * Check to see if xpc_handle_notify_IRQ_sn2() dropped any IRQs on the floor
 322 * because the write to their associated amo variable completed after the IRQ
 323 * was received.
 324 */
 325static void
 326xpc_check_for_dropped_notify_IRQ_sn2(struct timer_list *t)
 327{
 328        struct xpc_partition *part =
 329                from_timer(part, t, sn.sn2.dropped_notify_IRQ_timer);
 330
 331        if (xpc_part_ref(part)) {
 332                xpc_check_for_sent_chctl_flags_sn2(part);
 333
 334                t->expires = jiffies + XPC_DROPPED_NOTIFY_IRQ_WAIT_INTERVAL;
 335                add_timer(t);
 336                xpc_part_deref(part);
 337        }
 338}
 339
 340/*
 341 * Send a notify IRQ to the remote partition that is associated with the
 342 * specified channel.
 343 */
 344static void
 345xpc_send_notify_IRQ_sn2(struct xpc_channel *ch, u8 chctl_flag,
 346                        char *chctl_flag_string, unsigned long *irq_flags)
 347{
 348        struct xpc_partition *part = &xpc_partitions[ch->partid];
 349        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
 350        union xpc_channel_ctl_flags chctl = { 0 };
 351        enum xp_retval ret;
 352
 353        if (likely(part->act_state != XPC_P_AS_DEACTIVATING)) {
 354                chctl.flags[ch->number] = chctl_flag;
 355                ret = xpc_send_IRQ_sn2(part_sn2->remote_chctl_amo_va,
 356                                       chctl.all_flags,
 357                                       part_sn2->notify_IRQ_nasid,
 358                                       part_sn2->notify_IRQ_phys_cpuid,
 359                                       SGI_XPC_NOTIFY);
 360                dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
 361                        chctl_flag_string, ch->partid, ch->number, ret);
 362                if (unlikely(ret != xpSuccess)) {
 363                        if (irq_flags != NULL)
 364                                spin_unlock_irqrestore(&ch->lock, *irq_flags);
 365                        XPC_DEACTIVATE_PARTITION(part, ret);
 366                        if (irq_flags != NULL)
 367                                spin_lock_irqsave(&ch->lock, *irq_flags);
 368                }
 369        }
 370}
 371
 372#define XPC_SEND_NOTIFY_IRQ_SN2(_ch, _ipi_f, _irq_f) \
 373                xpc_send_notify_IRQ_sn2(_ch, _ipi_f, #_ipi_f, _irq_f)
 374
 375/*
 376 * Make it look like the remote partition, which is associated with the
 377 * specified channel, sent us a notify IRQ. This faked IRQ will be handled
 378 * by xpc_check_for_dropped_notify_IRQ_sn2().
 379 */
 380static void
 381xpc_send_local_notify_IRQ_sn2(struct xpc_channel *ch, u8 chctl_flag,
 382                              char *chctl_flag_string)
 383{
 384        struct xpc_partition *part = &xpc_partitions[ch->partid];
 385        union xpc_channel_ctl_flags chctl = { 0 };
 386
 387        chctl.flags[ch->number] = chctl_flag;
 388        FETCHOP_STORE_OP(TO_AMO((u64)&part->sn.sn2.local_chctl_amo_va->
 389                                variable), FETCHOP_OR, chctl.all_flags);
 390        dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
 391                chctl_flag_string, ch->partid, ch->number);
 392}
 393
 394#define XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(_ch, _ipi_f) \
 395                xpc_send_local_notify_IRQ_sn2(_ch, _ipi_f, #_ipi_f)
 396
 397static void
 398xpc_send_chctl_closerequest_sn2(struct xpc_channel *ch,
 399                                unsigned long *irq_flags)
 400{
 401        struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
 402
 403        args->reason = ch->reason;
 404        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_CLOSEREQUEST, irq_flags);
 405}
 406
 407static void
 408xpc_send_chctl_closereply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
 409{
 410        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_CLOSEREPLY, irq_flags);
 411}
 412
 413static void
 414xpc_send_chctl_openrequest_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
 415{
 416        struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
 417
 418        args->entry_size = ch->entry_size;
 419        args->local_nentries = ch->local_nentries;
 420        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENREQUEST, irq_flags);
 421}
 422
 423static void
 424xpc_send_chctl_openreply_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
 425{
 426        struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
 427
 428        args->remote_nentries = ch->remote_nentries;
 429        args->local_nentries = ch->local_nentries;
 430        args->local_msgqueue_pa = xp_pa(ch->sn.sn2.local_msgqueue);
 431        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENREPLY, irq_flags);
 432}
 433
 434static void
 435xpc_send_chctl_opencomplete_sn2(struct xpc_channel *ch,
 436                                unsigned long *irq_flags)
 437{
 438        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENCOMPLETE, irq_flags);
 439}
 440
 441static void
 442xpc_send_chctl_msgrequest_sn2(struct xpc_channel *ch)
 443{
 444        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_MSGREQUEST, NULL);
 445}
 446
 447static void
 448xpc_send_chctl_local_msgrequest_sn2(struct xpc_channel *ch)
 449{
 450        XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_MSGREQUEST);
 451}
 452
 453static enum xp_retval
 454xpc_save_remote_msgqueue_pa_sn2(struct xpc_channel *ch,
 455                                unsigned long msgqueue_pa)
 456{
 457        ch->sn.sn2.remote_msgqueue_pa = msgqueue_pa;
 458        return xpSuccess;
 459}
 460
 461/*
 462 * This next set of functions are used to keep track of when a partition is
 463 * potentially engaged in accessing memory belonging to another partition.
 464 */
 465
 466static void
 467xpc_indicate_partition_engaged_sn2(struct xpc_partition *part)
 468{
 469        unsigned long irq_flags;
 470        struct amo *amo = (struct amo *)__va(part->sn.sn2.remote_amos_page_pa +
 471                                             (XPC_ENGAGED_PARTITIONS_AMO_SN2 *
 472                                             sizeof(struct amo)));
 473
 474        local_irq_save(irq_flags);
 475
 476        /* set bit corresponding to our partid in remote partition's amo */
 477        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
 478                         BIT(sn_partition_id));
 479
 480        /*
 481         * We must always use the nofault function regardless of whether we
 482         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
 483         * didn't, we'd never know that the other partition is down and would
 484         * keep sending IRQs and amos to it until the heartbeat times out.
 485         */
 486        (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
 487                                                               variable),
 488                                                     xp_nofault_PIOR_target));
 489
 490        local_irq_restore(irq_flags);
 491}
 492
 493static void
 494xpc_indicate_partition_disengaged_sn2(struct xpc_partition *part)
 495{
 496        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
 497        unsigned long irq_flags;
 498        struct amo *amo = (struct amo *)__va(part_sn2->remote_amos_page_pa +
 499                                             (XPC_ENGAGED_PARTITIONS_AMO_SN2 *
 500                                             sizeof(struct amo)));
 501
 502        local_irq_save(irq_flags);
 503
 504        /* clear bit corresponding to our partid in remote partition's amo */
 505        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
 506                         ~BIT(sn_partition_id));
 507
 508        /*
 509         * We must always use the nofault function regardless of whether we
 510         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
 511         * didn't, we'd never know that the other partition is down and would
 512         * keep sending IRQs and amos to it until the heartbeat times out.
 513         */
 514        (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
 515                                                               variable),
 516                                                     xp_nofault_PIOR_target));
 517
 518        local_irq_restore(irq_flags);
 519
 520        /*
 521         * Send activate IRQ to get other side to see that we've cleared our
 522         * bit in their engaged partitions amo.
 523         */
 524        xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
 525                                  cnodeid_to_nasid(0),
 526                                  part_sn2->activate_IRQ_nasid,
 527                                  part_sn2->activate_IRQ_phys_cpuid);
 528}
 529
 530static void
 531xpc_assume_partition_disengaged_sn2(short partid)
 532{
 533        struct amo *amo = xpc_vars_sn2->amos_page +
 534                          XPC_ENGAGED_PARTITIONS_AMO_SN2;
 535
 536        /* clear bit(s) based on partid mask in our partition's amo */
 537        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
 538                         ~BIT(partid));
 539}
 540
 541static int
 542xpc_partition_engaged_sn2(short partid)
 543{
 544        struct amo *amo = xpc_vars_sn2->amos_page +
 545                          XPC_ENGAGED_PARTITIONS_AMO_SN2;
 546
 547        /* our partition's amo variable ANDed with partid mask */
 548        return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
 549                BIT(partid)) != 0;
 550}
 551
 552static int
 553xpc_any_partition_engaged_sn2(void)
 554{
 555        struct amo *amo = xpc_vars_sn2->amos_page +
 556                          XPC_ENGAGED_PARTITIONS_AMO_SN2;
 557
 558        /* our partition's amo variable */
 559        return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) != 0;
 560}
 561
 562/* original protection values for each node */
 563static u64 xpc_prot_vec_sn2[MAX_NUMNODES];
 564
 565/*
 566 * Change protections to allow amo operations on non-Shub 1.1 systems.
 567 */
 568static enum xp_retval
 569xpc_allow_amo_ops_sn2(struct amo *amos_page)
 570{
 571        enum xp_retval ret = xpSuccess;
 572
 573        /*
 574         * On SHUB 1.1, we cannot call sn_change_memprotect() since the BIST
 575         * collides with memory operations. On those systems we call
 576         * xpc_allow_amo_ops_shub_wars_1_1_sn2() instead.
 577         */
 578        if (!enable_shub_wars_1_1())
 579                ret = xp_expand_memprotect(ia64_tpa((u64)amos_page), PAGE_SIZE);
 580
 581        return ret;
 582}
 583
 584/*
 585 * Change protections to allow amo operations on Shub 1.1 systems.
 586 */
 587static void
 588xpc_allow_amo_ops_shub_wars_1_1_sn2(void)
 589{
 590        int node;
 591        int nasid;
 592
 593        if (!enable_shub_wars_1_1())
 594                return;
 595
 596        for_each_online_node(node) {
 597                nasid = cnodeid_to_nasid(node);
 598                /* save current protection values */
 599                xpc_prot_vec_sn2[node] =
 600                    (u64)HUB_L((u64 *)GLOBAL_MMR_ADDR(nasid,
 601                                                  SH1_MD_DQLP_MMR_DIR_PRIVEC0));
 602                /* open up everything */
 603                HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 604                                             SH1_MD_DQLP_MMR_DIR_PRIVEC0),
 605                      -1UL);
 606                HUB_S((u64 *)GLOBAL_MMR_ADDR(nasid,
 607                                             SH1_MD_DQRP_MMR_DIR_PRIVEC0),
 608                      -1UL);
 609        }
 610}
 611
 612static enum xp_retval
 613xpc_get_partition_rsvd_page_pa_sn2(void *buf, u64 *cookie, unsigned long *rp_pa,
 614                                   size_t *len)
 615{
 616        s64 status;
 617        enum xp_retval ret;
 618
 619        status = sn_partition_reserved_page_pa((u64)buf, cookie,
 620                        (u64 *)rp_pa, (u64 *)len);
 621        if (status == SALRET_OK)
 622                ret = xpSuccess;
 623        else if (status == SALRET_MORE_PASSES)
 624                ret = xpNeedMoreInfo;
 625        else
 626                ret = xpSalError;
 627
 628        return ret;
 629}
 630
 631
 632static int
 633xpc_setup_rsvd_page_sn2(struct xpc_rsvd_page *rp)
 634{
 635        struct amo *amos_page;
 636        int i;
 637        int ret;
 638
 639        xpc_vars_sn2 = XPC_RP_VARS(rp);
 640
 641        rp->sn.sn2.vars_pa = xp_pa(xpc_vars_sn2);
 642
 643        /* vars_part array follows immediately after vars */
 644        xpc_vars_part_sn2 = (struct xpc_vars_part_sn2 *)((u8 *)XPC_RP_VARS(rp) +
 645                                                         XPC_RP_VARS_SIZE);
 646
 647        /*
 648         * Before clearing xpc_vars_sn2, see if a page of amos had been
 649         * previously allocated. If not we'll need to allocate one and set
 650         * permissions so that cross-partition amos are allowed.
 651         *
 652         * The allocated amo page needs MCA reporting to remain disabled after
 653         * XPC has unloaded.  To make this work, we keep a copy of the pointer
 654         * to this page (i.e., amos_page) in the struct xpc_vars_sn2 structure,
 655         * which is pointed to by the reserved page, and re-use that saved copy
 656         * on subsequent loads of XPC. This amo page is never freed, and its
 657         * memory protections are never restricted.
 658         */
 659        amos_page = xpc_vars_sn2->amos_page;
 660        if (amos_page == NULL) {
 661                amos_page = (struct amo *)TO_AMO(uncached_alloc_page(0, 1));
 662                if (amos_page == NULL) {
 663                        dev_err(xpc_part, "can't allocate page of amos\n");
 664                        return -ENOMEM;
 665                }
 666
 667                /*
 668                 * Open up amo-R/W to cpu.  This is done on Shub 1.1 systems
 669                 * when xpc_allow_amo_ops_shub_wars_1_1_sn2() is called.
 670                 */
 671                ret = xpc_allow_amo_ops_sn2(amos_page);
 672                if (ret != xpSuccess) {
 673                        dev_err(xpc_part, "can't allow amo operations\n");
 674                        uncached_free_page(__IA64_UNCACHED_OFFSET |
 675                                           TO_PHYS((u64)amos_page), 1);
 676                        return -EPERM;
 677                }
 678        }
 679
 680        /* clear xpc_vars_sn2 */
 681        memset(xpc_vars_sn2, 0, sizeof(struct xpc_vars_sn2));
 682
 683        xpc_vars_sn2->version = XPC_V_VERSION;
 684        xpc_vars_sn2->activate_IRQ_nasid = cpuid_to_nasid(0);
 685        xpc_vars_sn2->activate_IRQ_phys_cpuid = cpu_physical_id(0);
 686        xpc_vars_sn2->vars_part_pa = xp_pa(xpc_vars_part_sn2);
 687        xpc_vars_sn2->amos_page_pa = ia64_tpa((u64)amos_page);
 688        xpc_vars_sn2->amos_page = amos_page;    /* save for next load of XPC */
 689
 690        /* clear xpc_vars_part_sn2 */
 691        memset((u64 *)xpc_vars_part_sn2, 0, sizeof(struct xpc_vars_part_sn2) *
 692               XP_MAX_NPARTITIONS_SN2);
 693
 694        /* initialize the activate IRQ related amo variables */
 695        for (i = 0; i < xpc_nasid_mask_nlongs; i++)
 696                (void)xpc_init_IRQ_amo_sn2(XPC_ACTIVATE_IRQ_AMOS_SN2 + i);
 697
 698        /* initialize the engaged remote partitions related amo variables */
 699        (void)xpc_init_IRQ_amo_sn2(XPC_ENGAGED_PARTITIONS_AMO_SN2);
 700        (void)xpc_init_IRQ_amo_sn2(XPC_DEACTIVATE_REQUEST_AMO_SN2);
 701
 702        return 0;
 703}
 704
 705static int
 706xpc_hb_allowed_sn2(short partid, void *heartbeating_to_mask)
 707{
 708        return test_bit(partid, heartbeating_to_mask);
 709}
 710
 711static void
 712xpc_allow_hb_sn2(short partid)
 713{
 714        DBUG_ON(xpc_vars_sn2 == NULL);
 715        set_bit(partid, xpc_vars_sn2->heartbeating_to_mask);
 716}
 717
 718static void
 719xpc_disallow_hb_sn2(short partid)
 720{
 721        DBUG_ON(xpc_vars_sn2 == NULL);
 722        clear_bit(partid, xpc_vars_sn2->heartbeating_to_mask);
 723}
 724
 725static void
 726xpc_disallow_all_hbs_sn2(void)
 727{
 728        DBUG_ON(xpc_vars_sn2 == NULL);
 729        bitmap_zero(xpc_vars_sn2->heartbeating_to_mask, xp_max_npartitions);
 730}
 731
 732static void
 733xpc_increment_heartbeat_sn2(void)
 734{
 735        xpc_vars_sn2->heartbeat++;
 736}
 737
 738static void
 739xpc_offline_heartbeat_sn2(void)
 740{
 741        xpc_increment_heartbeat_sn2();
 742        xpc_vars_sn2->heartbeat_offline = 1;
 743}
 744
 745static void
 746xpc_online_heartbeat_sn2(void)
 747{
 748        xpc_increment_heartbeat_sn2();
 749        xpc_vars_sn2->heartbeat_offline = 0;
 750}
 751
 752static void
 753xpc_heartbeat_init_sn2(void)
 754{
 755        DBUG_ON(xpc_vars_sn2 == NULL);
 756
 757        bitmap_zero(xpc_vars_sn2->heartbeating_to_mask, XP_MAX_NPARTITIONS_SN2);
 758        xpc_online_heartbeat_sn2();
 759}
 760
 761static void
 762xpc_heartbeat_exit_sn2(void)
 763{
 764        xpc_offline_heartbeat_sn2();
 765}
 766
 767static enum xp_retval
 768xpc_get_remote_heartbeat_sn2(struct xpc_partition *part)
 769{
 770        struct xpc_vars_sn2 *remote_vars;
 771        enum xp_retval ret;
 772
 773        remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer_sn2;
 774
 775        /* pull the remote vars structure that contains the heartbeat */
 776        ret = xp_remote_memcpy(xp_pa(remote_vars),
 777                               part->sn.sn2.remote_vars_pa,
 778                               XPC_RP_VARS_SIZE);
 779        if (ret != xpSuccess)
 780                return ret;
 781
 782        dev_dbg(xpc_part, "partid=%d, heartbeat=%lld, last_heartbeat=%lld, "
 783                "heartbeat_offline=%lld, HB_mask[0]=0x%lx\n", XPC_PARTID(part),
 784                remote_vars->heartbeat, part->last_heartbeat,
 785                remote_vars->heartbeat_offline,
 786                remote_vars->heartbeating_to_mask[0]);
 787
 788        if ((remote_vars->heartbeat == part->last_heartbeat &&
 789            !remote_vars->heartbeat_offline) ||
 790            !xpc_hb_allowed_sn2(sn_partition_id,
 791                                remote_vars->heartbeating_to_mask)) {
 792                ret = xpNoHeartbeat;
 793        } else {
 794                part->last_heartbeat = remote_vars->heartbeat;
 795        }
 796
 797        return ret;
 798}
 799
 800/*
 801 * Get a copy of the remote partition's XPC variables from the reserved page.
 802 *
 803 * remote_vars points to a buffer that is cacheline aligned for BTE copies and
 804 * assumed to be of size XPC_RP_VARS_SIZE.
 805 */
 806static enum xp_retval
 807xpc_get_remote_vars_sn2(unsigned long remote_vars_pa,
 808                        struct xpc_vars_sn2 *remote_vars)
 809{
 810        enum xp_retval ret;
 811
 812        if (remote_vars_pa == 0)
 813                return xpVarsNotSet;
 814
 815        /* pull over the cross partition variables */
 816        ret = xp_remote_memcpy(xp_pa(remote_vars), remote_vars_pa,
 817                               XPC_RP_VARS_SIZE);
 818        if (ret != xpSuccess)
 819                return ret;
 820
 821        if (XPC_VERSION_MAJOR(remote_vars->version) !=
 822            XPC_VERSION_MAJOR(XPC_V_VERSION)) {
 823                return xpBadVersion;
 824        }
 825
 826        return xpSuccess;
 827}
 828
 829static void
 830xpc_request_partition_activation_sn2(struct xpc_rsvd_page *remote_rp,
 831                                     unsigned long remote_rp_pa, int nasid)
 832{
 833        xpc_send_local_activate_IRQ_sn2(nasid);
 834}
 835
 836static void
 837xpc_request_partition_reactivation_sn2(struct xpc_partition *part)
 838{
 839        xpc_send_local_activate_IRQ_sn2(part->sn.sn2.activate_IRQ_nasid);
 840}
 841
 842static void
 843xpc_request_partition_deactivation_sn2(struct xpc_partition *part)
 844{
 845        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
 846        unsigned long irq_flags;
 847        struct amo *amo = (struct amo *)__va(part_sn2->remote_amos_page_pa +
 848                                             (XPC_DEACTIVATE_REQUEST_AMO_SN2 *
 849                                             sizeof(struct amo)));
 850
 851        local_irq_save(irq_flags);
 852
 853        /* set bit corresponding to our partid in remote partition's amo */
 854        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
 855                         BIT(sn_partition_id));
 856
 857        /*
 858         * We must always use the nofault function regardless of whether we
 859         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
 860         * didn't, we'd never know that the other partition is down and would
 861         * keep sending IRQs and amos to it until the heartbeat times out.
 862         */
 863        (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
 864                                                               variable),
 865                                                     xp_nofault_PIOR_target));
 866
 867        local_irq_restore(irq_flags);
 868
 869        /*
 870         * Send activate IRQ to get other side to see that we've set our
 871         * bit in their deactivate request amo.
 872         */
 873        xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
 874                                  cnodeid_to_nasid(0),
 875                                  part_sn2->activate_IRQ_nasid,
 876                                  part_sn2->activate_IRQ_phys_cpuid);
 877}
 878
 879static void
 880xpc_cancel_partition_deactivation_request_sn2(struct xpc_partition *part)
 881{
 882        unsigned long irq_flags;
 883        struct amo *amo = (struct amo *)__va(part->sn.sn2.remote_amos_page_pa +
 884                                             (XPC_DEACTIVATE_REQUEST_AMO_SN2 *
 885                                             sizeof(struct amo)));
 886
 887        local_irq_save(irq_flags);
 888
 889        /* clear bit corresponding to our partid in remote partition's amo */
 890        FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
 891                         ~BIT(sn_partition_id));
 892
 893        /*
 894         * We must always use the nofault function regardless of whether we
 895         * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
 896         * didn't, we'd never know that the other partition is down and would
 897         * keep sending IRQs and amos to it until the heartbeat times out.
 898         */
 899        (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
 900                                                               variable),
 901                                                     xp_nofault_PIOR_target));
 902
 903        local_irq_restore(irq_flags);
 904}
 905
 906static int
 907xpc_partition_deactivation_requested_sn2(short partid)
 908{
 909        struct amo *amo = xpc_vars_sn2->amos_page +
 910                          XPC_DEACTIVATE_REQUEST_AMO_SN2;
 911
 912        /* our partition's amo variable ANDed with partid mask */
 913        return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
 914                BIT(partid)) != 0;
 915}
 916
 917/*
 918 * Update the remote partition's info.
 919 */
 920static void
 921xpc_update_partition_info_sn2(struct xpc_partition *part, u8 remote_rp_version,
 922                              unsigned long *remote_rp_ts_jiffies,
 923                              unsigned long remote_rp_pa,
 924                              unsigned long remote_vars_pa,
 925                              struct xpc_vars_sn2 *remote_vars)
 926{
 927        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
 928
 929        part->remote_rp_version = remote_rp_version;
 930        dev_dbg(xpc_part, "  remote_rp_version = 0x%016x\n",
 931                part->remote_rp_version);
 932
 933        part->remote_rp_ts_jiffies = *remote_rp_ts_jiffies;
 934        dev_dbg(xpc_part, "  remote_rp_ts_jiffies = 0x%016lx\n",
 935                part->remote_rp_ts_jiffies);
 936
 937        part->remote_rp_pa = remote_rp_pa;
 938        dev_dbg(xpc_part, "  remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
 939
 940        part_sn2->remote_vars_pa = remote_vars_pa;
 941        dev_dbg(xpc_part, "  remote_vars_pa = 0x%016lx\n",
 942                part_sn2->remote_vars_pa);
 943
 944        part->last_heartbeat = remote_vars->heartbeat - 1;
 945        dev_dbg(xpc_part, "  last_heartbeat = 0x%016llx\n",
 946                part->last_heartbeat);
 947
 948        part_sn2->remote_vars_part_pa = remote_vars->vars_part_pa;
 949        dev_dbg(xpc_part, "  remote_vars_part_pa = 0x%016lx\n",
 950                part_sn2->remote_vars_part_pa);
 951
 952        part_sn2->activate_IRQ_nasid = remote_vars->activate_IRQ_nasid;
 953        dev_dbg(xpc_part, "  activate_IRQ_nasid = 0x%x\n",
 954                part_sn2->activate_IRQ_nasid);
 955
 956        part_sn2->activate_IRQ_phys_cpuid =
 957            remote_vars->activate_IRQ_phys_cpuid;
 958        dev_dbg(xpc_part, "  activate_IRQ_phys_cpuid = 0x%x\n",
 959                part_sn2->activate_IRQ_phys_cpuid);
 960
 961        part_sn2->remote_amos_page_pa = remote_vars->amos_page_pa;
 962        dev_dbg(xpc_part, "  remote_amos_page_pa = 0x%lx\n",
 963                part_sn2->remote_amos_page_pa);
 964
 965        part_sn2->remote_vars_version = remote_vars->version;
 966        dev_dbg(xpc_part, "  remote_vars_version = 0x%x\n",
 967                part_sn2->remote_vars_version);
 968}
 969
 970/*
 971 * Prior code has determined the nasid which generated a activate IRQ.
 972 * Inspect that nasid to determine if its partition needs to be activated
 973 * or deactivated.
 974 *
 975 * A partition is considered "awaiting activation" if our partition
 976 * flags indicate it is not active and it has a heartbeat.  A
 977 * partition is considered "awaiting deactivation" if our partition
 978 * flags indicate it is active but it has no heartbeat or it is not
 979 * sending its heartbeat to us.
 980 *
 981 * To determine the heartbeat, the remote nasid must have a properly
 982 * initialized reserved page.
 983 */
 984static void
 985xpc_identify_activate_IRQ_req_sn2(int nasid)
 986{
 987        struct xpc_rsvd_page *remote_rp;
 988        struct xpc_vars_sn2 *remote_vars;
 989        unsigned long remote_rp_pa;
 990        unsigned long remote_vars_pa;
 991        int remote_rp_version;
 992        int reactivate = 0;
 993        unsigned long remote_rp_ts_jiffies = 0;
 994        short partid;
 995        struct xpc_partition *part;
 996        struct xpc_partition_sn2 *part_sn2;
 997        enum xp_retval ret;
 998
 999        /* pull over the reserved page structure */
1000
1001        remote_rp = (struct xpc_rsvd_page *)xpc_remote_copy_buffer_sn2;
1002
1003        ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
1004        if (ret != xpSuccess) {
1005                dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
1006                         "which sent interrupt, reason=%d\n", nasid, ret);
1007                return;
1008        }
1009
1010        remote_vars_pa = remote_rp->sn.sn2.vars_pa;
1011        remote_rp_version = remote_rp->version;
1012        remote_rp_ts_jiffies = remote_rp->ts_jiffies;
1013
1014        partid = remote_rp->SAL_partid;
1015        part = &xpc_partitions[partid];
1016        part_sn2 = &part->sn.sn2;
1017
1018        /* pull over the cross partition variables */
1019
1020        remote_vars = (struct xpc_vars_sn2 *)xpc_remote_copy_buffer_sn2;
1021
1022        ret = xpc_get_remote_vars_sn2(remote_vars_pa, remote_vars);
1023        if (ret != xpSuccess) {
1024                dev_warn(xpc_part, "unable to get XPC variables from nasid %d, "
1025                         "which sent interrupt, reason=%d\n", nasid, ret);
1026
1027                XPC_DEACTIVATE_PARTITION(part, ret);
1028                return;
1029        }
1030
1031        part->activate_IRQ_rcvd++;
1032
1033        dev_dbg(xpc_part, "partid for nasid %d is %d; IRQs = %d; HB = "
1034                "%lld:0x%lx\n", (int)nasid, (int)partid,
1035                part->activate_IRQ_rcvd,
1036                remote_vars->heartbeat, remote_vars->heartbeating_to_mask[0]);
1037
1038        if (xpc_partition_disengaged(part) &&
1039            part->act_state == XPC_P_AS_INACTIVE) {
1040
1041                xpc_update_partition_info_sn2(part, remote_rp_version,
1042                                              &remote_rp_ts_jiffies,
1043                                              remote_rp_pa, remote_vars_pa,
1044                                              remote_vars);
1045
1046                if (xpc_partition_deactivation_requested_sn2(partid)) {
1047                        /*
1048                         * Other side is waiting on us to deactivate even though
1049                         * we already have.
1050                         */
1051                        return;
1052                }
1053
1054                xpc_activate_partition(part);
1055                return;
1056        }
1057
1058        DBUG_ON(part->remote_rp_version == 0);
1059        DBUG_ON(part_sn2->remote_vars_version == 0);
1060
1061        if (remote_rp_ts_jiffies != part->remote_rp_ts_jiffies) {
1062
1063                /* the other side rebooted */
1064
1065                DBUG_ON(xpc_partition_engaged_sn2(partid));
1066                DBUG_ON(xpc_partition_deactivation_requested_sn2(partid));
1067
1068                xpc_update_partition_info_sn2(part, remote_rp_version,
1069                                              &remote_rp_ts_jiffies,
1070                                              remote_rp_pa, remote_vars_pa,
1071                                              remote_vars);
1072                reactivate = 1;
1073        }
1074
1075        if (part->disengage_timeout > 0 && !xpc_partition_disengaged(part)) {
1076                /* still waiting on other side to disengage from us */
1077                return;
1078        }
1079
1080        if (reactivate)
1081                XPC_DEACTIVATE_PARTITION(part, xpReactivating);
1082        else if (xpc_partition_deactivation_requested_sn2(partid))
1083                XPC_DEACTIVATE_PARTITION(part, xpOtherGoingDown);
1084}
1085
1086/*
1087 * Loop through the activation amo variables and process any bits
1088 * which are set.  Each bit indicates a nasid sending a partition
1089 * activation or deactivation request.
1090 *
1091 * Return #of IRQs detected.
1092 */
1093int
1094xpc_identify_activate_IRQ_sender_sn2(void)
1095{
1096        int l;
1097        int b;
1098        unsigned long nasid_mask_long;
1099        u64 nasid;              /* remote nasid */
1100        int n_IRQs_detected = 0;
1101        struct amo *act_amos;
1102
1103        act_amos = xpc_vars_sn2->amos_page + XPC_ACTIVATE_IRQ_AMOS_SN2;
1104
1105        /* scan through activate amo variables looking for non-zero entries */
1106        for (l = 0; l < xpc_nasid_mask_nlongs; l++) {
1107
1108                if (xpc_exiting)
1109                        break;
1110
1111                nasid_mask_long = xpc_receive_IRQ_amo_sn2(&act_amos[l]);
1112
1113                b = find_first_bit(&nasid_mask_long, BITS_PER_LONG);
1114                if (b >= BITS_PER_LONG) {
1115                        /* no IRQs from nasids in this amo variable */
1116                        continue;
1117                }
1118
1119                dev_dbg(xpc_part, "amo[%d] gave back 0x%lx\n", l,
1120                        nasid_mask_long);
1121
1122                /*
1123                 * If this nasid has been added to the machine since
1124                 * our partition was reset, this will retain the
1125                 * remote nasid in our reserved pages machine mask.
1126                 * This is used in the event of module reload.
1127                 */
1128                xpc_mach_nasids[l] |= nasid_mask_long;
1129
1130                /* locate the nasid(s) which sent interrupts */
1131
1132                do {
1133                        n_IRQs_detected++;
1134                        nasid = (l * BITS_PER_LONG + b) * 2;
1135                        dev_dbg(xpc_part, "interrupt from nasid %lld\n", nasid);
1136                        xpc_identify_activate_IRQ_req_sn2(nasid);
1137
1138                        b = find_next_bit(&nasid_mask_long, BITS_PER_LONG,
1139                                          b + 1);
1140                } while (b < BITS_PER_LONG);
1141        }
1142        return n_IRQs_detected;
1143}
1144
1145static void
1146xpc_process_activate_IRQ_rcvd_sn2(void)
1147{
1148        unsigned long irq_flags;
1149        int n_IRQs_expected;
1150        int n_IRQs_detected;
1151
1152        spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1153        n_IRQs_expected = xpc_activate_IRQ_rcvd;
1154        xpc_activate_IRQ_rcvd = 0;
1155        spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1156
1157        n_IRQs_detected = xpc_identify_activate_IRQ_sender_sn2();
1158        if (n_IRQs_detected < n_IRQs_expected) {
1159                /* retry once to help avoid missing amo */
1160                (void)xpc_identify_activate_IRQ_sender_sn2();
1161        }
1162}
1163
1164/*
1165 * Setup the channel structures that are sn2 specific.
1166 */
1167static enum xp_retval
1168xpc_setup_ch_structures_sn2(struct xpc_partition *part)
1169{
1170        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1171        struct xpc_channel_sn2 *ch_sn2;
1172        enum xp_retval retval;
1173        int ret;
1174        int cpuid;
1175        int ch_number;
1176        struct timer_list *timer;
1177        short partid = XPC_PARTID(part);
1178
1179        /* allocate all the required GET/PUT values */
1180
1181        part_sn2->local_GPs =
1182            xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE, GFP_KERNEL,
1183                                          &part_sn2->local_GPs_base);
1184        if (part_sn2->local_GPs == NULL) {
1185                dev_err(xpc_chan, "can't get memory for local get/put "
1186                        "values\n");
1187                return xpNoMemory;
1188        }
1189
1190        part_sn2->remote_GPs =
1191            xpc_kzalloc_cacheline_aligned(XPC_GP_SIZE, GFP_KERNEL,
1192                                          &part_sn2->remote_GPs_base);
1193        if (part_sn2->remote_GPs == NULL) {
1194                dev_err(xpc_chan, "can't get memory for remote get/put "
1195                        "values\n");
1196                retval = xpNoMemory;
1197                goto out_1;
1198        }
1199
1200        part_sn2->remote_GPs_pa = 0;
1201
1202        /* allocate all the required open and close args */
1203
1204        part_sn2->local_openclose_args =
1205            xpc_kzalloc_cacheline_aligned(XPC_OPENCLOSE_ARGS_SIZE,
1206                                          GFP_KERNEL, &part_sn2->
1207                                          local_openclose_args_base);
1208        if (part_sn2->local_openclose_args == NULL) {
1209                dev_err(xpc_chan, "can't get memory for local connect args\n");
1210                retval = xpNoMemory;
1211                goto out_2;
1212        }
1213
1214        part_sn2->remote_openclose_args_pa = 0;
1215
1216        part_sn2->local_chctl_amo_va = xpc_init_IRQ_amo_sn2(partid);
1217
1218        part_sn2->notify_IRQ_nasid = 0;
1219        part_sn2->notify_IRQ_phys_cpuid = 0;
1220        part_sn2->remote_chctl_amo_va = NULL;
1221
1222        sprintf(part_sn2->notify_IRQ_owner, "xpc%02d", partid);
1223        ret = request_irq(SGI_XPC_NOTIFY, xpc_handle_notify_IRQ_sn2,
1224                          IRQF_SHARED, part_sn2->notify_IRQ_owner,
1225                          (void *)(u64)partid);
1226        if (ret != 0) {
1227                dev_err(xpc_chan, "can't register NOTIFY IRQ handler, "
1228                        "errno=%d\n", -ret);
1229                retval = xpLackOfResources;
1230                goto out_3;
1231        }
1232
1233        /* Setup a timer to check for dropped notify IRQs */
1234        timer = &part_sn2->dropped_notify_IRQ_timer;
1235        timer_setup(timer, xpc_check_for_dropped_notify_IRQ_sn2, 0);
1236        timer->expires = jiffies + XPC_DROPPED_NOTIFY_IRQ_WAIT_INTERVAL;
1237        add_timer(timer);
1238
1239        for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1240                ch_sn2 = &part->channels[ch_number].sn.sn2;
1241
1242                ch_sn2->local_GP = &part_sn2->local_GPs[ch_number];
1243                ch_sn2->local_openclose_args =
1244                    &part_sn2->local_openclose_args[ch_number];
1245
1246                mutex_init(&ch_sn2->msg_to_pull_mutex);
1247        }
1248
1249        /*
1250         * Setup the per partition specific variables required by the
1251         * remote partition to establish channel connections with us.
1252         *
1253         * The setting of the magic # indicates that these per partition
1254         * specific variables are ready to be used.
1255         */
1256        xpc_vars_part_sn2[partid].GPs_pa = xp_pa(part_sn2->local_GPs);
1257        xpc_vars_part_sn2[partid].openclose_args_pa =
1258            xp_pa(part_sn2->local_openclose_args);
1259        xpc_vars_part_sn2[partid].chctl_amo_pa =
1260            xp_pa(part_sn2->local_chctl_amo_va);
1261        cpuid = raw_smp_processor_id(); /* any CPU in this partition will do */
1262        xpc_vars_part_sn2[partid].notify_IRQ_nasid = cpuid_to_nasid(cpuid);
1263        xpc_vars_part_sn2[partid].notify_IRQ_phys_cpuid =
1264            cpu_physical_id(cpuid);
1265        xpc_vars_part_sn2[partid].nchannels = part->nchannels;
1266        xpc_vars_part_sn2[partid].magic = XPC_VP_MAGIC1_SN2;
1267
1268        return xpSuccess;
1269
1270        /* setup of ch structures failed */
1271out_3:
1272        kfree(part_sn2->local_openclose_args_base);
1273        part_sn2->local_openclose_args = NULL;
1274out_2:
1275        kfree(part_sn2->remote_GPs_base);
1276        part_sn2->remote_GPs = NULL;
1277out_1:
1278        kfree(part_sn2->local_GPs_base);
1279        part_sn2->local_GPs = NULL;
1280        return retval;
1281}
1282
1283/*
1284 * Teardown the channel structures that are sn2 specific.
1285 */
1286static void
1287xpc_teardown_ch_structures_sn2(struct xpc_partition *part)
1288{
1289        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1290        short partid = XPC_PARTID(part);
1291
1292        /*
1293         * Indicate that the variables specific to the remote partition are no
1294         * longer available for its use.
1295         */
1296        xpc_vars_part_sn2[partid].magic = 0;
1297
1298        /* in case we've still got outstanding timers registered... */
1299        del_timer_sync(&part_sn2->dropped_notify_IRQ_timer);
1300        free_irq(SGI_XPC_NOTIFY, (void *)(u64)partid);
1301
1302        kfree(part_sn2->local_openclose_args_base);
1303        part_sn2->local_openclose_args = NULL;
1304        kfree(part_sn2->remote_GPs_base);
1305        part_sn2->remote_GPs = NULL;
1306        kfree(part_sn2->local_GPs_base);
1307        part_sn2->local_GPs = NULL;
1308        part_sn2->local_chctl_amo_va = NULL;
1309}
1310
1311/*
1312 * Create a wrapper that hides the underlying mechanism for pulling a cacheline
1313 * (or multiple cachelines) from a remote partition.
1314 *
1315 * src_pa must be a cacheline aligned physical address on the remote partition.
1316 * dst must be a cacheline aligned virtual address on this partition.
1317 * cnt must be cacheline sized
1318 */
1319/* ??? Replace this function by call to xp_remote_memcpy() or bte_copy()? */
1320static enum xp_retval
1321xpc_pull_remote_cachelines_sn2(struct xpc_partition *part, void *dst,
1322                               const unsigned long src_pa, size_t cnt)
1323{
1324        enum xp_retval ret;
1325
1326        DBUG_ON(src_pa != L1_CACHE_ALIGN(src_pa));
1327        DBUG_ON((unsigned long)dst != L1_CACHE_ALIGN((unsigned long)dst));
1328        DBUG_ON(cnt != L1_CACHE_ALIGN(cnt));
1329
1330        if (part->act_state == XPC_P_AS_DEACTIVATING)
1331                return part->reason;
1332
1333        ret = xp_remote_memcpy(xp_pa(dst), src_pa, cnt);
1334        if (ret != xpSuccess) {
1335                dev_dbg(xpc_chan, "xp_remote_memcpy() from partition %d failed,"
1336                        " ret=%d\n", XPC_PARTID(part), ret);
1337        }
1338        return ret;
1339}
1340
1341/*
1342 * Pull the remote per partition specific variables from the specified
1343 * partition.
1344 */
1345static enum xp_retval
1346xpc_pull_remote_vars_part_sn2(struct xpc_partition *part)
1347{
1348        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1349        u8 buffer[L1_CACHE_BYTES * 2];
1350        struct xpc_vars_part_sn2 *pulled_entry_cacheline =
1351            (struct xpc_vars_part_sn2 *)L1_CACHE_ALIGN((u64)buffer);
1352        struct xpc_vars_part_sn2 *pulled_entry;
1353        unsigned long remote_entry_cacheline_pa;
1354        unsigned long remote_entry_pa;
1355        short partid = XPC_PARTID(part);
1356        enum xp_retval ret;
1357
1358        /* pull the cacheline that contains the variables we're interested in */
1359
1360        DBUG_ON(part_sn2->remote_vars_part_pa !=
1361                L1_CACHE_ALIGN(part_sn2->remote_vars_part_pa));
1362        DBUG_ON(sizeof(struct xpc_vars_part_sn2) != L1_CACHE_BYTES / 2);
1363
1364        remote_entry_pa = part_sn2->remote_vars_part_pa +
1365            sn_partition_id * sizeof(struct xpc_vars_part_sn2);
1366
1367        remote_entry_cacheline_pa = (remote_entry_pa & ~(L1_CACHE_BYTES - 1));
1368
1369        pulled_entry = (struct xpc_vars_part_sn2 *)((u64)pulled_entry_cacheline
1370                                                    + (remote_entry_pa &
1371                                                    (L1_CACHE_BYTES - 1)));
1372
1373        ret = xpc_pull_remote_cachelines_sn2(part, pulled_entry_cacheline,
1374                                             remote_entry_cacheline_pa,
1375                                             L1_CACHE_BYTES);
1376        if (ret != xpSuccess) {
1377                dev_dbg(xpc_chan, "failed to pull XPC vars_part from "
1378                        "partition %d, ret=%d\n", partid, ret);
1379                return ret;
1380        }
1381
1382        /* see if they've been set up yet */
1383
1384        if (pulled_entry->magic != XPC_VP_MAGIC1_SN2 &&
1385            pulled_entry->magic != XPC_VP_MAGIC2_SN2) {
1386
1387                if (pulled_entry->magic != 0) {
1388                        dev_dbg(xpc_chan, "partition %d's XPC vars_part for "
1389                                "partition %d has bad magic value (=0x%llx)\n",
1390                                partid, sn_partition_id, pulled_entry->magic);
1391                        return xpBadMagic;
1392                }
1393
1394                /* they've not been initialized yet */
1395                return xpRetry;
1396        }
1397
1398        if (xpc_vars_part_sn2[partid].magic == XPC_VP_MAGIC1_SN2) {
1399
1400                /* validate the variables */
1401
1402                if (pulled_entry->GPs_pa == 0 ||
1403                    pulled_entry->openclose_args_pa == 0 ||
1404                    pulled_entry->chctl_amo_pa == 0) {
1405
1406                        dev_err(xpc_chan, "partition %d's XPC vars_part for "
1407                                "partition %d are not valid\n", partid,
1408                                sn_partition_id);
1409                        return xpInvalidAddress;
1410                }
1411
1412                /* the variables we imported look to be valid */
1413
1414                part_sn2->remote_GPs_pa = pulled_entry->GPs_pa;
1415                part_sn2->remote_openclose_args_pa =
1416                    pulled_entry->openclose_args_pa;
1417                part_sn2->remote_chctl_amo_va =
1418                    (struct amo *)__va(pulled_entry->chctl_amo_pa);
1419                part_sn2->notify_IRQ_nasid = pulled_entry->notify_IRQ_nasid;
1420                part_sn2->notify_IRQ_phys_cpuid =
1421                    pulled_entry->notify_IRQ_phys_cpuid;
1422
1423                if (part->nchannels > pulled_entry->nchannels)
1424                        part->nchannels = pulled_entry->nchannels;
1425
1426                /* let the other side know that we've pulled their variables */
1427
1428                xpc_vars_part_sn2[partid].magic = XPC_VP_MAGIC2_SN2;
1429        }
1430
1431        if (pulled_entry->magic == XPC_VP_MAGIC1_SN2)
1432                return xpRetry;
1433
1434        return xpSuccess;
1435}
1436
1437/*
1438 * Establish first contact with the remote partititon. This involves pulling
1439 * the XPC per partition variables from the remote partition and waiting for
1440 * the remote partition to pull ours.
1441 */
1442static enum xp_retval
1443xpc_make_first_contact_sn2(struct xpc_partition *part)
1444{
1445        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1446        enum xp_retval ret;
1447
1448        /*
1449         * Register the remote partition's amos with SAL so it can handle
1450         * and cleanup errors within that address range should the remote
1451         * partition go down. We don't unregister this range because it is
1452         * difficult to tell when outstanding writes to the remote partition
1453         * are finished and thus when it is safe to unregister. This should
1454         * not result in wasted space in the SAL xp_addr_region table because
1455         * we should get the same page for remote_amos_page_pa after module
1456         * reloads and system reboots.
1457         */
1458        if (sn_register_xp_addr_region(part_sn2->remote_amos_page_pa,
1459                                       PAGE_SIZE, 1) < 0) {
1460                dev_warn(xpc_part, "xpc_activating(%d) failed to register "
1461                         "xp_addr region\n", XPC_PARTID(part));
1462
1463                ret = xpPhysAddrRegFailed;
1464                XPC_DEACTIVATE_PARTITION(part, ret);
1465                return ret;
1466        }
1467
1468        /*
1469         * Send activate IRQ to get other side to activate if they've not
1470         * already begun to do so.
1471         */
1472        xpc_send_activate_IRQ_sn2(part_sn2->remote_amos_page_pa,
1473                                  cnodeid_to_nasid(0),
1474                                  part_sn2->activate_IRQ_nasid,
1475                                  part_sn2->activate_IRQ_phys_cpuid);
1476
1477        while ((ret = xpc_pull_remote_vars_part_sn2(part)) != xpSuccess) {
1478                if (ret != xpRetry) {
1479                        XPC_DEACTIVATE_PARTITION(part, ret);
1480                        return ret;
1481                }
1482
1483                dev_dbg(xpc_part, "waiting to make first contact with "
1484                        "partition %d\n", XPC_PARTID(part));
1485
1486                /* wait a 1/4 of a second or so */
1487                (void)msleep_interruptible(250);
1488
1489                if (part->act_state == XPC_P_AS_DEACTIVATING)
1490                        return part->reason;
1491        }
1492
1493        return xpSuccess;
1494}
1495
1496/*
1497 * Get the chctl flags and pull the openclose args and/or remote GPs as needed.
1498 */
1499static u64
1500xpc_get_chctl_all_flags_sn2(struct xpc_partition *part)
1501{
1502        struct xpc_partition_sn2 *part_sn2 = &part->sn.sn2;
1503        unsigned long irq_flags;
1504        union xpc_channel_ctl_flags chctl;
1505        enum xp_retval ret;
1506
1507        /*
1508         * See if there are any chctl flags to be handled.
1509         */
1510
1511        spin_lock_irqsave(&part->chctl_lock, irq_flags);
1512        chctl = part->chctl;
1513        if (chctl.all_flags != 0)
1514                part->chctl.all_flags = 0;
1515
1516        spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1517
1518        if (xpc_any_openclose_chctl_flags_set(&chctl)) {
1519                ret = xpc_pull_remote_cachelines_sn2(part, part->
1520                                                     remote_openclose_args,
1521                                                     part_sn2->
1522                                                     remote_openclose_args_pa,
1523                                                     XPC_OPENCLOSE_ARGS_SIZE);
1524                if (ret != xpSuccess) {
1525                        XPC_DEACTIVATE_PARTITION(part, ret);
1526
1527                        dev_dbg(xpc_chan, "failed to pull openclose args from "
1528                                "partition %d, ret=%d\n", XPC_PARTID(part),
1529                                ret);
1530
1531                        /* don't bother processing chctl flags anymore */
1532                        chctl.all_flags = 0;
1533                }
1534        }
1535
1536        if (xpc_any_msg_chctl_flags_set(&chctl)) {
1537                ret = xpc_pull_remote_cachelines_sn2(part, part_sn2->remote_GPs,
1538                                                     part_sn2->remote_GPs_pa,
1539                                                     XPC_GP_SIZE);
1540                if (ret != xpSuccess) {
1541                        XPC_DEACTIVATE_PARTITION(part, ret);
1542
1543                        dev_dbg(xpc_chan, "failed to pull GPs from partition "
1544                                "%d, ret=%d\n", XPC_PARTID(part), ret);
1545
1546                        /* don't bother processing chctl flags anymore */
1547                        chctl.all_flags = 0;
1548                }
1549        }
1550
1551        return chctl.all_flags;
1552}
1553
1554/*
1555 * Allocate the local message queue and the notify queue.
1556 */
1557static enum xp_retval
1558xpc_allocate_local_msgqueue_sn2(struct xpc_channel *ch)
1559{
1560        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1561        unsigned long irq_flags;
1562        int nentries;
1563        size_t nbytes;
1564
1565        for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1566
1567                nbytes = nentries * ch->entry_size;
1568                ch_sn2->local_msgqueue =
1569                    xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL,
1570                                                  &ch_sn2->local_msgqueue_base);
1571                if (ch_sn2->local_msgqueue == NULL)
1572                        continue;
1573
1574                nbytes = nentries * sizeof(struct xpc_notify_sn2);
1575                ch_sn2->notify_queue = kzalloc(nbytes, GFP_KERNEL);
1576                if (ch_sn2->notify_queue == NULL) {
1577                        kfree(ch_sn2->local_msgqueue_base);
1578                        ch_sn2->local_msgqueue = NULL;
1579                        continue;
1580                }
1581
1582                spin_lock_irqsave(&ch->lock, irq_flags);
1583                if (nentries < ch->local_nentries) {
1584                        dev_dbg(xpc_chan, "nentries=%d local_nentries=%d, "
1585                                "partid=%d, channel=%d\n", nentries,
1586                                ch->local_nentries, ch->partid, ch->number);
1587
1588                        ch->local_nentries = nentries;
1589                }
1590                spin_unlock_irqrestore(&ch->lock, irq_flags);
1591                return xpSuccess;
1592        }
1593
1594        dev_dbg(xpc_chan, "can't get memory for local message queue and notify "
1595                "queue, partid=%d, channel=%d\n", ch->partid, ch->number);
1596        return xpNoMemory;
1597}
1598
1599/*
1600 * Allocate the cached remote message queue.
1601 */
1602static enum xp_retval
1603xpc_allocate_remote_msgqueue_sn2(struct xpc_channel *ch)
1604{
1605        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1606        unsigned long irq_flags;
1607        int nentries;
1608        size_t nbytes;
1609
1610        DBUG_ON(ch->remote_nentries <= 0);
1611
1612        for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1613
1614                nbytes = nentries * ch->entry_size;
1615                ch_sn2->remote_msgqueue =
1616                    xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL, &ch_sn2->
1617                                                  remote_msgqueue_base);
1618                if (ch_sn2->remote_msgqueue == NULL)
1619                        continue;
1620
1621                spin_lock_irqsave(&ch->lock, irq_flags);
1622                if (nentries < ch->remote_nentries) {
1623                        dev_dbg(xpc_chan, "nentries=%d remote_nentries=%d, "
1624                                "partid=%d, channel=%d\n", nentries,
1625                                ch->remote_nentries, ch->partid, ch->number);
1626
1627                        ch->remote_nentries = nentries;
1628                }
1629                spin_unlock_irqrestore(&ch->lock, irq_flags);
1630                return xpSuccess;
1631        }
1632
1633        dev_dbg(xpc_chan, "can't get memory for cached remote message queue, "
1634                "partid=%d, channel=%d\n", ch->partid, ch->number);
1635        return xpNoMemory;
1636}
1637
1638/*
1639 * Allocate message queues and other stuff associated with a channel.
1640 *
1641 * Note: Assumes all of the channel sizes are filled in.
1642 */
1643static enum xp_retval
1644xpc_setup_msg_structures_sn2(struct xpc_channel *ch)
1645{
1646        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1647        enum xp_retval ret;
1648
1649        DBUG_ON(ch->flags & XPC_C_SETUP);
1650
1651        ret = xpc_allocate_local_msgqueue_sn2(ch);
1652        if (ret == xpSuccess) {
1653
1654                ret = xpc_allocate_remote_msgqueue_sn2(ch);
1655                if (ret != xpSuccess) {
1656                        kfree(ch_sn2->local_msgqueue_base);
1657                        ch_sn2->local_msgqueue = NULL;
1658                        kfree(ch_sn2->notify_queue);
1659                        ch_sn2->notify_queue = NULL;
1660                }
1661        }
1662        return ret;
1663}
1664
1665/*
1666 * Free up message queues and other stuff that were allocated for the specified
1667 * channel.
1668 */
1669static void
1670xpc_teardown_msg_structures_sn2(struct xpc_channel *ch)
1671{
1672        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1673
1674        lockdep_assert_held(&ch->lock);
1675
1676        ch_sn2->remote_msgqueue_pa = 0;
1677
1678        ch_sn2->local_GP->get = 0;
1679        ch_sn2->local_GP->put = 0;
1680        ch_sn2->remote_GP.get = 0;
1681        ch_sn2->remote_GP.put = 0;
1682        ch_sn2->w_local_GP.get = 0;
1683        ch_sn2->w_local_GP.put = 0;
1684        ch_sn2->w_remote_GP.get = 0;
1685        ch_sn2->w_remote_GP.put = 0;
1686        ch_sn2->next_msg_to_pull = 0;
1687
1688        if (ch->flags & XPC_C_SETUP) {
1689                dev_dbg(xpc_chan, "ch->flags=0x%x, partid=%d, channel=%d\n",
1690                        ch->flags, ch->partid, ch->number);
1691
1692                kfree(ch_sn2->local_msgqueue_base);
1693                ch_sn2->local_msgqueue = NULL;
1694                kfree(ch_sn2->remote_msgqueue_base);
1695                ch_sn2->remote_msgqueue = NULL;
1696                kfree(ch_sn2->notify_queue);
1697                ch_sn2->notify_queue = NULL;
1698        }
1699}
1700
1701/*
1702 * Notify those who wanted to be notified upon delivery of their message.
1703 */
1704static void
1705xpc_notify_senders_sn2(struct xpc_channel *ch, enum xp_retval reason, s64 put)
1706{
1707        struct xpc_notify_sn2 *notify;
1708        u8 notify_type;
1709        s64 get = ch->sn.sn2.w_remote_GP.get - 1;
1710
1711        while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
1712
1713                notify = &ch->sn.sn2.notify_queue[get % ch->local_nentries];
1714
1715                /*
1716                 * See if the notify entry indicates it was associated with
1717                 * a message who's sender wants to be notified. It is possible
1718                 * that it is, but someone else is doing or has done the
1719                 * notification.
1720                 */
1721                notify_type = notify->type;
1722                if (notify_type == 0 ||
1723                    cmpxchg(&notify->type, notify_type, 0) != notify_type) {
1724                        continue;
1725                }
1726
1727                DBUG_ON(notify_type != XPC_N_CALL);
1728
1729                atomic_dec(&ch->n_to_notify);
1730
1731                if (notify->func != NULL) {
1732                        dev_dbg(xpc_chan, "notify->func() called, notify=0x%p "
1733                                "msg_number=%lld partid=%d channel=%d\n",
1734                                (void *)notify, get, ch->partid, ch->number);
1735
1736                        notify->func(reason, ch->partid, ch->number,
1737                                     notify->key);
1738
1739                        dev_dbg(xpc_chan, "notify->func() returned, notify=0x%p"
1740                                " msg_number=%lld partid=%d channel=%d\n",
1741                                (void *)notify, get, ch->partid, ch->number);
1742                }
1743        }
1744}
1745
1746static void
1747xpc_notify_senders_of_disconnect_sn2(struct xpc_channel *ch)
1748{
1749        xpc_notify_senders_sn2(ch, ch->reason, ch->sn.sn2.w_local_GP.put);
1750}
1751
1752/*
1753 * Clear some of the msg flags in the local message queue.
1754 */
1755static inline void
1756xpc_clear_local_msgqueue_flags_sn2(struct xpc_channel *ch)
1757{
1758        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1759        struct xpc_msg_sn2 *msg;
1760        s64 get;
1761
1762        get = ch_sn2->w_remote_GP.get;
1763        do {
1764                msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
1765                                             (get % ch->local_nentries) *
1766                                             ch->entry_size);
1767                DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
1768                msg->flags = 0;
1769        } while (++get < ch_sn2->remote_GP.get);
1770}
1771
1772/*
1773 * Clear some of the msg flags in the remote message queue.
1774 */
1775static inline void
1776xpc_clear_remote_msgqueue_flags_sn2(struct xpc_channel *ch)
1777{
1778        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1779        struct xpc_msg_sn2 *msg;
1780        s64 put, remote_nentries = ch->remote_nentries;
1781
1782        /* flags are zeroed when the buffer is allocated */
1783        if (ch_sn2->remote_GP.put < remote_nentries)
1784                return;
1785
1786        put = max(ch_sn2->w_remote_GP.put, remote_nentries);
1787        do {
1788                msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
1789                                             (put % remote_nentries) *
1790                                             ch->entry_size);
1791                DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
1792                DBUG_ON(!(msg->flags & XPC_M_SN2_DONE));
1793                DBUG_ON(msg->number != put - remote_nentries);
1794                msg->flags = 0;
1795        } while (++put < ch_sn2->remote_GP.put);
1796}
1797
1798static int
1799xpc_n_of_deliverable_payloads_sn2(struct xpc_channel *ch)
1800{
1801        return ch->sn.sn2.w_remote_GP.put - ch->sn.sn2.w_local_GP.get;
1802}
1803
1804static void
1805xpc_process_msg_chctl_flags_sn2(struct xpc_partition *part, int ch_number)
1806{
1807        struct xpc_channel *ch = &part->channels[ch_number];
1808        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1809        int npayloads_sent;
1810
1811        ch_sn2->remote_GP = part->sn.sn2.remote_GPs[ch_number];
1812
1813        /* See what, if anything, has changed for each connected channel */
1814
1815        xpc_msgqueue_ref(ch);
1816
1817        if (ch_sn2->w_remote_GP.get == ch_sn2->remote_GP.get &&
1818            ch_sn2->w_remote_GP.put == ch_sn2->remote_GP.put) {
1819                /* nothing changed since GPs were last pulled */
1820                xpc_msgqueue_deref(ch);
1821                return;
1822        }
1823
1824        if (!(ch->flags & XPC_C_CONNECTED)) {
1825                xpc_msgqueue_deref(ch);
1826                return;
1827        }
1828
1829        /*
1830         * First check to see if messages recently sent by us have been
1831         * received by the other side. (The remote GET value will have
1832         * changed since we last looked at it.)
1833         */
1834
1835        if (ch_sn2->w_remote_GP.get != ch_sn2->remote_GP.get) {
1836
1837                /*
1838                 * We need to notify any senders that want to be notified
1839                 * that their sent messages have been received by their
1840                 * intended recipients. We need to do this before updating
1841                 * w_remote_GP.get so that we don't allocate the same message
1842                 * queue entries prematurely (see xpc_allocate_msg()).
1843                 */
1844                if (atomic_read(&ch->n_to_notify) > 0) {
1845                        /*
1846                         * Notify senders that messages sent have been
1847                         * received and delivered by the other side.
1848                         */
1849                        xpc_notify_senders_sn2(ch, xpMsgDelivered,
1850                                               ch_sn2->remote_GP.get);
1851                }
1852
1853                /*
1854                 * Clear msg->flags in previously sent messages, so that
1855                 * they're ready for xpc_allocate_msg().
1856                 */
1857                xpc_clear_local_msgqueue_flags_sn2(ch);
1858
1859                ch_sn2->w_remote_GP.get = ch_sn2->remote_GP.get;
1860
1861                dev_dbg(xpc_chan, "w_remote_GP.get changed to %lld, partid=%d, "
1862                        "channel=%d\n", ch_sn2->w_remote_GP.get, ch->partid,
1863                        ch->number);
1864
1865                /*
1866                 * If anyone was waiting for message queue entries to become
1867                 * available, wake them up.
1868                 */
1869                if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1870                        wake_up(&ch->msg_allocate_wq);
1871        }
1872
1873        /*
1874         * Now check for newly sent messages by the other side. (The remote
1875         * PUT value will have changed since we last looked at it.)
1876         */
1877
1878        if (ch_sn2->w_remote_GP.put != ch_sn2->remote_GP.put) {
1879                /*
1880                 * Clear msg->flags in previously received messages, so that
1881                 * they're ready for xpc_get_deliverable_payload_sn2().
1882                 */
1883                xpc_clear_remote_msgqueue_flags_sn2(ch);
1884
1885                smp_wmb(); /* ensure flags have been cleared before bte_copy */
1886                ch_sn2->w_remote_GP.put = ch_sn2->remote_GP.put;
1887
1888                dev_dbg(xpc_chan, "w_remote_GP.put changed to %lld, partid=%d, "
1889                        "channel=%d\n", ch_sn2->w_remote_GP.put, ch->partid,
1890                        ch->number);
1891
1892                npayloads_sent = xpc_n_of_deliverable_payloads_sn2(ch);
1893                if (npayloads_sent > 0) {
1894                        dev_dbg(xpc_chan, "msgs waiting to be copied and "
1895                                "delivered=%d, partid=%d, channel=%d\n",
1896                                npayloads_sent, ch->partid, ch->number);
1897
1898                        if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)
1899                                xpc_activate_kthreads(ch, npayloads_sent);
1900                }
1901        }
1902
1903        xpc_msgqueue_deref(ch);
1904}
1905
1906static struct xpc_msg_sn2 *
1907xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
1908{
1909        struct xpc_partition *part = &xpc_partitions[ch->partid];
1910        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1911        unsigned long remote_msg_pa;
1912        struct xpc_msg_sn2 *msg;
1913        u32 msg_index;
1914        u32 nmsgs;
1915        u64 msg_offset;
1916        enum xp_retval ret;
1917
1918        if (mutex_lock_interruptible(&ch_sn2->msg_to_pull_mutex) != 0) {
1919                /* we were interrupted by a signal */
1920                return NULL;
1921        }
1922
1923        while (get >= ch_sn2->next_msg_to_pull) {
1924
1925                /* pull as many messages as are ready and able to be pulled */
1926
1927                msg_index = ch_sn2->next_msg_to_pull % ch->remote_nentries;
1928
1929                DBUG_ON(ch_sn2->next_msg_to_pull >= ch_sn2->w_remote_GP.put);
1930                nmsgs = ch_sn2->w_remote_GP.put - ch_sn2->next_msg_to_pull;
1931                if (msg_index + nmsgs > ch->remote_nentries) {
1932                        /* ignore the ones that wrap the msg queue for now */
1933                        nmsgs = ch->remote_nentries - msg_index;
1934                }
1935
1936                msg_offset = msg_index * ch->entry_size;
1937                msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
1938                    msg_offset);
1939                remote_msg_pa = ch_sn2->remote_msgqueue_pa + msg_offset;
1940
1941                ret = xpc_pull_remote_cachelines_sn2(part, msg, remote_msg_pa,
1942                                                     nmsgs * ch->entry_size);
1943                if (ret != xpSuccess) {
1944
1945                        dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
1946                                " msg %lld from partition %d, channel=%d, "
1947                                "ret=%d\n", nmsgs, ch_sn2->next_msg_to_pull,
1948                                ch->partid, ch->number, ret);
1949
1950                        XPC_DEACTIVATE_PARTITION(part, ret);
1951
1952                        mutex_unlock(&ch_sn2->msg_to_pull_mutex);
1953                        return NULL;
1954                }
1955
1956                ch_sn2->next_msg_to_pull += nmsgs;
1957        }
1958
1959        mutex_unlock(&ch_sn2->msg_to_pull_mutex);
1960
1961        /* return the message we were looking for */
1962        msg_offset = (get % ch->remote_nentries) * ch->entry_size;
1963        msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue + msg_offset);
1964
1965        return msg;
1966}
1967
1968/*
1969 * Get the next deliverable message's payload.
1970 */
1971static void *
1972xpc_get_deliverable_payload_sn2(struct xpc_channel *ch)
1973{
1974        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
1975        struct xpc_msg_sn2 *msg;
1976        void *payload = NULL;
1977        s64 get;
1978
1979        do {
1980                if (ch->flags & XPC_C_DISCONNECTING)
1981                        break;
1982
1983                get = ch_sn2->w_local_GP.get;
1984                smp_rmb();      /* guarantee that .get loads before .put */
1985                if (get == ch_sn2->w_remote_GP.put)
1986                        break;
1987
1988                /* There are messages waiting to be pulled and delivered.
1989                 * We need to try to secure one for ourselves. We'll do this
1990                 * by trying to increment w_local_GP.get and hope that no one
1991                 * else beats us to it. If they do, we'll we'll simply have
1992                 * to try again for the next one.
1993                 */
1994
1995                if (cmpxchg(&ch_sn2->w_local_GP.get, get, get + 1) == get) {
1996                        /* we got the entry referenced by get */
1997
1998                        dev_dbg(xpc_chan, "w_local_GP.get changed to %lld, "
1999                                "partid=%d, channel=%d\n", get + 1,
2000                                ch->partid, ch->number);
2001
2002                        /* pull the message from the remote partition */
2003
2004                        msg = xpc_pull_remote_msg_sn2(ch, get);
2005
2006                        if (msg != NULL) {
2007                                DBUG_ON(msg->number != get);
2008                                DBUG_ON(msg->flags & XPC_M_SN2_DONE);
2009                                DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
2010
2011                                payload = &msg->payload;
2012                        }
2013                        break;
2014                }
2015
2016        } while (1);
2017
2018        return payload;
2019}
2020
2021/*
2022 * Now we actually send the messages that are ready to be sent by advancing
2023 * the local message queue's Put value and then send a chctl msgrequest to the
2024 * recipient partition.
2025 */
2026static void
2027xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
2028{
2029        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2030        struct xpc_msg_sn2 *msg;
2031        s64 put = initial_put + 1;
2032        int send_msgrequest = 0;
2033
2034        while (1) {
2035
2036                while (1) {
2037                        if (put == ch_sn2->w_local_GP.put)
2038                                break;
2039
2040                        msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
2041                                                     local_msgqueue + (put %
2042                                                     ch->local_nentries) *
2043                                                     ch->entry_size);
2044
2045                        if (!(msg->flags & XPC_M_SN2_READY))
2046                                break;
2047
2048                        put++;
2049                }
2050
2051                if (put == initial_put) {
2052                        /* nothing's changed */
2053                        break;
2054                }
2055
2056                if (cmpxchg_rel(&ch_sn2->local_GP->put, initial_put, put) !=
2057                    initial_put) {
2058                        /* someone else beat us to it */
2059                        DBUG_ON(ch_sn2->local_GP->put < initial_put);
2060                        break;
2061                }
2062
2063                /* we just set the new value of local_GP->put */
2064
2065                dev_dbg(xpc_chan, "local_GP->put changed to %lld, partid=%d, "
2066                        "channel=%d\n", put, ch->partid, ch->number);
2067
2068                send_msgrequest = 1;
2069
2070                /*
2071                 * We need to ensure that the message referenced by
2072                 * local_GP->put is not XPC_M_SN2_READY or that local_GP->put
2073                 * equals w_local_GP.put, so we'll go have a look.
2074                 */
2075                initial_put = put;
2076        }
2077
2078        if (send_msgrequest)
2079                xpc_send_chctl_msgrequest_sn2(ch);
2080}
2081
2082/*
2083 * Allocate an entry for a message from the message queue associated with the
2084 * specified channel.
2085 */
2086static enum xp_retval
2087xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
2088                     struct xpc_msg_sn2 **address_of_msg)
2089{
2090        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2091        struct xpc_msg_sn2 *msg;
2092        enum xp_retval ret;
2093        s64 put;
2094
2095        /*
2096         * Get the next available message entry from the local message queue.
2097         * If none are available, we'll make sure that we grab the latest
2098         * GP values.
2099         */
2100        ret = xpTimeout;
2101
2102        while (1) {
2103
2104                put = ch_sn2->w_local_GP.put;
2105                smp_rmb();      /* guarantee that .put loads before .get */
2106                if (put - ch_sn2->w_remote_GP.get < ch->local_nentries) {
2107
2108                        /* There are available message entries. We need to try
2109                         * to secure one for ourselves. We'll do this by trying
2110                         * to increment w_local_GP.put as long as someone else
2111                         * doesn't beat us to it. If they do, we'll have to
2112                         * try again.
2113                         */
2114                        if (cmpxchg(&ch_sn2->w_local_GP.put, put, put + 1) ==
2115                            put) {
2116                                /* we got the entry referenced by put */
2117                                break;
2118                        }
2119                        continue;       /* try again */
2120                }
2121
2122                /*
2123                 * There aren't any available msg entries at this time.
2124                 *
2125                 * In waiting for a message entry to become available,
2126                 * we set a timeout in case the other side is not sending
2127                 * completion interrupts. This lets us fake a notify IRQ
2128                 * that will cause the notify IRQ handler to fetch the latest
2129                 * GP values as if an interrupt was sent by the other side.
2130                 */
2131                if (ret == xpTimeout)
2132                        xpc_send_chctl_local_msgrequest_sn2(ch);
2133
2134                if (flags & XPC_NOWAIT)
2135                        return xpNoWait;
2136
2137                ret = xpc_allocate_msg_wait(ch);
2138                if (ret != xpInterrupted && ret != xpTimeout)
2139                        return ret;
2140        }
2141
2142        /* get the message's address and initialize it */
2143        msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
2144                                     (put % ch->local_nentries) *
2145                                     ch->entry_size);
2146
2147        DBUG_ON(msg->flags != 0);
2148        msg->number = put;
2149
2150        dev_dbg(xpc_chan, "w_local_GP.put changed to %lld; msg=0x%p, "
2151                "msg_number=%lld, partid=%d, channel=%d\n", put + 1,
2152                (void *)msg, msg->number, ch->partid, ch->number);
2153
2154        *address_of_msg = msg;
2155        return xpSuccess;
2156}
2157
2158/*
2159 * Common code that does the actual sending of the message by advancing the
2160 * local message queue's Put value and sends a chctl msgrequest to the
2161 * partition the message is being sent to.
2162 */
2163static enum xp_retval
2164xpc_send_payload_sn2(struct xpc_channel *ch, u32 flags, void *payload,
2165                     u16 payload_size, u8 notify_type, xpc_notify_func func,
2166                     void *key)
2167{
2168        enum xp_retval ret = xpSuccess;
2169        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2170        struct xpc_msg_sn2 *msg = msg;
2171        struct xpc_notify_sn2 *notify = notify;
2172        s64 msg_number;
2173        s64 put;
2174
2175        DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
2176
2177        if (XPC_MSG_SIZE(payload_size) > ch->entry_size)
2178                return xpPayloadTooBig;
2179
2180        xpc_msgqueue_ref(ch);
2181
2182        if (ch->flags & XPC_C_DISCONNECTING) {
2183                ret = ch->reason;
2184                goto out_1;
2185        }
2186        if (!(ch->flags & XPC_C_CONNECTED)) {
2187                ret = xpNotConnected;
2188                goto out_1;
2189        }
2190
2191        ret = xpc_allocate_msg_sn2(ch, flags, &msg);
2192        if (ret != xpSuccess)
2193                goto out_1;
2194
2195        msg_number = msg->number;
2196
2197        if (notify_type != 0) {
2198                /*
2199                 * Tell the remote side to send an ACK interrupt when the
2200                 * message has been delivered.
2201                 */
2202                msg->flags |= XPC_M_SN2_INTERRUPT;
2203
2204                atomic_inc(&ch->n_to_notify);
2205
2206                notify = &ch_sn2->notify_queue[msg_number % ch->local_nentries];
2207                notify->func = func;
2208                notify->key = key;
2209                notify->type = notify_type;
2210
2211                /* ??? Is a mb() needed here? */
2212
2213                if (ch->flags & XPC_C_DISCONNECTING) {
2214                        /*
2215                         * An error occurred between our last error check and
2216                         * this one. We will try to clear the type field from
2217                         * the notify entry. If we succeed then
2218                         * xpc_disconnect_channel() didn't already process
2219                         * the notify entry.
2220                         */
2221                        if (cmpxchg(&notify->type, notify_type, 0) ==
2222                            notify_type) {
2223                                atomic_dec(&ch->n_to_notify);
2224                                ret = ch->reason;
2225                        }
2226                        goto out_1;
2227                }
2228        }
2229
2230        memcpy(&msg->payload, payload, payload_size);
2231
2232        msg->flags |= XPC_M_SN2_READY;
2233
2234        /*
2235         * The preceding store of msg->flags must occur before the following
2236         * load of local_GP->put.
2237         */
2238        smp_mb();
2239
2240        /* see if the message is next in line to be sent, if so send it */
2241
2242        put = ch_sn2->local_GP->put;
2243        if (put == msg_number)
2244                xpc_send_msgs_sn2(ch, put);
2245
2246out_1:
2247        xpc_msgqueue_deref(ch);
2248        return ret;
2249}
2250
2251/*
2252 * Now we actually acknowledge the messages that have been delivered and ack'd
2253 * by advancing the cached remote message queue's Get value and if requested
2254 * send a chctl msgrequest to the message sender's partition.
2255 *
2256 * If a message has XPC_M_SN2_INTERRUPT set, send an interrupt to the partition
2257 * that sent the message.
2258 */
2259static void
2260xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
2261{
2262        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
2263        struct xpc_msg_sn2 *msg;
2264        s64 get = initial_get + 1;
2265        int send_msgrequest = 0;
2266
2267        while (1) {
2268
2269                while (1) {
2270                        if (get == ch_sn2->w_local_GP.get)
2271                                break;
2272
2273                        msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
2274                                                     remote_msgqueue + (get %
2275                                                     ch->remote_nentries) *
2276                                                     ch->entry_size);
2277
2278                        if (!(msg->flags & XPC_M_SN2_DONE))
2279                                break;
2280
2281                        msg_flags |= msg->flags;
2282                        get++;
2283                }
2284
2285                if (get == initial_get) {
2286                        /* nothing's changed */
2287                        break;
2288                }
2289
2290                if (cmpxchg_rel(&ch_sn2->local_GP->get, initial_get, get) !=
2291                    initial_get) {
2292                        /* someone else beat us to it */
2293                        DBUG_ON(ch_sn2->local_GP->get <= initial_get);
2294                        break;
2295                }
2296
2297                /* we just set the new value of local_GP->get */
2298
2299                dev_dbg(xpc_chan, "local_GP->get changed to %lld, partid=%d, "
2300                        "channel=%d\n", get, ch->partid, ch->number);
2301
2302                send_msgrequest = (msg_flags & XPC_M_SN2_INTERRUPT);
2303
2304                /*
2305                 * We need to ensure that the message referenced by
2306                 * local_GP->get is not XPC_M_SN2_DONE or that local_GP->get
2307                 * equals w_local_GP.get, so we'll go have a look.
2308                 */
2309                initial_get = get;
2310        }
2311
2312        if (send_msgrequest)
2313                xpc_send_chctl_msgrequest_sn2(ch);
2314}
2315
2316static void
2317xpc_received_payload_sn2(struct xpc_channel *ch, void *payload)
2318{
2319        struct xpc_msg_sn2 *msg;
2320        s64 msg_number;
2321        s64 get;
2322
2323        msg = container_of(payload, struct xpc_msg_sn2, payload);
2324        msg_number = msg->number;
2325
2326        dev_dbg(xpc_chan, "msg=0x%p, msg_number=%lld, partid=%d, channel=%d\n",
2327                (void *)msg, msg_number, ch->partid, ch->number);
2328
2329        DBUG_ON((((u64)msg - (u64)ch->sn.sn2.remote_msgqueue) / ch->entry_size) !=
2330                msg_number % ch->remote_nentries);
2331        DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
2332        DBUG_ON(msg->flags & XPC_M_SN2_DONE);
2333
2334        msg->flags |= XPC_M_SN2_DONE;
2335
2336        /*
2337         * The preceding store of msg->flags must occur before the following
2338         * load of local_GP->get.
2339         */
2340        smp_mb();
2341
2342        /*
2343         * See if this message is next in line to be acknowledged as having
2344         * been delivered.
2345         */
2346        get = ch->sn.sn2.local_GP->get;
2347        if (get == msg_number)
2348                xpc_acknowledge_msgs_sn2(ch, get, msg->flags);
2349}
2350
2351static struct xpc_arch_operations xpc_arch_ops_sn2 = {
2352        .setup_partitions = xpc_setup_partitions_sn2,
2353        .teardown_partitions = xpc_teardown_partitions_sn2,
2354        .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_sn2,
2355        .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_sn2,
2356        .setup_rsvd_page = xpc_setup_rsvd_page_sn2,
2357
2358        .allow_hb = xpc_allow_hb_sn2,
2359        .disallow_hb = xpc_disallow_hb_sn2,
2360        .disallow_all_hbs = xpc_disallow_all_hbs_sn2,
2361        .increment_heartbeat = xpc_increment_heartbeat_sn2,
2362        .offline_heartbeat = xpc_offline_heartbeat_sn2,
2363        .online_heartbeat = xpc_online_heartbeat_sn2,
2364        .heartbeat_init = xpc_heartbeat_init_sn2,
2365        .heartbeat_exit = xpc_heartbeat_exit_sn2,
2366        .get_remote_heartbeat = xpc_get_remote_heartbeat_sn2,
2367
2368        .request_partition_activation =
2369                xpc_request_partition_activation_sn2,
2370        .request_partition_reactivation =
2371                xpc_request_partition_reactivation_sn2,
2372        .request_partition_deactivation =
2373                xpc_request_partition_deactivation_sn2,
2374        .cancel_partition_deactivation_request =
2375                xpc_cancel_partition_deactivation_request_sn2,
2376
2377        .setup_ch_structures = xpc_setup_ch_structures_sn2,
2378        .teardown_ch_structures = xpc_teardown_ch_structures_sn2,
2379
2380        .make_first_contact = xpc_make_first_contact_sn2,
2381
2382        .get_chctl_all_flags = xpc_get_chctl_all_flags_sn2,
2383        .send_chctl_closerequest = xpc_send_chctl_closerequest_sn2,
2384        .send_chctl_closereply = xpc_send_chctl_closereply_sn2,
2385        .send_chctl_openrequest = xpc_send_chctl_openrequest_sn2,
2386        .send_chctl_openreply = xpc_send_chctl_openreply_sn2,
2387        .send_chctl_opencomplete = xpc_send_chctl_opencomplete_sn2,
2388        .process_msg_chctl_flags = xpc_process_msg_chctl_flags_sn2,
2389
2390        .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_sn2,
2391
2392        .setup_msg_structures = xpc_setup_msg_structures_sn2,
2393        .teardown_msg_structures = xpc_teardown_msg_structures_sn2,
2394
2395        .indicate_partition_engaged = xpc_indicate_partition_engaged_sn2,
2396        .indicate_partition_disengaged = xpc_indicate_partition_disengaged_sn2,
2397        .partition_engaged = xpc_partition_engaged_sn2,
2398        .any_partition_engaged = xpc_any_partition_engaged_sn2,
2399        .assume_partition_disengaged = xpc_assume_partition_disengaged_sn2,
2400
2401        .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_sn2,
2402        .send_payload = xpc_send_payload_sn2,
2403        .get_deliverable_payload = xpc_get_deliverable_payload_sn2,
2404        .received_payload = xpc_received_payload_sn2,
2405        .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_sn2,
2406};
2407
2408int
2409xpc_init_sn2(void)
2410{
2411        int ret;
2412        size_t buf_size;
2413
2414        xpc_arch_ops = xpc_arch_ops_sn2;
2415
2416        if (offsetof(struct xpc_msg_sn2, payload) > XPC_MSG_HDR_MAX_SIZE) {
2417                dev_err(xpc_part, "header portion of struct xpc_msg_sn2 is "
2418                        "larger than %d\n", XPC_MSG_HDR_MAX_SIZE);
2419                return -E2BIG;
2420        }
2421
2422        buf_size = max(XPC_RP_VARS_SIZE,
2423                       XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES_SN2);
2424        xpc_remote_copy_buffer_sn2 = xpc_kmalloc_cacheline_aligned(buf_size,
2425                                                                   GFP_KERNEL,
2426                                              &xpc_remote_copy_buffer_base_sn2);
2427        if (xpc_remote_copy_buffer_sn2 == NULL) {
2428                dev_err(xpc_part, "can't get memory for remote copy buffer\n");
2429                return -ENOMEM;
2430        }
2431
2432        /* open up protections for IPI and [potentially] amo operations */
2433        xpc_allow_IPI_ops_sn2();
2434        xpc_allow_amo_ops_shub_wars_1_1_sn2();
2435
2436        /*
2437         * This is safe to do before the xpc_hb_checker thread has started
2438         * because the handler releases a wait queue.  If an interrupt is
2439         * received before the thread is waiting, it will not go to sleep,
2440         * but rather immediately process the interrupt.
2441         */
2442        ret = request_irq(SGI_XPC_ACTIVATE, xpc_handle_activate_IRQ_sn2, 0,
2443                          "xpc hb", NULL);
2444        if (ret != 0) {
2445                dev_err(xpc_part, "can't register ACTIVATE IRQ handler, "
2446                        "errno=%d\n", -ret);
2447                xpc_disallow_IPI_ops_sn2();
2448                kfree(xpc_remote_copy_buffer_base_sn2);
2449        }
2450        return ret;
2451}
2452
2453void
2454xpc_exit_sn2(void)
2455{
2456        free_irq(SGI_XPC_ACTIVATE, NULL);
2457        xpc_disallow_IPI_ops_sn2();
2458        kfree(xpc_remote_copy_buffer_base_sn2);
2459}
2460