drivers/misc/sgi-xp/xpc_uv.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (c) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
   7  */
   8
   9 /*
  10  * Cross Partition Communication (XPC) uv-based functions.
  11  *
  12  *     Architecture specific implementation of common functions.
  13  *
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/mm.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/delay.h>
  20 #include <linux/device.h>
  21 #include <linux/cpu.h>
  22 #include <linux/module.h>
  23 #include <linux/err.h>
  24 #include <linux/slab.h>
  25 #include <asm/uv/uv_hub.h>
  26 #if defined CONFIG_X86_64
  27 #include <asm/uv/bios.h>
  28 #include <asm/uv/uv_irq.h>
  29 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
  30 #include <asm/sn/intr.h>
  31 #include <asm/sn/sn_sal.h>
  32 #endif
  33 #include "../sgi-gru/gru.h"
  34 #include "../sgi-gru/grukservices.h"
  35 #include "xpc.h"
  36
  37 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
  38 struct uv_IO_APIC_route_entry {
  39         __u64   vector          :  8,
  40                 delivery_mode   :  3,
  41                 dest_mode       :  1,
  42                 delivery_status :  1,
  43                 polarity        :  1,
  44                 __reserved_1    :  1,
  45                 trigger         :  1,
  46                 mask            :  1,
  47                 __reserved_2    : 15,
  48                 dest            : 32;
  49 };
  50 #endif
  51
  52 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
  53
  54 #define XPC_ACTIVATE_MSG_SIZE_UV        (1 * GRU_CACHE_LINE_BYTES)
  55 #define XPC_ACTIVATE_MQ_SIZE_UV         (4 * XP_MAX_NPARTITIONS_UV * \
  56                                          XPC_ACTIVATE_MSG_SIZE_UV)
  57 #define XPC_ACTIVATE_IRQ_NAME           "xpc_activate"
  58
  59 #define XPC_NOTIFY_MSG_SIZE_UV          (2 * GRU_CACHE_LINE_BYTES)
  60 #define XPC_NOTIFY_MQ_SIZE_UV           (4 * XP_MAX_NPARTITIONS_UV * \
  61                                          XPC_NOTIFY_MSG_SIZE_UV)
  62 #define XPC_NOTIFY_IRQ_NAME             "xpc_notify"
  63
  64 static int xpc_mq_node = -1;
  65
  66 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
  67 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
  68
  69 static int
  70 xpc_setup_partitions_uv(void)
  71 {
  72         short partid;
  73         struct xpc_partition_uv *part_uv;
  74
  75         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
  76                 part_uv = &xpc_partitions[partid].sn.uv;
  77
  78                 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
  79                 spin_lock_init(&part_uv->flags_lock);
  80                 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
  81         }
  82         return 0;
  83 }
  84
  85 static void
  86 xpc_teardown_partitions_uv(void)
  87 {
  88         short partid;
  89         struct xpc_partition_uv *part_uv;
  90         unsigned long irq_flags;
  91
  92         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
  93                 part_uv = &xpc_partitions[partid].sn.uv;
  94
  95                 if (part_uv->cached_activate_gru_mq_desc != NULL) {
  96                         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
  97                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
  98                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
  99                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 100                         kfree(part_uv->cached_activate_gru_mq_desc);
 101                         part_uv->cached_activate_gru_mq_desc = NULL;
 102                         mutex_unlock(&part_uv->
 103                                      cached_activate_gru_mq_desc_mutex);
 104                 }
 105         }
 106 }
 107
 108 static int
 109 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
 110 {
 111         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
 112
 113 #if defined CONFIG_X86_64
 114         mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
 115                         UV_AFFINITY_CPU);
 116         if (mq->irq < 0)
 117                 return mq->irq;
 118
 119         mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
 120
 121 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
 122         if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
 123                 mq->irq = SGI_XPC_ACTIVATE;
 124         else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
 125                 mq->irq = SGI_XPC_NOTIFY;
 126         else
 127                 return -EINVAL;
 128
 129         mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
 130         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
 131 #else
 132         #error not a supported configuration
 133 #endif
 134
 135         return 0;
 136 }
 137
 138 static void
 139 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
 140 {
 141 #if defined CONFIG_X86_64
 142         uv_teardown_irq(mq->irq);
 143
 144 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
 145         int mmr_pnode;
 146         unsigned long mmr_value;
 147
 148         mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
 149         mmr_value = 1UL << 16;
 150
 151         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
 152 #else
 153         #error not a supported configuration
 154 #endif
 155 }
 156
 157 static int
 158 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
 159 {
 160         int ret;
 161
 162 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
 163         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
 164
 165         ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
 166                                     mq->order, &mq->mmr_offset);
 167         if (ret < 0) {
 168                 dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
 169                         ret);
 170                 return -EBUSY;
 171         }
 172 #elif defined CONFIG_X86_64
 173         ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
 174                                          mq->order, &mq->mmr_offset);
 175         if (ret < 0) {
 176                 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
 177                         "ret=%d\n", ret);
 178                 return ret;
 179         }
 180 #else
 181         #error not a supported configuration
 182 #endif
 183
 184         mq->watchlist_num = ret;
 185         return 0;
 186 }
 187
 188 static void
 189 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
 190 {
 191         int ret;
 192         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
 193
 194 #if defined CONFIG_X86_64
 195         ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
 196         BUG_ON(ret != BIOS_STATUS_SUCCESS);
 197 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
 198         ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
 199         BUG_ON(ret != SALRET_OK);
 200 #else
 201         #error not a supported configuration
 202 #endif
 203 }
 204
 205 static struct xpc_gru_mq_uv *
 206 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
 207                      irq_handler_t irq_handler)
 208 {
 209         enum xp_retval xp_ret;
 210         int ret;
 211         int nid;
 212         int nasid;
 213         int pg_order;
 214         struct page *page;
 215         struct xpc_gru_mq_uv *mq;
 216         struct uv_IO_APIC_route_entry *mmr_value;
 217
 218         mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
 219         if (mq == NULL) {
 220                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
 221                         "a xpc_gru_mq_uv structure\n");
 222                 ret = -ENOMEM;
 223                 goto out_0;
 224         }
 225
 226         mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
 227                                   GFP_KERNEL);
 228         if (mq->gru_mq_desc == NULL) {
 229                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
 230                         "a gru_message_queue_desc structure\n");
 231                 ret = -ENOMEM;
 232                 goto out_1;
 233         }
 234
 235         pg_order = get_order(mq_size);
 236         mq->order = pg_order + PAGE_SHIFT;
 237         mq_size = 1UL << mq->order;
 238
 239         mq->mmr_blade = uv_cpu_to_blade_id(cpu);
 240
 241         nid = cpu_to_node(cpu);
 242         page = __alloc_pages_node(nid,
 243                                       GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
 244                                       pg_order);
 245         if (page == NULL) {
 246                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
 247                         "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
 248                 ret = -ENOMEM;
 249                 goto out_2;
 250         }
 251         mq->address = page_address(page);
 252
 253         /* enable generation of irq when GRU mq operation occurs to this mq */
 254         ret = xpc_gru_mq_watchlist_alloc_uv(mq);
 255         if (ret != 0)
 256                 goto out_3;
 257
 258         ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
 259         if (ret != 0)
 260                 goto out_4;
 261
 262         ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
 263         if (ret != 0) {
 264                 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
 265                         mq->irq, -ret);
 266                 goto out_5;
 267         }
 268
 269         nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
 270
 271         mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
 272         ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
 273                                      nasid, mmr_value->vector, mmr_value->dest);
 274         if (ret != 0) {
 275                 dev_err(xpc_part, "gru_create_message_queue() returned "
 276                         "error=%d\n", ret);
 277                 ret = -EINVAL;
 278                 goto out_6;
 279         }
 280
 281         /* allow other partitions to access this GRU mq */
 282         xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
 283         if (xp_ret != xpSuccess) {
 284                 ret = -EACCES;
 285                 goto out_6;
 286         }
 287
 288         return mq;
 289
 290         /* something went wrong */
 291 out_6:
 292         free_irq(mq->irq, NULL);
 293 out_5:
 294         xpc_release_gru_mq_irq_uv(mq);
 295 out_4:
 296         xpc_gru_mq_watchlist_free_uv(mq);
 297 out_3:
 298         free_pages((unsigned long)mq->address, pg_order);
 299 out_2:
 300         kfree(mq->gru_mq_desc);
 301 out_1:
 302         kfree(mq);
 303 out_0:
 304         return ERR_PTR(ret);
 305 }
 306
 307 static void
 308 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
 309 {
 310         unsigned int mq_size;
 311         int pg_order;
 312         int ret;
 313
 314         /* disallow other partitions to access GRU mq */
 315         mq_size = 1UL << mq->order;
 316         ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
 317         BUG_ON(ret != xpSuccess);
 318
 319         /* unregister irq handler and release mq irq/vector mapping */
 320         free_irq(mq->irq, NULL);
 321         xpc_release_gru_mq_irq_uv(mq);
 322
 323         /* disable generation of irq when GRU mq op occurs to this mq */
 324         xpc_gru_mq_watchlist_free_uv(mq);
 325
 326         pg_order = mq->order - PAGE_SHIFT;
 327         free_pages((unsigned long)mq->address, pg_order);
 328
 329         kfree(mq);
 330 }
 331
 332 static enum xp_retval
 333 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
 334                  size_t msg_size)
 335 {
 336         enum xp_retval xp_ret;
 337         int ret;
 338
 339         while (1) {
 340                 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
 341                 if (ret == MQE_OK) {
 342                         xp_ret = xpSuccess;
 343                         break;
 344                 }
 345
 346                 if (ret == MQE_QUEUE_FULL) {
 347                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
 348                                 "error=MQE_QUEUE_FULL\n");
 349                         /* !!! handle QLimit reached; delay & try again */
 350                         /* ??? Do we add a limit to the number of retries? */
 351                         (void)msleep_interruptible(10);
 352                 } else if (ret == MQE_CONGESTION) {
 353                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
 354                                 "error=MQE_CONGESTION\n");
 355                         /* !!! handle LB Overflow; simply try again */
 356                         /* ??? Do we add a limit to the number of retries? */
 357                 } else {
 358                         /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
 359                         dev_err(xpc_chan, "gru_send_message_gpa() returned "
 360                                 "error=%d\n", ret);
 361                         xp_ret = xpGruSendMqError;
 362                         break;
 363                 }
 364         }
 365         return xp_ret;
 366 }
 367
 368 static void
 369 xpc_process_activate_IRQ_rcvd_uv(void)
 370 {
 371         unsigned long irq_flags;
 372         short partid;
 373         struct xpc_partition *part;
 374         u8 act_state_req;
 375
 376         DBUG_ON(xpc_activate_IRQ_rcvd == 0);
 377
 378         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 379         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
 380                 part = &xpc_partitions[partid];
 381
 382                 if (part->sn.uv.act_state_req == 0)
 383                         continue;
 384
 385                 xpc_activate_IRQ_rcvd--;
 386                 BUG_ON(xpc_activate_IRQ_rcvd < 0);
 387
 388                 act_state_req = part->sn.uv.act_state_req;
 389                 part->sn.uv.act_state_req = 0;
 390                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 391
 392                 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
 393                         if (part->act_state == XPC_P_AS_INACTIVE)
 394                                 xpc_activate_partition(part);
 395                         else if (part->act_state == XPC_P_AS_DEACTIVATING)
 396                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 397
 398                 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
 399                         if (part->act_state == XPC_P_AS_INACTIVE)
 400                                 xpc_activate_partition(part);
 401                         else
 402                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
 403
 404                 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
 405                         XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
 406
 407                 } else {
 408                         BUG();
 409                 }
 410
 411                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 412                 if (xpc_activate_IRQ_rcvd == 0)
 413                         break;
 414         }
 415         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 416
 417 }
 418
 419 static void
 420 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
 421                               struct xpc_activate_mq_msghdr_uv *msg_hdr,
 422                               int part_setup,
 423                               int *wakeup_hb_checker)
 424 {
 425         unsigned long irq_flags;
 426         struct xpc_partition_uv *part_uv = &part->sn.uv;
 427         struct xpc_openclose_args *args;
 428
 429         part_uv->remote_act_state = msg_hdr->act_state;
 430
 431         switch (msg_hdr->type) {
 432         case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
 433                 /* syncing of remote_act_state was just done above */
 434                 break;
 435
 436         case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
 437                 struct xpc_activate_mq_msg_activate_req_uv *msg;
 438
 439                 /*
 440                  * ??? Do we deal here with ts_jiffies being different
 441                  * ??? if act_state != XPC_P_AS_INACTIVE instead of
 442                  * ??? below?
 443                  */
 444                 msg = container_of(msg_hdr, struct
 445                                    xpc_activate_mq_msg_activate_req_uv, hdr);
 446
 447                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 448                 if (part_uv->act_state_req == 0)
 449                         xpc_activate_IRQ_rcvd++;
 450                 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
 451                 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
 452                 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
 453                 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
 454
 455                 if (msg->activate_gru_mq_desc_gpa !=
 456                     part_uv->activate_gru_mq_desc_gpa) {
 457                         spin_lock(&part_uv->flags_lock);
 458                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
 459                         spin_unlock(&part_uv->flags_lock);
 460                         part_uv->activate_gru_mq_desc_gpa =
 461                             msg->activate_gru_mq_desc_gpa;
 462                 }
 463                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 464
 465                 (*wakeup_hb_checker)++;
 466                 break;
 467         }
 468         case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
 469                 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
 470
 471                 msg = container_of(msg_hdr, struct
 472                                    xpc_activate_mq_msg_deactivate_req_uv, hdr);
 473
 474                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 475                 if (part_uv->act_state_req == 0)
 476                         xpc_activate_IRQ_rcvd++;
 477                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
 478                 part_uv->reason = msg->reason;
 479                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 480
 481                 (*wakeup_hb_checker)++;
 482                 return;
 483         }
 484         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
 485                 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
 486
 487                 if (!part_setup)
 488                         break;
 489
 490                 msg = container_of(msg_hdr, struct
 491                                    xpc_activate_mq_msg_chctl_closerequest_uv,
 492                                    hdr);
 493                 args = &part->remote_openclose_args[msg->ch_number];
 494                 args->reason = msg->reason;
 495
 496                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 497                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
 498                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 499
 500                 xpc_wakeup_channel_mgr(part);
 501                 break;
 502         }
 503         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
 504                 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
 505
 506                 if (!part_setup)
 507                         break;
 508
 509                 msg = container_of(msg_hdr, struct
 510                                    xpc_activate_mq_msg_chctl_closereply_uv,
 511                                    hdr);
 512
 513                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 514                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
 515                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 516
 517                 xpc_wakeup_channel_mgr(part);
 518                 break;
 519         }
 520         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
 521                 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
 522
 523                 if (!part_setup)
 524                         break;
 525
 526                 msg = container_of(msg_hdr, struct
 527                                    xpc_activate_mq_msg_chctl_openrequest_uv,
 528                                    hdr);
 529                 args = &part->remote_openclose_args[msg->ch_number];
 530                 args->entry_size = msg->entry_size;
 531                 args->local_nentries = msg->local_nentries;
 532
 533                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 534                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
 535                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 536
 537                 xpc_wakeup_channel_mgr(part);
 538                 break;
 539         }
 540         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
 541                 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
 542
 543                 if (!part_setup)
 544                         break;
 545
 546                 msg = container_of(msg_hdr, struct
 547                                    xpc_activate_mq_msg_chctl_openreply_uv, hdr);
 548                 args = &part->remote_openclose_args[msg->ch_number];
 549                 args->remote_nentries = msg->remote_nentries;
 550                 args->local_nentries = msg->local_nentries;
 551                 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
 552
 553                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 554                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
 555                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 556
 557                 xpc_wakeup_channel_mgr(part);
 558                 break;
 559         }
 560         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
 561                 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
 562
 563                 if (!part_setup)
 564                         break;
 565
 566                 msg = container_of(msg_hdr, struct
 567                                 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
 568                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
 569                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
 570                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 571
 572                 xpc_wakeup_channel_mgr(part);
 573         }
 574         case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
 575                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 576                 part_uv->flags |= XPC_P_ENGAGED_UV;
 577                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 578                 break;
 579
 580         case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
 581                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 582                 part_uv->flags &= ~XPC_P_ENGAGED_UV;
 583                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 584                 break;
 585
 586         default:
 587                 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
 588                         "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
 589
 590                 /* get hb checker to deactivate from the remote partition */
 591                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 592                 if (part_uv->act_state_req == 0)
 593                         xpc_activate_IRQ_rcvd++;
 594                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
 595                 part_uv->reason = xpBadMsgType;
 596                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 597
 598                 (*wakeup_hb_checker)++;
 599                 return;
 600         }
 601
 602         if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
 603             part->remote_rp_ts_jiffies != 0) {
 604                 /*
 605                  * ??? Does what we do here need to be sensitive to
 606                  * ??? act_state or remote_act_state?
 607                  */
 608                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 609                 if (part_uv->act_state_req == 0)
 610                         xpc_activate_IRQ_rcvd++;
 611                 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
 612                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 613
 614                 (*wakeup_hb_checker)++;
 615         }
 616 }
 617
 618 static irqreturn_t
 619 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
 620 {
 621         struct xpc_activate_mq_msghdr_uv *msg_hdr;
 622         short partid;
 623         struct xpc_partition *part;
 624         int wakeup_hb_checker = 0;
 625         int part_referenced;
 626
 627         while (1) {
 628                 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
 629                 if (msg_hdr == NULL)
 630                         break;
 631
 632                 partid = msg_hdr->partid;
 633                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
 634                         dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
 635                                 "received invalid partid=0x%x in message\n",
 636                                 partid);
 637                 } else {
 638                         part = &xpc_partitions[partid];
 639
 640                         part_referenced = xpc_part_ref(part);
 641                         xpc_handle_activate_mq_msg_uv(part, msg_hdr,
 642                                                       part_referenced,
 643                                                       &wakeup_hb_checker);
 644                         if (part_referenced)
 645                                 xpc_part_deref(part);
 646                 }
 647
 648                 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
 649         }
 650
 651         if (wakeup_hb_checker)
 652                 wake_up_interruptible(&xpc_activate_IRQ_wq);
 653
 654         return IRQ_HANDLED;
 655 }
 656
 657 static enum xp_retval
 658 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
 659                                 unsigned long gru_mq_desc_gpa)
 660 {
 661         enum xp_retval ret;
 662
 663         ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
 664                                sizeof(struct gru_message_queue_desc));
 665         if (ret == xpSuccess)
 666                 gru_mq_desc->mq = NULL;
 667
 668         return ret;
 669 }
 670
 671 static enum xp_retval
 672 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
 673                          int msg_type)
 674 {
 675         struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
 676         struct xpc_partition_uv *part_uv = &part->sn.uv;
 677         struct gru_message_queue_desc *gru_mq_desc;
 678         unsigned long irq_flags;
 679         enum xp_retval ret;
 680
 681         DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
 682
 683         msg_hdr->type = msg_type;
 684         msg_hdr->partid = xp_partition_id;
 685         msg_hdr->act_state = part->act_state;
 686         msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
 687
 688         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
 689 again:
 690         if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
 691                 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
 692                 if (gru_mq_desc == NULL) {
 693                         gru_mq_desc = kmalloc(sizeof(struct
 694                                               gru_message_queue_desc),
 695                                               GFP_KERNEL);
 696                         if (gru_mq_desc == NULL) {
 697                                 ret = xpNoMemory;
 698                                 goto done;
 699                         }
 700                         part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
 701                 }
 702
 703                 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
 704                                                       part_uv->
 705                                                       activate_gru_mq_desc_gpa);
 706                 if (ret != xpSuccess)
 707                         goto done;
 708
 709                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
 710                 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
 711                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
 712         }
 713
 714         /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
 715         ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
 716                                msg_size);
 717         if (ret != xpSuccess) {
 718                 smp_rmb();      /* ensure a fresh copy of part_uv->flags */
 719                 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
 720                         goto again;
 721         }
 722 done:
 723         mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
 724         return ret;
 725 }
 726
 727 static void
 728 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
 729                               size_t msg_size, int msg_type)
 730 {
 731         enum xp_retval ret;
 732
 733         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
 734         if (unlikely(ret != xpSuccess))
 735                 XPC_DEACTIVATE_PARTITION(part, ret);
 736 }
 737
 738 static void
 739 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
 740                          void *msg, size_t msg_size, int msg_type)
 741 {
 742         struct xpc_partition *part = &xpc_partitions[ch->partid];
 743         enum xp_retval ret;
 744
 745         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
 746         if (unlikely(ret != xpSuccess)) {
 747                 if (irq_flags != NULL)
 748                         spin_unlock_irqrestore(&ch->lock, *irq_flags);
 749
 750                 XPC_DEACTIVATE_PARTITION(part, ret);
 751
 752                 if (irq_flags != NULL)
 753                         spin_lock_irqsave(&ch->lock, *irq_flags);
 754         }
 755 }
 756
 757 static void
 758 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
 759 {
 760         unsigned long irq_flags;
 761         struct xpc_partition_uv *part_uv = &part->sn.uv;
 762
 763         /*
 764          * !!! Make our side think that the remote partition sent an activate
 765          * !!! mq message our way by doing what the activate IRQ handler would
 766          * !!! do had one really been sent.
 767          */
 768
 769         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 770         if (part_uv->act_state_req == 0)
 771                 xpc_activate_IRQ_rcvd++;
 772         part_uv->act_state_req = act_state_req;
 773         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 774
 775         wake_up_interruptible(&xpc_activate_IRQ_wq);
 776 }
 777
 778 static enum xp_retval
 779 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
 780                                   size_t *len)
 781 {
 782         s64 status;
 783         enum xp_retval ret;
 784
 785 #if defined CONFIG_X86_64
 786         status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
 787                                           (u64 *)len);
 788         if (status == BIOS_STATUS_SUCCESS)
 789                 ret = xpSuccess;
 790         else if (status == BIOS_STATUS_MORE_PASSES)
 791                 ret = xpNeedMoreInfo;
 792         else
 793                 ret = xpBiosError;
 794
 795 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
 796         status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
 797         if (status == SALRET_OK)
 798                 ret = xpSuccess;
 799         else if (status == SALRET_MORE_PASSES)
 800                 ret = xpNeedMoreInfo;
 801         else
 802                 ret = xpSalError;
 803
 804 #else
 805         #error not a supported configuration
 806 #endif
 807
 808         return ret;
 809 }
 810
 811 static int
 812 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
 813 {
 814         xpc_heartbeat_uv =
 815             &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
 816         rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
 817         rp->sn.uv.activate_gru_mq_desc_gpa =
 818             uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
 819         return 0;
 820 }
 821
 822 static void
 823 xpc_allow_hb_uv(short partid)
 824 {
 825 }
 826
 827 static void
 828 xpc_disallow_hb_uv(short partid)
 829 {
 830 }
 831
 832 static void
 833 xpc_disallow_all_hbs_uv(void)
 834 {
 835 }
 836
 837 static void
 838 xpc_increment_heartbeat_uv(void)
 839 {
 840         xpc_heartbeat_uv->value++;
 841 }
 842
 843 static void
 844 xpc_offline_heartbeat_uv(void)
 845 {
 846         xpc_increment_heartbeat_uv();
 847         xpc_heartbeat_uv->offline = 1;
 848 }
 849
 850 static void
 851 xpc_online_heartbeat_uv(void)
 852 {
 853         xpc_increment_heartbeat_uv();
 854         xpc_heartbeat_uv->offline = 0;
 855 }
 856
 857 static void
 858 xpc_heartbeat_init_uv(void)
 859 {
 860         xpc_heartbeat_uv->value = 1;
 861         xpc_heartbeat_uv->offline = 0;
 862 }
 863
 864 static void
 865 xpc_heartbeat_exit_uv(void)
 866 {
 867         xpc_offline_heartbeat_uv();
 868 }
 869
 870 static enum xp_retval
 871 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
 872 {
 873         struct xpc_partition_uv *part_uv = &part->sn.uv;
 874         enum xp_retval ret;
 875
 876         ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
 877                                part_uv->heartbeat_gpa,
 878                                sizeof(struct xpc_heartbeat_uv));
 879         if (ret != xpSuccess)
 880                 return ret;
 881
 882         if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
 883             !part_uv->cached_heartbeat.offline) {
 884
 885                 ret = xpNoHeartbeat;
 886         } else {
 887                 part->last_heartbeat = part_uv->cached_heartbeat.value;
 888         }
 889         return ret;
 890 }
 891
 892 static void
 893 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
 894                                     unsigned long remote_rp_gpa, int nasid)
 895 {
 896         short partid = remote_rp->SAL_partid;
 897         struct xpc_partition *part = &xpc_partitions[partid];
 898         struct xpc_activate_mq_msg_activate_req_uv msg;
 899
 900         part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
 901         part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
 902         part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
 903         part->sn.uv.activate_gru_mq_desc_gpa =
 904             remote_rp->sn.uv.activate_gru_mq_desc_gpa;
 905
 906         /*
 907          * ??? Is it a good idea to make this conditional on what is
 908          * ??? potentially stale state information?
 909          */
 910         if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
 911                 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
 912                 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
 913                 msg.activate_gru_mq_desc_gpa =
 914                     xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
 915                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
 916                                            XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
 917         }
 918
 919         if (part->act_state == XPC_P_AS_INACTIVE)
 920                 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
 921 }
 922
 923 static void
 924 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
 925 {
 926         xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
 927 }
 928
 929 static void
 930 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
 931 {
 932         struct xpc_activate_mq_msg_deactivate_req_uv msg;
 933
 934         /*
 935          * ??? Is it a good idea to make this conditional on what is
 936          * ??? potentially stale state information?
 937          */
 938         if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
 939             part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
 940
 941                 msg.reason = part->reason;
 942                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
 943                                          XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
 944         }
 945 }
 946
 947 static void
 948 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
 949 {
 950         /* nothing needs to be done */
 951         return;
 952 }
 953
 954 static void
 955 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
 956 {
 957         head->first = NULL;
 958         head->last = NULL;
 959         spin_lock_init(&head->lock);
 960         head->n_entries = 0;
 961 }
 962
 963 static void *
 964 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
 965 {
 966         unsigned long irq_flags;
 967         struct xpc_fifo_entry_uv *first;
 968
 969         spin_lock_irqsave(&head->lock, irq_flags);
 970         first = head->first;
 971         if (head->first != NULL) {
 972                 head->first = first->next;
 973                 if (head->first == NULL)
 974                         head->last = NULL;
 975
 976                 head->n_entries--;
 977                 BUG_ON(head->n_entries < 0);
 978
 979                 first->next = NULL;
 980         }
 981         spin_unlock_irqrestore(&head->lock, irq_flags);
 982         return first;
 983 }
 984
 985 static void
 986 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
 987                       struct xpc_fifo_entry_uv *last)
 988 {
 989         unsigned long irq_flags;
 990
 991         last->next = NULL;
 992         spin_lock_irqsave(&head->lock, irq_flags);
 993         if (head->last != NULL)
 994                 head->last->next = last;
 995         else
 996                 head->first = last;
 997         head->last = last;
 998         head->n_entries++;
 999         spin_unlock_irqrestore(&head->lock, irq_flags);
1000 }
1001
1002 static int
1003 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
1004 {
1005         return head->n_entries;
1006 }
1007
1008 /*
1009  * Setup the channel structures that are uv specific.
1010  */
1011 static enum xp_retval
1012 xpc_setup_ch_structures_uv(struct xpc_partition *part)
1013 {
1014         struct xpc_channel_uv *ch_uv;
1015         int ch_number;
1016
1017         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1018                 ch_uv = &part->channels[ch_number].sn.uv;
1019
1020                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1021                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1022         }
1023
1024         return xpSuccess;
1025 }
1026
1027 /*
1028  * Teardown the channel structures that are uv specific.
1029  */
1030 static void
1031 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1032 {
1033         /* nothing needs to be done */
1034         return;
1035 }
1036
1037 static enum xp_retval
1038 xpc_make_first_contact_uv(struct xpc_partition *part)
1039 {
1040         struct xpc_activate_mq_msg_uv msg;
1041
1042         /*
1043          * We send a sync msg to get the remote partition's remote_act_state
1044          * updated to our current act_state which at this point should
1045          * be XPC_P_AS_ACTIVATING.
1046          */
1047         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1048                                       XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1049
1050         while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1051                  (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1052
1053                 dev_dbg(xpc_part, "waiting to make first contact with "
1054                         "partition %d\n", XPC_PARTID(part));
1055
1056                 /* wait a 1/4 of a second or so */
1057                 (void)msleep_interruptible(250);
1058
1059                 if (part->act_state == XPC_P_AS_DEACTIVATING)
1060                         return part->reason;
1061         }
1062
1063         return xpSuccess;
1064 }
1065
1066 static u64
1067 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1068 {
1069         unsigned long irq_flags;
1070         union xpc_channel_ctl_flags chctl;
1071
1072         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1073         chctl = part->chctl;
1074         if (chctl.all_flags != 0)
1075                 part->chctl.all_flags = 0;
1076
1077         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1078         return chctl.all_flags;
1079 }
1080
1081 static enum xp_retval
1082 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1083 {
1084         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1085         struct xpc_send_msg_slot_uv *msg_slot;
1086         unsigned long irq_flags;
1087         int nentries;
1088         int entry;
1089         size_t nbytes;
1090
1091         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1092                 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1093                 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1094                 if (ch_uv->send_msg_slots == NULL)
1095                         continue;
1096
1097                 for (entry = 0; entry < nentries; entry++) {
1098                         msg_slot = &ch_uv->send_msg_slots[entry];
1099
1100                         msg_slot->msg_slot_number = entry;
1101                         xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1102                                               &msg_slot->next);
1103                 }
1104
1105                 spin_lock_irqsave(&ch->lock, irq_flags);
1106                 if (nentries < ch->local_nentries)
1107                         ch->local_nentries = nentries;
1108                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1109                 return xpSuccess;
1110         }
1111
1112         return xpNoMemory;
1113 }
1114
1115 static enum xp_retval
1116 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1117 {
1118         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1119         struct xpc_notify_mq_msg_uv *msg_slot;
1120         unsigned long irq_flags;
1121         int nentries;
1122         int entry;
1123         size_t nbytes;
1124
1125         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1126                 nbytes = nentries * ch->entry_size;
1127                 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1128                 if (ch_uv->recv_msg_slots == NULL)
1129                         continue;
1130
1131                 for (entry = 0; entry < nentries; entry++) {
1132                         msg_slot = ch_uv->recv_msg_slots +
1133                             entry * ch->entry_size;
1134
1135                         msg_slot->hdr.msg_slot_number = entry;
1136                 }
1137
1138                 spin_lock_irqsave(&ch->lock, irq_flags);
1139                 if (nentries < ch->remote_nentries)
1140                         ch->remote_nentries = nentries;
1141                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1142                 return xpSuccess;
1143         }
1144
1145         return xpNoMemory;
1146 }
1147
1148 /*
1149  * Allocate msg_slots associated with the channel.
1150  */
1151 static enum xp_retval
1152 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1153 {
1154         static enum xp_retval ret;
1155         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1156
1157         DBUG_ON(ch->flags & XPC_C_SETUP);
1158
1159         ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1160                                                    gru_message_queue_desc),
1161                                                    GFP_KERNEL);
1162         if (ch_uv->cached_notify_gru_mq_desc == NULL)
1163                 return xpNoMemory;
1164
1165         ret = xpc_allocate_send_msg_slot_uv(ch);
1166         if (ret == xpSuccess) {
1167
1168                 ret = xpc_allocate_recv_msg_slot_uv(ch);
1169                 if (ret != xpSuccess) {
1170                         kfree(ch_uv->send_msg_slots);
1171                         xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1172                 }
1173         }
1174         return ret;
1175 }
1176
1177 /*
1178  * Free up msg_slots and clear other stuff that were setup for the specified
1179  * channel.
1180  */
1181 static void
1182 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1183 {
1184         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1185
1186         DBUG_ON(!spin_is_locked(&ch->lock));
1187
1188         kfree(ch_uv->cached_notify_gru_mq_desc);
1189         ch_uv->cached_notify_gru_mq_desc = NULL;
1190
1191         if (ch->flags & XPC_C_SETUP) {
1192                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1193                 kfree(ch_uv->send_msg_slots);
1194                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1195                 kfree(ch_uv->recv_msg_slots);
1196         }
1197 }
1198
1199 static void
1200 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1201 {
1202         struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1203
1204         msg.ch_number = ch->number;
1205         msg.reason = ch->reason;
1206         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1207                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1208 }
1209
1210 static void
1211 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1212 {
1213         struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1214
1215         msg.ch_number = ch->number;
1216         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1217                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1218 }
1219
1220 static void
1221 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1222 {
1223         struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1224
1225         msg.ch_number = ch->number;
1226         msg.entry_size = ch->entry_size;
1227         msg.local_nentries = ch->local_nentries;
1228         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1229                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1230 }
1231
1232 static void
1233 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1234 {
1235         struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1236
1237         msg.ch_number = ch->number;
1238         msg.local_nentries = ch->local_nentries;
1239         msg.remote_nentries = ch->remote_nentries;
1240         msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1241         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1242                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1243 }
1244
1245 static void
1246 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1247 {
1248         struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1249
1250         msg.ch_number = ch->number;
1251         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1252                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1253 }
1254
1255 static void
1256 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1257 {
1258         unsigned long irq_flags;
1259
1260         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1261         part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1262         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1263
1264         xpc_wakeup_channel_mgr(part);
1265 }
1266
1267 static enum xp_retval
1268 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1269                                unsigned long gru_mq_desc_gpa)
1270 {
1271         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1272
1273         DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1274         return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1275                                                gru_mq_desc_gpa);
1276 }
1277
1278 static void
1279 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1280 {
1281         struct xpc_activate_mq_msg_uv msg;
1282
1283         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1284                                       XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1285 }
1286
1287 static void
1288 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1289 {
1290         struct xpc_activate_mq_msg_uv msg;
1291
1292         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1293                                       XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1294 }
1295
1296 static void
1297 xpc_assume_partition_disengaged_uv(short partid)
1298 {
1299         struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1300         unsigned long irq_flags;
1301
1302         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1303         part_uv->flags &= ~XPC_P_ENGAGED_UV;
1304         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1305 }
1306
1307 static int
1308 xpc_partition_engaged_uv(short partid)
1309 {
1310         return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1311 }
1312
1313 static int
1314 xpc_any_partition_engaged_uv(void)
1315 {
1316         struct xpc_partition_uv *part_uv;
1317         short partid;
1318
1319         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1320                 part_uv = &xpc_partitions[partid].sn.uv;
1321                 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1322                         return 1;
1323         }
1324         return 0;
1325 }
1326
1327 static enum xp_retval
1328 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1329                          struct xpc_send_msg_slot_uv **address_of_msg_slot)
1330 {
1331         enum xp_retval ret;
1332         struct xpc_send_msg_slot_uv *msg_slot;
1333         struct xpc_fifo_entry_uv *entry;
1334
1335         while (1) {
1336                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1337                 if (entry != NULL)
1338                         break;
1339
1340                 if (flags & XPC_NOWAIT)
1341                         return xpNoWait;
1342
1343                 ret = xpc_allocate_msg_wait(ch);
1344                 if (ret != xpInterrupted && ret != xpTimeout)
1345                         return ret;
1346         }
1347
1348         msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1349         *address_of_msg_slot = msg_slot;
1350         return xpSuccess;
1351 }
1352
1353 static void
1354 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1355                      struct xpc_send_msg_slot_uv *msg_slot)
1356 {
1357         xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1358
1359         /* wakeup anyone waiting for a free msg slot */
1360         if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1361                 wake_up(&ch->msg_allocate_wq);
1362 }
1363
1364 static void
1365 xpc_notify_sender_uv(struct xpc_channel *ch,
1366                      struct xpc_send_msg_slot_uv *msg_slot,
1367                      enum xp_retval reason)
1368 {
1369         xpc_notify_func func = msg_slot->func;
1370
1371         if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1372
1373                 atomic_dec(&ch->n_to_notify);
1374
1375                 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1376                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1377                         msg_slot->msg_slot_number, ch->partid, ch->number);
1378
1379                 func(reason, ch->partid, ch->number, msg_slot->key);
1380
1381                 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1382                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1383                         msg_slot->msg_slot_number, ch->partid, ch->number);
1384         }
1385 }
1386
1387 static void
1388 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1389                             struct xpc_notify_mq_msg_uv *msg)
1390 {
1391         struct xpc_send_msg_slot_uv *msg_slot;
1392         int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1393
1394         msg_slot = &ch->sn.uv.send_msg_slots[entry];
1395
1396         BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1397         msg_slot->msg_slot_number += ch->local_nentries;
1398
1399         if (msg_slot->func != NULL)
1400                 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1401
1402         xpc_free_msg_slot_uv(ch, msg_slot);
1403 }
1404
1405 static void
1406 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1407                             struct xpc_notify_mq_msg_uv *msg)
1408 {
1409         struct xpc_partition_uv *part_uv = &part->sn.uv;
1410         struct xpc_channel *ch;
1411         struct xpc_channel_uv *ch_uv;
1412         struct xpc_notify_mq_msg_uv *msg_slot;
1413         unsigned long irq_flags;
1414         int ch_number = msg->hdr.ch_number;
1415
1416         if (unlikely(ch_number >= part->nchannels)) {
1417                 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1418                         "channel number=0x%x in message from partid=%d\n",
1419                         ch_number, XPC_PARTID(part));
1420
1421                 /* get hb checker to deactivate from the remote partition */
1422                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1423                 if (part_uv->act_state_req == 0)
1424                         xpc_activate_IRQ_rcvd++;
1425                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1426                 part_uv->reason = xpBadChannelNumber;
1427                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1428
1429                 wake_up_interruptible(&xpc_activate_IRQ_wq);
1430                 return;
1431         }
1432
1433         ch = &part->channels[ch_number];
1434         xpc_msgqueue_ref(ch);
1435
1436         if (!(ch->flags & XPC_C_CONNECTED)) {
1437                 xpc_msgqueue_deref(ch);
1438                 return;
1439         }
1440
1441         /* see if we're really dealing with an ACK for a previously sent msg */
1442         if (msg->hdr.size == 0) {
1443                 xpc_handle_notify_mq_ack_uv(ch, msg);
1444                 xpc_msgqueue_deref(ch);
1445                 return;
1446         }
1447
1448         /* we're dealing with a normal message sent via the notify_mq */
1449         ch_uv = &ch->sn.uv;
1450
1451         msg_slot = ch_uv->recv_msg_slots +
1452             (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1453
1454         BUG_ON(msg_slot->hdr.size != 0);
1455
1456         memcpy(msg_slot, msg, msg->hdr.size);
1457
1458         xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1459
1460         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1461                 /*
1462                  * If there is an existing idle kthread get it to deliver
1463                  * the payload, otherwise we'll have to get the channel mgr
1464                  * for this partition to create a kthread to do the delivery.
1465                  */
1466                 if (atomic_read(&ch->kthreads_idle) > 0)
1467                         wake_up_nr(&ch->idle_wq, 1);
1468                 else
1469                         xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1470         }
1471         xpc_msgqueue_deref(ch);
1472 }
1473
1474 static irqreturn_t
1475 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1476 {
1477         struct xpc_notify_mq_msg_uv *msg;
1478         short partid;
1479         struct xpc_partition *part;
1480
1481         while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1482                NULL) {
1483
1484                 partid = msg->hdr.partid;
1485                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1486                         dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1487                                 "invalid partid=0x%x in message\n", partid);
1488                 } else {
1489                         part = &xpc_partitions[partid];
1490
1491                         if (xpc_part_ref(part)) {
1492                                 xpc_handle_notify_mq_msg_uv(part, msg);
1493                                 xpc_part_deref(part);
1494                         }
1495                 }
1496
1497                 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1498         }
1499
1500         return IRQ_HANDLED;
1501 }
1502
1503 static int
1504 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1505 {
1506         return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1507 }
1508
1509 static void
1510 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1511 {
1512         struct xpc_channel *ch = &part->channels[ch_number];
1513         int ndeliverable_payloads;
1514
1515         xpc_msgqueue_ref(ch);
1516
1517         ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1518
1519         if (ndeliverable_payloads > 0 &&
1520             (ch->flags & XPC_C_CONNECTED) &&
1521             (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1522
1523                 xpc_activate_kthreads(ch, ndeliverable_payloads);
1524         }
1525
1526         xpc_msgqueue_deref(ch);
1527 }
1528
1529 static enum xp_retval
1530 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1531                     u16 payload_size, u8 notify_type, xpc_notify_func func,
1532                     void *key)
1533 {
1534         enum xp_retval ret = xpSuccess;
1535         struct xpc_send_msg_slot_uv *msg_slot = NULL;
1536         struct xpc_notify_mq_msg_uv *msg;
1537         u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1538         size_t msg_size;
1539
1540         DBUG_ON(notify_type != XPC_N_CALL);
1541
1542         msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1543         if (msg_size > ch->entry_size)
1544                 return xpPayloadTooBig;
1545
1546         xpc_msgqueue_ref(ch);
1547
1548         if (ch->flags & XPC_C_DISCONNECTING) {
1549                 ret = ch->reason;
1550                 goto out_1;
1551         }
1552         if (!(ch->flags & XPC_C_CONNECTED)) {
1553                 ret = xpNotConnected;
1554                 goto out_1;
1555         }
1556
1557         ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1558         if (ret != xpSuccess)
1559                 goto out_1;
1560
1561         if (func != NULL) {
1562                 atomic_inc(&ch->n_to_notify);
1563
1564                 msg_slot->key = key;
1565                 smp_wmb(); /* a non-NULL func must hit memory after the key */
1566                 msg_slot->func = func;
1567
1568                 if (ch->flags & XPC_C_DISCONNECTING) {
1569                         ret = ch->reason;
1570                         goto out_2;
1571                 }
1572         }
1573
1574         msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1575         msg->hdr.partid = xp_partition_id;
1576         msg->hdr.ch_number = ch->number;
1577         msg->hdr.size = msg_size;
1578         msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1579         memcpy(&msg->payload, payload, payload_size);
1580
1581         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1582                                msg_size);
1583         if (ret == xpSuccess)
1584                 goto out_1;
1585
1586         XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1587 out_2:
1588         if (func != NULL) {
1589                 /*
1590                  * Try to NULL the msg_slot's func field. If we fail, then
1591                  * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1592                  * case we need to pretend we succeeded to send the message
1593                  * since the user will get a callout for the disconnect error
1594                  * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1595                  * error returned here will confuse them. Additionally, since
1596                  * in this case the channel is being disconnected we don't need
1597                  * to put the the msg_slot back on the free list.
1598                  */
1599                 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1600                         ret = xpSuccess;
1601                         goto out_1;
1602                 }
1603
1604                 msg_slot->key = NULL;
1605                 atomic_dec(&ch->n_to_notify);
1606         }
1607         xpc_free_msg_slot_uv(ch, msg_slot);
1608 out_1:
1609         xpc_msgqueue_deref(ch);
1610         return ret;
1611 }
1612
1613 /*
1614  * Tell the callers of xpc_send_notify() that the status of their payloads
1615  * is unknown because the channel is now disconnecting.
1616  *
1617  * We don't worry about putting these msg_slots on the free list since the
1618  * msg_slots themselves are about to be kfree'd.
1619  */
1620 static void
1621 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1622 {
1623         struct xpc_send_msg_slot_uv *msg_slot;
1624         int entry;
1625
1626         DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1627
1628         for (entry = 0; entry < ch->local_nentries; entry++) {
1629
1630                 if (atomic_read(&ch->n_to_notify) == 0)
1631                         break;
1632
1633                 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1634                 if (msg_slot->func != NULL)
1635                         xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1636         }
1637 }
1638
1639 /*
1640  * Get the next deliverable message's payload.
1641  */
1642 static void *
1643 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1644 {
1645         struct xpc_fifo_entry_uv *entry;
1646         struct xpc_notify_mq_msg_uv *msg;
1647         void *payload = NULL;
1648
1649         if (!(ch->flags & XPC_C_DISCONNECTING)) {
1650                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1651                 if (entry != NULL) {
1652                         msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1653                                            hdr.u.next);
1654                         payload = &msg->payload;
1655                 }
1656         }
1657         return payload;
1658 }
1659
1660 static void
1661 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1662 {
1663         struct xpc_notify_mq_msg_uv *msg;
1664         enum xp_retval ret;
1665
1666         msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1667
1668         /* return an ACK to the sender of this message */
1669
1670         msg->hdr.partid = xp_partition_id;
1671         msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
1672
1673         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1674                                sizeof(struct xpc_notify_mq_msghdr_uv));
1675         if (ret != xpSuccess)
1676                 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1677 }
1678
1679 static struct xpc_arch_operations xpc_arch_ops_uv = {
1680         .setup_partitions = xpc_setup_partitions_uv,
1681         .teardown_partitions = xpc_teardown_partitions_uv,
1682         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1683         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1684         .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1685
1686         .allow_hb = xpc_allow_hb_uv,
1687         .disallow_hb = xpc_disallow_hb_uv,
1688         .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1689         .increment_heartbeat = xpc_increment_heartbeat_uv,
1690         .offline_heartbeat = xpc_offline_heartbeat_uv,
1691         .online_heartbeat = xpc_online_heartbeat_uv,
1692         .heartbeat_init = xpc_heartbeat_init_uv,
1693         .heartbeat_exit = xpc_heartbeat_exit_uv,
1694         .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1695
1696         .request_partition_activation =
1697                 xpc_request_partition_activation_uv,
1698         .request_partition_reactivation =
1699                 xpc_request_partition_reactivation_uv,
1700         .request_partition_deactivation =
1701                 xpc_request_partition_deactivation_uv,
1702         .cancel_partition_deactivation_request =
1703                 xpc_cancel_partition_deactivation_request_uv,
1704
1705         .setup_ch_structures = xpc_setup_ch_structures_uv,
1706         .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1707
1708         .make_first_contact = xpc_make_first_contact_uv,
1709
1710         .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1711         .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1712         .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1713         .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1714         .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1715         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1716         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1717
1718         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1719
1720         .setup_msg_structures = xpc_setup_msg_structures_uv,
1721         .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1722
1723         .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1724         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1725         .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1726         .partition_engaged = xpc_partition_engaged_uv,
1727         .any_partition_engaged = xpc_any_partition_engaged_uv,
1728
1729         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1730         .send_payload = xpc_send_payload_uv,
1731         .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1732         .received_payload = xpc_received_payload_uv,
1733         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1734 };
1735
1736 static int
1737 xpc_init_mq_node(int nid)
1738 {
1739         int cpu;
1740
1741         get_online_cpus();
1742
1743         for_each_cpu(cpu, cpumask_of_node(nid)) {
1744                 xpc_activate_mq_uv =
1745                         xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
1746                                              XPC_ACTIVATE_IRQ_NAME,
1747                                              xpc_handle_activate_IRQ_uv);
1748                 if (!IS_ERR(xpc_activate_mq_uv))
1749                         break;
1750         }
1751         if (IS_ERR(xpc_activate_mq_uv)) {
1752                 put_online_cpus();
1753                 return PTR_ERR(xpc_activate_mq_uv);
1754         }
1755
1756         for_each_cpu(cpu, cpumask_of_node(nid)) {
1757                 xpc_notify_mq_uv =
1758                         xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
1759                                              XPC_NOTIFY_IRQ_NAME,
1760                                              xpc_handle_notify_IRQ_uv);
1761                 if (!IS_ERR(xpc_notify_mq_uv))
1762                         break;
1763         }
1764         if (IS_ERR(xpc_notify_mq_uv)) {
1765                 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1766                 put_online_cpus();
1767                 return PTR_ERR(xpc_notify_mq_uv);
1768         }
1769
1770         put_online_cpus();
1771         return 0;
1772 }
1773
1774 int
1775 xpc_init_uv(void)
1776 {
1777         int nid;
1778         int ret = 0;
1779
1780         xpc_arch_ops = xpc_arch_ops_uv;
1781
1782         if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1783                 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1784                         XPC_MSG_HDR_MAX_SIZE);
1785                 return -E2BIG;
1786         }
1787
1788         if (xpc_mq_node < 0)
1789                 for_each_online_node(nid) {
1790                         ret = xpc_init_mq_node(nid);
1791
1792                         if (!ret)
1793                                 break;
1794                 }
1795         else
1796                 ret = xpc_init_mq_node(xpc_mq_node);
1797
1798         if (ret < 0)
1799                 dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1800                         -ret);
1801
1802         return ret;
1803 }
1804
1805 void
1806 xpc_exit_uv(void)
1807 {
1808         xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1809         xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1810 }
1811
1812 module_param(xpc_mq_node, int, 0);
1813 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");